Axel Gelfert, Mathematical Formalisms in Scientific Practice, From Denotation To Model Based Representation
Axel Gelfert, Mathematical Formalisms in Scientific Practice, From Denotation To Model Based Representation
Axel Gelfert, Mathematical Formalisms in Scientific Practice, From Denotation To Model Based Representation
Axel Gelfert
Department of Philosophy
National University of Singapore
3 Arts Link, 117570 Singapore
email:
axel@gelfert.net
Abstract
The present paper argues that 'mature mathematical formalisms' play
a central role in achieving representation via scientic models. A close discussion of two contemporary accounts of how mathematical models apply
- the DDI account (according to which representation depends on the
successful interplay of denotation, demonstration and interpretation) and
the 'matching model' account reveals shortcomings of each, which, it is
argued, suggests that scientic representation may be ineliminably heterogeneous in character. In order to achieve a degree of unication that is
compatible with successful representation, scientists often rely on the existence of a 'mature mathematical formalism', where the latter refers to a
mathematically formulated and physically interpreted notational system of locally applicable rules that derive from (but need not be reducible
to) fundamental theory. As mathematical formalisms undergo a process
of elaboration, enrichment, and entrenchment, they come to embody theoretical, ontological, and methodological commitments and assumptions.
Since these are enshrined in the formalism itself, they are no longer readily obvious to either the novice or the procient user. At the same time
as formalisms constrain what may be represented, they also function as
inferential and interpretative resources.
Keywords: Representation, denotation, mathematical formalism, inferential utility, scientic models.
Financial support via an AcRF Tier 1 grant (WBS R-106-000-016-112) from the Ministry of Education (MOE) Singapore is gratefully acknowledged. I would like to thank Tarja
Knuuttila and Mauricio Surez for helpful comments on an earlier draft of this paper.
Introduction
Idealization, abstraction, and approximation have long been at the heart of much
of the physical sciences, and the deliberate construction of scientic models has
an only slightly shorter history. And yet there is a sense that the systematic and
pervasive deployment of scientic modelling techniques is a fairly recent phenomenon; in some disciplines especially those that deal with complex systems
scientic models have become the default mode of how to approach scientic
problems.
As Adam Morton puts it, `what is new and distinctive in the sci-
ence of our time is the existence of complex mediating models which themselves
have explanatory power and which embody techniques of modelling which can
be rened and passed down to successor models, even though the models never
themselves can function as background theories' [Morton 1993, p. 664]. In the
present paper, I focus on the role of mathematics in this process; in particular,
I shall argue that `mature mathematical formalisms' embody inferential strategies and theoretical commitments that contribute to, and sometimes ensure, the
representational success of scientic models.
In the philosophical literature, the success of models continues, for better
or worse, to be discussed in representational terms: in short, scientic models
represent, and successful models represent successfully. In the present paper, I
do not attempt to explain what scientic representation is
in general.
Instead,
mathematical formalisms
in achieving representa-
In
order to avoid misunderstanding, it is perhaps worth pointing out that, as I understand the term, `mathematical formalism' does not refer to (nor entail any
preference with respect to) any particular debate or position within the philosophy of mathematics; my concern in this paper is with how scientists deploy
(physically interpreted) mathematical formalisms as inferential and interpretative strategies.
The present paper can be thought of as divided into two parts, Sections
1-3 and Section 4.
In the present section, I discuss two recent accounts of how mathematical models
represent, each of which is indicative of a certain conception of how mathematical models apply to the world.
Philosophers
(1)
Whenever we apply a particular equation to a physical phenomenon, we can inquire into (1) the equation's relation to the theory that is presumed to cover the
phenomenon in question, and (2) the adequacy of the equation as a representation of the phenomenon. In the present example, this leads to the following
questions: (1) `under what conditions are scientists warranted in adding this
equation to their `scientic theory of heat', and (2) `under what conditions are
scientists warranted in using this equation to describe a particular physical system?'
[Pincock 2005, p.
derivation conditions,
application conditions.
derivation conditions rst, with the caveat that the specic conditions and assumptions that feature in the context of a derivation will, of course, vary from
one example to another. Given that the mathematical details of the derivation
of the 1-D heat equation are discussed in Pincock's case study, I shall conne
myself to pointing out some of the derivation's general features, especially those
that illustrate the kinds of questions that arise from the demand for derivation
conditions.
In deriving the equations that are constitutive of a mathematical model, one
may, for example, start from empirically established regularities that hold across
a wide range of experimental data.
x1 , x2 , ...)
and
1 Other
x 0,
imal changes. Their ubiquity across so many areas of science should not belie
the fact that such limiting procedures raise some very fundamental problems.
Apart from the mathematical question of whether the limiting procedure successfully converges or not, there is the fundamental worry that such limiting
procedures may cast doubt on whatever representational signicance the quantities in question might have had in the rst place. Consider the example of the
1-D heat equation. Deriving the partial dierential equation, in its form given
in equation (1) above, requires, amongst other steps, evaluating the heat loss
across an innitesimal distance
in the limit
x 0.
x 0
less: there is no such thing as the heat at a given point in space, only the heat
in a region of space that is large enough to actually accommodate an ensemble
of atoms. Yet, for the mathematical derivation of the heat equation it is crucial
that it be possible to take the limit stopping at some nite minimum length
derivation in this case, the heat equation simply on its own terms, by its
empirical performance or its predictive success. However, such an instrumentalist approach would make the success of our derivations, if not a miracle,
then certainly a mystery. Derivations play an important part as aides to our
understanding of mathematical models and to the resolution of interpretative
problems; dismissing the value of actual derivations in favour of brute instrumental success would at the very least be premature. A less radical approach
might acknowledge the relevance of the derivation, but regard the latter as a
primarily
pretation
mathematical exercise, which need not lend itself to a physical interat every step. This, in eect, is the position advocated by Pincock,
who holds that in deriving mathematical models, we typically move back and
forth between a
about physical systems and physical magnitudes' [Pincock 2005, p. 70] and a
mathematical attitude,
x 0
Each attitude alone typically does not suce: from the perspective of the physical attitude, many steps of a derivation will strictly speaking be false simply
prima facie
in-
compatible attitudes, is not intended as a merely descriptive claim about scientic practice.
