P O I N C A R É O N T H E FO U N D AT I O N S O F

A R I T H M E T I C A N D G E O M E T RY. PA RT 1 :
A G A I N S T “D E P E N D E N C E - H I E R A R C H Y ”
I N T E R P R E TAT I O N S
Katherine Dunlop

The main goal of part 1 is to challenge the widely held view that Poincaré orders the sci-
ences in a hierarchy of dependence, such that all others presuppose arithmetic. Com-
mentators have suggested that the intuition that grounds the use of induction in arith-
metic also underlies the conception of a continuum, that the consistency of geometrical
axioms must be proved through arithmetical induction, and that arithmetical induction
licenses the supposition that certain operations form a group. I criticize each of these
readings. More fully, I argue that the justification Poincaré offers for the use of the group
notion in geometry appears to extend to set-theoretic notions that would suffice to put
arithmetic on a logical foundation, thus undermining his own case for the necessity of
intuition in arithmetic. In part 2, I offer an interpretation of intuition’s role on which it
justifies the use of group-theoretic, but not set-theoretic, notions.

1. Introduction
The ostensible aim of Science and Hypothesis, Poincaré’s first book of philosoph-
ical essays, is to put forth a comprehensive view of scientific thought. In the in-
troduction, Poincaré proposes to “review the series of sciences from arithmetic
and geometry to mechanics and experimental physics” (1902/1929, 28). This
article and its sequel seek to show how Poincaré’s account of arithmetical and

Contact Katherine Dunlop at Department of Philosophy, University of Texas at Austin, Austin, TX

78712 ([email protected]).
Earlier versions of this paper were presented at Midwest Phil Math Workshop (Notre Dame) and
PSA (San Diego). I thank members of those audiences, especially Michael Detlefsen, for helpful feed-
back. I also thank Colin McLarty, Sahotra Sarkar, Mojtaba Soltani, and two anonymous referees for
their helpful comments.

HOPOS: The Journal of the International Society for the History of Philosophy of Science, vol. 6 (Fall 2016).
2152-5188/2016/0602-0006$10.00. © 2016 by the International Society for the History of Philosophy of Science.
All rights reserved. Electronically published September 16, 2016.

274

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

geometrical knowledge serves his broader aims, which are to explain how hy-
potheses are used in the sciences and show that their use does not revoke contact
with reality.
Poincaré holds that in mathematics, hypotheses occur as “disguised defini-
tions” or conventions. Although Poincaré’s defense of a role for intuition in
arithmetic composes the first chapter of Science and Hypothesis, his general view
of mathematics does not easily accommodate it. For with respect to arithme-
tic, Poincaré is often taken to hold the Kantian view that its objects (the natural
numbers) are directly grasped through intuition. Since (on this reading) the
numbers are thus already given without any conceptual or linguistic interme-
diary, there seems to be no way for definitions of the numbers to further arith-
metical knowledge. Poincaré decisively rejects, in particular, attempts to define
the natural numbers in what we now recognize as set theory, partly because he
thinks that this background theory illicitly tries to establish the existence of its
objects by implicit definition. This brings the seeming conflict into sharp relief.
For to show that experience cannot determine which geometry correctly de-
scribes space, Poincaré argues that any empirical finding can be made consistent
with either Euclidean or Lobachevskian geometry by assigning appropriate
physical referents to geometry’s primitive terms. His account suggests that the
referents are fixed by implicit definition and has the explicit consequence that
the principles of geometry are conventional.
A clear sign of the incongruity between Poincaré’s philosophy of arithmetic
and his overall view of mathematics is that the introduction of Science and Hy-
pothesis (which lays out the overall view) does not even contain the word “in-
tuition”; conversely, the word “hypothesis” does not appear in the first two
chapters (on arithmetic and analysis, respectively; see Heinzmann 2009, 177).
In place of a unitary philosophy of mathematics, Poincaré thus seems to offer
two views in considerable tension with one another.
The prevalent understanding of mathematics’ place in Poincaré’s “series of
sciences” affords a resolution of the tension.1 In mathematics, hypotheses func-
tion as conventions (as mentioned above), and experience steers their adoption
only indirectly, by making one or another convenient. Mechanics’ principles

1. What I call “dependence-hierarchy” interpretations are sometimes proposed specifically in order

to reconcile Poincaré’s Kantian view of arithmetic with his conventionalism about geometry. For
instance, Michael Friedman appeals to Poincaré’s view of “the importance of arithmetical intuition (the
intuition of succession or indefinite iteration) in geometrical reasoning” to show that, despite Poincaré’s
conventionalism about geometry, the connection he finds between “the group of motions in a space of
constant curvature and an iterative procedure of geometrical construction” is a generalization of Kant’s
view (Friedman 2000, 204–5). Similarly, Janet Folina argues that Poincaré “had to” reconceive ge-
ometry “and its relation to science” in order “to retain the spirit of Kant’s philosophy” (Folina 1994,
217).

