Pergamon

Stud. Hist. Phil. Sci., Vol. 29, No. 3, pp. 409424, 1998
1998 Elsevier Science Ltd. All rights reserved
Printed in Great Britain
00393681/98 $19.00+0.00

What is Structural Realism?

James Ladyman*

[T]he structure of this physical world consistently moved farther and farther away
from the world of sense and lost its former anthropomorphic character . Thus the
physical world has become progressively more and more abstract; purely formal mathematical operations play a growing part (Planck, 1996, p. 41).

1. Introduction
In the debate about scientific realism, arguably the two most compelling arguments
around are the no miracles argument, and the pessimistic meta-induction. Unfortunately these pull in opposite directions (towards realism and antirealism
respectively). In an attempt to break this impasse, and have the best of both
worlds, John Worrall introduced structural realism (although he attributes its original formulation to Poincare) (Worrall, 1989). Using the case of the transition in
nineteenth-century optics from Fresnels elastic solid ether theory to Maxwells
theory of the electromagnetic field, Worrall argues that:
There was an important element of continuity in the shift from Fresnel to Maxwell
and this was much more than a simple question of carrying over the successful empirical content into the new theory. At the same time it was rather less than a carrying
over of the full theoretical content or full theoretical mechanisms (even in approximate form). There was continuity or accumulation in the shift, but the continuity
is one of form or structure, not of content (1989, p. 117).

According to Worrall, we should not accept full-blown scientific realism, which

asserts that the nature of things is correctly described by the metaphysical and
physical content of our best theories. Rather we should adopt the structural realist
emphasis on the mathematical or structural content of our theories. Since there is
(says Worrall) retention of structure across theory change, structural realism both
(a) avoids the force of the pessimistic meta-induction (by not committing us to

* Department of Philosophy, University of Bristol, 9 Woodland Road, Bristol BS8 1TB, U.K.
Received 5 February 1997; in revised form 11 July 1997; in revised form 11 July 1997

PII: S0039-3681(98)00014-4
409

410

Studies in History and Philosophy of Science

belief in the theorys description of the furniture of the world), and (b) does not
make the success of science (especially the novel predictions of mature physical
theories) seem miraculous (by committing us to the claim that the theorys structure, over and above its empirical content, describes the world).
This paper addresses the interpretation and further elaboration of structural
realism. Although structural realism has attracted considerable attention, and similar views seem to be advocated by Michael Redhead and Elie Zahar among others,
Worralls proposal needs to be developed if it is to provide a substantive alternative
to traditional scientific realism.1 In particular, there is a fundamental question about
the nature of structural realism that should be answered: is it metaphysics or epistemology? Worralls paper is ambiguous in this respect. At times his proposal
seems to be that we put an epistemic constraint on realism to the effect that we
should only commit ourselves to believing in the structural content of a theory,
while remaining agnostic about the rest. This is suggested by Poincare who talks
of the redundant theories of the past capturing the true relations between the real
objects which Nature will hide for ever from our eyes (Poincare, 1905, p. 161).
What this might mean is explored below.
On the other hand, Worralls position is not explicitly an epistemic one, and
other comments suggest a departure from the metaphysics of standard scientific
realism. For example, he says:
On the structural realist view what Newton really discovered are the relationships
between phenomena expressed in the mathematical equations of his theory (1989, p.
122; my emphasis).

If the continuity in scientific change is of form or structure, then perhaps it is

being suggested that we should abandon commitment to even the putative the reference of theories to objects and properties, and account for the success of science
in other terms. Redhead says:
[R]ealism about what? Is it the entities, the abstract structural relations, the fundamental laws or what? My own view is that the best candidate for what is true about a
physical theory is the abstract structural aspect (1996, p.2).

Others who have contributed to structural realism have also implied a significant
departure from traditional realist metaphysics. According to Zahar, science can
never tell us more than the structure of the noumenal world; the nature of the
entities and properties of which it consists is inaccessible to us. This seems to
concur with structural realist sentiments expressed by Howard Stein:
[O]ur science comes closest to comprehending the real, not in its account of substances and their kinds, but in its account of the Forms which phenomena imitate
(for Forms read theoretical structures, for imitate, are represented by) (Stein,
1989, p. 57).

Redhead (1996, p. 3); Zahar (1994).

What is Structural Realism?