The point is not that some scientists lean more towards one
attitude rather than the other, or that there is perpetual vacillation between the
two; instead, both are essential components of any complete account of mathematical modelling, and each contributes in distinct ways to the representational
capacity of mathematical models. Whether or not a model represents a physical
system, or phenomenon, cannot be a question of whether or not it accurately
mirrors every single detail; such a criterion would be far too strong, not only
because it could hardly ever be met (except in the most simplistic and uninteresting cases), but also because many phenomena for example in thermodynamics
are quite robust with respect to minor variations (e.g., among microstates).
Some degree of idealisation and approximation is thus entirely legitimate, yet,
for the reasons described earlier, these are also likely to give rise to steps in
the derivation that are incompatible with the physical attitude. At the same
time, some approximations or idealisations, while mathematically admissible,
may simply be too crude, thereby undermining the representational capacity of
the resulting equations. What is needed, then, is a framework which not only
accommodates both attitudes in such a way as to assign each its proper place,
but which also recognises that both are indispensable the physical attitude for
giving empirical content to what would otherwise be a bare mathematical structure, and the mathematical attitude for ensuring the validity of the intermediate
derivational steps.
Pincock attempts to develop such a framework by making mathematical
modelling that is, the successful derivation of model equations that represent
a physical system a two-step process. Given that model equations are typically the outcome of idealisation and abstraction, any attempt to imbue them
with representational content retrospectively, in a way that does not respect
how they were derived, would seem arbitrary, at best loosely aimed at lling in
details that may have been lost in the course of the derivation. Instead of trying
to vindicate representation on the basis of a set of (post-derivation) equations,
one might look for a more direct criterion for when a model has the potential to
represent. One criterion that might make representation seem inevitable, would
be the existence of an isomorphism between the model and the physical system
or phenomenon it is supposed to represent.
argue, a model would not only represent, it would also represent truly. (On this
2 On
closer inspection, however, isomorphism at least in its naive form suers from
numerous problems, not least its inability to make sense of the asymmetry of the representation
relation: after all, a successful model represents its target system, not vice versa. (On this
point, see [Surez 2003, p. 232f.].) For a discussion of more elaborate proposals in the same
vein as isomorphism, see the second half of Section 4.3.
point, see [Pincock 2005b].) However, a model that has been derived via a series
of idealisations and approximations will typically not be isomorphic to any real
system. In order to nonetheless save the idea that `representation must involve
an isomorphism between some mathematical model and the physical situation
represented', Pincock suggests `that the model described in the derivation (the
derivation model) be distinct from the model that is isomorphic to the physical
situation (the matching model)'. [Pincock 2005, p. 71] On this account, when
dealing with the
ical attitude and treat it as a wholly mathematical object, since it is not itself
isomorphic to the physical situation
to the latter, via its relation to the
P,
matching model MM .
At rst sight this seems to be a mere shift in the burden of proof, since
it is now the matching model
isomorphy to
place for
MD
P.
MM ,
to be related to
MM
were, the representational capacity of the matching model, and what the nature
of this relation would have to be. Pincock's suggestion that what is needed is
an `acceptable mathematical transformation' from
the condition that
MM
MD
to
MM ,
combined with
true ) representation.3
MD )
faithful
4 Here, it suces
to note that, at the very least, the distinction between two kinds of models,
and
MD ,
provides a
possible
MM
3 On
complete
4 Paul
an object that make it a model' is misguided: `Once this is fully appreciated it becomes clear
that we can get on with the project on the strength of a good supply of clear cases of things
which are used to represent. These will adequately support study the variety of such uses, the
way they function in the scientic enterprise, their interrelations, and so on.' [Teller 2001, p.
397]
tion.
Denotation (more about which later) accounts for the fact that theoretical
denotation,
demonstration, and interpretation (henceforth: DDI) suggests the following picture: denotation, as the basic relation whereby certain elements of a model can
successful account of mathematical models must meet, the triad of
`stand for', or `be a symbol of ', elements in the physical world, may explain
how physical magnitudes can be `picked out' by mathematical objects, thereby
shedding light on the puzzling fact of the physical world's apparent amenability
to mathematical description.
Demonstration
and
interpretation,
by contrast,
jointly account for the derivation of specic results, such as empirically testable
predictions.
physical
and a
mathematical
atti-
tude to the DDI account, one might say that, whereas the mathematical attitude
may well be appropriate as long as one is dealing with issues of
demonstration
that is, as long as one views the mathematical model `from within' both
denotation
and
interpretation
as the latter supplies empirically meaningful content. The DDI account, thus,
claims to elucidate both the general character of theoretical models and their
application in specic empirical contexts. It does not, however, purport to be an
exhaustive account of theoretical representation. Hughes is careful to point out
that he is `not arguing that denotation, demonstration, and interpretation [are]
individually necessary and jointly sucient for an act of theoretical representation to take place'. [Hughes 1996, p. 329] Rather, he makes `the more modest
suggestion that, if we examine a theoretical model with these three activities in
mind, we shall achieve some insight into the kind of representation it provides'
(ibid.). It therefore makes sense to analyse which central aspect of theoretical
models each of the components brings out.