275

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

share the conventionality of geometry’s but are “more directly based on experi-
ence” (Poincaré 1902/1929, 29), whereas hypotheses capable of experimental
confirmation (or invalidation) first appear in physics. Poincaré thus places the
sciences in a serial ordering, which (on any account) reflects increasing scope for
experience to dictate principles. On what I will call “dependence-hierarchy” in-
terpretations, Poincaré orders the sciences in relations of dependence: arith-
metic is presupposed by analysis, which is presupposed by geometry, which is
presupposed by mechanics, and so on. Since the principle of induction founds
arithmetic, it is (on this reading) presupposed in all mathematical reasoning, but
as we ascend through the levels, there is increasing room for choice among
principles. This understanding of Poincaré’s view is now so widely endorsed that
it appears in reference works.2 It has attracted special notice because it promises
to explain why Einstein’s theory of relativity would not have been acceptable to
Poincaré.3
This paper concerns only the relationship of arithmetic to geometry and will
not discuss mechanics or physics. On dependence-hierarchy interpretations,
geometry has been seen to depend on arithmetic (or at least on a structure sat-
isfying the principle of induction) in at least three ways. After setting out the
textual evidence for these construals, I will argue against them, in part by situ-

2. See Folina (2006) and Heinzmann and Stump (2013). Heinzmann and Stump claim that the
hierarchical view “implied in” Science and Hypothesis is “explicitly seen in” The Value of Science, but they
give no specific reference. They may have in mind the introduction, which gives Poincaré’s rationale for
proceeding from mathematical reasoning in general to mathematical analysis and its subject matter, the
“frames” time and space, and thence to mathematical physics, but does not in fact assert dependence
relations among these. Chap. 10 asserts that conventions are of decreasing importance as we move from
geometry, via mechanics, to physics but does not claim that the latter depend on the former.
3. Poincaré was not confronted with the general theory of relativity, but Friedman (1995–96/1999)
argues that Poincaré’s view is incompatible with the “conceptual framework” of (general) relativity on
two counts. First, the general theory of relativity “essentially employs a geometry of variable curvature,”
but Poincaré limits geometrical consideration to manifolds of constant curvature, by requiring that the
entire domain of geometry be constructible by iterated operations (Friedman 1995–96/1999, 81).
Secondly, in the general theory of relativity the description of the motion of bodies in a gravitational
field as geodesics in space-time both determines a geometry and “takes over the role of the law of
inertia,” in the sense that the geodesics “have precisely the role previously played by the inertial motions”
(80). Thus, the priority Poincaré assigns to geometry over mechanics, and to mechanics over physics,
cannot be sustained. Robert DiSalle similarly explains the difference between Poincaré’s and Einstein’s
views in terms of how, on Einstein’s principle, physical considerations determine “the metrical structure
that is to play the role of the gravitational field” (DiSalle 2006, 87).
With respect to special relativity, both DiSalle (2006, 91) and Friedman contend that what Poin-
caré calls “the law of relativity” differs from Einstein’s principle of 1905 in that only the latter involves
raising a measurement procedure or physical generalization to the level of a convention or disguised def-
inition. Poincaré’s is “not so much a physical as a purely geometrical principle,” insofar as it “rests squarely
on the homogeneity and isotropy of space, which is a necessary presupposition for” having a geometry in
place, and therefore prior to any physical or empirical theory (quoting Friedman 2010, 655).

276

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

ating the arguments of Poincaré (1902/1929) in their original settings (in papers
published prior to the book) and in relation to Poincaré’s other writings.
Section 2 presents Poincaré’s argument that arithmetical reasoning is based
on intuition. Section 3 shows how the intuition that justifies the use of induc-
tion in arithmetic can also be seen to underlie the conception of a continuum,
on which geometry imposes a metric. I contend, however, that Poincaré claims
that intuition of a different kind is needed to conceive of irrational numbers and
thus a true continuum. As this takes away a role (in geometry) for the intuition
that grounds arithmetic, it raises the question of what function that intuition
could have.
Section 4 introduces Poincaré’s “translatability” argument for the conven-
tionality of geometry, which (as I explain) supports the view that Poincaré re-
gards geometrical axioms as implicit definitions. This suggests a second way in
which geometry might depend on arithmetic, namely, if mathematical induc-
tion is necessary for proving the consistency of geometrical axiom systems. In
section 5 I argue that this reading overlooks Poincaré’s reservations about for-
malization and formalist views of mathematics.
Poincaré’s deepest explanation of geometry’s application to perceptual ex-
perience—and his strongest argument for its conventionality—draws on the
notion of a group, as I explain in section 6. As it appears in Poincaré (1902/
1929), the view that certain geometrical operations form a group appears to
rest on the intuition that grounds arithmetical reasoning, and this is the third
way in which geometry has been seen to depend on arithmetic. But, as I argue
in section 7, in Poincaré (1902/1929) the argument for basing geometry on the
choice of a group (rather than a single, given group) is incomplete. Section 8
shows how the gaps in the argument are filled by considerations set forth in
Poincaré’s other writings.
As I argue in section 9, Poincaré’s complete account of our warrant for
applying the notion of group to geometrical operations stresses the power and
importance of linguistic innovation. The immediate problem this raises for
dependence-hierarchy interpretations is to find a role in geometry for the in-
tuition that grounds arithmetical reasoning, since one of the roles they assign to
it now appears to be filled by linguistic innovation.
There is also a deeper problem. Poincaré’s full account of the warrant for the
application of the group notion in geometry seems to provide equal warrant for
the introduction of set-theoretic notions (in the foundations of arithmetic),
which would place arithmetic on a logical foundation. In section 10 I consider
Poincaré’s objections to the definitional stipulations by which these notions are
introduced. In section 11 I show that the application of the notion of group in
geometry carries the same logical commitments as the definitions Poincaré re-

