You are on page 1of 38

F. P.

OGORMAN

POINCARl?S CONVENTIONALISM
APPLIED GEOMETRY

OF

1. Physical and Mathematical Spaces


tend to look on pure geometry as an uninterpreted formal
system, or what Frege calls a formal theory. Applied mathematicians, on the
other hand, are often said to be concerned with the question of which
interpreted formal geometry is true of our world, or, to use Freges terminology,
which of the first-level geometrical propositions is true. From this latter point
of view, while the statement-forms of a purely formal system of geometry
are, as such, neither true nor false, the question of the truth of an interpreted
system does arise. Nicod criticizes Poincare for having overlooked this in
arguing for the geometrical conventionalism
of applied geometryZ, and
Nagel holds a similar viewa.
Poincare, however, did not overlook this point; rather he held, at least
implicitly, that this kind of criticism does not apply in the case of the
interpretation
of geometry within physics. Indeed Nagel himself in his
comments on Poincares conventionalism of applied geometry, seems to have
seen this, but he failed to realise its significance because of the dichotomy he
himself makes between pure and applied geometry. Thus he remarks that
PoincarC sometimes wrote as if the grounds for the conventional or definitional
status of applied geometry were identical with those for pure geometrf , but
he did not devote sufficient attention to this aspect of Poincarts thought and,
therefore, failed to see its true import. As we shall see, Poincare held that the
space studied in physics is the space of classical mechanics, and that this
space is the mathematical continuum. In other words, the mathematical
continuum was, for Poincare, part of the model or interpretation of geometry
in physics, i.e. any assertion about physical space within classical mechanics
MATHEMATICIANS

Strictly speaking, in Freges opinion, one should not speak of the interpretation of a for&
theory. However, since this manner of speaking is standard, I will retain it.
'Cf.Nicod, Geometry and Induction. p. 17.
Cf. Nagel, The Structure of Science, p. 261.
Cf. Nagel, op. cit. p. 262.
Stud. Hist. Phil. Sci. 8, (1977), No. 4. Printed in Great Britain.
303

304

Studies in History and Philosophy

of Science

presupposes that it is mathematically continuous. If Poincare was right in this,


then his argument for the conventionalism
of pure geometry, (viz. the
metrical amorphousness of the real continuum5) holds also for applied
geometry, and despite Nagels views to the contrary, Poincare was perfectly
entitled to argue in this manner.
(a) Poincar& and Lobachewsky s Experiment
In his discussion of Lobachewskys parallax experiment, Poincare states
that, even if it were proved by experiment that the parallax of a distant star is
negative, one could choose either to abandon Euclidean geometry or to modify
the laws of optics by supposing that light waves are not strictly propagated in
a straight line, and that therefore Euclidean geometry has nothing to fear
from new experiments6.
Prescinding for the moment from the kind of change necessary in optics
for the rentention of Euclidean geometry in such a contingency, the question
may be asked whether Poincares statement above is intended to be an
argument for geometrical conventionalism
in applied geometry. Despite
Nagels interpretation to the contrary, I believe that Poincare did not ground
his geometrical conventionalism on considerations of this kind. If he did so
his argument is invalid since, as Russell points out, the question as to
whether the metrical axioms of Euclidean geometry are conventions, or are
true or false, is logically distinct from the question as to whether we can verify
whether they are true or false. If Poincares statement above is taken as his
only argument for geometrical conventionalism, all it proves is at most that
we cannot verify the truth of any geometry. It also implies that verifiability is
the ultimate ground of Poincares geometrical conventionalism. But this is to
read too much of logical positivism into Poincares works. If Poincare intended
to argue from verifiability, one would expect him to explicitly say so in his
reply to Russells criticism. But in this reply Poincare does not refer to
verifiability; on the contrary, he explicitly rejects Russells contention that
his arguments merely show that some one geometry is true but we cannot
verify which onelo. Indeed Poincare, in his reply to Russell, expressly rules
out the possibility of arguing from verifiability to geometrical conventionalism,
and therefore his claim that one can retain Euclid despite apparently adverse
Poincare maintained that, topologically speaking, there is nothing in the nature of the real
continuum which singles out the Euclidean metric from the other possible metrics indicated by
the distance function d(x,y), and hence the real continuum is metrically amorphous.
Poincart, Science and Hypothesis, p. 73.
Cf. Nagel, op. cit. p. 262.
Cf. Russell, Sur les Axiomes de la Cikometrie, p. 685.
sin particular one would be reading too much of Reichenbach into Poincart, since there is no
doubt that Reichenbach rests his own thesis of geometrical conventionalism on testability (c$ The
Phitosophy of Space and Time, p. 16).
Cf. Poincare, Sur les Principes de la Geometric, p. 74.
Cf. ibid.

Poincarbs Conventionalism

of Applied

Geometry

305

experimental evidence cannot be interpreted


as an argument for the
conventionalism of applied geometry.
In making such a claim, Poincare is assuming that he has already established
the thesis of geometrical conventionalism in the case of applied geometry,
and is merely spelling out its consequences. What he is saying is that, if one
were to discover, for instance, a negative parallax for a distant star it would
not be necessary to conclude that physical space is Riemannian, for, since
congruence is a matter of definition, it would still be possible to retain Euclid
provided one made the necessary modifications
in our optics. If this
interpretation is correct, we have yet to discover Poincarts reason for
geometrical conventionalism in the case of applied geometry. His reason is the
following. Within classical mechanics, and hence within classical physics,
space is taken by convention to be mathematically continuous; and, because
such a continuum is metrically amorphous, congruence is a matter of
convention. We shall now discuss these points more in detail.
(b) Classical Space and the Mathematical Continuum
In his discussion of Russells geometrical empiricism, Poincare states that
the distance, for example, between London and Paris is not an absolute datum
of experiencej2. Poincare is here clearly talking about the material world,
and, to use Russells words, is contending that distance is not an absolute
datum preexisting measurement, as America, for instance, pre-existed its
discovery. In other words, Poincare is assuming that physical space is
metrically amorphous, and that therefore the choice of congruence is a
matter of definition13. He makes the same assumption in Science and
Hypothesis, when he states that the question of the self-congruence of a
transported rod is a matter of definition14. The reason why Poincare held
physical space to be metrically amorphous is because he interpreted the
properties of physical space in the light of classical mechanics. While
Poincare held that classical mechanics is an experimental science, he also
maintained that it contains certain conventional elements, and among these
elements he mentions Euclidean geometry (which he caIls a kind of convention
of language) with its claim that space is continuous and therefore metrically
amorphous.5 In other words, Poincare accepted the physical space of classical
mechanics as being, by convention, mathematically continuous and as such
metrically amorphous.
In this context it is quite clear what Poincare meant when he said that
Euclidean geometry has nothing to fear from new experiments. Because the
Euclidean or continuous space of classical mechanics is conventional, it has,
Poincart, op. cit. p. 81.
He explicitly makes this claim in The Value ofScience, p. 37.
Cf. Poincart, Science and Hypothesis, p. 45.
Cf. Poincart, op. cit. p. 89.

306

Studies in History and Philosophy of Science

according to Poincare, no causal, or any other empirical influence on physical


bodies and as such is not open to experimental verification. Thus Poincart
holds that there is an absolute dichotomy between physical space and physical
bodies: because of this dichotomy, experiments, which of their nature have to
do with physical bodies, cannot give us any information about the relations
between these bodies and space or, a fortiori, betweenthe points of space thus
understood6. That Poincare did in fact hold this kind of dualism between
space and matter, and that it is not merely a possible explanation of his
position, is clear from the fact that he explains how this dualism comes about*.
The relationships between bodies as given to us in experience are, he says,
highly complex and therefore, instead of directly considering the complex
relation of one body A to another body B, we introduce an intermediary, space,
thereby envisaging three distinct relations, viz. that of body A with the figure
A of space, that of body B with the figure B of the same space, and that of
the two figures A and B to each other. The advantage of this, according to
Poincart, is that the relations between A and B are simple in comparison with
the relations between A and B, the aggregate of principles governing the
former being expressed in some geometry. Here Poincare expressly adopts
the dualistic theory of matter and space mentioned above, and this precisely is
why he maintained that experiment cannot give us any information about the
relationship between bodies and space and the mutual relations between the
different parts of space.
Against this interpretation of Poincare, however, it could be objected that
Poincare himself clearly distinguishes between laws which he considered to
be experimental, and principles, which he considered to be conventionaPO,
and calls the expression of a relation between a body A and its corresponding
spatial figure A a law, and the expression of a relation between the spatial
figures A and B a principle. Hence, it could be argued, his dichotomy
between space and matter proves, at best, that experiment cannot give us any
information about the mutual relations of the various points of space, but
not that it cannot give us any information about the relations between bodies
and space.
In my opinion, however, Poincarts conception of the relation between a
body and its corresponding spatial figure does not bear out this objection.
Cf. Poincare, Science and Z-Zypothesis,
pp. 79-84, and Sur les Principes de la Geometric,
pp. 79-86.
It is interesting to note that Quine also draws our attention to this distinction and refers to it
as the dualistic theory of spatio-temporal reality (cJ Word und Object, p. 252). However,
Poincare, unlike Quine, was not concerned with the problem of the ontological status of the
points of such a space, since for him mathematical existence simply meant freedom from
contradiction.
Cf. Poincare, The Value of Science, pp. 125-126.
Cf. ibid.
Cf. op. cit. pp. 123-124.
Cf. op. cit. p. 125.

Poincarek Conventionalism of Applied Geometry

307

The law expressing the relationship between, for example, a natural solid and
its corresponding spatial figure is that the natural solid moves approximately
according to the structures of the Euclidean group of transformations. But, in
Poincares view, this does not express a relationship between the natural solid
and some actual metrical space studied by physicists. He gives two reasons for
this. First, one may encounter the Euclidean group of transformations
in
disciplines which have nothing at all to do with metrical spaces; hence one
does not have to know, or to assume, the metrical properties of physical
space in studying the structure of the group formed by the movements of
natural solids. Secondly, while it can be verified experimentally that solids
move approximately according to the structure of the Euclidean group of
transformations without making any assumption about the metrical structures
of space, it is logically possible to construct a non-Euclidean solid and to verify
experimentally that it moves approximately according to the structure of, for
instance, Lobachewskian geometry. In such a case the scientist would conclude,
not that physical space is both Euclidean and Lobachewskian, but rather that
such experiments give us information only about physical bodies, and not
about their relationship to the metrical space of classical mechanicsz3.
It has recently been claimed by Griinbaum that the theory that the continuity
of physical space is conventional is unfounded24. Though he admits the close
relationship between the theory that physical space is continuous and
Poincares theory of geometrical conventionalism,
he maintains that
there is broad inductive evidence to support the former theory, and that
therefore this theory is not conventional. The principal reason he gives for
this view is the lack of any alternative convention which would express the
same total body of experimental findings, i.e. the fact that, in principle, no
mathematically discontinuous set of theories has been shown to be as
empirically viable as those based on the mathematical continuumz5. He fails,
however, to show that, in the case of classical physics, there is any inductive
evidence for the continuity of physical space. Indeed Grtinbaum himself
admits that this is not directly verifiable, but he offers no criteria for its indirect
Cf. SW les Principes de la Geometric, p. 82.
*Cf. Science and Hypothesis, pp. 80-84. In these pages Poincare gives an account of how we
may experimentally verify that a natural solid moves according to the structure of the Euclidean
group of transformations and of the difference between its movements and the movements of a
non-Euclidean solid.
Grtinbaum, Philosophical Problems of Space and Time, pp. 334-337. Since he justifies this
claim prior to his discussion of Einsteins general theory of relativity, it is clear that he envisages it
to be applicable to pre-relativity physics, which is the domain of Poincares geometrical conventionalism.
Cf. Grttnbaum, op. cit. p. 337. He concedes, however, that his argument merely shows the
unfoundedness, and not the falsity, of the conventionalist conception of the continuity of space;
indeed he appears to weaken his objection to the gratuitousness of the conventionalist conception
of continuity, by admitting that his argument shows merely that the advocates of the conventionalism of continuity are merely offering a programme to be completed (c$ ibid).