At the heart of the DDI account is the idea of
Languages of Art
the relation between a painting and what it represents. Having dismissed the
proposal that
similarity
a picture of
after all,
`a Constable painting of Marlborough Castle is more like any other picture than
it is like the Castle' [Goodman 1976, p.
`the relation between a predicate and what it applies to' (ibid.). Importantly,
then, denotation is conceived of as a
connection between elements in the world and elements of a model a connection that does not rely on problematic notions of similarity. Goodman's claim
that `denotation is the core of representation and is independent of resemblance'
[Goodman 1976, p. 5] may seem to suggest that denotation automatically entails representational success, but this is not the case.
certain kinds of
25].)
To the extent that theoretical elements denote real features of the world,
the relation they bear to those features is, in a phrase endorsed by Hughes
(and borrowed from Duhem), `only the relation of sign to the thing signied'
[Hughes 1996, p.
330].
cisely this reason; they argue that `the notion of denotation seems to imply that
the relation between model and target is purely arbitrary' [Ducheyne 2006, p.
216].
be
To be sure, denotation
may
retical models, denotation will more often than not be guided by background
assumptions and theoretical frameworks, as model-building is typically part of
a larger project of inquiry. This naturally imposes limits on the degree to which
denotation can be arbitrarily stipulated. Furthermore, the criticism overlooks
that, according to the DDI account, it is the triad of denotation, demonstration,
and interpretation, rather than denotation alone, which jointly establishes representation. By following Goodman's original advice that `we must examine the
characteristics of representation as a special kind of denotation' [Goodman 1976,
p. 5], the DDI account does not simply equate denotation and representation.
Rather, the idea is that, in order for representation to be attained, denotation
must be put to the test by successful demonstration and interpretation.
Some degree of arbitrariness in denotation is not only to be expected, but
may even be considered necessary for the purpose of model-building. Devising
a theoretical model requires selecting the
relevant characteristics
of a physical
the DDI account claims. The point is not that representation can somehow be
established by mere stipulation, but rather that denotation provides a tentative,
prima facie
demonstration
is perhaps the
While such
techniques may have their own practical and conceptual problems, they do not
directly touch upon the question of how theoretical models represent a reality
external to it; demonstration, to paraphrase Hughes, takes place entirely within
the model in its mathematical formulation. Nonetheless, demonstration is essential, both insofar as it allows for the derivation of results and predictions from
a model, and because it makes salient that a mathematical model, by virtue of
its being a mathematical object, has an internal dynamic.
bring out the internal dynamic of a model, there are also the considerable deductive resources contained within mature mathematical formalisms, which will
be discussed in detail in the second half of this paper.
Interpretation, like the other two components of the DDI account, is constitutive of the way theoretical models represent. Without it, demonstrated results
would remain merely formal results within a deductive mathematical structure,
lacking empirical meaning.
what we have demonstrated [...] back into the world of things' [Hughes 1996, p.
333], and interpretation plays this part. Whereas denotation picks out features
in the world, which are then denoted by elements within a model, interpretation projects internally derived results back on to the world, where they must
be assessed in terms of their empirical adequacy. There is no guarantee of success in either denotation or interpretation. However, when both are successful
when a model picks out the right features in the world, and interpretation
assigns empirically correct meanings to demonstrated results denotation and
interpretation may indeed be said to be the inverse of each other, and the model
as a whole may be deemed empirically adequate.
Interpretation is, of course, no simple aair, as it may require considerable
ingenuity and imagination. In the case of a mathematical model, even when a
result has been successfully derived within the formalism of the model equations,
its empirical interpretation may not always be self-evident.
consider the case of mathematical divergencies:
10
As an example,
variables diverge for certain parameter values, the interpreter may be faced
with the choice of either dismissing it as an `unphysical' result for example
because the corresponding physical magnitude is recognised as necessarily nite
for any nite physical system under consideration or interpreting it as an
indicator of a real feature in the world (e.g., a phase transition), though perhaps
one that the model is unable to capture (e.g., a phase transition).
Just as
not merely passively presented with results which it must make sense of; rather,
its operation will be informed, and at the same time constrained, by what has
gone into the construction of the model up to that point.
stipulative
element.
It would, how-
ever, be hasty to conclude that the DDI account fails on this ground, since
the DDI account construes successful representation as the
joint
eect of the
That
is, demonstration and interpretation may provide the DDI account with the
internal resources to keep `in check' whatever element of arbitrariness is introduced by denotation. Still, there remains the very real worry that, by making
representational force 6 (or directedness ), the DDI account cannot exclude spurious
denotation the decisive element by which a model fundamentally acquires
cases, where a failure of denotation for example, because it `misses the target' (or targets the wrong sort of entity or phenomenon) may go undetected
during the later stages of demonstration and interpretation, simply as the result of fortuitous but non-accidental circumstances of the case at hand. If this
sounds too abstract, the same point can be put more concretely, in the form
of a question: Why should only denotation, not demonstration and interpretation, be involved in establishing the fundamental representational directedness
of a model? After all, scientic practice appears to be replete with examples
where demonstration and interpretation determine to a signicant degree what
a model is
a model of
pretation constrain the direction of t. Whether model and target stand in a
in this volume.
6 The
11
A successfully represents
A's inferential utility, where
that, without inferential utility as a constraint on model-construction, denotation may fall foul of problems of indeterminacy.
One might worry whether the indeterminacy of denotation and the resulting ambiguity of denotational arrangements are all that signicant. Surely, one
might argue, `unnatural' denotational arrangements such as the paper's (instead of the pens') representing the ships in the pens-and-paper model can
easily be ruled out on the basis of simple criteria, such as similarity, which
do not require deployment of the full inferential apparatus. While some basic
recognitional capacity is needed in order to recognise a at piece of paper as
more similar to the sea's surface than a compact pointed object such as a pen,
such a capacity hardly requires much in terms of the active drawing of inferences. It is, after all, what children do intuitively when they play `battleships'
7 Responses
12
Fermi sphere.