277

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

jects in arithmetic. This is specifically how the considerations that justify the
notion’s application in geometry appear to undermine Poincaré’s arguments for
the inadequacy of logic to found (and thus the need for intuition in) arithmetic.
Hence, Poincaré’s full account of the justification for applying the group notion
seems not only to leave no place for intuition in geometry but also to make in-
tuition unnecessary even in arithmetic.
I will ultimately argue (in part 2 of this article) that according to Poincaré’s
view, intuition justifies the application of the group notion in geometry, but
not the introduction of set-theoretic notions in arithmetic. Here I claim that it
will not be possible to draw a (principled) distinction between these cases so long
as intuition is supposed to ground all mathematics by justifying mathematical
induction. But if this understanding of intuition’s role cannot be sustained,
neither can the dependence-hierarchical account (which I set out in sec. 6) of
how intuition justifies the assertion that certain operations form a group.
Part 2 of this article will present an alternative to the dependence-hierarchy
construal of intuition’s role in mathematics. Briefly, I propose that for Poincaré,
intuition’s role is to acquaint us with the unity or harmony that characterizes
some collections and orderings, and thereby justify rules, practices, and gener-
alizations.

2. Poincaré’s View of Arithmetical Reasoning

Chapter 1 of Science and Hypothesis, “On the Nature of Mathematical Reason-
ing,” addresses the puzzle of how mathematical reasoning can have its acknowl-
edged “perfect rigor” without reducing to an “immense tautology” (Poincaré
1902/1929, 31–32). Poincaré claims that on account of their “analytic charac-
ter,” the techniques of formal reasoning cannot teach us anything “essentially
new.” To say that mathematical reasoning is informative because the axioms
from which it begins are synthetic a priori is, he maintains, “not to solve the
difficulty, but only to baptize it,” because the notion of synthetic apriority has
not been properly explained. (And while Poincaré ultimately defends the view
that arithmetic is synthetic a priori, he does not locate its syntheticity in its
axioms, but rather in the reasoning by which theorems are derived; see Detlef-
sen 1992, 349.) Poincaré’s solution to the puzzle is that “mathematical reason-
ing has of itself a sort of creative virtue” (1902/1929, 31–32). Specifically, it im-
parts new knowledge by proceeding from the particular to the general.
To better understand the nature of mathematical reasoning, Poincaré claims,
we “must seek mathematical thought where it has remained pure,” that is, with-
out complications peculiar to geometry and analysis; hence, we must seek it in
arithmetic, and specifically at “the beginning,” where the “primitive notions” have

278

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

not “already undergone an elaboration so profound that it becomes difficult to

analyze them” (1902/1929, 37). But we are brought up short by the fact that in
the “classic treatises,” the demonstration of the “most elementary theorems” is the
“least rigorous.” Thus, new proofs must be supplied for the associativity and
commutativity of addition and the distributivity and commutativity of multi-
plication. Poincaré proves them by induction, based on recursive definitions of
the operations. On the basis of its ubiquity in these arguments, he identifies
“reasoning by recurrence” as “the mathematical reasoning par excellence” (37).
Poincaré finds it impossible to “escape the conclusion that the rule of rea-
soning by recurrence is irreducible to the principle of contradiction” (1902/
1929, 39). His argument appears to be that any attempt to derive the principle
of mathematical induction (using the principle of contradiction) will inevitably
“be arrested” at “an undemonstrable axiom,” which is really “only the propo-
sition to be proved translated into another language.” The example he gives is
the claim that “in an infinite collection of whole numbers there is always one
which is less than all the others” (39). Having thus argued that the rule or
principle of reasoning by recurrence is not analytic, Poincaré argues that it is not
empirical. He claims that “the essential characteristic of reasoning by recurrence
is that it contains, condensed, so to speak, in a single formula, an infinity of
syllogisms” (37; and likewise, the recursion clause of a definition “contains an
infinity of distinct definitions, each having a meaning only when one knows the
preceding” [35]). But experience could only show a rule to be valid over a finite
domain of application. Poincaré concludes that the rule is “the veritable type of
the synthetic a priori judgment” (39). Specifically, its evidence is to be explained
in terms of “a direct intuition of the power of the mind” to “conceive the in-
definite repetition of the same act when once this act is possible,” a power that
the rule “affirms” (39).
Because Poincaré here speaks of an intuition of a mental power, this passage
may seem to conflict with the view, often attributed to him, that intuition pre-
sents us with objects (natural numbers, or a structure in which they are places).4
But it seems clear that Poincaré is not concerned with a psychological capacity
to “conceive the indefinite repetition of the same act.” He does not raise or
address the issue, for instance, of how many instantiations of a sentence “If the
theorem is true for n, it is true for n+1” could be thought in a human lifetime.
He is instead concerned with our justification for taking a rule to be valid, or
an operation to yield results, over an indefinitely extended domain. And it is

4. “Poincaré . . . sometimes equates the intuition of iteration with the sequence of natural numbers”
(Folina 2008, 40). See also Wilder (1967, 605) and Da Silva (1997, 279).