308

Studies in History and Philosophy of Science

verificationz6, so that it is difficult to see in what his broad inductive evidence


consists. Moreover, if in classical physics the postulate of the continuity of
space is, as Poincare maintained, simply logically conjoined to the basic
empirical propositions of physics, then, while it could be argued that the
propositions deduced from this combination are empirical, this empirical
character would not constitute broad inductive evidence for the continuity of
space. Hence the primary issue is whether Poincares account of the relationship
between space and physical objects in classical physics is correct, and
Grtinbaum does not discuss this. Finally, even if we agree with Grtinbaum
that noncontinuous spaces have not been shown, even in principle, to be as
empirically viable as continuous ones, this does not imply that the theory
that the continuity of space is conventional is gratuitous. For the continuum
used by classical physics is the real number continuum and, as Poincare
maintains, this continuum is merely one kind of continuum among others. In
other words, there are other kinds of continua than that of the real number
system, and any of these may be adequate for the construction of classical
physics*. Grtinbaums argument fails to take account of this.
(c) Models and Interpretations of Formal Systems
We saw above the distinction made by Nicod and Nagel between the
statement-forms of geometry understood as a formal system, which are neither
true nor false, and these same statement-forms
as interpreted, which are
geometrical propositions and as such either true or false. This distinction has
been used as an argument against Poincarts theory of (applied) geometrical
conventionalism but, in my opinion, it cannot be used against Poincarts
theory. In the first place, his argument in pure geometry from implicit
definitions to geometrical conventionalism has nothing to do with purely
formal systems. For example, the so-called axioms of congruence (as given by
Hilbert) combined with Euclids parallel postulate, used by Poincare to define
congruence, are not statement-forms; rather, to use Freges terminology, they
are defining characteristics of the concept equality of length. Secondly,
geometry, as a purely formal system, is not given a physical interpretation in
classical physics. In the case of three dimensional space, for instance, real
number co-ordinates are substituted for the point-variables of the formal
system, which is a mathematical and not a physical, interpretation. This is
clear from the following considerations. First, there is a rigid dichotomy
Ayer, for instance, in Language, Truth and Logic does offer us such a criterion but, as is well
known, any proposition whatsoever, according to this criterion, is indirectly verifiable.
Cf. Poincare, Science and Hypothesis, p. 29. He actually points out that the work of Du Bois
Reymond offers us an account of a continuum of higher order than the real number continuum.
Also Rogers argues that the continuum of the rational numbers may suffice for the construction of
a space adequate for classical physics (cJ Rogers, On Discrete Spaces, American Philosophical
Quurterly, (1968), 118-120).

PoincarPs Conventionalism of Applied Geometry

309

within classical physics between matter and space, and thus the nonempirical
kind of interpretation just mentioned is possible. Secondly, one of the basic
claims within classical physics is that space is continuous, and that each point
of space may be assigned a Cartesian co-ordinate. Thus the formal system of
geometry is given a non-empirical interpretation. Finally, since within classical
physics there is no causal or other empirical relation between space and matter,
one must interpret the point-variables in the above nonempirical fashion. It
would seem, therefore,
that Nicods and Nagels distinction between
geometry as a formal system and as an interpreted empirical system is
inapplicable to classical physics, and thus misses the point involved in
Poincares theory of geometrical conventionalism.
The first reason mentioned above, viz. that in classical physics there is a
rigid dichotomy between space and matter may be stated in a more general
way as follows. Any empirical theory may be considered, from the point of
views of syntax, to be a formalized language. From this point of view, the
distinguishing characteristics of an empirical language are given by semantical
considerations, especially by the condition that experience must decide the
truth-value of some at least of its interpreted theorems, which presupposes
that we have a material model for an empirical language. Now, if we apply
these considerations to classical physics, which in Poincares view is an
empirical language, we find the situation is not as simple as it might seem at
first. Nicod and Nagel assume that there is one and only one material model (or
fragment of reality) of which the language of classical physics speaks. But
Poincares theory of the dichotomy between space and matter implies that
this assumption does not hold. The factors which decide what the language of
classical physics actually speaks about do not determine a unique model;
rather, to use Przeleckis phrase, they determine a family of models. This
family consists of at least two distinct models, namely the mathematical
continuum, with its nonempirical
relations between its elements, and the
natural solids and their motionsSo. Hence, the formalized language of
geometry is not, contrary to Nicods and Nagels assumption, given a physical
interpretation within classical physics, and hence the continuity of the space
of classical physics is not an empirical issue.
In connection with formal systems and their interpretation, it is interesting
to compare the view of Popper in The Logic of Scientific Discovery with that
of Poincare. Popper maintains that there are two distinct ways of viewing any
8Przelecki, The Logic of Empiricul Theories, p. 18. Przelecki maintains that this seems to be
true of all empirical languages.
AS we mentioned above, once we assert that space is continuous the argument for the
geometrical conventionalism of pure geometry holds also for applied geometry. Moreover, the
fact that these two models are so different is, as we shall see later, Poincares principal reason for
distinguishing between geometrical conventionalism and the conventionalism of the principles
of mechanics.

310

Studies in History and Philosophy of Science

system of axioms, viz. either as conventions or as hypothese?.


Taken as
conventions, the axioms may be compared to algebraic equations (e.g.
x + y = 12), and when numbers are substituted for the variables in such
equations we sometimes get true, and sometimes false, propositions. For this
reason Popper calls the axioms statement-functions, and he maintains that if we
decide, with respect to some one statement-function,
to admit only such
substitutional values as turn this function into a true statement, a definite class of
admissible value-systems is defined, and we may regard the axioms in question
as implicit definitions of this c1a.s~~~.
Each class which satisfies a system of axioms
is a model of that system of axioms, and the substitution of such a model will
result in a system of analytic statements, since, as Popper points out, it will be
true by convention. According to Popper, Poincares conventionalism of applied
geometry is a particular instance of this kind of approach. In this Popper is
manifestly wrong. In the first place, Poincare explicitly maintains that it is not
always possible to view any system of axioms as either conventions or hypotheses.
He insists that, for example, Peanos postulates cannot be conventions (in his
sense of the word). This results from the fact that Poincart, unlike Popper,
imposes restrictions on the notion of an implicit definition: an axiomatic system
may be viewed as an implicit definition of a notion only if it is consistent and is the
one and only definition of the notion in questionJ3. Secondly, if as we maintained
above, Poincares implicit definition of congruence is in no way based on the
logical notion of a purely formal system it is unlikely that it is based on
Poppers statement-functions,
assuming the latter to be distinct from the
former. Finally, while Poppers conventions may define a family of models,
it cannot, unlike Poincares conventions, define the predicates indicated by the
predicate variables of the statement functions. For instance, in the elementary
case where each member of the family of models defined by a formal axiomatic
system is isomorphic with every other member, the predicates indicated by the
predicate-variables
of the system are, to use Przeleckis phrase, merely
determined up to an isomorphism, which amounts to saying that only their
structure is determined. In such a case, however, the models in question must
be finite34, and in a more complicated case, the models do not even determine
the structural properties of the predicates and hence it is impossible to say that
they define the predicates indicated by the predicate variables of the system. On
the other hand, the axioms of congruence combined with Euclids postulate
are used by Poincare to define distance which is a predicate indicated by a
predicate variable of the formalized language of Euclidean geometry.

3CJ The Logic of Scientific Discovery, p. 12.


Cf. ibid.
33Scienceand Method, pp. 151-154.
Przelecki, op. cit. p. 27.

Poincarks Conventionalism of Applied Geometry

311

(d) The Empirical Element in Geometry


So far we have argued that, for Poincare, the continuity of physical space is
a matter of convention and as such is not empirically verifiable, and that
within this continuum the choice of congruence is also conventional. Because
of this latter thesis Reichenbach claims that Poincare was an extreme conventionalist, i.e. that he held that it is impossible to make objective metrical
statements about real space35. The implications of this claim may be understood
if we consider Reichenbachs own theory of qualified geometrical conventionalism. Reichenbach agrees with Poincart that the choice of congruence is a
definitional, and not an empirical issue, but he maintains that, once this
co-ordinative definition is fixed, the geometrical structure of the physical
space relative to it is an empirical one. As he himself puts it, once the definitions
have been formulated, it is determined through objective reality alone which is
the actual geometry36. Thus Reichenbachs view of the the nature of physical
geometry emphasizes, on the one hand, the empirical character of this geometry
and, on the other, the limited, but important role of conventions in its actual
ascertainment. Poincare, according to Reichenbach, failed to recognise this
empirical character and he quotes the following passage from Science and
Hypothesis in justification of this: To sum up, whichever way we look at it, it
is impossible to discover in geometrical empiricism a rational meaning37.
Carnap and Griinbaum, however, disagree with this interpretation
of
Poincares geometrical conventionalism. According to Carnap, while Poincare
emphasized the conventional aspect of the structure of physical space in
stating that the rules of measurement should be adjusted in the light of
experimental results, he also clearly saw that, if the rules for the measurement
of length are defined, the geometrical structure of physical space is an empirical
issue38. He does not, however, produce any evidence to substantiate this claim,
and while it would seem that he is right in holding that the rules of measurement of length were, for Poincare, a matter of definition, this does not imply
that Poincare held that the question of the structure of physical space is an
empirical one relative to ones rule of measurement. Unlike Carnap, Grtinbaum
discusses in detail Reichenbachs interpretation of Poincare as an extreme, or,
if one prefers, pure geometrical conventionalisF.
He points out that
Reichenbach has taken Poincares statement quoted above out of its proper

35Cf. Reichenbach, The Philosophy of Space and Time, p. 36.


Op. cit. p. 47.
Science and Hypothesis, p. 79. Nagel (c$ op. cit., p. 261) and Weyl (cf. The Philosophy of
Mathematicsand Natural Science, D. 34) for examele. also interpret Poincart in this way.
Cf. Carnapsintroductory re&rks;o Reichenbachs; The Philosophy of Space and Time, p. 6.
J8Griinbaum,op. cit. pp. 127-131.

312

Studies in History and Philosophy of Science

context, namely Russells extreme geometrical empiricism40, and argues that


in this context, the statement does not imply that Poincart was a pure
geometrical coventionalist. On the positive side Grtinbaum quotes a long
passage from Poincarts reply to Russell as evidence for his own interpretation
of Poincare as a qualified geometrical empiricist. However, he does admit
that there are certain passages which do not fit in with this interpretation of
Poincart, notably Poincares statement in Science and Hypothesis that the
axioms of geometry are not experimental truths, and he suggests that there
is an apparent contradiction in Poincares works. In order to avoid this
contradiction he gives the following as a possible interpretation of Poincarts
position. There are practical rather than logical obstacles which frustrate the
complete elimination of perturbational
distortions,
and the resulting
vagueness as well as the finitude of the empirical data provide scope for the
exercise of a certain measure of convention in the determination of a metric
tensor43.
There is, however, no need for any such interpretation. Grtinbaum, as we
have already seen, does not take sufficient account of Poincares view of the
relationship between physical space and geometrical space, a view which
provides the solution to Griinbaums apparent contradiction. As we saw
above Poincare held for a rigid dichotomy between the relations between the
elements of the space of classical physics and the relations between material
bodies, and thus thought it possible to distinguish between the structural
relations of physical space and the metrical relations between the material
bodies of our universe. Therefore, since he held that physical space is the
mathematical continuum, he believed, on the one hand, that the question of
the geometrical structure of such a space is not an empirical issue4, and, on
the other, that the assertion (or denial) of a metrical relation between material
bodies is an empirical issue4. Thus in saying, for instance, that the distance
between London and Paris is not an absolute datum of experience, Poincart
did not mean that such statements are wholly conventional. Statements about
ORussell maintained that the ascertainment of the geometry of physical space is an immediate
in no way dependent on the prior stipulation of a co-ordinative definition. Similarly,
Gauss and Lobachewsky were also unaware of this Reichenbachian-type of qualified geometrical
empiricism.
Cf. Grtinbaum, op. cit. pp. 129-130, and Poincart, Sur les Principes de la G&rm&ie, pp. 8586.
The passage in question is the following. Should we conclude that the axioms of geometry are
experimental truths? - If geometry were an experimental science, it would be subject to continual
revision. Nay, it would from this very day be convicted of error, since we know that no rigorously
invariable solid exists. The axioms of geometry therefore are - conventions - Thus it is that the
postulates can remain rigorously true - (Science andHypothesis, pp. 49-50).
3Grtinbaum, op. cit. p. 130.
Thus he can claim that no geometry is either true or false and that the axioms of geometry are
not experimental truths.
45PoincarC, Science and Hypothesis, p. 97.