In accordance with fundamental constraints such as the Pauli exclusion principle, correlated electrons in a solid can only occupy certain (mutually exclusive)
energy states. If a certain single-particle state is fully occupied, a new electron
that is added to the system must occupy the next-highest energy state that is
available. In a system of many electrons, for example a solid with of the order
23
of 10
(k, )
Ek
of an electron varies
= 0),
the visual
model that is typically invoked is that of a sphere containing all occupied electron states, centered around the single-particle ground state
k = 0.
At zero
temperature, when the system occupies the many-body ground state of lowest
total energy, this
surface, all occupied states inside, none outside. In the presence of interactions
and uctuations, however in other words, when the system is at nite (nonzero) temperature the sphere will be deformed, and some electrons will be
excited into states outside the original sphere, leaving behind an unoccupied
state inside.
Due to the strong correlations that exist between the constituent electrons in
a solid, various disturbances and excitations from the ground state can give rise
to a range of collective responses of the system, perhaps the most important of
which are instances of quasi-particle behaviour. Such behaviour is characterised
by the (chimerical) appearance of `new' particles, which appear to have measurable properties such as a virtual `eective mass', average life-time, electric
charge, and so forth. These so-called
quasi-particles
particles across a wide range of measurements and experiments, and their eects
can be reproduced experimentally in a robust and reliable manner. For example,
removing an electron from the system, and with it its negative electrical charge,
inferential utility, can only go so far. As I shall suggest in the next section, a more fruitful
place to look for constraints on denotational arrangements would be in their material or
formal characteristics, insofar as these embody certain inferential strategies and theoretical
commitments.
As I will claim in Section 4, this is one of the features that make mature
13
would appear to leave an `electron hole' behind which may then `act' as a positively charged quasi-particle, and as such may, for example, form bound states
with other electrons (forming so-called `excitons').
those involving ctional objects, it may still turn out to be a central ingredient
in
scientic
it
14
else' [Goodman 1976, p. 5], it does not have a role to play in scientic models.
After all, `if the appropriate relationships are not in place between the relevant
properties then the model will not be deemed scientic' [French 2003, p. 1478].
But this is a platitude, not a valid criticism:
mere
15
[French 2003, p.
note to the same paper, French refers to work by Hendry and Psillos in which
they develop an `interactive view' [Hendry 1999], according to which theories
and models combine elements from dierent representational media: `On such
a view, representation might be analyzed in terms of some combination of denotation and isomorphism although it is not clear quite how this would work.'
[French 2003, p. 1478] Such rhetorical questions and quick dismissals, however,
only carry weight if one is already operating on the assumption that denotation,
interpretation and, by extension, considerations of inferential utility in spite of
their being conceptually distinct from isomorphism are incapable of giving rise
to any interesting questions of their own. What needs to be realised, however,
and what has perhaps not yet been adequately acknowledged, is that the multiple accounts of representation discussed in the present paper can be thought of
much like Hendry and Psillos's `interactive view' as attempts to make sense
of representation as a
heterogeneous phenomenon.
an ongoing project which does not argue towards a xed conclusion and still
needs to address many open questions, should not be held against it; such is the
nature of live philosophical projects. The real challenge, then, consists not in
concentrating on cases where one of the candidate sources of representational
force where it be denotation, inferential utility, or any of the other possible
sources fails, but in investigating the interplay between them.
Recognising that successful representation is typically the joint eect of an
interplay of multiple factors, is necessary, not least in order to account for the
diversity of actual scientic models. Most philosophical discussions of scientic
models this paper included tend to focus on theoretical or mathematical
models. However, just as it would be hasty to dismiss examples that cannot be
assimilated to one's preferred (e.g., structuralist) account of representation, so
it would be careless to dismiss a whole tradition of philosophical thinking about
mechanical and material models, simply on the ground that such models are
specied directly in their materials, rather than in their relation to theory. The
denotational and inferential accounts of representation can be usefully applied to
such material models, as has recently been argued by Davis Baird (2004). Baird
discusses both the example of orreries devices which mechanically represent
the motion of various planets and moons and Watson and Crick's (physical)
model of the DNA double helix. In those cases it is entirely appropriate to say
that `the sticks in Watson and Crick's [double-helix] model
not rigid metallic connections', and that one can use an orrery in a hands-on
way, by setting it in motion `to
demonstrate [...]
(1968), manipulating
the physical model became not just a convenient shortcut for theoretical reasoning based on empirical data, but an independent source of inferential utility:
8 It
is again telling that some structuralist critics relegate denotation to, at best, the seman-
tic realm of `how words represent' [French 2003: 1478] i.e., of linguistic meaning thereby
ignoring the manifestly denotational aspects of some of our non-linguistic ways of interacting
with the world.
16
`The
the essential
trick, instead, was to ask which atoms like to sit next to each other. In place of
pencil and paper, the main working tools were a set of molecular models supercially resembling the toys of preschool children.' [Watson 1981, p. 34] From
the philosophical armchair, tactile manipulation of a physical model may look
like a poor substitute for theoretical derivation, but it may acquire paramount
importance `when conceptual manipulations are impossible either for lack of a
theory or because analytical manipulations would be too dicult' [Baird 2004,
p. 39]. Either way, material models as well as theoretical models allow their
user to articulate and infer knowledge of the part of the world they represent;
successful denotation is an important component in this relation, as is inferential
utility, which allows for the extraction of information about the target system.