411

Below I will argue that structural realism gains no advantage over traditional
scientific realism if it is understood as merely an epistemological refinement of it,
and that instead it ought to be developed as a metaphysical position. I explain why
the semantic approach to scientific theories offers the natural framework for this,
and what a metaphysical structural realism must involve if it is to do justice to the
intuition behind the no-miracles argument.
2. Epistemological Structural Realism
This section explores how we can understand the claim that we should believe
in the structural content of theories as an epistemic constraint. It will be argued
that the most obvious way of doing so leads to a position that is both familiar
(from the writings of Russell and of Grover Maxwell) and incapable of solving the
problem of ontological discontinuity. Another epistemological structural realism, an
interpretation due to Stathis Psillos, will also be considered and found to be inadequate. However, it is argued that structural realism might yet offer the best of
both worlds if it is understood in metaphysical terms.
2.1. Russell, Maxwell and Ramsey Structure
A position called structural realism, that amounts to an epistemological gloss on
traditional scientific realism, was, as Worrall mentioned, detailed and advocated
by Maxwell (1962, 1970a, b, 1972). Maxwell wanted to defend a robust scientific
realism, but he wanted this to be compatible with what he called concept empiricism about the meaning of theoretical terms, and he also wanted to explain how
we can have epistemic access to unobservable entities. The problem as Maxwell
saw it was that theories talk about all sorts of entities and processes with which
we are not acquainted. How, he wondered, can we then know about and refer to
them and their properties? The answer that Maxwell gave, following Russells
emphasis on structure and relations in his later philosophy (see, for example, Russell, 1927), was that we can know about them by description, that is we can know
them via their structural properties. In fact, he argues, this is the limit of our
knowledge of them, and the meanings of theoretical terms are to be understood
purely structurally.
The way that Maxwell (and others such as David Lewis, 1970) explicate this
notion that the structure of the theory exhausts the cognitive content of its theoretical terms, is to consider the Ramsey sentence of the theory. As is well known,
Ramseys method allows the elimination of theoretical terms from a theory, and
their replacement with a collection of existentially-quantified bound predicate variables (or names, in Lewis). If one replaces the conjunction of assertions of a firstorder theory with its Ramsey sentence, the observational consequences of the
theory are carried over, but direct reference to unobservables is eliminated.
If we formalise a theory in a first order language: (O1, , On; T1, , Tm),
then the corresponding Ramsey sentence is t1, , tm(O1, , On;t1, , tm).
Thus the Ramsey sentence only asserts that there are some objects, properties and

412

Studies in History and Philosophy of Science

relations that have certain logical features, satisfying certain implicit definitions. It
is a higher-order description, but ultimately connects the theoretical content of the
theory with observable behaviour. However, it is a mistake to think that the Ramsey
sentence allows us to eliminate theoretical entities, for it still states that these exist.
It is just that they are referred to not directly, with theoretical terms, but by description or by logical form, that is via variables, connectives, quantifiers and predicate
terms whose direct referents are (allegedly) known by acquaintance. Thus Maxwell
(and Russell) claimed that knowledge of the unobservable realm is of its structural
rather than intrinsic properties, or, as is sometimes said, is limited to knowledge
of its second order properties.2 It is arguable that this is the purest structuralism
possible, for the notion of structure employed refers to the higher-order properties
of a theory, those that are only expressible in purely formal terms.
This is an epistemological structural realism meant to vindicate and not to revise
the ontological commitments of scientific realism. On this view the objective world
is composed of unobservable objects between which certain properties and relations
obtain; but we can only know the properties and relations of these properties and
relations, that is the structure of the objective world.
There are serious difficulties with this view, which were originally raised by
Newman (1928) and which have been recently discussed by Demopoulos and Friedman (1985). The basic problem is that structure is not sufficient to uniquely pick
out any relations in the world. Suppose that the world consists of a set of objects
whose structure is W with respect to some relation R, about which nothing else is
known. Any collection of things can be organised to have structure W provided
there is the right number of them. As Demopoulos and Friedman point out, this
is because if is consistent, and if all its purely observational consequences are
true, then the truth of the corresponding Ramsey sentence follows as a theorem of
second-order logic or set theory (provided the initial domain has the right cardinality, and if it does not then consistency implies that there exists one that does).
The formal structure of a relation can easily be obtained with any collection of
objects provided there are enough of them, so having the formal structure cannot
single out a unique referent for this relationin order to do so we must stipulate
that we are talking about the intended relation, which is to go beyond the structural
description. Thus on this view, only cardinality questions are open to discovery!
(ibid., p. 188); everything else will be known a priori.
This leads Demopoulos and Friedman to conclude that reducing a theory to its
Ramsey sentence is equivalent to reducing it to its empirical consequences, and
thus that:
Russells realism collapses into a version of phenomenalism or strict empiricism after
all: all theories with the same observational consequences will be equally true (p.
635).
2
Eddington also defended the view that knowledge of the unobservable is structural but added, unlike
Russell, that it was a priori. Eddington seemed to think that the canonical form of structure was grouptheoretic, though not for the same reasons as those influenced by phenomenology like Weyl (see Solomon, 1989). Weyls views are discussed briefly below.

What is Structural Realism?