empiricalissue,

Poincarb

Conventionalism

of Applied

Geometry

313

the metrical relations between material bodies contain a conventional and an


empirical element: once the choice of congruence is made, the ascertainment
of the distance is an empirical issue46. In this sense, Poincark may be called a
qualified geometrical empiricist, but, because of the manner in which he
conceived the relationship between physical space and geometrical space, this
does not imply that the ascertainment of the geometrical structure of physical
space was for him an empirical issue. In this way the apparent contradiction,
mentioned by Grtinbaum, disappears.
2. Geometrical Conventionalism and the Conventionalism of Classical
and Special Relativity Physics
We have seen that PoincarC understood physical space to mean the
mathematical continuous space of classical mechanics, and that he combined
this with a dualistic theory of space and matter, thereby implying that the
conventionalism of pure geometry also holds good of applied geometry. We
shall now develop PoincarCs thesis of applied geometrical conventionalism by
comparing and contrasting it with the conventionalism he attributed to the
principles of mechanics, to the measurement of time, and, more generally, to
the principles of physics. We shall see that, while PoincarC maintained that
the principles of physics share the conventional character of the geometrical
postulates, he held the former are more directly based on experience47. Thus
PoincarCs applied geometrical conventionalism is not, as Popper maintained48,
simply a particular instance of the general conventionalism of the principles or
laws of science. By way of conclusion, we shall examine PoincarCs interpretation
of the impact of the special theory of relativity on his thesis of geometrical
conventionalism.
(a) Geometrical Conventionalism and the Conventionalism of the Laws of
Science
As is well known, Poincark not only defended the thesis of the conventionalism
of geometry, but also developed the more general thesis of the conventionalism
of the principles of classical mechanics and of the other principles of classical
physics4g. Despite Poppers interpretation
to the contrary,
this latter
conventionalism cannot be identified with the conventionalism of the laws of
science, i.e. with the thesis that the laws of science are conventions or implicit
definitions. Indeed PoincarC himself expressly states that the laws of science
Cf. PoincarC, Sur Les Principes de la GComttrie, p. 81. This point is also borne out by the
fact that Poincare maintained that mechanics is an empirical science. If particular distance statements are completely conventional then the concepts of velocity and acceleration, which are
essential to mechanics, are also completely conventional.
Poincart, Science and Hypothesis, p. 26.
Cf. Popper, The Logic of Scientific Discovery, pp. 78-84.
Cf. Poincart, op. cit. pp. 91-105. The classical mechanical principles in question are Newtons
laws of inertia, of the equality of action and reaction, and of acceleration.

314

Studies in History and Philosophy

of Science

are not conventions or implicit definitions, and he goes so far as to contrast his
own thesis of geometrical conventionalism
with a typical example of a
scientific law in order to bring out the differences between the two5.
For the present let us concentrate on this contrast. The typical example used
by Poincare is the law that phosphorus melts at 44 degrees centrigrade5. He
makes three points of distinction between this law and Euclids postulate which
he holds to be a genuine convention. First, he admits that both Euclids
postulate and the above law are known to be free from contradiction, and
that therefore both fulfil the mathematical formulation of the first condition
for the procedure of definition by postulates52. However, he insists on Mills
formulation of this condition as applied to the domain of material objects,
i.e. that a definition must guarantee the existence of the object defiiedS3. In the
case of phosphorus, since existence means material existence, the statement
phosphorus melts at 44 degrees centigrade, however consistent, cannot guarantee
thiss4. His second point is that if a scientist accepts the above law as a definition
of phosphorus, he is contravening the second condition for the procedure
of definition by postulates, since, in using laboratory samples to define
phosphorus, he is using two different definitions of the same symbo155.
Cf. Poincart, The Value of Science, p. 125, and Science and Method, pp. 171-176. Also
Poincare makes it abundantly clear that his conception of a scientific principle is quite different
to that of a scientific law (cJ Scienceand Hypothesis, pp. xxvi, 150-153).
He borrows the law concerning phosphorus from Le Roy, who uses it as a typical illustration of
a law which functions as a disguised definition in order to justify his own thesis of nominalism
(cJ Poincart, The Value ofScience, p. 122).
5zCJ supra, p. 8.
Cf. Poincart, Science and Method, p. 172.
5While Poincares argument here is valid, its truth depends on the identification of mathematical
existence with freedom from contradiction and, in the opinion of the logisticians, this is too narrow
a view.
Cf. Poincare, op. cit. p. 174. Poincares point here appears to be vague and inconclusive. It is
not clear whether he understands the definition by means of laboratory samples to be a particular
instance of the procedure of ostensive definition, or to indicate some other kind of procedure. In
other words, he does not explain the kind of definition involved in the use of laboratory samples,
and such an explanation is crucial to his point. It could beargued that the samples play the following
role: the label marked phosphorus on the sample is used as, what Geach calls, a name (cf.
Reference and Generality) i.e. it is used to acknowledge the presence of the thing. In addition,
phosphorus is also what Geach calls a substantival term, i.e. the expression the same phosphorus
supplies a criterion of identity or, if one wishes, the term phosphorus conveys a nominal essence
(c$ Geach, op. cit. pp. 38-42). Moreover, if one holds that a statement of the nominal essence
required to identify a specific object, e.g. phosphorus, must contain one or more of what Locke
calls secondary qualities (c$ Swinburne, Space and Time, p. 17), it is possible that, even if the
above law (viz. that phosphorus melts at 44 degrees centigrade) does not express a real essential
property of phosphorus, it may, nevertheless, figure in the nominal essence and thus be required
as a part of the definition. In this connection Poincart makes the point that the above law is open to
verification in the following sense: all bodies which possess such and such properties in finite
number (namely, the properties of phosphorus given in chemistry books with the exception of its
melting point) melt at 44 degrees centigrade. But this point is more compatible with the nominal
essence interpretation of the naming of the laboratory samples than with an ostensive definition
of phosophorus in terms of these same samples. This beingthe case, Poincare should have explained
why the law in question is excluded from the defining characteristics of phosphorus, while other
laws are, at least implicitly, included in it. This he fails to do.

Poincarks

Conventionalism

of Applied

Geometry

315

Thirdly, Poincare contrasts the reaction of a scientist who, for instance,


discovered a negative parallax of a distant star, thereby apparently showing
that a light ray does not satisfy Euclids postulate with his reaction to the
discovery that phosphorus melts at 43.9 degrees centigrade, and not at 44
degrees centigrade, as stated in the above law. In the former case he might
conclude either that a straight line is, by definition, the path of a light ray and
thus does not satisfy Euclids postulate or, on the contrary, that, since a
straight line, by definition, satisfies Euclids postulate, the path of a light ray is
not rectilinear. Similarly, in the latter case, he might conclude either that
phosphorus melts at 43.9 degrees centigrade or, on the contrary, that, since
phosphorus is, by definition, that which melts at 44 degrees centigrade, the
substance called phosphorus and which melts at 43.9 degrees centigrade is not
really phosphorus. But in fact Poincare states that the scientist would adopt
the second alternative in the first case, and the first alternative in the second56.
The reason he gives is that the scientist does not (and cannot) change the name
of a substance every time he adds or subtracts a decimal to its melting points.
This reason, however, is limited in scope and is not typical of all the laws of
science: while it is clearly applicable to laws concerning the melting points of
substances and the boiling and freezing points of liquids etc., its relevance to
other laws of science (for instance, laws which assert a relation of functional
dependence between two or more variable magnitudes associated with stated
properties or processes) is by no means apparent. It is possible that Poincare
himself recognized this limitation, for in The Value of Science he gives us a
more general account of the conventional and empirical aspects of the laws of
science, and it is to this account which we now turn5,
According to Poincare, when a law of science has received sufficient
experimental confirmation, one may adopt two attitudes towards it: one may
accept it as a law and, as such, open to future verification and revision, or
else one may elevate it into a principle, which by definition (and therefore by
convention) is not open to revision. For Poincare, however, the latter step is
not accomplished simply by stating that the law is a convention; rather, where
the original law expresses a relation between observable terms A and B, one
introduces a more or less fictitious and abstract term C. In this way one
obtains two relations: one between A and C which is assumed to be rigorous,
and this is the principle; the other between C and B which remains a law subiect

Cf.op. cit. pp. 174-176.


%I this instance Poincare is, in my opinion, correct. Usually scientists do not change the name
of a substance simply because a decimal point has been added to its melting point, and if one assumes,
as indeed Poincare does, that generally speaking classical definitions are used within the positive
sciences (Max Black, for instance, argues against this view, cJ Problems of Anulysis, pp. g-14),
then this law does not form part of the definition of phosphorus.
Cf. Poincare, The Value of Science, pp. 122-127.

316

Studies in History and Philosophy

of Science

to constant testing5. He illustrates this procedure as follows: the law the stars
obey Newtons law may be analysed into the principle gravitation obeys
Newtons law, and the experimental law gravitation is the only force acting on
the stars. In addition, he maintains that it is considerations of convenience
which decide whether or not one introduces these conventional principles into
the sciences. Hence, for Poincare, there appears to be no essential difference
between the conventional status of geometry and that of the principles of
science. Brunschvicg, for example, thus maintains that this kind of analysis of
scientific principles enables us to interpret without any fear of equivocation
Poincares remarks about the conventionalism of geometry in Poincares SW
les hypothkses fondamentales de la GkomktrieO, and Nagel makes the same
pointE . However, Brunschvicg (but not Nagel) draws attention to the
limitations, acknowledged by Poincare himself, in the analogy between the
principles of geometry and those of science, and especially those of mechanicssZ.
According to Poincare the limitations in question are due, first, to the fact
that, even though the principles of mechanics are conventional, they are
initially verified by mechanical experiments, and secondly, to the fact that,
despite the conventional nature of these principles, mechanics remains an
experimental science. The situation is different in the case of geometry: the
experiences which initially suggest Euclidean geometry are much more indirect,
consisting, for example, of physiological, kinematical and optical experiments;
and secondly, geometry itself is not an experimental science, but the study of
specific groups of transformationss3. Hence, while in mechanics the separation
of principles from laws is artificial, in the case of geometry it is necessary to
recognize that it would have been difficult not to draw this distinction that is
pretended to be artificial. And Poincare remarks that as one moves from
geometry to mechanics and from mechanics to physics, the radius of action,
so to speak, of principles diminishes, and thus in the latter two cases, unlike
the former, there is no reason either for separating the principles proper to these
sciences from the sciences themselves, or for considering these sciences to be
solely deductive.
Poincarts insight, expressed in this latter point, may be formulated in a
more technical fashion in the following way, not unlike what came to be called
the hypotheticodeductive method, and which has itself come to be the subject
of searching critiques in recent years. Generally speaking the extra-logical
5BPoincarks postulation
of the continuous
space of classical mechanics,
Which we discussed
above, fits into this schema; indeed he introduced
it in this same context.
Cf. Brunschvicg,
Henri PoincarC: Le Philosophe,
Revue de Mktaphysique et de Morale
(1913), 595. Poincarts
article may be found in Bulletin de la Sock% mathdmatiques de France

(1887), X3-216.
@CJNagel, The Structure of Science, pp. 260-261.
Cf. Brunschvicg, op. cit. p. 596.
Cf. Poincark, op. cit. p. 216.
La Valeur de la Science, p. 242; my own translation.

Poincarks Conventionalism of Applied Geometry

317

vocabulary of the empirical language of mechanics and physics may be divided


into two parts: an observational vocabulary, i.e. the predicates of the language
which are interpreted in a non-verbal (e.g., ostensive) way, and a theoretical
vocabulary, consisting of predicates which cannot be directly interpreted in
this way. For example, to use Poincares own illustration, the predicate
gravitational, while it directly applies to observable objects, does not directly
ascribe to them observable properties. However, since the interpretation of the
theoretical predicates, unlike that of the observational predicates, is not
direct, the question arises as to how exactly they are to be interpreted. Now,
Poincares point is that the languages of mechanics and physics are empirical,
and that consequently their theoretical predicates must be connected in some
way with their already directly interpreted observational predicates. According
to certain logicians this connection is effected by a set of statements called the
meaning postulates of the theoretical predicates. These postulates contain all
the theoretical predicates and all or some of the observational ones. In addition,
they must fulfil the following condition: the theoretical terms must be interpreted
in such a way that the meaning postulates be trues5. If this is so, then meaning
postulates, to use Poincares phrase, are removed from the fray, i.e. they are
not empirically verifiable, and as such conventional.
Poincares principles of mechanics and physics are meaning postulates of this
kind. But, in his opinion, while some of the geometrical predicates (for
example continuous) are theoretical, i.e. cannot be given a non-verbal
interpretation,
they are not interpreted in the above manner, i.e. their
interpretation
within classical mechanics does not connect them with
observational terms; rather their model is the mathematical continuum of the
type of the real number system as understood in pure mathematics. Hence
applied geometry, unlike mechanics, has no logical connection with observational
procedures, and so, to use Poincares terminology, its radius of action is
unlimited in this sense. In other words, while applied geometry, in so far as it
makes use of meaning postulates, shares in the general character of the
conventionalism of mechanics and physics, the kind of meaning postulate used
in applied geometry is essentially different to that of mechanics and physics.
For this reason, as we stated above, Poincares geometrical conventionalism is
not simply a particular instance of the general thesis of the conventionalism of
the principles of science.
(b) Geometrical and Chronometrical Conventionalism
In the last section we were concerned with the analogy between Poincares
geometrical conventionalism
and the conventionalism
of mechanics and
physics; in this section we shall discuss the analogy between his geometrical
and chronometrical conventionalism. For Poincare, this latter conventionalism
65Cf.Przelecki,

The Logic of Empirical Theories, p. 48.