Let us step back for a moment and take stock. In the preceding sections, I argued
that representation needs to be understood as a heterogeneous phenomenon,
which, except in trivial cases, is brought about by the presence of more than
one source of representational force.
support of the thesis that no single source denotation, inferential utility, let
alone similarity is enough to
guarantee
alone,
and
failure to come up with universal necessary and sucient conditions for representation, we should conclude that there are `no deeper feature to scientic
representation other than its surface features' [Surez 2004, p. 769] where the
dominant `surface feature' of a representational model is its inferential utility
as a source of `surrogate reasoning'.
This
inferential
conception of scientic
17
A is itself `a
pragmatic skill that depends on the aim and context of the particular inquiry'
[Surez 2004, p.
773].
successful
case, beyond
the mere fulllment of these two necessary conditions, is not something that can
be prejudged by abstract analysis: `In every specic context of inquiry, given
a putative target and source, some stronger condition will typically be met;
but which one specically will vary from case to case.' [Surez 2004, p. 776]
While I concur with Surez that the search for a substantial account of
representation that is, an account in terms of
universal
scientic
representational success
the deployment
and articulation, plays an important role in achieving the kind and degree of
unication required for scientic representation.
most physicists do not regard models as literal descriptions of nature, but as standing in a relation of analogy to nature ' [Hesse 1953,
is the suggestion that `
p. 201; emphasis original]. It is her detailed defence of this second claim that
sparked a long-lasting debate over the analogical character of models. Much less
mathematical
formalisms, when used as hypotheses in the description of physical phenomena, may function like the mechanical models of an earlier stage in physics,
has been made of Hesse's rst claim, however, which asserts that `
9 This
specic
dierence is partly a matter of whether one wishes to emphasise the role played by
features of a given model, or whether one is more optimistic about the possibility of
identifying
general
18
on in the remainder of this paper. In particular, I shall argue that the existence
of mature mathematical formalisms often plays an important role in bringing
about successful representation by mathematical models. In order to make this
case, I shall have more to say about what the term `mathematical formalism' is
meant to capture, and how Hesse's suggestion can be adapted to the contemporary debate. Before delivering on this promise, however, I shall try to unpack
Hesse's position in a little more detail.
Hesse's early work must be understood against the backdrop of a tradition that largely equated scientic models with either
material
or
mechanical
models. On this view, which derives from 19th-century scientic usage, a typical scientic model is `a (real or imagined) concrete, material representation
of something' [De Regt 2005, p.
215].
point towards
progress and
suggest
new steps of modication and generalisation, rather than making progress inevitable or entailing specic modications. At the same time this is the second
19
point, which is at least implicitly present in the quote from Hesse mathematical formalism not merely
accommodates
ism `
branch
the sense discussed here, will typically be more local than any branch of mathematics such as topology, number theory, algebra, etc. For the same reason, not
just
any
itself.
recognisable
its value lies in its utility to the user, for example in the extent to which it
enables the user to derive empirical results or to pursue those modications and
generalisations that Hesse identies as `pointers towards further progress'.
10 This
11 Hence,
in this section, I do not intend this to be synonymous with Eugene Wigner's phrase of the
`unreasonableness eectiveness of mathematics in the natural sciences'.
20
culus', `partial dierential equations', `method of least squares', `Green's function techniques', `discretization', and so forth.
usage is of limited usefulness in assigning a precise meaning to the term. However, even though the individual items on this (or any similar) list may not seem
to have much in common with one another, the investigative contexts in which
scientists help themselves to such `mathematical techniques' nonetheless share
certain general characteristics. Typically, and by no means trivially, references
to `mathematical techniques' usually occur in contexts where the generation of
(often numerical) results is the main concern. Thus, mathematical techniques
may be employed to reduce complexity, predict numerical outcomes, extract
data, or calculate expectation values. What counts as a `result' that is: as an
outcome that will bring inquiry, at least momentarily, to an end depends on
the overall goal of the investigator: for a mathematical physicist, being able to
write down a solution in a mathematically closed form might suce, whereas a
computational physicist might not be content until a numerical value (perhaps
within a certain margin of error) has been calculated.
12
For the purpose of the present argument, I shall understand by `mathematical techniques' those outcome-oriented uses of mathematics that aim at
generating empirically adequate results, either numerically or derived through
approximations, which may or may not be considered
ad hoc
by the lights of
fundamental theory. With mathematical techniques, given the wide range of approximations and methods employed across the sciences, it is dicult to single
out any one example in particular. Most approximation techniques for systems
of dierential and integral equations (including discretization and nite element
methods) would fall into this category, as would variational methods which, in
a purely formal way, map analytically intractable problems onto optimization
problems that may then be solved numerically. It is not untypical, within the
class of `mathematical techniques', to nd a lot of continuity between techniques
for
deriving results
and methods of
data analysis.
plied factor analysis may be understood equally as `[1] a tool for uncovering the
order, patterns or regularity in data or [2] as a mathematical mold of casting
scientic theory' [Rummel 1970, p. 12]. Indeed, one could argue that it is in
the nature of the `outcome-orientedness' of mathematical techniques, as dened
here, that little distinction is made between
fundamental theory, and
data
results
formal framework.
cal techniques', scientists do not themselves typically use the term `formal (or
mathematical) framework'; hence, whereas in the former case, it was necessary
to `clean up' scientic usage in order to narrow down the meaning of the term,
in the case of the notion of `formal framework' some degree of `extrapolation'
12 Tarja
21
will be necessary in order to make clear what element of scientic practice the
term is supposed to capture. To this end, it will be useful to begin by drawing a few contrasts. Whereas mathematical techniques are typically tailored to
specic empirical contexts, formal frameworks are usually established in a more
`global' way, without much regard for the details of any particular empirical situation. Also, whereas the choice of mathematical techniques often takes place
well into
the process of inquiry for example, when it turns out that a deriva-
tion from `rst principles' is too complex and that numerical approximations
need to be made the choice of a formal framework often
precedes
the formu-
symmetry.