413

This conclusion can be reached by a different path. It has been proved, by Jane
English (1973), that any two Ramsey sentences that are incompatible with one
another cannot have all their observational consequences in common. In other
words, if we treat a theory just as its Ramsey sentence then the notion of theoretical
equivalence collapses onto that of empirical equivalence. Therefore, equating the
structure of a theory to what is embodied in its Ramsey sentence cannot do justice
to Worralls intention in proposing structural realism, since he is quite clear that
commitment to the structure of a theory goes beyond commitment to its strictly
empirical level.
In any case, it is hard to see how Maxwells epistemology and semantics could
help deal with the problem of theory change that structural realism is intended to
solve. After all, as Maxwell himself pointed out, his structural realism does not
dispense with reference to theoretical entities, but makes that reference a function
of the (place of the theoretical terms in the) overall structure of the theory, as
manifested in the Ramsey sentence. The problem of ontological discontinuity is
left untouched by simply adopting Ramsification. In fact, it seems even worse if
contextualism about the meaning of theoretical terms is adopted to this extent.
Hence, this epistemic reading of structural realism is of no help with the problem
of theory change.
2.2. The Problem of Theory Change
The most powerful element of Worralls highly suggestive paper is the demand
for a view that can accommodate both the key arguments in the realism debate.
Of course, many have argued that traditional realism can be defended against the
supposed problem of theory change.3 Let us suppose, however, that the realist
accepts that theory change presents them with serious difficulties, and, in particular,
that they admit that parts of current theories may well not refer. Realists may claim
to be able to modify scientific realism in the light of this. If they were to sanction
belief only in those parts of theories that have been or will be retained in cases of
radical theory change, then realism would not be undermined by the pessimistic
meta-induction.
What is needed for this strategy is a principled way of distinguishing those parts
of theories that will be retained from those that may be abandoned, so that a nonad hoc differentiated attitude towards theories can be advocated. Realists would
then not believe in all of the theories currently accepted, in the hope of avoiding
ontological commitments that will be relinquished by their future counterparts.4
Perhaps then we should interpret structural realism as offering a way of differen3
Discussion of radical theory change centres around Laudan (1981). Notable attempts to defend
referential continuity and traditional realism in the face of historical problems are Hardin and Rosenberg
(1982), Psillos (1992, 1994, 1996), and Kitcher (1993). (For a response to Psillos see Ladyman,
forthcoming.) It is worth noting that we do not need to construct a meta-induction, as Laudan does, to
use theory change against traditional realism. Instead we can treat the cases of the ether and caloric as
counterexamples to the (version of the no-miracles) argument that theories could not be novelly predictively successful if their central theoretical terms did not refer.
4
This is how Papineau sees the issue (1996, p. 12), as well as Psillos (1996).

414

Studies in History and Philosophy of Science

tiating ontological commitment so that we can be sure of achieving referential

success. Papineau and Psillos argue, quite rightly I think, that structural realism
cannot do this.
Psillos (1995) claims that structural realism presupposes distinctions between
the form and content of a theory, or between our ability to know the structure and
our ability to know the nature of the world. The (sensible) realist will not accept
these distinctions, according to Psillos. He argues that one success of the scientific
revolution was the banishing of mysterious forms and substances that might not
be fully describable in structural terms, and the consequent concentration of the
mechanical philosophers on quantitative descriptions of the properties of things.
For Psillos, properties in mature science are defined by the laws in which they
feature: the nature of something consists in its basic properties, and in the equations
expressing the laws these obey.
Similarly, Ernan McMullin says:
What are electrons? Just what the theory of electrons says they are, no more, no
less, always allowing for the likelihood that the theory is open to further refinement
(McMullin, 1984, p. 15).

We have a theory of electrons that describes their behaviour in terms of the

laws, interactions, and so on, to which they are subject. This description, which
may be termed a structural one, gives us the last word on electrons, says the realist.
This is what Psillos means when he says that:
[T]he nature and the structure of a physical entity form a continuum (1995, p. 15).

Thus, for Psillos, structural realism cannot be distinguished from traditional

realism without a dubious distinction between structure and nature.5
So it seems the realist can claim that structural realism is no different from
scientific realism in so far as it advises a structural understanding of theoretical
entities, because restriction of belief to structural claims is in fact no restriction
at all (Papineau, 1996, p. 12). Thus construed, it gains no advantage over traditional realism with the problem of theory change because it fails to make any
distinction between parts of theories that we should and shouldnt believe. Realists
may well think that all their knowledge of the unobservable is structural, whether
they explicate this in the Ramsey way or not, but this will not help them with
ontological discontinuity.
In fact, I will argue later that traditional realism is committed to more than
purely structural knowledge. In any case, suppose that Worrall is correct in his
claim that there is retention of mathematical structure across theory change which
is in other respects radical. I do not think this too controversial a claim, for there
are many examples other than the structural continuity between Fresnels wave
optics and Maxwells theory of the electromagnetic field. Friedmans (1983) presentation of the structure of Newtonian and relativistic spacetime theories shows
5
This is just the complaint of Braithwaite against Eddingtons structuralism (Braithwaite, 1940, p.
463).

What is Structural Realism?

415

that there is more continuity between them than is obvious. The use of the correspondence principle in the development of quantum mechanics gives us many
examples of structural continuity. For instance, the Hamiltonian operators used in
quantum mechanics for systems such as a simple harmonic oscillator or a particle
subject to a Coulomb potential are of the same form as their classical counterparts
(the difference being that they feature the appropriate Hermitian operators instead
of the variables that classically represent the values of the quantities of position and
momentum).6 There is also the very important analogy between quantum operator
commutation relations and classical Poisson brackets developed by Dirac. Again
there is a common mathematical form. It seems that Worrall is indeed on to something when he points out the continuity in the mathematics of very different
theories. If Psillos reading of structural realism does not allow us to take advantage
of Worralls insight, then we should look for a different interpretation of structural realism.
The traditional realist may also be heartened by the claim that there is structural
or mathematical continuity in the history of science, hoping that where there is
continuity of structure this can be construed as continuity of ontology. However,
there are many cases of theory change where although relations are preserved, the
relata and even their logical type is not. For example, it is widely held that the
equation (known as Ehrenfests theorem) F(r) = md2r/dt2 exhibits continuity
between classical and quantum mechanics; it certainly has a similar form to the
equation F = ma. But the quantum equation has as its arguments the expectation
values of Hermitian operators, whereas the classical equation features continuous
real variables. Another case is the transition from classical to special relativistic
mechanics where mass is transformed to relativistic mass, and thus goes from being
a simple property of objects to being a relation between the simple property of
rest mass, the two-place relation of the speed of light, and the three-place relation
of relative velocity. The suggestion that the extension of these classical theories is
the same as that of their successors is implausible in the extreme. Worralls paper
suggests that we accept the fact of radical ontological replacement and thus that
we abandon our commitment to the ontology of scientific theories (standardly
construed).
3. Towards a Metaphysical Structural Realism
3.1. The Semantic Approach and Structural Realism
To be an alternative to both traditional realism and instrumentalism, structural
realism must incorporate epistemic commitment to more than the empirical content
of a scientific theory, namely to the structure of the theory, while stopping short
of realists commitment to the full ontology postulated by the theory. Hence, the
debate about how to characterise theories and their structure is of central concern
6
Zahar (1994) appeals to the Correspondence Principle in his defence of a neo-Kantian species of
structural realism.