318

Studies in History and Philosophy

of Science

was the outcome of his reflections on two problems: first, can we transform
psychological time, which is qualitative, into a quantitative time?, and,
secondly, can we reduce to one and the same measure facts which transpire
in different worlds?66
The manner in which Poincare poses these two problems is admittedly
psychological in tone and partially results from his belief that psychological
time is a datum of each individual consciousness. Nevertheless, under the first
problem he discusses the question of the equality of two intervals of physical
time, and under the second, the question of the simultaneity of events occurring
at different places, i.e. simultaneity at a distance. In connection with the former
question, he argues that psychological time is discontinuous, whereas physical
time is mathematically continuous. Hence, he implies that the continuum of
physical time, like that of physical space, is metrically amorphous, i.e. that it
lacks an intrinsic metric, and that therefore there is no unique standard of
equality of length imposed upon us by the nature of this continuum. Thus,
for instance, he says that we have not a direct intuition of the equality of
two intervals of time. However, unlike the case of geometry, there is no
distinction between pure and applied time; hence Poincare proceeds to discuss
the problem of the physical determination of the equality of two intervals of
time. In this discussion he points out that certain difficulties arises8 and, to use
a phrase coined by Putnam, he concludes that the concept of physical time is a
law-cluster or a multiple-criterion concepts, i.e. that there is a multiplicity
of compatible physical criteria, rather than one single criterion, by which
physical time can be measured7. He maintains, however, that the choice of
these criteria is governed by considerations of convenience, and not of truth;
in other words, unlike Putnam, Poincare considered the multiple-criterion
nature of the concept of physical time to be irrelevant to the question of its
conventional or empirical status.
Poincare is not as explicit about the physical criteria used in the measurement
of distance, but it is reasonable to assume that he also considered the concept

Poincare, The Value of Science, (Chapter two), p. 27. This chapter consists of his article La
Mesure du Temps, Revue deM&aphysique et de Morale (1898), l-13.
Cf. Poincare, op. cit. p. 26.
He does not claim originality in pointing out these difficulties, but attributes them to Calinons
&tudesur les diverses grandeurs, and Andrades Lefons de Mkhaniquephysique.
Cf. Putnam, The Analytic and Synthetic in Minnesota Studies in the Philosophy ofScience,
III, pp. 376-381.
Cf. Poincare, op. cit. p. 30.
In this connection we may say that Poincare would agree with the following remark made by
Griinbaum . ..much as attention to the multiple-criterion character of concepts in physics may be
philosophically salutary in other contexts, it constitutes an intrusion of a pedantic irrelevancy in the
consideration of the consequences of alternative metrizability (of physical space and time)
(Griinbaum, Philosophical Problems of Space and Time, p. 15).

Poincares Conventionalism of Applied Geometry

319

of physical distance to be a multiple-criterion one. Thus, entirely apart from


the distinction between pure and applied geometry and its consequences, it is
reasonable to say that Poincares thesis of the conventionalism of the measurement of the equality of two spatial intervals is analogous to the conventionalism
of the measurement of the equality of two temporal intervals.
However, Poincare saw that not all the problems concerning physical time
are solved with the solution of the measurement of the equality of two intervals.
In his opinion, the logically prior and neglected problem of simultaneity
remains unsolved, i.e. the problem of the meaning of the statement that two
physical events which occur at a distance are simultaneous7. He held that we
have not a direct intuition of simultaneity at a distance; this is a matter of
definition or convention.
However, though according to Poincare the
conventionalism of simultaneity at a distance is, not only in this sense, but also
fully analogous to the conventionalism of the measurement of the equality of
two temporal intervah?,
the former, unlike the latter, is not based on
considerations of the metrical amorphousness of the continuum of time and
so, it could be argued, is not entirely analogous to the latter. More specifically,
while in the case of geometrical conventionalism the possible choices of
congruence standards are indicated by the function d(x,y), there is no such
indication of the possible standards of simultaneity. Poincares point, however,
is that there is no inherent property in the continuum of time from which one
may abstract the concept of simultaneity at a distance and for this reason it is
analogous to the concept of equality of time. Moreover, we have no direct
intuition of either concept, and there is nothing in the nature of events in the
material world which imposes any set of criteria of either concept upon us.
Finally, both concepts are multiplecriterion concepts given by the application of
certain rules. Thus, according to Poincare, simultaneity at a distance, like the
equality of spatial and temporal intervals, is neither an absolute datum of the
mathematical continuum, nor of experience, and therefore the ascertainment
of the simultaneity of spatially separate events is not an immediate empirical
issue but depends on our prior conventions. However, once these conventions
are fixed the issue is an empirical one relative to them75. But, according to
Poincare, this is alsohe case with geometry. Hence his geometrical conventionalism is analogous to his conventionalism of simultaneity.

Thus, for example, at times Poincare mentions intervals given by the coincidence behaviour of
unperturbed transported solid rods as a criterion of spatial congruence, while at other times he
mentions intervals for which light rays require equal transit times.
Cf. Poincare, The Value of Science, pp. 30-32.
Cf. op. cit. p. 30.
However, if we examine more closely Poincares conventionalism of simultaneity, we shall
find indications of standards of simultaneity (cj. Poincare, op. cit. pp. 32-35). though these have
not as concise a form as those indicated by the distance function d(x,y).

320

Studies in History and Philosophy

of Science

(c) Geometrical Conventionalism and the Special Theory of Relativity


The difficulty raised by the special theory of relativity against Poincares
geometrical conventionalism was stated succintly by Poincart himself: will
not the principle of relativity, as conceived by Lorentz, impose upon us an
entirely new conception of space and time and thus force us to abandon some
conclusions which might seem to have been established? Have we not said that
geometry was devised by the mind as a result of experience, but without having
been imposed upon us by experience, so that, once constituted, it is secure
from all revision and beyond the reach of new assaults from experience? And
yet do not the experiments on which the new mechanics is based seem to have
shaken it?76 To this question Poincare himself answered in the negative. His
conventionalism, he held, was not affected by the special theory of relativity,
since, in his opinion, scientists are not constrained by reality to adopt the
conventions of this theory. Rather, just as scientists in the past found the
conventions of classical mechanics to be the most convenient, so now other
scientists find the new conventions of the special theory of relativity to be the
most convenienF7.
To see Poincarts reasons for this, let us first consider his view of the
relationship between space and time in the special theory of relativity. According
to Poincare an essential element of this theory is that in it space and time are
no longer two entirely distinct entities which can be considered apart, but two
parts of the same whole (space-time), two parts which are so closely knit that
they cannot be easily separated. Thus, as Poincare himself admits, if one
attributes an ontological or physical status to space-time one must maintain
that the connection between, or the unity of, space and time is not merely
conventional, but is, as Dingle puts it, an association that is more fundamental
than a mere ad hoc union to be recognized in some physical problems but
ePoincarC, Last &says, p, 15. In the above quotation Poincare mentions Lorentzs, and not
Einsteins, theory of relativity and, indeed, he continues this practice throughout his works. Thus
he appears not to credit Einstein with the discovery of the special theory of relativity. However,
there is no doubt that he was aware of Einsteins work (cf. M. Born, Physics in my Generation,
p. 192). Also some commentators group together Poincares, Lorentzs and Einsteins works on
relativity, and consider them to be intimately connected (cJ Keswani, The Origin and Concept of
Relativity I and II, The British Joumalfor Philosophy of Science, 15 (1965) 286306 and 16 (1966),
19-32), whereas others maintain that they are quite distinct (cJ H. Dingle, Note on Mr. Keswanis
Article, The Origin and Concept of Relativity, The British Journalfor the Philosophy of Science,
16 (1965), 242-246). From our point of view, viz. the conventionalism of geometry and of the
measurement of time, we can group Lorentzs and Einsteins theories together, and discuss their
implications vis d vis Poincarts thesis of geometrical conventionalism, and thus we shall simply
speak of the special theory of relativity.
Cf. Poincart, op. cit. p. 24.
81bid. This point is also noted by other commentators. For example, Earman maintains that
space-time is the basic spatio-temporal entity (cJ Space-time, The Journal of Philosophy
(1970), 259). Similarly Russell maintains that events, and not particles, must be the stuff of physics
and that each event has to each other a relation called interval which could be analysed in various
ways into a time-element and a space-element (cf. History of Western Philosophy, pp. 860-861).

PoincarPs Conventionalism

of Applied

Geometry

321

abandoned in others if it becomes inconvenientv. From this point of view,


Earman argues that Poincares geometry of space, on which he grounds his
geometrical conventionalism,
is taken out of its proper context, namely
space-time, whereas this question should be seen relative to the slicing off of
space from space-time, which contrary to Poincarts assumption, can be
accomplished in not just one, but in many, ways.
Poincare, however, did not agree that the space-time of the special theory
of relativity is a basic physical entity, in the sense that it is empirically or
otherwise necessary to view nature in this way, but regarded this unity merely
as a convention. He does not expressly state his reasons for this view, but these
reasons, I believe, are analogous to those he gives for the conventional nature
of geometry in classical mechanics, viz. the manner in which the concepts of
space and geometry are used in it. Thus he would agree, for example, with
Dingle that an examination of the way in which the concept of space-time has
appeared in physics shows that the implied association between space and time
has been chosen for a definite limited purpose, namely the derivation of the
laws of motion. Dingle illustrates this in a manner which would, in all
probability, be acceptable to Poincare. He points out that, in considering the
problem of, say, the derivation of the laws of radiation of energy by hot bodies,
one may choose to associate energy with time and to speak of energy-time in
connection with this problem, in exactly the same way as one chooses to
speak of space-time in connection with the problem of motion*. Poincare
goes further than Dingle in pointing out the limitation of the physicists
choice of space-time. He maintains that, even though space-time is used by
scientists to express the laws of motion, these same scientists still continue to
use classical mechanics, and thus space and time separately, in their investigations of the motions of terrestrial and other bodies whose velocities are
small relative to the velocity of lightaz. Moreover, when scientists do use spacetime in physics, they continue to maintain that space-time is mathematically
continuous, and therefore does not possess an intrinsic metric, which implies
(though Poincare does not expressly say so) that geometrical conventionalism
is not invalidated by the introduction of space-time into physics.
Cf. Dingle, The Philosophical Significance of Space-Time, The Proceedings of the
Aristotelian Society, (1947-Q, 155.
OCJ Earman, op. cit. pp. 261-262.
Cf. Dingle, op. cit. p. 156.
Cf. Poincart, The Principles of Mathematical Physics, The Monist (l!IOS), 24. While this
article, which consists of an address delivered by Poincare before the International Congress of
Arts and Science in St. Louis, in September 1904, was written before the publication of Einsteins
work, it, nevertheless, contains in germ some of Einsteins principles. Thus it is relevant to the
above discussion. Moreover, Poincarts point above is correct, as may be seen from an examination
of most texts on the special theory of relativity. For instance, Landau and Lifshitz point out that
in the limiting case when the velocities of moving bodies are small compared with the velocity of
light - relativistic mechanics goes over into the usual mechanics - which is called Newtonian or
classical (The Classical Theory of Fields p. 2).