279

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

natural to suppose that intuition delivers this justification by acquainting us

with what is in the domain.
To sum up, Poincaré holds that arithmetical reasoning is distinguished from
tautologous substitutions by a rule, namely, reasoning by recurrence, which
itself rests on an intuition. The intuition has for its object the power to indef-
initely repeat (with justification) the same operation. We will now see how this
intuition can be regarded as foundational for analysis and geometry.

3. Poincaré on the Conception of the Continuum

Poincaré understands “analysis” as a theory of continua that abstracts from
spatiality. He reports that the analysts themselves have given an account of the
continua about which they reason: a continuum is an infinite collection formed
by “intercalating terms” between members of a series. Poincaré distinguishes
the continuum so conceived, as “mathematical,” from the ordinary or “physical”
continuum.5 His epistemological investigation of this notion appears to con-
clude that it, and thereby analysis as a whole, is founded on the intuition just
discussed.
Poincaré rejects the view that the mathematical continuum is freely created
by the mind without the involvement of experience. This view is suggested by
the work of “the mathematicians of the Berlin school, Kronecker in particu-
lar,” who “have chiefly striven to define the incommensurable [i.e., irrational]
number” (Poincaré 1902/1929, 44). Poincaré assumes that it is straightforward
to construct the rationals by combinatorial operations on the natural numbers.6
On this view an irrational number is “nothing but the symbol of [a] particular
mode of partition,” for example, a Dedekind cut, of rational numbers. Poincaré
charges that “to be content with this would be to forget too far the origin of
these symbols” and would leave no way to explain “how we have been led to
attribute to them a sort of concrete existence” (45). He also rejects the view that
the mathematical continuum is “simply drawn from experience,” on the grounds
that the “raw results of experience” violate the transitivity of equality, when there
is a perceptible difference between sensations (e.g., of weight) at the extremes of a
series, the intermediate members of which cannot be distinguished by the senses
(47). Poincaré maintains that it is precisely to avoid such “contradictions” that we

5. In the latter, “between the elements of the continuum a sort of intimate bond [is supposed] which
makes of them a whole, where the point does not exist before the line, but the line before the point”
(Poincaré 1902/1929, 43).
6. Although he later insinuates that even the fractional numbers are subject to the “difficulty” that
we would not have the notion of them “if we had not previously known a matter that we conceive as
infinitely divisible, that is, a continuum” (Poincaré 1902/1929, 46).

280

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

are led to insert (“intercalate”), between members of such a series, terms corre-
sponding to intermediate values that we find ourselves unable to discriminate by
sense. That the number of intercalated terms should be infinite is required to
avoid the “new contradiction” that arises when, through the use of magnifying
instruments, parts of lengths are made to appear indiscriminable from the wholes
(perceived without the instruments; 47).
Poincaré claims that the full explanation of how the mind is led to create the
concept of a continuum is that “as soon as we have been led to intercalate means
between two consecutive terms of a series, we feel that this operation can be
continued beyond all limit and that there is, so to speak, no intrinsic reason for
stopping” (1902/1929, 48). But this is just as it “happens in the sequence of
whole numbers”: experience gives us occasion to exercise a faculty (of increas-
ing a collection of units by one), on becoming conscious of which we “feel that
our power has no limit and that we can count indefinitely,” even though we
have never counted “more than a finite number of objects” (47). It thus appears
that our conception of the (mathematical) continuum is based on the same
intuition that grounds arithmetic.
Yet in this way we obtain only the dense set of rationals, or, in Poincaré’s
words, an “aggregate of terms formed according to the same law as the scale of
commensurable numbers” (1902/1929, 48). He argues that the formation of
irrational numbers, as a law for intercalating terms in a series, is necessary to
avoid a further threat of contradiction. This argument is overlooked in some
“dependence-hierarchy” interpretations,7 and it indeed seems to show that anal-
ysis is founded on an intuition different from that of indefinite iteration.
Poincaré claims that the geometer, “without entirely renouncing the aid of
the senses,” tries to conceive of breadthless lines and extensionless points. Since
we are unable to imagine a line literally without breadth, our only recourse is to
regard the line “as the limit towards which tends an ever narrowing band, and
the point as the limit towards which tends an ever lessening area” (Poincaré
1902/1929, 48). Then the claim that lines that intersect have a point in com-
mon is justified, and made intuitive, by the fact that “our two bands, however
narrow they may be, will always have a common area, the smaller as they are the
narrower,” and its limit is “what the pure geometer calls a point.” But this claim
“would imply contradiction if . . . on the lines traced by the geometer should
be found only points having for coordinates rational numbers” (48). Poincaré

7. See Friedman (1995–96/1999, 75) and Brading and Crull (2013, 8). Friedman (2000) sidesteps
the issue by basing (metrical) geometry directly on arithmetical intuition. Cognizant of the difficulty,
Folina argues that to avoid circularity, Poincaré must appeal to “intuition of the n-dimensional con-
tinuum” (1994, 222).