416

Studies in History and Philosophy of Science

for the structural realist. I have argued that conceiving the structure of a theory in
terms of its Ramsey sentence cannot be appropriate for structural realism. Implicit
in that discussion was the so-called syntactic or received conception of theories.
The alternative semantic or model-theoretic approach to theories, which is to
be preferred on independent grounds, is particularly appropriate for the structural
realist.7 This is because the semantic approach itself contains an emphasis on structures. That is, theories are to be thought of as presenting structures or models that
may be used to represent systems, rather than as partially-interpreted axiomatic
systems. Theories are not collections of propositions or statements, but are extralinguistic entities which may be described or characterised by a number of different
linguistic formulations (Suppe, 1974, p. 221).
In the context of the syntactic approach, within which a theory is taken to be a
set of sentences, realism amounts to the commitment to standard (correspondence)
referential semantics, and to truth, for the whole theory. How should we understand
questions about the relationship between theoretical objects and the world in terms
of the semantic conception of theories? Ronald Giere addresses this issue in his
writings on the semantic approach. Unlike van Fraassen, he seems to accept that
the semantic approach transforms the terms of the scientific realism debate. In
particular, Giere argues that it is unreasonable to expect all theoretical representation in science to fit the mould in which philosophers cast linguistic representation; Tarskian semantics is not appropriate for a consideration of the representative role of spacetime structure or Hilbert space.8 One reason for this is that Gieres
constructive realism stops short of asserting that real systems resemble models
in all respects. Instead, he claims that models are at best similar to real systems
in many or most aspects.
So, Gieres version of the semantic approach relies on a notion of similarity in
specified respects and to specified degrees, and of course it is a notorious complaint
that similarity is a uselessly vague concept; everything is similar to everything else
in some respect or other. It may therefore be better to use the notion of partial
isomorphism to represent the relationships between theoretical structures and models of the phenomena.9 However, this is not the crucial issue here, for, even if we
reject the claim that models are similar to anything in the world, Giere is right
that once the semantic approach is adopted the crucial issue is whether or not
theoretical models tell us about modalities.10
Recall that the motivation for the realist element in the structural realist programme is the no-miracles argument. We demand some explanation of the success
7
That the semantic approach to scientific theories is preferable is obviously controversial, but I take
it for granted in the following. For a defence and articulation of the semantic approach see French and
Ladyman (forthcoming).
8
See Giere (1995).
9
This is introduced in French and Ladyman (forthcoming), and applied to the history of superconductivity in French and Ladyman (1997).
10
He argues that they do and states that (his) [c]onstructive realism is thus a model-theoretic analogue
of the view advocated by Grover Maxwell (Giere, 1985, p. 83). Therefore, he says, constructive realism
is a species of structural realism (ibid.).

What is Structural Realism?

417

of sciencein particular, of the novel predictive success of theories. Consider van

Fraassens explanation for the success of science: theories are born into a jungle
red in tooth and claw, that is, a fiercely competitive environment, and hence they
have to be successful to survive. This is as if we were to explain consciousness
by saying that selection pressures are such that organisms that do not develop
sophisticated representational systems will not survive as well as those that do.
This does not explain how it is that any particular organism has the capacity for
consciousness. To answer that question we need a theory that explains how particular physical constitutions and functionality can give rise to consciousness. The
equivalent question for scientific theories is: how is it possible for a particular
theory to make novel predictions that are then vindicated? Think of Eddingtons
detection of light from a star behind the sun, the example that so impressed Popper.
How can general relativity have predicted this correctly on van Fraassens account?
It is examples like this that seem miraculous without some realist ingredient in
ones philosophy of science. Yet this need not be the same as asking how is this
possible if the theory is not true (or approximately true).
In an everyday context our ability to predict relations between phenomena is
grounded in our acceptance of the existence of objects and their properties. In the
case of the unobservable world, realism about unobservable objects attempts to
provide a similar grounding for our predictive success. But, if we reject traditional
scientific realism yet wish to avoid scepticism (if structural realism is to be any
kind of realism at all), we must provide some account of the world as unobserved
by us. If structural realism is motivated by the no-miracles argument then it must
explain how the relations in a model can correctly predict some previously
unknown modal relations among phenomena in the world. This is a question that
van Fraassen cannot answer at all, for he says that the locus of possibility is the
model, not a reality behind the phenomena (van Fraassen, 1980, p. 202). Although
agnostic about scientific realism, he is atheistic about modality and believes in no
objective causality or necessity in reality.11
Of course the traditional realist has such an explanation: theories that are
approximately true express laws of nature that describe how the unobservable entities and processes described by the theories must behave. However, this account
is undermined not merely by cases of radical ontological discontinuity, but also
by the work of Cartwright and others which has emphasised the manifest falsity
and ideal nature of theoretical models in science. Cartwright argues that the (socalled) laws that constrain abstract models, such as the law of universal gravitation
or Schrodingers equation, if understood as factual descriptions of how things in
the world behave, are either false or irrelevant to the empirical domains to which
they are supposed to apply. This is because the conditions under which they are
11
There is therefore a tension in van Fraassens position, for he thinks that unobservable entities may
exist, but that laws and natural kinds definitely do not. Yet if, for example, electrons do indeed exist
but do not form a natural kind subject to laws of interaction and so on, then we are wrong about a
very important aspect of them, making it hard to see on what grounds we can say that they exist rather
than some other things entirely.