322

Studies in History and Philosophy of Science

Another characteristic of the special theory of relativity, according to


Poincare, is the importance it attributes to the principle of the finiteness of the
velocity of the propagation of interactions for the notion of metrical time,
the maximum velocity being that of light. While a detailed discussion of the
conventionalism of time is beyond the scope of this paper, it is both useful and
interesting to note the analogy between the chronometrical conventionalism of
the special theory of relativity and Poincares geometrical conventionalism. As
we mentioned above, in classical mechanics there is no functional or algebraic
way of indicating the possible choices of standards of simultaneity and in this
respect the conventionalism of geometry differs from that of time. In the
special theory of relativity, however, there is such a wag4. For this reason
Grtinbaum, for instance, maintains that Poincarts geometrical conventionalism
is fully analogous to Einsteins conventionalism of simultaneitf5, though in
all probability, Poincare himself would maintain that there is no essential
difference between the conventionalism
of simultaneity in classical and
relativity mechanicsa6.
Poincare finally considers the importance of the Lorentz transformations of
the special theory of relativity, with particular reference to their bearing on his
geometrical conventionalism.
In classical mechanics rigid bodies undergo
Galilean transformations, which is compatible with the claim that they move
(in any system whatsoever) approximately according to the structure of the
Euclidean group of transformations used by Poincart to define spaces7. In the
special theory of relativity, however, the motions of solid bodies are governed
by the Lorentz transformations which are incompatible with the Euclidean.
Hence, it would seem, the classical notion of space as defined by Poincare
must be changed in the light of the special theory of relativity. Thus, it would
appear, Poincares thesis of the conventionalism of geometry is false, since it
implies that the geometry of space should not be changed for any empirical
considerations.
Cf.Last Essay, pp. 23-24.
This functional indication of simultaneity may be explained as follows: let us consider an event
E, namely the departure of a light ray from a point A at a time t, measured by a clock at A, and
let us suppose that this light ray is reflected from a point B (an event we shall call E,) and that it
returns to A at a time Is measured by the same clock at A (an event we shall call E,). Finally, let us
consider any event E, at A between the times t, and t,. The problem is how can an observer at A
know whether E, is simultaneous or not with E,. This is not an empirical issue, since E, and E,
are not connectible by interacting velocities less than, or equal to, the velocity of light; rather, as
Einstein says, it is a matter of definition (CA The Principles ofRelativity, p. 40). Its answer depends
on how one synchronizes the clocks at A and B and, as Reichenbach clearly points out (c$ The
Philosophy of Space and Time, pp. 126-127), this may be done in an infinite number of ways.
The time t, assigned to the event E2 is indicated by the function t2 = t, + E(1,- t,), such that E is
greater than zero and less than one. This function also indicates the possible choices of simultaneity.
In other words, depending on the scientists choice of .r, E2 is said to be simultaneous or not with E,.
Cf. Grttnbaum, Philosophical Problems of Space and Time, p. 28.
Cf.supra p. 17.
8 Cf. Poincare, op. cit. p. 25.

Poincarks Conventionalism

of Applied

Geometry

323

According to Poincare this objection does not hold*. First he says, it takes
the concept of space in a different, and more limited, sense than the sense it
has in his argument for geometrical conventionalism. In the objection space is
considered solely from the point of view of the principle of relativity, according
to which the laws of mechanics (which, in Poincares view are expressed by
means of differential equations) are identical in all inertial frames of references,
whereas in Poincares argument the concept of space is not limited to the
mechanics of inertial frames of reference. In other words the notion of space
defined by the group of Lorentz transformations, which is assumed to be the
correct notion of space in the above objection, is limited to considerations of
the invariance of the laws of mechanics vis d vis the Einstein principle of
relativity. But, if we look on space from the broader perspective of the
mathematical continuum, which is assumed both in the special theory of
relativity and in the other positive sciences, we see that the notion of space
defined by the group of Lorentz transformations is only one of the possible
metrical spaces compatible with the metrical amorphousness of this continuum.
Moreover, even within the limited point of view of the principle of relativity,
the above objection, according to Poincart, can be shown to be invalid. It
assumes that the metrical spaces of classical mechanics and of the special
theory of relativity are defined respectively by the groups of Galilean and
Lorentz transformations. But, in the first place, there is no essential difference
between these definitions, and secondly, these definitions result respectively
from the Galileo and Einstein principles of relativity understood as conventional
postulates, and not as experimental truths. Poincares argument for the first
point may be summed up as follows. In classical mechanics we adopt the
convention that two figures are equal if the same solid body can be superimposed
on these figures such that it coincides first with one and then with the other. But
the solid body in question may be considered to be a mechancial system in
equilibrium under the influence of the various forces acting on its constituent
molecules, in which case the above convention is equivalent to an agreement
that the laws describing the equilibrium of the mechanical system of molecules
constituting the solid body remain invariant in all inertial systems. On the
other hand, in the special theory of relativity we agree to call two figures equal
if any mechancial system is placed in such a way that it coincides with these
figures in any inertial system, i.e. we agree that the laws describing the equili-

Cf. Last Essays, pp. 15, 18-22.


Cf . op . cit. pp. 22, 23. One must distinguish, as indeed PoincarC himself does, between this
principle of relativity and what may be called the Einstein principle of relativity, which consists
of the former principle combined with the principle of the finiteness of velocity of propagation of
interactions, and also between it and what may be called the Galileo principle of relativity of
classical mechanics, viz. the principle of relativity as explained above combined with the principle
of the infinity of the velocity of propagation of mechanical interactions.

324

Studies in History and Philosophy of Science

brium of any mechanical system are invariant in all inertial systemsgo. But this
latter convention is merely a more general version of the former and hence there
is no essential difference between the definitions of space in terms of the
Galilean and Lorentz transformations.
His second point, viz. that the
different definitions result from the Galileo and Einstein principles taken as
conventional principles, is more complex. As we saw above, Poincare held that,
if a scientist decides to elevate a corroborated general empirical proposition to
the status of a conventional principle, he must do so by transforming it into a
conventional principle and an experimental law. Thus the Einstein principle of
relativity can be transformed into the conventional principle that the differential
equations of dynamics satisfy the Lorentz group of transformations,
and
experimental laws among which we have the law that the velocities of
propagation of mechanical interactions are finite. Now, according to Poincare,
it is the conventional principle (which includes the conventional principle of
relativity common to the Galileo and Einstein principles) that enables the
relativity physicist to define metrical space in terms of the Lorentz group of
transformations.
Hence the definition of metrical space in terms of the
Lorentz group of transformations results from the conventional, and not the
experimental, dimension of the special theory of relativity.
Against this, it could be argued that the definition of metrical space cannot
be isolated from the context of the special theory of relativity, and that in this
context it is imposed by the total empirical theory, and as such is not
conventional. In other words, since Einsteins law of relativity, unlike the
classical law, is an experimental truth, and since the definition of metrical
space in terms of the Lorentz group of transformations is associated with the
former, and not the latter, this definition of space is based on empirical, as
well as conventional, grounds. In Poincarts view, however, this objection does
not hold. First, since the dualism of metrical space and matter holds both in
the special theory of relativity and in classical mechanics, the definition of
metrical space in both cases is, logically speaking, prior to empirical
investigations, and thus cannot be influenced by the empirical results. It is true,
of course, that the special theory of relativity and classical mechanics are
mutually inconsistent and that the former is better corroborated on empirical
grounds than the latter, but this implies merely that the special theory of
relativity as a whole is, in some sense, more probable than classical mechanics;
it does not imply that the definition of metrical space in terms of the Lorentz
Cf. Poincare, op. cit. p. 22.
BLandau and Lifshitz maintain that the finiteness of the velocity of propagation of mechanical
interactions is an empirical truth (cJ The CIossicul Theory of Fields, p. #I).
It is worth noting, as Giedymin points out, that Poincare, in opposition to Le Roy, denied the
truth of the incommensurability thesis in general (cJ Giedymin, Logical Comparability and
Conceptual Disparity between Newtonian and Relativistic Mechanics, The British Journal for the
Philosophy of Science, (1973), 271).

Poincarks Conventionalism of Applied Geometry

325

transformations
is more probable than that used in classical mechanics.
Secondly, despite the fact that the law of the special theory of relativity,
(namely, that the velocities of mechanical interactions are finite) is better
corroborated than the laws of classical mechanics, nevertheless, if one takes into
account Einsteins assumption that the maximum velocity of interaction is
equal to the velocity of light in vacua then one can maintain that the choice
between the definitions of metrical space in terms of the Galilean or of the
Lorentz transformations
is solely a matter of convenience. For Einsteins
assumption contains a conventional element, which can be highlighted by
Reichenbachs view of simultaneity mentioned above, namely that the
constancy of the velocity of light in vacua depends on ones choice of e in the
equation
t2 = t, + E&-t,)

and that this choice is a matter of conventionB3. It follows that, for certain
purposes, the velocity of light in vacua may be taken to exceed its normal value.
For this reason it is legitimate to interpret the velocity of light in the Lorentz
transformations
as tending towards infinity, in which case the Lorentz
transformations degenerate into those of GalileoB. Thus is would seem that the
choice of either set of transformations
as a basis for ones definition of
metrical space is a matter of convenience.
3. The Retention of Euclid
Poincare held that the scientist, in his empirical descriptions of the physical
world, could retain Euclidean geometry in the face of apparently adverse
experimental evidence. Poincart, however, went further than this: he held that
the scientist not only could, but should, retain Euclid in such an hypothesis.
We shall now examine his reasons for holding this, and his view of the required
modification of physics necessary for its accomplishment. As regard the first
point (the retention of Euclid), it is necessary to distinguish between Poincares
view prior to, and following upon, the development of the special theory of
relativity. In his earlier works he maintained that scientists always do, as a
matter of fact, find it more convenient to retain Euclid, but in his later
works he conceded that scientists might find it more convenient to adopt some
other geometry, but only for specialised purposes. As regards the kind of
change he thought necessary in physics for the retention of Euclid in the
hypothesis of adverse experimental evidence, we shall see that Poincart did not,
contrary to one widely held opinion, conceive this change as the introduction
of ad hoc hypotheses, but simply as a change in the (geometrical) language of
physics. Such a change, however, is not simply a matter of semantics, but is
Cf. supra note 84.
Cf. Landau and Lifshitz op. cit. p. 13.

326

Studies in History and Philosophy

of Science

based on the metrically amorphousness of space, and on the peculiar kind of


relationship he held to exist between geometry and mechanics mentioned
aboves5.
(a) The Reasons for Retaining Euclid
In the passage from Science and Hypothesis quoted above Poincart states
that, if the scientist were to discover a negative parallax for a distant star, he
could choose either to give up Euclidean for Riemannian geometry, or retain
Euclid by modifying the laws of optics. In both Science and Hypothesis and
Science and Method he states that this choice is a matter of convenience, and
that it is more convenient to retain Euclid, even at the expense of modifying
the laws of physics.
In Science and Hypothesis he gives two reasons for this, viz. that Euclidean
geometry is the simplest in itself, just as a polynomial of the first degree is
simpler than a polynomial of the second degree, and that it is the best
approximation to those properties of natural solids, which we can compare
and measure by means of our senses96. As regards the first reason, Poincare
has been accused of giving too much weight to the analytical simplicity of
Euclidean geometry, and of neglecting the physics in which that geometry is
usedg7. Hence an argument frequently used against Poincare here is the overall
simplicity of the general theory of relativity, which demands the use of
Riemannian, rather than of Euclidean geometry. This however, cannot be
used as an argument against Poincare. As we saw above, Poincare held a
dualistic theory of space and matter in classical mechanics, in the sense that
there is no empirical relationship between them, and therefore his choice of
Euclidean geometry on the grounds that it is analytically simpler than the other
metrical geometries does not, in this particular case, conflict with the overall
simplicity of classical mechanics. In other words, Poincart could argue that,
since space and matter are empirically independent, the question of the overall
simplicity of classical physics entails two independent questions, viz. the
question of which metrical geometry is the simplest and the question of which
formulation of the physical laws is the simplest. From this point of view the
reference to the general theory of relativity as a counter argument to Poincares
argument for retaining Euclid is irrelevant, since in this latter theory the
dualism of space and matter no longer holds, and hence it is impossible to
divide the overall simplicity of this theory into the simplicity of independent
constituent element?.
As regards his second reason, viz. that the Euclidean
Cf.SlqxYlpp. 34.
g6Scienceand Hypothesis, p. 50.
Cf. Grtinbaum, op. cit. pp. 21-22, 121.
The dualism of space and matter of classical mechanics no longer holds in the general theory
of relativity, since in this latter theory it is legitimate to enquire about the influence of a gravitational
field upon the metrical properties of space. Also one should note that Poincare died before the
development of this theory and he had not foreseen this kind of development.