281

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

concludes that the capacity for creating symbols (through which analysts ex-
plain the formation of the continuum) is exercised only as experience furnishes
the occasion. The initial stimulus provided by experience, which is the “rough
data” to be serially ordered, gave rise to contradictions; since the necessity of
avoiding contradiction is the only limit on the power in question, we are “forced
to imagine a more and more complicated system of symbols,” until we arrive at
one exempt from both “internal contradiction” and “contradiction with various
propositions called intuitive, which are derived from empirical notions more
or less elaborated” (49).
In a lecture given in 1900 (“Intuition and Logic in Mathematics,” cited as
reprinted in The Value of Science [1905]), Poincaré makes clear that he regards
this way of representing a line as intuition of a fundamentally different kind
(than the arithmetical). For him, the belief that every curve has a tangent ex-
emplifies the error to which sensory or imaginative intuition leads. On his ac-
count, because straight lines and curves must be visualized as having finite
width, any two such “narrow bands, one straight, one curved,” can always be
“pictured” as “encroach[ing] slightly one upon the other without crossing” (but
not in the “limit” of narrowness, at which both widths vanish; Poincaré 1905/
1929, 213). Hence, intuition of a common width, which was taken to show
that any two intersecting lines have a point in common (and thereby stimulate
us to conceive of points with irrational coordinates), is here supposed to estab-
lish that any straight line and curve have a unique point in common. Poincaré
then explicitly distinguishes such intuition from that “of natural number,” on
the grounds that only the latter can provide certainty (216). And in an 1897
essay also reprinted in The Value of Science, Poincaré claims explicitly that “the
sole natural object of mathematical thought is the whole number,” whereas the
continuum is imposed on us by “the external world” (285) or “physical nature”
(286). The context makes clear that he means the continuum (rather than the
dense set of rationals), so it is plausible to think that experience imposes the
continuum on us precisely by giving rise to intuition of lines as “narrow bands,”
thereby forcing us to enrich our symbol system.
The difference between the intuition of natural number and the intuition
through which we grasp arithmetical objects raises the question of what role
the former can have in arithmetic, and so calls into question one construal of
geometry’s dependence on arithmetic. But the problem it poses can be resolved.
In the introduction to The Value of Science, Poincaré says that in Science and
Hypothesis he spoke “of scarcely more than space, and particularly quantitative
space, so to say, that is of the mathematical relations whose aggregate constitutes
geometry” (1905/1929, 207). The continuity required for analysis could be
thought to pertain instead to “qualitative space,” the more so because—as Poin-

282

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

caré (1902/1903, 8) realizes—the axioms of geometry (in Hilbert’s formulation)

are satisfied in a space containing only points constructible by ruler and compass
(from certain given points), whose coordinates are therefore algebraic numbers.8
This suggests that the symbolic construction of the continuum serves at most to
ensure the consistency of geometers’ ideas, not to make geometry’s axioms true.
Most importantly, since metrical geometry is to this extent independent of
analysis, it could still fit into a hierarchy of dependence even if analysis does not.9
It has seemed particularly important to determine metrical geometry’s place, in
order to illuminate Poincaré’s view of physics (see n. 3 above).

4. The Axioms of Geometry as Implicit Definitions

Poincaré’s argument for the conventionality of geometrical principles is the ob-
ject of considerable interpretative disagreement and puzzlement. In this sec-
tion, I focus on those aspects of Poincaré (1902/1929) that suggest that ge-
ometry is founded on implicit definition. Group-theoretic considerations will
be introduced in section 6.
Non-Euclidean geometry makes its appearance in chapter 3. Poincaré ob-
serves that the consistency of two-dimensional Riemannian and Lobachevskian
geometry is not subject to doubt because each can be “correlated” with “a branch
of ordinary,” that is, Euclidean, geometry. To show that three-dimensional Lo-
bachevskian geometry is likewise consistent relative to Euclidean geometry,
Poincaré gives a “dictionary” by which its primitives can be interpreted in the
“fundamental plane” (1902/1929, 59), so as to translate its theorems into those
of ordinary geometry. He thereby specifies what is now known as the Poin-
caré half-plane model of hyperbolic geometry.10 At the close of the chapter,
Poincaré notes that the possibility of alternative geometries calls into question
whether “ours is the true one” (63). He first concludes that the axioms of Eu-
clidean geometry are not synthetic a priori judgments, for if they were, they
would “impose themselves on us with such force that . . . there would be no

8. As Friedman explains, in analytical terms the domain of (Euclidean) geometry “is represented . . .
by a Cartesian space over the . . . Euclidean subfield of the reals, which results by closing the rationals
under the operation of extracting real square roots” (2000, 206).
9. Indeed, when (in 1912) Poincaré affirms the existence of an intuition of the continuum (distinct
from the intuition of physical space), he makes clear that its use is limited to analysis situs, whereas the
“intuitive ideas” operative in metrical geometry are “of another kind, analogous to those which play the
essential role in arithmetic and algebra” (1913/1963, 27).
10. As Elie Zahar observes, to call Poincaré’s manual a “dictionary” is “misleading,” because the
manual does not supply “a translation of [synthetic hyperbolic geometry] into [synthetic Euclidean
geometry],” but rather “the construction of a model of [synthetic hyperbolic geometry] within the
complex plane” (1997, 187–88), which constitutes a different vocabulary (as it were).