418

Studies in History and Philosophy of Science

supposed to obtain are counterfactual, and also because they are applied in the
models to idealised or abstract entities.12
Cartwright takes this to show that theoretical models are not genuinely representational at all; rather, they are to be understood as tools for the construction of
phenomenological models which are representational.13 However, from the perspective of the semantic approach we do not expect the high-theoretical structure
to be a law-like description of the world, rather we envisage a complex hierarchy
of models or structures employed by the theory. As has been argued elsewhere,
contra Cartwright et al., abstract theoretical structure plays an important role in
the development of low-level models of the phenomena, and some of the most
profound novel predictions of science are directly obtained from highly theoretical
models.14 What is more, the highly abstract structures of modern physical theories
do furnish us with explanations of a sort, what Hughes has called structural explanations (Hughes, 1989). Hence, the most theoretical parts of a theory, the abstract
mathematical structures it employs at the greatest level of generality, must have
some grip on reality. It is clear that the grip on reality in question must go
beyond a correct description of the actual phenomena to the representation of modal
relations between them.
3.2. Structure
According to Zahar (1994, p. 14) the continuity in science is in the intension
not the extension of its concepts. Perhaps, if we are to believe that the mathematical
structure of theories is what is important, then as Zahar suggests, we need a different semantics for theories: one that addresses the representative role of mathematics
directly. The advantage of adopting such a view is that we would then be content
with the continuity of mathematical structure that is found even between theories
that differ radically if taken realistically, and so would not be confounded by theory
change.15 This would seem to entail a corresponding shift from a metaphysics of
objects, properties, and relations, to one that takes structure as primitive.
To return to the claim that traditional realism is itself operating with a structuralist conception of its objects; this is quite misleading. Although theoretical terms
are indeed understood in terms of the laws, equations and so on in which they
feature, traditional realism does involve acceptance of more than the structural
properties of theoretical entities. Theories are construed in terms of metaphysical
commitments that are underdetermined by the evidence for them.
Underdetermination has been thought to show both that theory choice in science
is not rational, and that we are unjustified in believing that the world is just how
our best theories say it is. These are related but can be distinguished. Hence, recent
work has sought to show that the existence of empirically equivalent theories does
12

Cartwright (1983, 1989).

Cartwright et al. (1996).
14
French and Ladyman (1997).
15
For example, Saunders (1993) has shown that the method of epicycles may be regarded as a Fourier
approximation to an elliptical representation of planetary orbits.
13

What is Structural Realism?

419

not undermine the rationality of choosing one theory over another at a particular
stage in the development of a science. It is pointed out that the empirical equivalence of two theories need not imply their evidential equivalence. This issue may
be orthogonal to that of whether underdetermination undermines ontological commitment to a theory. For example, if, like Larry Laudan, one is a pragmatist and
not a scientific realist, then one may well think that theory choice is thoroughly
rational given pragmatic constraints, but that nonetheless ontological commitment
to the theories thus chosen is unjustified given the possibility of empirically equivalent theories with alternative ontologies (Laudan, 1981). Furthermore, Roger Jones
(1991) has argued that where we have a theory admitting alternative formulations
each of which suggests a different metaphysics (his example is Newtonian mechanics and its action at a distance, variational, field-theoretic, and even curved
spacetime formulations), we cannot be realists about such a theory because we
do not know which version to believe about how the world really is.
Even if we are able to decide on a canonical formulation of our theory, there
is the further problem of metaphysical underdetermination with respect to, for
example, whether the entities postulated by a theory are individuals or not.16 There
is, of course, much dispute about whether or not quantum particles, or spacetime
points, are individuals. Realists would perhaps say that deciding these issues is a
matter for the special sciences involved, and that these problems do not have to
be solved for us to be realists.17 But the recent history of the philosophy of physics
shows us the impossibility of simply reading ones metaphysics off ones physics.
One example is the debate about nonlocality in the context of the BohmEPR
experiment, which showed that whether the world is nonlocal is in a sense underdetermined, and may be proved or disproved according to the choice of assumptions
about possessed values, determinism and so on.18 In the case of individuality, it
has been shown in the work of Steven French and Michael Redhead (French and
Redhead, 1988; French, 1989), that electrons may be interpreted as either individuals or nonindividuals.19
We need to recognise the failure of our best theories to determine even the most
16
Recent work by Decio Krause (1992) suggests that standard model theory presupposes that the
basic objects of first-order predication are individuals because the axioms of ZF set theory ensure that
the elements of sets are distinguishable individuals.
17
Peter Forrest (1994) argues that we need not have solved the metaphysical disputes about the nature
of the entities involved to accept scientific realism. In this he claims to follow the practice of the
working scientist of talking as neutrally as possible about, say, particles. But of course the working
scientist must be allowed to believe in scientific realism on any plausible, even non-realist, view. The
neutrality of that discourse is prephilosophical and in particular premetaphysical, and provides merely
temporary solace before the metaphysical nature of the entities involved becomes a scientific issue.
18
See Redhead (1987), for example.
19
It is worth remarking that Paul Teller (see, for example, Teller, 1989) has proposed the introduction
of non-supervenient relations, that is relations that do not supervene on the monadic properties of any
individuals, in order to understand quantum entanglement. On this view, facts about relations must be
understood as irreducible to facts about the non-relational properties of individuals. The non-supervenient relations are in fact there to partially salvage a classical ontology, with quantum entities as individuals (see French, 1989). However, this is an interesting shift away from an ontology that takes objects
as primary, for a relation that does not supervene must be understood as having reality of itself, and
to this extent Tellers proposal amounts to realism about structure.