of many fundamental laws of contemporary physics, so it may seem that arguments from symmetry constitute simply an appeal to fundamental theory rather
than an instance of imposing a (mathematical) formal framework. However, if
one looks at the history of modern physics, it becomes clear that in crucial periods of, for example, the evolution of modern quantum theory,
group theory
An important
historical role is played by the mathematician Hermann Weyl. (For an insightful discussion of the relation between mathematics in physics in Weyl's work,
see [French 1999].) In his monograph The Theory of Groups and Quantum Mechanics (rst published in German in 1928), Weyl provides what is perhaps best
22
features which are not contingent on a special form of the dynamical laws nor
on special considerations concerning the forces involved' [Weyl 1928, p. xxi]. Its
very independence from perhaps even disregard for specic empirical detail,
along with an at best loose relationship to what was known about fundamental
theory at the time, allowed group theory to function as a formal framework
for atomic physics and, later, nuclear physics, where the pursuit of symmetry
relations (for example via the standard model of elementary particles) became
13
The two uses of mathematics distinguished thus far as, on the one hand,
a set of techniques for the `local' generation of specic results and, on the other
hand, a formal framework for physical theorising dene dierent ends of a
spectrum of deploying mathematics in scientic inquiry. I now wish to turn to
a third mode of using mathematics in science, which occupies a middle position between the two extremes outlined above, but which at the same time has
characteristic features that cannot be reduced to an admixture of elements from
either mathematical techniques or formal frameworks. I shall refer to this use
of mathematics in science as the deployment of a
mathematical formalism.
As
with the two previous cases, some care must be taken to give a specic meaning
to the term `mathematical formalism'. `Mathematical formalism', in the present
context, is to be understood as a
physically interpreted
a range of
It must
same time, it will typically fall short of universality, in the sense that certain
cases though logically and physically permissible nonetheless do not lend
themselves to the application of the formalism.
ematical formalism will typically include rules and heuristic assumptions that
are not themselves part of either the mathematical framework or entailed by
the `underlying' fundamental theory.
The general features mentioned so far are inferred from the assumption that
mathematical formalisms inhabit a middle ground between outcome-oriented
mathematical techniques and formal mathematical frameworks. What needs to
be shown is that such a middle ground does indeed exist, and this can be done
by way of example. In the remainder of this subsection, I wish to discuss in some
detail what I take to be a good example of a mature
mathematical formalism
in physics, in the sense that has been discussed so far. The example I have in
mind is the formalism of creation and annihilation operators in quantum manybody physics. I shall rst give a theoretical justication (or rather: motivation)
for this formalism, before giving examples of how it is deployed in practice.
However, it is worth emphasising beforehand that the mere possibility of giving a
theoretical justication for a given formalism does not entail that any particular
application of that formalism is best understood as a `direct' application of
fundamental theory to a specic problem.
be set up in such a way that its output will `automatically' satisfy certain
13 This
applies to the Standard Model of particle physics as well as its proposed supersym-
metric extensions.
23
mediates
between theoretical
).
nihilation operators,
k,
and
k, ,
creation
and
respectively.
an-
This
14 As representations of fermions
(spin-1/2 particles such as electrons), the operators must satisfy certain anticommutation relations in order to satisfy the basic constraints of fundamental
theory.
So far, I have only indicated which mathematical framework in this case,
operator algebra the formalism draws on, and how its symbolic conventions are
to be understood. However, for a physically interpreted mathematical formalism
to emerge, it is necessary that certain theoretical, or other (e.g., methodological),
commitments are embodied in the rules and criteria for what constitutes a meaningful application of the formalism. As I have already indicated, mathematical
formalisms occupy a middle ground between global mathematical frameworks
and specic local applications of mathematical techniques; their usefulness to,
for example, a scientist interested in modelling a specic class of physical systems, consists in narrowing down the wide range of what is logically or physically
permissible (as judged from fundamental theory), without thereby precluding
the application of the
14 Thanks
same
formalism to
dierent
to the regularity of the crystal lattice, however, one can perform a (discrete)
Fourier transformation, which maps the representation back onto individual electrons associated with lattice sites. Hence, even though the electrons are delocalised, we may nonetheless
speak of the creation
lattice site
i.
( )
ai,
and annihilation
(i, )
a
24
of an electron of spin
associated with
in quantum many-body
per se
However,
simultaneously
cal formalism does not include a variable for time, there is no `time lag' (so
to speak) between the removal of an electron at one place and the addition of
another electron at a dierent place in the system.
temporal dimension, physicists nonetheless choose to give a dynamic interpretation to the eect of the joint application of an annihilation and a creation
operator (never mind the absence of a dynamic component from the model):
they claim perhaps: pretend that what is `being modelled' is the movement
of an electron from the place where it was `annihilated' to the place where it
was (re-)`created'.
I have here given only the briefest sketch of the example of creation and
annihilation operators in many-body physics.
rules that are neither entailed by fundamental theory nor tailored to any specic
empirical problem in particular. Such substantive assumptions and constraints
including whatever theoretical justication it might be possible to give for
them are not always explicitly stated, but are often implicit, for example in
25
mapping )
approach (see [da Costa and French 2003] and references therein). It needs to
be emphasised that, by drawing contrasts between these three accounts of the
role of mathematics in science, I am not suggesting that they are necessarily mutually exclusive. Rather, each emphasises dierent formal aspects of the mathematics employed, and each is characterised by qualitative dierences between
the kinds of theoretical activities involved in achieving model-based representation i.e., between mathematical
the
deployment
moulding,
the
mapping
of structures, and
26
`an adaptive strategy until all ingredients [are] integrated'; often, one nds that
`in the integration process, tuning is essential': This ne-tuning `is essential
to ensure that the mathematical representation has empirical signicance'. In
the particular case study, an essential part of mathematical moulding consist in
the models' parameter values being `chosen such that the model could precisely
mimic specic facts about the cycle' [Boumans 2005, p. 50].