283

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

non-Euclidean geometry.” He then maintains that they are not “experimental

verities,” for in that case geometry “would not be an exact science, [but] subject
to continual revision” (64). Poincaré finally concludes that the axioms are con-
ventions and the question of their truth “has no meaning,” any more than
“whether the metric system is true and the old measures false” (65).
Here Poincaré sharply contrasts geometrical axioms with the principle of in-
duction, claiming specifically that it is not possible to deny the latter and “found
a false arithmetic analogous to non-Euclidean geometry” (1902/1929, 64). But
it is not clear how the considerations about translatability support his conclu-
sion that geometry’s axioms are conventions. In elaborating his reasoning, com-
mentators have found an important link (between geometry and arithmetic)
underlying the contrast.
An initial problem is that if pure and applied geometry are distinguished,
Poincaré’s argument is seen to concern only pure geometry, but it does not
make clear sense to call this conventional (Black 1942; Torretti 1978, 326). For
it is not clear how to understand the special status accorded to the chosen un-
interpreted geometry (whichever it is). Poincaré himself holds that distinct
notions of existence apply to “mathematical entities” and “material objects.” A
mathematical entity exists “provided its definition implies no contradiction”
(Poincaré 1902/1929, 61), while the existence of “real and concrete objects”
can be known only empirically (Poincaré 1905/1929, 216–17). This is close to
the view that for pure mathematical theories, truth is a matter of consistency.
But on that view Poincaré’s argument, so far from showing that the question of
their truth is meaningless, would show that the geometries are all true.
Yemima Ben-Menahem has explained how the translatability argument can
be extended to show the conventionality of (at least) applied geometry, via Poin-
caré’s subargument that geometry is not experimentally “verified.” Poincaré
observes that experiments (testing geometry’s truth) would have to be carried
out on “material objects,” of which the most natural candidates are solid bodies
and “light and its propagation” (1902/1929, 64). But as he argues in succeed-
ing chapters of Science and Hypothesis, geometrical terms have no unique physical
interpretation, and different assignments of material bodies to the terms per-
mit the same experiential finding to verify different geometries.11 In chapter 4,
Poincaré famously asks us to conceive “a world enclosed in a great sphere” in
which temperature varies as a certain function of distance from the center (of

11. Poincaré argues that there is no property that uniquely characterizes, e.g., the straight line (in
both Euclidean and non-Euclidean space), which would be “an absolute criterion enabling us to
recognize the straight line and distinguish it from every other line,” so as to test whether it has other
properties specific to Euclidean or non-Euclidean straights (1902/1929, 82–83).

284

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

the sphere) and makes all bodies expand and contract, immediately upon be-
ing moved, in the same proportion (76). When these bodies are taken as mea-
sures of distance, the geometry of the world will be non-Euclidean. Poincaré
claims that beings “educated” in such a world “would doubtless find it more
convenient to create” such a geometry, but that we can describe theirs “with-
out ceasing to employ the language of ordinary geometry,” which we “should not
have to change . . . if transported thither” (80). If we understand a convention
as the assignment of physical meanings to geometrical terms, the example shows
how experience can weigh, without militating, in favor of a certain convention.
Finally, in chapter 5, Poincaré refutes the suggestion that measurements of par-
allax could decide which geometry is true of physical space. He argues that an
experimental confirmation of non-Euclidean geometry could be dealt with by
either renouncing the laws of Euclidean geometry or “modify[ing] the laws of
optics and suppos[ing] that light does not travel in a straight line” (81). The lat-
ter course of action would be to assign a different physical interpretation to the
geometrical notion of straight line.
Ben-Menahem very plausibly suggests that the termwise correlation of (pure)
non-Euclidean and Euclidean geometries helps to show that they are empiri-
cally equivalent (as interpreted) by indicating how to construct equivalent de-
scriptions.12 On her reading, Poincaré’s “argument from intertranslatability . . .
goes beyond the methodological argument from the holistic nature of confir-
mation” precisely in that it involves “a method for actually producing” the equiv-
alent descriptions that Duhem and Quine merely claim to be possible (Ben-
Menahem 2006, 57–58). This reading thus satisfies an important desideratum
(stressed particularly in Stump 1989; Friedman 1995–96/1999), namely, to ex-
plain why the conclusion Poincaré draws in chapter 5 is restricted to geometry,
in contrast to the “global” confirmational holism associated with Duhem and
Quine.
Ben-Menahem further observes that “the construal of axioms as implicit def-
initions is in harmony with this understanding of conventionalism” (2006,
58). The idea seems to be that we are always free to change the assignment of
physical referents to the geometrical primitives because referents can be assigned
in any manner that makes the theory true.
This interpretation has clear textual support, as well as the merit of ex-
plaining why Poincaré’s conventionalism does not extend to arithmetic, in par-

12. As Ben-Menahem puts it, without “the modeling of Lobatschewsky’s geometry within Eu-
clidean geometry, it is extremely unlikely that Poincaré (or his hypothetical beings) would have dis-
covered such peculiar laws of thermal expansion and light refraction, but given the modeling, physics
can be easily adjusted to mediate between seemingly incompatible geometries” (2006, 57).