420

Studies in History and Philosophy of Science

fundamental ontological characteristic of the purported entities they feature. It is

an ersatz form of realism that recommends belief in the existence of entities that
have such ambiguous metaphysical status. What is required is a shift to a different
ontological basis altogether, one for which questions of individuality simply do
not arise. Perhaps we should view the individuals and nonindividuals packages,
like particle and field pictures, as different representations of the same structure.
There is an analogy here with the debate about substantivalism in general relativity.
Recently it has been suggested that this issue also calls for a different fundamental
ontology within which to assess the reality of spacetime. Robert DiSalle (1994)
has suggested that the structure of spacetime be accepted as existent without being
supervenient on the existence of spacetime points. This is a restatement of the
position developed by Stein in his famous exchange with Grunbaum, according to
which spacetime is neither a substance, nor a set of relations between substances,
but a structure in its own right.
So we should seek to elaborate structural realism in such a way that it can
diffuse the problems of traditional realism, with respect to both theory change and
underdetermination. This means taking structure to be primitive and ontologically
subsistent. I want briefly to turn to the use of group theory in modern science to
suggest further how we can explicate structural realism.
Weyls approach to relativity theory is essentially group-theoretic. That is, in
common with his approach in geometry, he takes the objects of the theory to be
characterised by invariance under particular groups of transformations. Similarly,
in the case of quantum mechanics, he asserts that:
All quantum numbers, with the exception of the so-called principle quantum number,
are indices characterising representations of groups (1931, p. xxi).

The central point of philosophical relevance is that the mathematical notion of

invariance is taken by Weyl to characterise the notion of objectivity.20 It is this
that liberates physics from the parochial confines of a particular coordinate system.
For Weyl, appearances are open only to intuition (in the Kantian sense of subjective
perception) and therefore agreement is obtained by giving objective status only to
those relations that are invariant under particular transformations. What is particularly striking is the way that Weyl uses the insights gathered from his work on
transformations and invariants in relativity theory to make his crucial contribution
to the development of quantum mechanics.21 The point is that the choice of the
20
Weyls views have recently been revived by Sunny Auyang (1996) in an explicitly neoKantian
project which attempts to solve the problem of objectivity in quantum mechanics. Auyang seeks to
extract the primitive conceptual structure in physical theories and she too finds it in what she calls
the representationtransformationinvariant structure. This is essentially group-theoretic structure.
Auyang, like Born and Weyl, thinks that this is what separates an objective state of affairs from its
various representations, or manifestations to observers under different perceptual conditions, and such
invariant structure under transformations.
21
Weyls fundamental role in the history of quantum mechanics is strangely absent from most histories of the subject. Some of the contributors to Deppert et al. (1988) seek to redress this. In particular
see the papers by Julian Schwinger, George Mackey, and David Speiser. It is surprising that even the
recent very detailed and scholarly study of the relationship between Heisenbergs and Schrodingers

What is Structural Realism?

421

momentum space representation of Schrodinger amounts to the choice of a coordinate system. Weyl saw immediately that the proto-theories of Schrodinger and
Heisenberg had fundamental mathematical similarities. He therefore took them to
be in all important respects different versions of the same theory:
[T]he essence of the new HeisenbergSchrodingerDirac quantum mechanics is to be
found in the fact that there is associated with each physical system a set of quantities,
constituting a non-commutative algebra in the technical mathematical sense, the
elements of which are the physical quantities themselves (Weyl, 1931, p. viii).

Thus Weyl here anticipates von Neumanns unification of the theories in his
classic text on quantum mechanics, and indeed Dirac in his book on the theory
cites Weyl as the only previous author to employ the same symbolic method for
the presentation of the theory as he does himself. This method according to Dirac
deals directly in an abstract way with the quantities of fundamental importance
(the invariants etc. of the transformations) and therefore it goes more deeply into
the nature of things (Dirac, 1930, p. viii). Dirac uses vectors, and not rays in
Hilbert space like Weyl and von Neumann, in his treatment of quantum mechanics,
but both he and Weyl recognised that the mathematical status of the two rival
theories of quantum mechanics as alternative representations of the same mathematical structure, makes preference for either eliminable once a unified framework
is available.22
The idea then is that we have various representations which may be transformed
or translated into one another, and then we have an invariant state under such
transformations which represents the objective state of affairs. Representations are
extraneous to physical states but they allow our empirical knowledge of them.
Objects are picked out by individuating invariants with respect to the transformations relevant to the context. Thus, on this view, elementary particles are just sets
of quantities that are invariant under the symmetry groups of particle physics.
Consider the following remark by Howard Stein:
But if one examines carefully how phenomena are represented by the quantum
theory then interpretation in terms of entities and attributes can be seen to
be highly dubious . I think the live problems concern the relation of the Forms
to phenomena, rather than the relation of (putative) attributes to (putative) entities
(1989, p. 59).