Poincarb

Conventionalism

of Applied Geometry

327

metric is the best approximation to the natural solids used in measurements,


Poincare here is clearly correct. We do continue to use rulers, chains etc. in
our ordinary terrestrial measurements of length, and we do retain the
Euclidean metric in accomplishing these measurement@.
In Science and Method Poincare gives a more detailed account of the choices
open to the physicist in the event of the discovery of negative parallaxesoo. In
such an event the physicist, according to Poincare, may choose between the
following two positions: a straight line is, by definition, the path of a light ray,
and therefore a straight line does not satisfy Euclids postulate; or a straight
line is, by definition, that which satisfies Euclids postulate, and therefore the
path of a light ray is not a straight line. Poincart maintained that it would be
foolish, though not false, to opt for the former alternative. He gives
three reasons for this. He argues, first, that a light ray probably satisfies in a
most imperfect way not only Euclids postulate but the other properties of the
straight line. Poincare here implies that the choice of the Euclidean geodesics
as straight lines is a matter of definition and, all things considered, is the best
definition O.
Poincares second reason is more difficult to interpret. He states simply that
it would be foolish to adopt the path of a light ray as the definition of a straight
line, because it not only deviates from the Euclidean straight but also from the
axis of rotation of solid bodies which is another imperfect image of the straight
line. As I see it Poincare is arguing here that the physicist has a choice between
the Euclidean definition of a straight line and its definition as a cluster or
multiple criterion concept, and that, if he chooses the second definition, the
multiple criteria in question should render approximately the same results02.
But in fact this is not the case, since the paths of a light ray and some other
criteria of a straight line (e.g. the axis of rotation of a solid body ) do not give
the same resultio3. Against this, however, it could be argued that, while
Poincare is correct in noting the possibility of what may be called a range
definitionO of the concept straight line as a cluster concept, he fails to note
that not all the characteristics of a cluster concept have equal importance in the
formulation of a range definition. It is possible (at least in certain cases, and
We shall discuss Einsteins interpretation of this fact later.
Cf. ScienceandMethod, pp. 175-176.
Cf. ibid. and The Vatueof Science, pp. 125-126. In The Valueof Science he also adds that the
elaboration of relationships between straight lines defined as paths of rays would be extremely
complex, whereas, if one accepts Euclids definition, such an elaboration is already accomplished
in Euclidean geometry. He also argues that the functional description of the Euclidean goedesics is
analytically simpler than that of the geodesics of the other metrical geometries.
Cf. The Value of Science, pp. 27-30,36.
One reason for this lack of approximation may be seen from Poincarts third reason for
choosing the Euclidean definition of a straight line, viz,. that the path of a light ray is subject to
change in passing through media of varying densities.
OI am borrowing the expression range definition from Black (cJ Problems of Analysis,
p. 13), even though I am not using it in exactly his sense.

328

Studies in History and Philosophy of Science

especially in the case of the concepts of distance and straight lines), to distinguish
between what may be called the primary and secondary characteristics of
cluster concepts05. The primary characteristics give the meaning of the cluster
concept: they constitute a set of criteria whose application always result in the
same value. The secondary characteristics, however, are used only because it is
assumed that they give the same result as the primary characteristics. Now in
the case of a straight line the primary characteristics are given by the
shortest-distance and the three-surface criteria. The path of a light ray is
merely a secondary criterion, since, as Poincare himself notes, it changes due
to the influence of media of varying densities. Hence, while Poincare is correct
in rejecting the possibility of defining a straight line in terms of the path of a
light ray because the results given by this criterion do not coincide with the
results given by others, this is not a legitimate reason for opting for Euclids
purely mathematical definition of a straight line: one could equally opt for
some other possible definition in terms of the primary characteristics mentioned
above. Indeed, it has been argued that such a definition is the only correct
definition of our ordinary concept of a straight linelO.
Unlike his second reason, PoincarCs third reason for rejecting the possibility
of defining a straight line in terms of a path of a light ray presents no difficulty.
He points out simply that the path of a light ray is subject to change, so that a
line which was straight yesterday will no longer be so tomorrow, if some
physical circumstance has altered08. Here Poincare is certainly correct, since
in fact the path of a light ray changes if the ray travels through media of
varying densities, and (if one takes the results of the general theory of relativity
into account) may be deflected by a gravitational field. Hence, if, as is assumed
by Poincart, the choice is simply between a pure mathematical definition and a
definition in terms of the path of a light ray, it is clearly more convenient to
opt for the former definition.
In the above discussion no mention was made of the special theory of
relativity and its influence on Poincares conception of the physicists choice
of geometry. Poincare admitted that certain physicists might find it more
convenient to define space in terms of the Lorentz, rather than the Euclidean,
group of transformations, and he actually mentions what might be considered
to be an advantage of the former over the latter definition, viz. that it not only
defines space but also timeOg. He states, however, that the choice of the
Lorentz group of transformations is still a matter of convenience, and that, for
Y3winburne, (cJ Space and Time, p. 71), uses the terminology of primary and secondary
tests of properties. We have altered this to primary and secondary characteristics of cluster
concepts.
losA detailed explanation of these two criteria may be found in Swinburne, op. cit. pp. 72-75.
Cf. Swinburne, op. cit. pp. 95-97. We shall discuss this later.
08Poincark, ibid.
Cf. Last Essays,pp. 22-24.

Poincarks Conventionalism

of Applied

Geometry

329

ordinary purposes, it is more convenient to use the Euclidean definition. In


my opinion, Poincare is correct in this. In certain areas the Lorentz contractions
of the special theory of relativity must be taken into account, but for the
scientists ordinary purposes of describing events close to, and on, the earth he
continues to use Euclidean geometry since he finds it more convenient.
(b) The Retention of Euclid and Ad Hoc Hypotheses
We now turn to Poincares view of the kind of change required in physics in
order to retain Euclidean geometry in the face of prima facie adverse experimental evidence. As we saw above, Poincart states that, in such an hypothesis,
the scientist may retain Euclidean geometry if he modifies the laws of optics.
Poincart himself does not here specify the kind of modification required, but
certain commentators claim that what he had in mind was the introduction into
optics of ad hoc hypotheses, thus implying that, for Poincare, the relationship
between geometry and physics was in some way empirical. Such is the opinion,
for example, of Nagel, who equates the ad hoc hypotheses in question with the
universal forces of Reichenbach. Similarly Lucas maintains that what
Poincare meant by modifying the laws of physics was the introduction into
physics of various gravitational and other forces at the expense of its empirical
content *.
The passage from Poincare most often quoted in favour of this interpretation
is that from Science and Hypothesis in which he describes the efforts of
imaginary two-dimensional intelligent beings living in a Suclidean sphere to
ascertain the geometry of their imaginary world3. Poincare assumes that the
following conditions hold of this world: (a) the beings in question use rigid rods
to ascertain the geometry of their world; (b) the world if subject to a temperature
gradient such that, if R is the radius of the sphere, and r the distance of any
given point from the centre of the sphere, the absolute temperature of such a
point is proportional to R-ti.2; (c) any body transported from one point to
If therefore, we were to discover negative parallaxes, or to prove that all parallaxes are higher
than a certain limit, we should have a choice between two conclusions: we could give up
Euclidean geometry, or modify the laws of optics, and suppose that light is not rigorously
propagated in a straight line.
Science and Hypothesis, p. 73.
Cf. Nagel, The Structure of Science, pp. 262-265. Reichenbach distinguishes between universal
forces, these being forces which affect all material bodies in the same way and against which
there are no insulating walls, and differential forces, these being forces which affect different
materials in different ways and from which bodies can be protected by insulating walls (cJ
Reichenbach, The Philosophy of Space and Time, p. 13). Nagel instances gravitation in Newtonian
mechanics as an example of a universal force.
CJ Lucas, Euclides ab omni naevo vindicates, The British Journal for the Philosophy of
Science, (1969), 2. Also it should be noted that Nagel and Lucas are not the only philosophers of
science holding this opinion; there are many others, among whom we find Robertson (cf. Geometry
as a Branch of Physics, in: J. J. C. Smarts (ed.) Problems of Space and Time, p. 239) and Sklar
(cJ The Conventionality of Geometry, in: Studies in the Philosophy of Science, pp. 50-52).
Cf. Science and Hypothesis, pp. 64-68.

330

Studies in History and Philosophy

of Science

another is instantaneously in thermal equilibrium with its new environment;


(d) light rays pass through media of different refractive indices such that
the index of refraction is inversely proportional to R*-r2. Poincare argues,
given these conditions, that the rigid rods used as measuring instruments
would become colder, and therefore smaller, as they approached the surface
of the sphere, and that therefore this world, though finite to us, would appear
infinite to its inhabitants, who, therefore, would conclude that the geometry of
their world was Euclidean. Now, according to Nagel, the second and third
conditions mentioned above presuppose universal forces, and for this reason
he interprets the modification of physics referred to by Poincart to mean the
introduction of such forces as ad hoc hypotheses14.
However, Nagel and other commentators exaggerate the importance of this
passage from Poincares works - due most probably to the influence of
Reichenbachs interpretation
of geometrical conventionalism in terms of
universal forcesli5. This passage*is taken from chapter four of Science and
Hypothesis which is practically identical with Poincares article LEspace et
la GkomCtrie. But in this article Poincare is primarily concerned with the
possibility of imagining a non-Euclidean world and not with geometrical
conventionalism6, and it is in this context that he introduces his imaginary
world: his intention was to show that intelligent beings living in such a world,
with senses similar to ours, would have no more difficulty in imagining a
non-Euclidean world than we have in imagining a Euclidean one.
Moreover, the above interpretation of Poincart, which is based chiefly on
the passage from Science and Hypothesis, is incompatible with other equally
important passages from Poincares works. For example, in The Value of
Science he says:
In space we know rectilinear triangles the sum of whose angles is equal to two right
angles; but equally we know curvilinear triangles the sum of whose angles is less than
Even Grtinbaum, despite his own rejection of this kind of interpretation of Poincare, holds
that the second condition presupposes a universal force (c$ Grtinbaum, The Philosophy of Space
and Time, p. 85-86). Moreover, the remarkable similarity between Poincarts fictitious world and
Reichenbachs use of universal forces in his own fictitious hemisphere (cJ Reichenbach, op. cit.
pp. 11-12) lends additional weight to this interpretation. Also Sklar maintains that Poincart
introduced this parable to convince us that one may retain Euclid despite prima facie adverse
experimental evidence (cJ. Sklar, op. cit. p. 50). However, he appears to disagree with Nagels
interpretation in terms of universal forces in maintaining that one should speak of universal
stretching, and optical medium distorting fields. This difference, however, is, in my opinion,
linguistic.
Y agree with Grtinbaum (cJ op. cit. pp. 82-89) that one should distinguish between literal and
metaphorical uses of universal forces in Reichenbachs account of geometrical conventionalism.
Reichenbach uses these forces in a literal sense merely to suggest his thesis of the conventionalism
of congruence (cf. Reichenbach op. cit. pp. 1l-13), whereas he uses them in a metaphorical sense
only in his actual establishment of this thesis.
Wtis point is much more explicit in Poincarts introduction to this article than in his introduction
to chapter four of Science and Hypothesis (cf. LEspace et la C%omCtrie, Revue de M&aphysique
etdeMorule(1895),631).
Cf. Science and Hypothesis, pp. 64-65.

Poincarks Conventionalism

of Applied Geometry

331

two right angles. The existence of one is not more doubtful than the other. To give
the name of straights to the sides of the fist is to adopt Euclidean geometry; to

give the name of straights to the sides of the latter is to adopt the non-Euclidean
geometry. Therefore to ask which geometry is correct, is to ask, to what line is it
proper to give the name straight.
Here Poincare sees the choice between Euclidean and non-Euclidean geometry
solely as a matter of which line the scientist calls straight, and this, as Poincare
points out, is not an empirical issue. In the light of this passage it is clear that,
for Poincare, the retention of Euclid in the face of prima facie adverse
experimental evidence, is accomplished simply by a process of renaming, and
the introduction of ad hoc hypotheses or universal forces is in no way
necessary for this. This is also clear in the analogy Poincare draws between
the question of the truth of geometry and the truth of other questions which
can be decided only on purely linguistic grounds. For instance, he says that
the question of the truth of geometry is analogous to the question of whether
the Continental or the English system of measurement is true, or whether
Cartesian or polar co-ordinates are true18. Similarly, when discussing the
opinion that certain phenomena are possible in Euclidean space and impossible
in a non-Euclidean space, he says that this opinion is entirely analogous
to the opinion that there are lengths which may be expressed in metres but not
in fathoms, feet or incheslzo. Now it is clear that in changing from one unit of
measurement to another the scientist does not introduce universal forces, or
any other ad hoc hypothesis, and since Poincare believed that this kind of
change is completely analogous to the change in physics from one metrical
geometry to another, there can be no question for Poincart of any necessity to
have recourse to such devices in order to retain Euclid in the face of adverse
experimental evidence. This can be done simply by a process of renaming.
There is one other passage in which Poincare is admirably clear on this point.
When discussing the claim that dynamics is impossible without geometry he
says:
It is quite clear that, if one chooses Lobachewskian in place of Euclidean geometry,
one would be obliged to change ones statements of the laws of dynamics. Likewise,
the statement of these laws is not the same in so far as one uses English rather than
French...However, if a Frenchman, having learned dynamics through French,
were to experimentally verify one of these laws would he believe that he had
verified French?
Once again this passage shows that Poincart did not justify the retention of
Euclidean geometry in physics by the introduction of universal forces or other
ad hoc hypotheses; rather it shows that he considered the interdependence
8The Value of Science, Q. 37.
Cf. Science and Hypothesis, Q. 50.
Cf. Des Fondements de la Gkomttrie, pp. 265-266.
lzlSur les Principes de la Gtomkie, Q. 79.