285

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

ticular. In his 1905 “Mathematics and Logic: I” (reprinted in Science and

Method ), Poincaré claims that it is “legitimate” to regard “Euclid’s postulate” as
a “disguised definition.” He elaborates that the “other assumptions of geometry
do not suffice to completely determine distance; . . . distance will then be, by
definition, among all the magnitudes which satisfy these other assumptions, that
which is such as to make Euclid’s postulate true” (1908/1929, 453). But Poincaré
then insists that the principle of mathematical induction cannot be considered a
disguised definition (of natural number). Ben-Menahem concisely sums up his
rationale: “implicit definitions must be shown to be consistent, but demon-
strating the consistency of this putative definition . . . would land us in an infi-
nite regress, for the demonstration would have to invoke the principle” itself
(2006, 58; cf. Poincaré 1908/1929, 453–55).
While Ben-Menahem does not claim that the consistency of implicit defi-
nitions of geometrical terms must be proved by mathematical induction, others
point out that such a requirement would clearly establish the dependence of
geometry on arithmetic. This view is most fully developed (to my knowledge)
by David Stump. Stump takes Poincaré to accept that “a complete proof of the
consistency of the metric geometries can be given formally” (1996, 486) and to
claim on that basis that the metric geometry of space cannot be determined a
priori (a separate argument is needed to rule out determining it empirically;
481–82). Stump further holds that on Poincaré’s “formalist” view of (pure) ge-
ometry, the (metric) primitives are defined specifically in terms of natural num-
bers, which is obviously another way in which geometry depends on arithmetic
(484). Stump’s summary claim that arithmetical intuition is “necessary to be-
gin” the formalization through which consistency is proved (490) accords closely
with Janet Folina’s view that arithmetical intuition is “foundational for logic,”
in that “we must appeal to the indefinite repetition of an act in order to under-
stand even the setting up of a formal system” (1994, 218).13

5. The Context of Poincaré’s Remarks on Formalization

Despite its elegance and initial plausibility, I do not think that this interpre-
tation succeeds in relating Poincaré’s view of arithmetic to his view of geometry,
for its textual support appears weak on closer scrutiny. Stump usefully locates
Poincaré’s views against the backdrop of the nineteenth-century trend toward
“arithmetization” and Hilbert’s axiomatization of geometry, but I think he over-
estimates Poincaré’s sympathy for these developments.

13. Folina maintains, further, that this intuition is required “to understand any rule of inference in
general” and is thus foundational for all “systematic” or rational thinking (1994, 218).

286

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

Stump takes Poincaré to assert in the following passage that “whole [sc. nat-
ural] number is the only mathematical concept we need” (1996, 484): “[One]
has . . . achieved rigor . . . [b]y restraining more and more the role of intuition in
science, and increasing the role of formal logic. Previously one started from a
large number of notions, which were regarded as irreducible and intuitive prim-
itives; such [as] whole number, fraction, continuous magnitude, space, point,
line, surface, etc. Today only one remains, that of whole number; all the others
are merely combinations” (Poincaré 1889, 157). Now for Poincaré in 1889 (just
as for Frege in 1884), the notion of natural number represented a “last frontier”
in the extirpation of intuition from mathematical reasoning: epsilon-delta def-
initions of continuity and limit had freed (real) analysis from dependence on
spatial intuition, and this grounding was made fully rigorous by defining the reals
in terms of natural numbers. I submit that in passages such as the one quoted,
Poincaré acknowledges only that the natural numbers suffice for expressing these
other notions.14 He does not concede that the eliminative project thereby sup-
plies an adequate foundation for analysis. For Poincaré denies that rigorization
can supply all that we want from a foundation. Thus, the quoted passage ends
“at this price we attain perfect rigor” (157). That Poincaré regards the price as
too high is suggested by his claiming, two paragraphs later, that mathematical
notions acquire “this perfect purity” only as they are separated (s’éloigner) from
reality (158). The continuation anticipates the better-known discussion in The
Value of Science: “In becoming rigorous, mathematical science takes a character
so artificial as to strike every one; it forgets its historical origins; we see how the
questions can be answered, we no longer see how and why they are put. This
shows us that logic is not enough; that the science of demonstration is not all
science and that intuition must retain its role as complement . . . of logic”
(Poincaré 1905/1929, 217).15
Of course, Poincaré also contends that intuition cannot be completely elim-
inated from mathematical reasoning because it is indispensable for arithmetic.
It is important for my reading of Poincaré that his argument that intuition
is necessary for arithmetic is directed specifically against attempts to “logicize”
arithmetic, by deducing all its truths (in a background theory that we would

14. This was widely acknowledged around the turn of the century. For instance, Felix Klein writes
that “the discrete magnitude, as conceptually the simplest, has come into the foreground” in “recent
times.” Specifically, the natural numbers are “regarded as the simplest given concepts,” and “we derive
from them, in the familiar way, rational and irrational numbers”; we then “construct the complete
apparatus for the control of geometry by means of analysis, namely analytic geometry.” This tendency to
“carry back” the “geometric idea of continuity” to the idea of natural number can be called “arithmetiz
[ing] geometry” (Klein 1908/1932, 266).
15. It bears noting that this passage is Poincaré’s articulation of what is “left of ” an objection, by “the
philosophers” (1905/1929, 217), which Poincaré has answered; thus, it clearly expresses his own view.