The suggestion is that the forms in question are given by the invariance structure
of theories. This notion of structure may give us one way of explicating structural
realism, although this is a matter for further study.

mechanics, and their successor von Neumanns quantum mechanics, completely omits Weyl from the
story (Muller, 1997).
22
Although it may not have been until the empirical success resulting from Diracs relativistic theory
of the electron, based as it was on the underlying group structure, that it was widely acknowledged by
physicists that the abstract structure of a group is of greater significance than particular representations
of it. Indeed, in the early days of quantum mechanics Weyl reports that some talked of eliminating the
group pest (Weyl, 1931, p. x).

422

Studies in History and Philosophy of Science

4. Conclusion
Ernan McMullin has said:
[I]maginability must not be made the test for ontology. The realist claim is that the
scientist is discovering the structures of the world; it is not required in addition that
these structures be imaginable in the categories of the macroworld (1984, p. 14).

The demand for an individuals-based ontology may be criticised on the grounds

that it is the demand that the structure of the mind-independent world be imaginable
in terms of the categories of the world of experience. Traditional realism should
be replaced by an account that allows for a global relation between models and
the world, which can support the predictive success of theories, but which does
not supervene on the successful reference of theoretical terms to individual entities,
or the truth of sentences involving them. It is interesting to note that Benacerraf
(1965) argued that objects to be properly called such must be individuals, and that,
therefore, a structuralist construal of abstract objects like numbers must fail,
because an object with only a structural character could be identified with any
object in the appropriate place in any exemplary structure. If he is right about this
then structural realism amounts to the claim that theories tell us not about the
objects and properties of which the world is made, but directly about structure
and relations.
AcknowledgementsFor detailed comments on an ancestor of this paper I am very grateful to Steven
French, Katherine Hawley, Robin Hendry, David Papineau, Stathis Psillos, and an anonymous reader
for this journal. Thanks also to Decio Krause, Mike McGuire, Mauricio Suarez, Paul Tappenden, John
Worrall and Elie Zahar.

References
Auyang, S. Y. (1996) How is Quantum Field Theory Possible? (Oxford: Oxford University Press).
Benacerraf, P. (1965) What Numbers Could Not Be, in P. Benecerraf and H. Putnam
(eds), Philosophy of Mathematics: Selected Readings (Cambridge: Cambridge University
Press), pp. 272294.
Braithwaite, R. B. (1940) Critical Notice: The Philosophy of Physical Science, Mind 49,
455466.
Cartwright, N. (1983) How the Laws of Physics Lie (Oxford: Oxford University Press).
Cartwright, N. (1989) Natures Capacities and Their Measurement (Oxford: Oxford University Press).
Cartwright, N., Shomar, T. and Suarez, M. (1996) The Tool Box of Science: Tools for
Building of Models with a Superconductivity Example, in W. E. Herfel et al., (eds),
Theories and Models in Scientific Processes (Amsterdam: Rodopi), pp. 137149.
Demopoulos, W. and Friedman, M. (1985) Critical Notice: Bertrand Russells The Analysis
of Matter: Its Historical Context and Contemporary Interest, Philosophy of Science 52,
621639.
Deppert, W. et al. (eds) (1988) Exact Sciences and their Philosophical Foundations (New
York: Verlag Peter Lang).
Dirac, P. A. M. (1930) The Principles of Quantum Mechanics (Oxford: Oxford University Press).

What is Structural Realism?

423

DiSalle, R. (1994) On Dynamics, Indiscernibility, and Spacetime Ontology, British Journal