The idea that models need to be `moulded' to t specic empirical contexts coheres well with other approaches that draw their inspiration from case
studies of actual scientic practice. Thus Margaret Morrison, in a contribution
to the same volume that contains Boumans' 1999 paper, argues that `model
construction involves a complex activity of integration' [Morrison 1999, p. 44];
like Boumans, she elsewhere emphasises that `the proof or legitimacy of the
representation arises as a result of the model's performance in experimental,
engineering and other kinds of interventionist contexts' [Morrison 1998, p. 81].
A similar sentiment is expressed by Surez when he writes that models are `inherently intended for specic phenomena' [Surez 1999, p.
75].
What these
remarks indicate is that models are frequently employed in specic (often explanatory) contexts, when a phenomenon cannot be adequately described, or
perhaps cannot even qualitatively accounted for, by (limiting cases of ) fundamental theory.
Hence,
phenomenological models
rather than the exception, across much of science. However, while this shift in
perspective, from models as an intermediate step in applying theory to the world
to models as representations of specic phenomena, has generally sharpened an
awareness of how models are being used in actual scientic practice, I believe
it does not adequately capture the role of mathematics in achieving representation. `Mathematical moulding', with its goal of enforcing empirical adequacy of
a model by ne-tuning its ingredients, appears to put too much emphasis on the
empirical performance of a model in a specic situation. The second element
identied by Boumans calibration as `the choice of the parameter values'
conveys the connotation of `curve-tting', which many model-builders seek to
avoid. Indeed, integrating `all the ingredients' in many cases of scientic modelling may neither be feasible nor desirable; much model-building is motivated
by the recognition that the `adding of details with the goal of improving [...
the] model is self-defeating such improvement is illusory' [Batterman 2002,
p. 22]. This is particularly salient in the case of what Robert Batterman calls
`minimal models', which are meant to shed light on the
universality
of certain
of phenomena,
class
formalisms are an obvious candidate for providing the requisite balance between
specicity and generality.
Let us turn to the second contrast case, which, for want of a better designation, I shall refer to as `the structuralist approach'. I here have in mind a cluster
27
and
M. Typically,
and
onto a substructure
of
M. Introducing
assuming a way to patch this calculus up with the semantic resources for
then allows one to properly regard
).
M,
[Redhead 1975, p.
is to be
picture attractive, for the present argument, is its ability to make plausible how
mathematics more specically: the `surplus' mathematical structure that is
made available by embedding a theoretical structure into a larger mathematical
framework may be of heuristic value, insofar as new mathematical resources
become available that may eventually suggest new theoretical developments.
The structuralist approach has since been extended and rened, most recently
through the inuential `partial structures' programme (for a review and defence, see [da Costa and French 2003]). Such renements are partly a response
to claims that a structuralist account (of the general kind sketched above) is
unable to account for the diversity of the kinds of models including analogies, iconic models, material models that scientists actually deploy in their
everyday work. One move has been to relax some of the requirements of earlier
structuralist accounts, for example by demanding only partial identity between
models and situations and by allowing partial isomorphisms to take the place of
full isomorphisms. Some critics, however, have argued that, while partial structures and partial isomorphisms bring greater exibility, they also are in `danger
of trivializing our representational relationships' [Pincock 2005b, p.
1254]; in
15 Mathematical for-
15 This
need not indicate a lack of interest in scientic practice generally, since within the
28
used
real
physical
as far as it goes,
I believe that the quote from Hendry and Psillos conveys a justied sense of
dissatisfaction, insofar as the focus on mere structures and their relations obscures the value of mathematics as something that is
Mature mathematical
As I suggested at the beginning of this section, my goal in considering mathematical moulding and the structuralist approach as contrast cases to mathematical formalism was not to argue for any incompatibility between the latter and
either of the former two. Rather, I intended to situate mathematical formalism
as an integral part of model-building between the two. For, as I hope to
have made plausible, mathematical moulding and the structuralist approach lie
at opposite ends of a whole spectrum of conceptions of how mathematics applies
to the world. Loosely speaking, whereas mathematical moulding accounts for
the
local
global
embedded in `mathematics-at-large', without necessarily much concern for specic features of a given model beyond its `embeddability', as it were. Both the
local and the global roles of mathematics are important and interesting in their
own right.
between the two ends of the spectrum dened by mathematical moulding and
the structuralist approach; it is this middle ground, I am suggesting, that is
16
structuralist approach there has been a fair bit of discussion of the history of science, especially
in relation to theory change.
themselves more easily to subsumption under a structuralist framework than does the activity
of model-building.
16 In
his book
How Economists Model The World Into Numbers, Boumans hints at a similar
29
embodies
theoretical
constraints and relationships in much the same way as, say, an orrery embodies
the dynamics of the planetary system it represents (see the discussion at the
into a formalism need not always be explicit; many mature formalisms involve
an element of `self-regulating conventionality', where such conventionality `is
not the result of any explicit attempt to conventionalise a practice', but arises
from successive applications, for specic representational goals, in a way that is
role of mathematical formalism when he writes that `a mathematical formalism is sought that
is able to
or described' [Boumans 2005, p. 58, italics added]. However, mathematical formalism is still
understood as `one of the ingredients that should be integrated' (ibid., p. 17), rather than as
an integrating force itself.