332

Studies in History and Philosophy of Science

between geometry and physics to be purely linguistic and not, as assumed by


the above commentators, as in some way empirical**.
(c) The Retention of Euclid and the Duhem-Quine Thesis
It is perhaps worth noting that the influence of Reichenbach is not the only
source of error in the above interpretation of Poincares arguments for the
retention of Euclid. There is another equally important source, namely the
Duhem-Quine thesis that an isolated hypothesis may be retained in the face of
prima facie adverse experimental evidence by a suitable alteration in other
parts of the theorylz3. From this point of view, it is the whole theory, and not
merely some of its specific isolated hypotheses, which is open to empirical
refutation. Both Nagel and Sklar, for instance, assume that this is also another
of Poincares underlying assumptions in his argument for geometrical
conventionalism24. However, as Grtinbaum clearly shows, this interpretation
of Poincare implies an inductive, i.e. empirical, interdependence between
geometry and physics, whereas Poincare himself conceived this interdependence
to be solely linguistic 25. Moreover, Poincare held that isolated hypotheses are
open to empirical verification apart entirely from the theory to which they
belong 28. And finally, as Mary Hesse points out, if Poincares geometrical
conventionalism means, as it appears to mean, that any given total theoretical
system (of geometry) can be imposed upon any logically possible experience,
it does not fall within the scope of the Duhem-Quine type of empiricism27.
It is interesting in this connection to consider Einsteins view of Poincares
geometrical conventionalism28. He depicts an imaginary conversation between
Poincart and Reichenbach in which Poincare defends geometrical conventionalism by arguing that geometry in itself is not open to empirical falsification; it
is so only in relation to the physical laws and theories by means of which the
physicist calculates the corrections necessary to account for the actual distortions
of his measuring rods due to thermal and other effects, since these laws and
theories presuppose Euclidean, or some other metrical, geometry. Einstein here
implies that, for Poincare, geometry and physics taken together are open to
empirical falsification, and that therefore his geometrical conventionalism is a
Grtlnbaum (cJ op. cit. p. 115-l 16) though he does not mention the above passages, draws
the same conclusion.
Cf. Duhem, The Aim and Structure of Physical Theory, part II, pp. 183-190; and Quine,
From a Logical Point of View, p. 14.
Cf. Nagel, op. cit. pp. 261-265, and Sklar, op. cit. pp. 51-52. Popper also is inclined to this
view (cf. The Logic of Scientific Discovery, p. 78).
2Cf. Grtinbaum, op. cit. pp. 115-131.
His discussion of Newtons laws shows this (cf. Scienceand Hypothesis, pp. 91-105).
Mary Hesse, Duhem, Quine and a New Empiricism, Knowledge and Necessity, p. 195. In
making this point, she is assuming that the whole or most of a system should not be considered as
an isolated hypothesis. In my opinion this assumption is valid: a system of geometry is much more
sophisticated than, for instance, a particular law or hypothesis.
Cf. P. A. Schilpp, Albert Einstein, Philosopher-Scientist,

pp. 677-679.

PoincarPs Conventionalism

of Applied Geometry

333

particular instance of the Duhem-Quine empiricism just referred to.


For Poincare, however, physics and geometry are only linguistically
interdependent, and are taken to be such even in the physicists corrections of
thermal and other physical distortions of rigid rods. Hence Poincares
geometrical conventionalism is entirely independent of any necessity to make
such corrections. For Einstein, on the other hand, the conventionalism of
geometry results from the fact that, in calculating the required corrections due
to thermal and other influences, the physicist must first assume the validity of
say, Euclidean geometry, since the laws and theories used to make these
corrections presuppose this geometry. It follows that, for Einstein, geometrical
conventionalism would not hold in determining the geometry of a logically
possible region free from deforming influences, which is clearly contrary to
Poincares position. Of course Einstein could argue that the above logical
possibility is empirically impossible, since it is empirically impossible to
determine, in isolation from the larger theoretical framework of physics,
whether in fact any given region is free from deforming influences. But this
would be to adopt what Mary Hesse calls the new empiricism of DuhemQuine, and Poincare, we saw, did not, even implicity, agree with this position2g.
It can be argued, however, that Poincares thesis of the linguistic interdependence between geometry and physics, while developed within classical
empiricism and thus outside the scope of the new Duhem-Quine type of
empiricism, is yet compatible with this empiricism. In other words, whether one
holds, with the classical empiricist, that hypotheses in isolation from a total
theory are empirically falsifiable, or, with the new Duhem-Quine empiricist,
that the total theory alone is falsifiable, Poincares thesis of the linguistic
interdependence between geometry and physics can still be justified. The
possibility of its justification within the new empiricism may be seen from
considerations of Hesses model of such an empiricist in terms of a computer
or machinelJO. In this model the machines receptor is causally modified by
the physical input from the environment, and the machine uses its coding
system to represent this information. The possibility of Poincares thesis may
be seen from an analysis of this representational stage of the mechanical model.
*gGrtlnbaum (c$ op. cit. p. 133) also maintains that, in this particular case, Einstein is clearly
wrong in the light of Poincares claim that one actually neglects the kind of variations which
Einstein has in mind in laying the foundations of geometry because they are both small and
irregular, and consequently appear to us to be accidental (Scienceand Hypothesis, p. 66). My own
opinion, however, is that Grtlnbaum is here neglecting the context of Poincares statement: in
making it, Poincart is not discussing the possibility of verifying, or falsifying Euclidean geometry;
rather the foundations which he has in mind here are the psychological foundations, and he is
maintaining that, because these distortions are small and irregular, we neglect them. Nevertheless,
in the light of the last paragraph, and the fact that Poincare does not discuss the possible influence
of the theories implicit in the above corrections, I think that Einsteins attribution of the
Duhem-Quine thesis to Poincart is not correct.
lCJ Hesse, op. cit. pp. 198-199.

334

Studies in History and Philosophy

of Science

From Poincares point of view, we may say that some definite metrical
geometry is part of the machines coding system, and therefore the interdependence between its geometry and physics, unlike the inductive interdependence between the various empirical elements of a theory subsequently
constructed by the machine, is solely linguistic. If this model is correct, then
Nagel, Sklar, and even Einstein have misjudged the point at which Poincares
thesis of the interdependence between geometry and physics becomes relevant
for the new empiricist: they fail to see that a metrical geometry may simply
form part of the coding language of the new empiricist.
(d) Semantic Conventionalism
In the last section we saw that Poincare conceived the interdependence
between geometry and physics to be purely linguistic. Because of this certain
philosophers of science have interpreted PoincarCs geometrical conventionalism
as merely a particular instance of what Grtinbaum calls trivial semantical
conventionalism, viz. that all it means is that geometry is conventional in the
way that words are conventional, i.e. are chosen arbitrarily as symbols for
certain meanings31. For instance Russell, Eddington and Swinburne assume
this. Also Mar, maintains that the kind of argument used by PoincarC to prove
the metrical amorphousness of the mathematical continuum is based on trivial
semantic conventionalism combined with other irrelevant issues32, while
Whitbeck holds that Poincares position, though not a case of trivial semantic
conventionalism, is still a semantic one.
Apart from Poincares statement about the linguistic interdependence
between geometry and physics mentioned above, the most common source of
this interpretation
is Poincares dictionary for translating terms from
Lobachewskian into Euclidean geometry M. This formal inter-translatibility
between the two geometries, as Nagel calls it, is interpreted as constituting the
essence of Poincares geometrical conventionalism. In this interpretation it is
assumed, as Russell points out in connection with the various definitions of
congruence compatible with the principles of projective geometry35, that
Poincare held that the meaning of such words as congruence may be changed
at will. But as Swinburne points out one cannot decide what ordinary words
mean. Words in common use mean what they do, and there are recognized tests
for determining their applicability36. Hence, by changing these tests, one is
analysing a concept of ones own invention. For this reason he rejects Poincares
thesis of geometrical conventionalism as, at best, an unenlightening triviality,
true of any language whatsoever and, at worst, as entailing an unacceptable
Cf. Griinbaum, op. cit. p. 24.
Cf. Sklar, op. cit. p. 49.
Cf. Whitbeck, op. cit. p. 33.
Cf. Scienceand Hypothesis, pp. 41-43.
Cf. Russell, An Essay on the Foundations of Geometry,
%winburne, op. cit. pp. 96-97.

p. 33.

Poincart!s Conventionalism of Applied Geometry

335

change in the ordinary meanings of words.


The first point to note about all this is that Poincare himself did not regard
his geometrical conventionalism as based on a trivial semantic conventionalism
in the above sense. This is clear from his reply to Russells criticism, which he
considered to be completely irrelevant. He illustrates this by means of the
following example. From the point of view of trivial semantic conventionalism,
I may say that a triangle has four sides, and, even if it be objected that I am not
using the word triangle in its ordinary sense, this still does not, as Poincare
points out, imply that the proposition a triangle has three sides is not a
definitioni3. He makes the same point in speaking about the trivial semantic
conventionalism of language in our descriptions of ordinary events. He points
out that, for instance, during an eclipse everyone agrees it becomes dark, but
that in a community where dark is called bright everyone would say it is
getting bright. But, as Poincare points out, this change of language is of no
consequence138. Secondly, Poincares dictionary does not prove that his
geometrical conventionalism is but a trivial semantic conventionalism. This
may be seen from the context in which Poincare uses the dictionary - he uses
it, not to prove the conventionalism of geometry, but rather to prove the
consistency and practical applicability of non-Euclidean geometries 3e.
It is interesting in this connection to consider Whitbecks conception of
Poincares geometrical conventionalism. In her opinion, this conventionalism
is not trivial but is an interesting species of linguistic conventionalism, its
principal source of interest being the fact that it denies the existence of apriori
structures in nature. It is not, however, she continues, an interesting claim
about congruence and simultaneity relations as such, since these are not shown
to be matters of convention in any way that other matters of fact are not, but it
is interesting in so far as experience could make it convenient to adopt another
way of speaking and thinking, and that the change of meanings for terms is
systematic40. The point Whitbeck makes here is that Poincares geometrical
conventionalism is logically contrary to the Kantian thesis that Euclidean
geometry is synthetic apriori and that the metrical structures of Euclid are not
apriori structures as understood by Kant. I think Whitbeck is right in this, but I
cannot accept her claim that Poincarts geometrical conventionalism is not
interesting in itself. We saw above that Poincares geometrical conventionalism
is in certain important respects different from the conventionalism of the
principles of mechanics. Also his attitude to trivial semantic conventionalism
mentioned in the last paragraph shows that he himself distinguishes between
Des Fondements de la Cibomktrie, p. 273.
Cf. The Value of Science, p. 118.
Cf. Science and Hypothesis, p. 43.
OSimultaneity and Distance, The Journulof Philosophy (19691, p. 337.