for the Philosophy of Science 45, 265287.
English, J. (1973) Underdetermination: Craig and Ramsey, Journal of Philosophy 70,
453462.
Forrest, P. (1994) Why Most of us should be Scientific Realists, The Monist 77, 4770.
French, S. (1989) Identity and Individuality in Classical and Quantum Physics, Australasian Journal of Philosophy 67, 432446.
French, S. and Ladyman, J. Reinflating the Semantic Approach, forthcoming.
French, S. and Ladyman, J. (1997) Superconductivity and Structures: Revisiting the London
Account, Studies in History and Philosophy of Modern Physics 28, 363393.
French, S. and Redhead, M. (1988) Quantum Physics and the Identity of Indiscernibles,
British Journal for the Philosophy of Science 39, 233246.
Friedman, M. (1983) Foundations of SpaceTime Theories: Relativistic Physics and Philosophy of Science (Princeton: Princeton University Press).
Giere, R. N. (1985) Constructive Realism, in P. Churchland and C. Hooker (eds), Images
of Science (Chicago: University of Chicago Press), pp. 7598.
Giere, R. N. (1995) Viewing Science, in D. Hull, H. Forbes and R. H. Buriam (eds), PSA
1994, Vol. 2 (East Lansing, MI: Philosophy of Science Association), pp. 316.
Hardin, C. L. and Rosenberg, A. (1982) In Defence of Convergent Realism, Philosophy
of Science 49, 604615.
Hughes, R. I. G. (1989) Bells Theorem, Ideology and Structural Explanation, in J. Cushing
and E. McMullin (eds), Philosophical Consequences of Quantum Theory: Reflections on
Bells Theorem (Notre Dame: University of Notre Dame Press), pp. 195207.
Jones, R. (1991) Realism about What?, Philosophy of Science 58, 185202.
Kitcher, P. (1993) The Advancement of Science (Oxford: Oxford University Press).
Krause, D. (1992) On a Quasi Set Theory, Notre Dame Journal of Formal Logic 33,
402411.
Ladyman, J. A Reply to Psillos on the Meta-induction, forthcoming.
Laudan, L. (1981) A Confutation of Convergent Realism, Philosophy of Science 48, 1949.
Lewis, D. (1970) How to Define Theoretical Terms, Journal of Philosophy 67, 427446.
Maxwell, G. (1962) The Ontological Status of Theoretical Entities, in H. Feigl and G.
Maxwell (eds), Scientific Explanation, Space and Time: Minnesota Studies in the Philosophy of Science, Volume III (Minneapolis: University of Minnesota Press), pp. 327.
Maxwell, G. (1970a) Structural Realism and the Meaning of Theoretical Terms, in S.
Winokur and M. Radner (eds), Analyses of Theories, and Methods of Physics and Psychology: Minnesota Studies in the Philosophy of Science, Volume IV (Minneapolis: University of Minnesota Press), pp. 181192.
Maxwell, G. (1970b) Theories, Perception and Structural Realism, in R. Colodny (ed.),
Nature and Function of Scientific Theories (University of Pittsburgh Series in the Philosophy of Science, Volume IV) (Pittsburgh: University of Pittsburgh Press), pp. 334.
Maxwell, G. (1972) Scientific Methodology and the Causal Theory of Perception, in H.
Feigl, W. Sellans and K. Lehrer (eds), New Readings in Philosophical Analysis (New
York: AppletonCenturyCrofts), pp. 289314.
McMullin, E. (1984) A Case for Scientific Realism, in J. Leplin (ed.), Scientific Realism
(Berkeley, CA: University of California Press), pp. 840.
Muller, F. (1997) The Equivalence Myth of Quantum Mechanics. To appear in Studies in
History and Philosophy of Modern Physics.
Newman, M. H. A. (1928) Mr Russells Causal Theory of Perception, Mind 37, 137148.
Papineau, D. (ed.) (1996) The Philosophy of Science (Oxford: Oxford University Press).
Planck, M. (1996) The Universe in the Light of Modern Physics, in W. Schirmacher (ed.),
German Essays on Science in the 20th Century (New York: Continuum), pp. 3857.
Poincare, H. (1905) Science and Hypothesis (Dover, New York).
Psillos, S. (1992) Conceptions and Misconceptions of Ether, in M. C. Duffy (ed.), Proceedings of the International Conference Physical Interpretations of Relativity Theory
(Sunderland: University of Sunderland Press), pp. 544556.

424

Studies in History and Philosophy of Science

Psillos, S. (1994) A Philosophical Study of the Transition from the Caloric Theory of Heat
to Thermodynamics, Studies in History and Philosophy of Science 25, 159190.
Psillos, S. (1995) Is Structural Realism the Best of Both Worlds?, Dialectica 49, 1546.
Psillos, S. (1996) Scientific Realism and the "Pessimistic Induction", Philosophy of Science
63, S306S314.
Redhead, M. L. G. (1987) Incompleteness, Nonlocality and Realism (Oxford: Oxford University Press).
Redhead, M. L. G. (1996) Quantum Field Theory and the Philosopher. Unpublished transcript of paper.
Russell, B. (1927) The Analysis of Matter (London: Kegan Paul).
Saunders, S. (1993) To what Physics Corresponds, in S. French and H. Kamminga (eds),
Correspondence, Invariance and Heuristics: Essays in Honour of Heinz Post (Dordrecht:
Kluwer), pp. 295325.
Solomon, G. (1989) Discussion: an Addendum to Demopoulos and Friedman (1985),
Philosophy of Science 56, 497501.
Stein, H. (1989) Yes, But Some Skeptical Remarks on Realism and Antirealism, Dialectica 43, 4765.
Suppe, F. (1974) The Structure of Scientific Theories (Chicago: University of Illinois Press).
Teller, P. (1989) Relativity, Relational Holism, and the Bell Inequalities, in J. Cushing
and E. McMullin (eds), Philosophical Consequences of Quantum Theory: Reflections on
Bells Theorem (Notre Dame: University of Notre Dame Press), pp. 208223.
van Fraassen, B. C. (1980) The Scientific Image (Oxford: Oxford University Press).
Weyl, H. (1931) The Theory of Groups and Quantum Mechanics (trans. H. P. Robertson),
(New York: Dover, 1950).
Worrall, J. (1989) Structural Realism: the Best of Both Worlds?, Dialectica 43, 99124.
Zahar, E. (1994) Poincares Structural Realism and his Logic of Discovery, in G.
Heinzmann et al. (eds) Henri Poincare: Akten Des Internationale Kongresses, Nancy,
1994 (Berlin: Akademie Verlag), pp. 4568.