17 While this is, of course, only an analogy, philosophers of mathematics, too, have recognised
the parallels between mathematics and activities and phenomena in the material world. Thus,
Imre Lakatos writes: `Mathematics, this product of human activity, alienates itself from the
human activity which has been producing it.
from the activity which has produced it; it develops its own
30
18
Mature formalisms of the sort discussed here are not conned to, and indeed will usually go beyond, the notational devices standardly used in pure
mathematics.
that straddle the boundary between the mathematical and the merely diagrammatic.
which describes quantum states using angle brackets and vertical bars as notational devices (thus allowing for the quick evaluation of expectation values, or
the spectral analysis of a Hamiltonian), are merely short-hand for more complex
mathematical
agrams, have been developed with the goal of representing a potentially indefinite number of specic kinds of
physical
in quantum electrodynamics consists of points (`vertices') and arrows (of different orientation) attached to the vertices, representing interacting electrons
and positrons, as well as wavy lines signifying photons that may be emitted or
absorbed. Enshrined in the formalism of Feynman diagrams are both rules for
the
construction
and momentum among the interacting particles is required'), as well as for the
in the diagram represent virtual particles, which may temporarily violate the
relativistic energy-momentum relation, but which are in-principle unobservable
if they do not'). While the formalism of Feynman diagrams was developed on
the basis of an overarching theoretical conception according to which each
diagram is to be interpreted as a contribution to the total amplitude for a (multiply realizable) quantum process it has taken on `a life of its own' in certain
areas of high-energy physics, where it has developed from a mere shorthand
to what one might call a notational `lingua franca'. As physicists are keen to
emphasise, `Feynman diagrams provide deep physical insight into the nature of
particle interactions', such as scattering processes, `in addition to their value as
19
a mathematical tool'.
18 This
is not to say that formalisms do not include explicit rules and conventions, only that
these do not
19 See
exhaust
the entry `Feynman diagram' on Wikipedia (7 March 2010 version, accessed 16 April
2010). It seems fair to say that, while not an authoritative source of knowledge, Wikipedia
usually provides a good sense of what some of those who feel most strongly about a topic (in
this case: theoretical physicists) tend to think.
31
constrain what may be represented (and how), they also function as inferential and interpretative resources.
tions to the overall Hamiltonian, they also often suggest `concrete' interpretations of individual terms (as in the `hopping' of electrons from one lattice site
to another, discussed in Section 4.2.). It bears repeating that a mature mathematical formalism (in the sense discussed here) even though it may aord its
user considerable latitude in concrete applications is infused with theoretical
commitments, which determine, at least in part, which physically meaningful
interpretations of a specic application are legitimate. This is precisely what
makes mature formalisms so convenient to their users, who would otherwise
have to settle on one amongst a plethora of
prima facie
permissible interpre-
constraining
and
References
Thing Knowledge: A Philosophy of Scientic Instruments, Berkeley: University of California Press 2004.
20 As
Marion Vorms argues in her contribution to this volume, the dierent formats and
media in which models are presented, play a similar dual role in scientic representation.
32
[Batterman 2002] Batterman, Robert W.: Asymptotics and the Role of Minimal Models,
53
and Margaret Morrison (eds.), Models as Mediators. Perspectives on Natural and Social Science, Cambridge: Cambridge
University Press 1999, pp. 66-96.
Numbers, London:
Routledge 2005.
Oxford:
Philosophy of Science
72 (2007)
pp. 48-68.
[da Costa and French 2003] da Costa, Newton C.A., and Steven French,
[De Regt 2005] De Regt, Henk: Scientic Realism in Action: Molecular Models
and Boltzmanns's
Bildtheorie , Erkenntnis
230.
[Ducheyne 2006] Ducheyne, Steen:
The Pragmatic
Philosophia Naturalis
[French 1999] French, Steven: Models and Mathematics in Physics: The Role
of Group Theory, in Jeremy Buttereld and Constantine Pagonis (eds.),
Cambridge
Philosophy of Science
Interna-
263.
[Gelfert 2009] Gelfert, Axel: Rigorous Results, Cross-Model Justication, and
the Transfer of Empirical Warrant: The Case of Many-Body
33
[Hendry 1999] Hendry, Robin: Theories and Models: the Interactive View, in
Ray Paton and Irene Neilson (eds.),
Interpretations, Berlin:
How to Do
The Courage of Doing Philosophy: Essays Dedicated to Leszek Nowak, Amsterdam: Rodopi 2007, pp.
Piotr Przybysz (eds.),
123-157.
[Hesse 1963]
ence
Philosophy of Sci-
[Lakatos 1976]
An Arte-
Studies in
[Lewis 2004] Lewis, David: Void and Object, in John Collins, Ned Hall, and
L.A. Paul (eds.), Causation and Counterfactuals, Cambridge
(Mass.): The MIT Press 2004.
[Morrison 1998] Morrison, Margaret: Modelling Nature: Between Physics and
the Physical World,
pp. 659-674.
[Netz 1999]
Netz, Raviel:
Press 1999.
[Pincock 2005] Pincock, Christopher:
34
Essays on the
Monza:
Polimetrica
[Pincock 2005b] Pincock, Christopher: Overextending Partial Structures: Idealization and Abstraction,
1248-1259.
Synthese
Rudolph
J.,
Evanston:
Magnani,
Nancy
J.
Nersessian,
&
Paul
Thagard
Science
Philosophy of Science
pass
Philosophy Com-
Erkenntnis
(this is-
sue).
Gesammelte Abhandlungen,
1968.
35
Berlin:
Springer