336

Studies in History and Philosophy of Science

geometrical conventionalism and the linguistic conventionalism in our expression


of ordinary events. Finally, the identification of his geometrical conventionalism with a kind of linguistic conventionalism resulting from the availability of
a systematically constructed dictionary fails to take into account the manner in
which Poincare viewed the relationship between space and matter in his
argument for geometrical conventionalism. Thus, while it may be granted that
Poincares thesis of geometrical conventionalism is a semantical one, it is
neither trivial nor uninteresting.
Finally, it is particularly instructive to look at the arguments by which Sklar
and Swinburne try to show that Poincares conventionalism was in fact purely
semantic. Let us first consider Sklars argument. He maintains that Poincares
argument from the metrical amorphousness of the continuum to the conventionalism of distance merely shows that one can change ones meaning of
distance without modifying the already specified topology, and that the
ordering relations of the continuum do not suffice for a definition of distance.
He concludes that, once again, what we appear to have is merely trivial
semantic conventionality, with a few interesting, but quite irrelevant, facts
about the properties of dense manifolds superadded4. If Sklar is maintaining
here that the simple metrical amorphousness of the mathematical continuum
does not prove the conventionalism of distance both pure and applied, I think
he is correct. However, Poincare himself does not argue in this way: his
argument, as we have seen, is complex and elaborate. In evaluating this
argument we must, for example, take into account his view that the concept of
distance is neither intuitive in the Kantian sense nor abstract. Also we cannot
afford to neglect his view of the relationship between the mathematical
continuum and the empirical elements of classical mechanics. Finally, Poincarts
approach to the ordinary concept of distance must be viewed as a mathematicological as well as an empirical one: the scientist normally treats of physical
space as mathematically continuous, but also normally rigid rods are used in
measuring distance. There is a tension between these two elements, but if we
consider the empirical element, we see that the self-congruence of a rigid rod
is not merely an empirical issue. For this reason, and taking into consideration
the nature of the continuum itself, Poincare concluded that congruence is a
matter of definition.
Swinburnes argument, unlike that of Sklar, hinges upon the ordinary
meaning of distance, which, he claims, is given by what we called above the
primary characteristics of this term 142.For him the meaning of an ordinary
word is defined by its use, and therefore one cannot arbitrarily decide what an
ordinary word means. He argues from this that geometrical conventionalists
Sklar, op. cit. p. 49.
.Cf. supra, p. 26.

Poincarks Conventionalism

of Applied Geometry

337

fail to see that there are recognized tests for congruence as ordinarily understood, and that in proposing new tests, they are really substituting new concepts
for the ordinary concept of congruence. In particular, when Poincare argues
from the fact that congruence is logically an equivalence relation to the metrical
amorphousness of space, he is assuming that the meaning of the ordinary
word congruent is simply given by its logical relations to other terms, whereas
in reality it is determined, not merely by the fact that it is a spatial equality
predicate, but by such considerations as that a wooden ruler is approximately
congruent with itself when moved across a room43. In my opinion Swinborne
here exaggerates the role of ordinary terms in science, and neglects the many
innovations introduced by scientists into our conceptual scheme.
While science, as Poincare himself admits, has its starting point in our
ordinary descriptions of the world, it is not limited to the concepts used in these
descriptions. Moreover, some at least of the new concepts introduced by
science, while they differ from ordinary concepts, are still related to these
concepts in various complex ways. For instance, when a scientist explicates the
word hot in terms of temperature, these concepts, though not the same,
are related. In his argument against geometrical conventionalism Swinburne,
in my opinion, concentrates too much on the differences between our ordinary
and mathematical concepts of congruence and neglects the connection between
them. For example, he emphasizes the fact that the ordinary meaning of
congruence is given by what he calls its primary tests, but he neglects the
connection between the mathematical concept of congruence and this
ordinary meaning. It is true that, in mathematics, a congruence relation is taken
to mean a relation which is symmetrical, reflexive and transitive, and since this
usage is different to the ordinary usage emphasized by Swinburne, we
must agree that the mathematical and ordinary concepts of congruence are
different. Nevertheless, they are related in various ways. For instance, it could
be argued that, if we logically analyse the ordinary concept of congruence
used by Swinburne, we should discover its mathematical meaning. Indeed, if
we follow the practice of scientists in treating physical space as a mathematical
continuum, the mathematical meaning of congruence becomes preponderant.
Thus, even if geometrical conventionalism is primarily concerned with the new
scientific concept of congruence, this concept is itself connected by complex
relations to the ordinary concept of congruence as understood, for instance,
by Swinburne himself. For this reason I believe that Swinburnes argument
against geometrical conventionalism is not as convincing as it first seems.
By way of conclusion, it is helpful to illustrate by some practical examples
the semantical nature of PoincarCs geometrical conventionalism. Take, for
%winburne, op. cit. p. %.
Cf. The Valueof Science,

pp. 115-126.

338

Studies in History and Philosophy of Science

example, the case of the blackboard. Relative to the standard Euclidean


definition of congruence, the curves on the blackboard are assigned various
names, and the movements of a rigid rod are described. But relative to the
definition of congruence given by ds = f(d.9 + df)/Y, these same curves
are assigned different names, and now the language of hyperbolic geometry is
used in describing the same movements of the rigid rods. In other words, the
same empiricial facts are described in different geometrical languages. Or,
again consider the case of Schwarzschilds experiment. If the parallax of a
distant star is shown to be negative, the scientist may describe this in nonEuclidean terms, by saying that the light rays are geodesics, or he may describe
it in Euclidean terms thus implying that the light rays are not geodesics. This
latter possibility results from the choice of the customary definition of
congruence, or some other definition which results in Euclidean geometry45.
In this connection it is worth comparing this account of how Euclid could be
retained in the above hypothesis with Griinbaums account, which he held to
be that of Poincare. He states:
If the paths of light rays are geodesics on the customary definition of congruence,
as indeed they are in the Schwarzschild procedure cited by Robertson, and if the
paths of light rays are found parallacticahy to sustain non-Euclidean relations on
that metrization, then we need only choose a different definition of congruence
such that these same paths will no longer be geodesics and that the geodesics of the
newly chosen congruence are Euclideanly related8.

If this passage is taken literally, it is contradictory. Literally, what Grtinbaum


is asserting is that, if the paths of light rays are geodesics relative to the
customary (i.e. the Euclidean) definition of congruence, they may be shown to
sustain non-Euclidean relations relative to this definition. But this is impossible:
if the paths of light rays are geodesics on the Euclidean definition of congruence,
they must necessarily sustain Euclidean relations since, once a particular
definition of congruence has been decided upon, both the geodesics associated
with this definition and its geometry are determined14. Hence, if we accept the
Euclidean definition of congruence, the geodesics are necessarily Euclidean
and no experiment can show them to be otherwise. What Griinbaum should
have said is that, if the paths of the light rays are shown to sustain non-Euclidean
relations, one may choose either to call these paths geodesics, thus introducing
a non-Euclidean geometry, or else retain Euclid and say that the paths are

45Grtinbaum has shown that there is an infinite number of definitions of congruence which
result in a Euclidean geometry for a semi-plane (cJ op. cit. pp. 93-101).
60p. cit. p. 119.
Cf. op. cit. pp. 13-14.

Poincarks Conventionalism of Applied Geometry

339

not geodesics. From these examples the linguistic nature of the interdependence
between geometry and physics in PoincarCs geometrical conventionalism is
clearly seen.
University

College

Gal way
Ireland

Bibliography
J. Andrade, LecoonsdeMPchaniquePhysique (Paris: Alcan, 1911).
A. J. Ayer, Language, Truth andLogic. 2nd edn, (London: Gollancz, 1948).
M. Black, Conventionality in Geometry and the Interpretation of Necessary Statements,
Philosophy of Science (1942), 335-349.
M. Black, Problems of Analysis (London: Routledge & Kegan Paul, 1954).
M. Born, Physics in My Generation (London, New York: Pergamon Press, 1956).
L. Brunschvicg, LOeuvre de Henri PoincarC: Le Philosophe, Revue de Metaphysique et de Morale
(1913), 585-616.
A. Calinon, Etude sur les Diverses Grandeurs en Mathematiques (Paris: Gauthier-Villars, 1897).
R. Carnap, Introductory Remarks to the English Edition of Reichenbachs Philosophy of Space
and Time, (1958), v-vii.
H. Dingle, The Philosophical Significance of Space-Time, Proceedings of the Aristotelian
Society, (1947-48) 153-164.
H. Dingle, Note on Mr. Keswanis articles, Origin and Concept of Relativity, The British Journal
for the Philosophy of Science (1%5), 242-246.
P. Duhem, The Aim and Structure of Physical Theory. Translated by Philip P. Wiener, (Princeton:
Princeton University Press, 1954).
J. Earman, Space-Time, or how to solve philosophical problems and dissolve philosophical
muddles without really trying The Journal of Philosophy (1970), 287-319.
H. Feigl and G. Maxwell (eds.) Minnesota Studies in Philosophy of Science, III (Minneapolis: University of Minnesota Press, 1971).
J. Giedymin, Logical Comparability and conceptual Disparity between Newtonian and Relativistic
Mechanics, The British Journalfor the Philasophy of Science (1973), 270-276.
A. Grtlnbaum, Philosophical Problemsof Spaceand Time (New York: Knopf, 1963).
M. Hesse, Duhem, Quine and a New Empiricism, Knowledge and Necessity. Royal Institute of
Philosophy Lectures, 3,1968-69. (London: Macmillan, 1970).
G. H. Keswani, Origin and Concept of Relativity I and II, The British Journalfor the Philosophy
of Science (1964), 19-32; 286-306.
L. D. Landau and E. M. Lifshitz, The Classical Theory of Fields, Translated by M. Hamermesh
(Oxford: Pergamon Press, 1962).
J. R. Lucas. Euclides ab Omni Naevo Vindicates, The British Journal for the Philosophy of
Science(lb69), l-11.
J. R. Lucas, A. Treatise on Time and Space (London: Methuen, 1972).
J. J. A. Mooij, La Philosophie des Mathematiques de Henri Poincare (Louvain: Nauwelaerts,
1966).
E. Nagel, The Structure of Science (London: Routledge & Kegan Paul, 1968).
J. Nicod, Geometry and Induction, translated by John Bell and Michael Woods (London:
Routledge & Kegan Paul, 1970).
H. Pomcart, Sur les hypotheae fondamentales de la Geometric, Bulletin de la So&W MathematiquedeFrance(l887),203-216.
H. Poincart, LEspace et la Geometric Revue de Metaphysique et de Morale (1895), 631-646.
H. Poincare, On the Foundations of Geometry, TheMon& (1898), l-43.
H. Poincare, La Mesure du Temps, Revue deM&aphysqueet deMorule (1898), l-13.
H. Poincare, Des Fondements de la Geomttrie, Revue de Metaphysique et de Morale (1899),
251-279.
H. Poincare, Sur les Principes de la Gtometrie, Revue de Metaphysique et de Morale (1900),
73-86.
H. Poincare, Science and Hypothesis, translated by W. J. G. (New York: Dover, 1952).

340
H.
H.
H.
H.

Studies in History and Philosophy

of Science

Poincart, The Principles of Mathematical Physics, TheMon& (1905), l-24.


Poincare, The Value of Science, translated by George Bruce Halsted (New York: Dover, 1958).
Poincare, Science and Method, translated by Francis Maitland (New York: Dover).
Poincare, Mathematics and Science: Last Essays, translated by John W. Bolduc (New York:
Dover, 1963).
K. P. Popper, The Logic ofkientificfliscovery
(London: Hutchinson, 1968).
M. Przelecki, The Logic of Empirical Theories (London: Routledge & Kegan Paul, 1969).
H. Putnam, The Analytic and the Synthetic, Minnesota Studies in the Philosophy of Science,
Vol. 3, Feigl and Maxwell (eds.), (1971), 358-397.
W. V. 0. Quine, Wordand Object (New York: M.I.T. Press, 1969).
H. Reichenbach, The Philosophy of Space and Time, translated by Maria Reichenbach and John
Freund (New York: Dover, 1958).
H. Reichenbach, The Rise of Scientific Philosophy, (California: University of California Press,
1968).
H. P. Robertson, Geometry as a Branch of Physics, Problems ofspaceand Time, J. J. C. Smart
(ed.) (New York: Macmillan, 1968) 231-247.
B. Rogers, On Discrete Spaces, American Philosophical Quarterly (1968), 117-123.
L. Rougier, La Philosophic GkomPtriaue de Henri Poincark (Paris: Alcan, 1920).
B. Russell, An Essay on the Foundatidns of Geometry, (New York: Dover, 1956).
B. Russell, Sur les axiomes de la G&m&e, Revue de Mktaphysique et de Morale (1890), 684-707.
B. Russell, A History of Western Philosophy, (London: Allen & Unwin, 1945).
P. A. Schilpp (ed.), Albert Einstein: Philosopher-Scientist,
(Illinois: The Library of Living
Philosophers, 1949).
L. Solar, The Conventionality of Geometry, Studies in the Philosophy of Science, N. Rescher (ed.)
(Oxford: Blackwell, 1%9), 42-60.
R. Swinburne, Spaceand Time, (London: Macmillan, 1968).
H. Weyl, The Philosophy of Mathematics and Natural Science, (Princeton: Princeton University
Press, 1949).
C. Whitbeck, Simultaneity and Distance The Journal of Philosophy, (1969), 329-340.

You might also like