287

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

regard as set-theoretic) from the Dedekind–Peano axioms.16 I believe that this

context is typically overlooked because the first chapter of Science and Hy-
pothesis seems to consider arithmetic’s status from a much broader historical
perspective. Poincaré claims to address the “old” question of whether arithmetic
is “purely analytic or could be analytically derived from a small number of syn-
thetic judgments,” and the only thinker he mentions is Leibniz (1902/1929,
32). He claims that it is imperative to “recast the demonstrations of the most
elementary theorems [into] the form that will satisfy a skilled geometer” sim-
ply because they are customarily presented in a form suitable for beginners, who
“are not prepared for real mathematical rigor” (37). But the chapter first ap-
peared (under the same title, “On the Nature of Mathematical Reasoning”) in
response to Ballue (1894), which seeks to base arithmetic on a theory of “plu-
ralities.” In a passage omitted from Science and Hypothesis, Poincaré claims that
the need to recast the most basic proofs will be apparent from a comparison be-
tween his and Ballue’s proofs of the commutativity of multiplication—more
precisely, from the latter’s inadequacy (Poincaré 1894, 374).17
The debate over logicism is also the context for Poincaré’s argument that
mathematical induction is required to prove the consistency of axiom systems
(construed as implicit definitions). Poincaré makes explicit that it is necessary
to “consider all the propositions deducible from [proposed axioms] considered
as premises, and . . . show that, among these propositions, no two are con-
tradictory,” only in case the axioms in question cannot be “directly” proved
consistent by finding an “example” of which they are provably true (1908/
1929, 454). Logicist treatments of number must take the first route because
the axiom “The successor of an integer is an integer” cannot be proved true of
any example, that is, finite initial segment of the natural number series (463).
But when a mathematical notion is to “apply to the physical sciences, . . . it is no
longer a question of” the notion as “rigorously” defined (by the axioms), “but of
a concrete object which is often only a rough image of it.” The claim that “this
object satisfies the definition, at least approximately,” while vulnerable to em-
pirical refutation, implies the consistency of the axioms-cum-definition (455).
Since geometrical notions are supposed to have physical application, the con-
sistency of their defining axioms can be proved this way; induction on the
axioms’ consequences is not necessary. Hence, geometry need not depend on
arithmetic for proof of its consistency.

16. A main target for Poincaré was Louis Couturat, who sees this sort of project as an obvious
continuation of rigorization in analysis (1905/1979, 5).
17. Poincaré takes issue specifically with Ballue’s view that the theory of pluralities, and thus all of
arithmetic, is empirically justified.

288

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

Stump finds further evidence of a formalist view in Poincaré’s admiration of

Hilbert’s treatment of geometry. As Stump observes (1996, 485), in “Mathe-
matics and Logic: II” Poincaré claims that “Hilbert has given a complete proof ”
that the Euclidean axioms, considered as a definition of “straight line,” are con-
sistent (1908/1929, 468). Stump also cites Poincaré’s largely positive review of
Hilbert’s Foundations of Geometry. But Poincaré’s praise is tempered in ways
Stump does not discuss.
In “Mathematics and Logic: I,” Poincaré claims that Hilbert’s approach
is fitting given his aim, which is to “reduce to a minimum the number of the
fundamental assumptions of geometry and completely enumerate them” (1908/
1929, 451). Poincaré clearly considers this aim worth pursuing; he argues at
the start of his review that efforts to “construct a universal logical symbolism
and to reduce mathematical reasoning to purely mechanical rules” yield both
clarity and important generalizations (1902/1903, 1–3). But he also stresses
how “disastrous” this approach would be “in teaching and how hurtful to men-
tal development; how deadening it would be for investigators, whose originality
it would nip in the bud” (5). These constituencies, presumably, need to un-
derstand “how and why” mathematical questions arise. Their concerns are ad-
dressed by the philosopher, who retains “the right to investigate the origin” of the
first principles even in case “all the theorems [can] be deduced by procedures
purely analytic, by simple logical combinations of a finite number of assump-
tions” (Poincaré 1908/1929, 451). In other words, formalization of geometrical
proof does not obviate “philosophical” investigation of the origin and rationale
of the axioms. Now such investigation is bound to consider whether the axioms
are exemplified by any objects. But then this sort of inquiry, necessary in its
own right, makes it unnecessary to formally prove the axioms’ deductive con-
sistency. I think that this is the point of Poincaré’s contrast between Hilbert’s
proof of the axioms’ consistency and “our” knowledge that they are consistent,
which is based on the “existence” of geometry.18
Poincaré also suggests that Hilbert could have taken account of the “actual
psychological origin” of the axioms, without “sacrificing” their “logical char-
acter,” if he had defined a congruence relation for figures. This would “pre-
serve,” in its “natural role” and with the “importance [that] Helmholtz was the
first to understand,” the idea of the displacement of a rigid figure (Poincaré
1902/1903, 7–8).19 Relatedly, Poincaré complains that Hilbert’s lack of re-

18. “We know well that [the axioms are consistent], since geometry exists; and Professor Hilbert also
answers in the affirmative” (Poincaré 1902/1903, 8).
19. It would also obviate the axiom asserting that if an angle and the sides enclosing it are equal in
two triangles, so are the remaining angles (Poincaré 1902/1903, 8).

289

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

gard for the origin and foundation of the axioms leads him to neglect group-
theoretic formulations of the geometries he characterizes and to “slur over” the
relationships among these groups (22). We now turn to the group-theoretic
considerations, whose importance for the foundations of geometry is stressed
here. They seem to show another relation of dependence between arithmetic
and geometry, which would reconcile Poincaré’s accounts of these sciences; I
will argue, however, that in fact they deepen the tension between his views.

6. Displacements and the Concept of Group

Beginning with Roberto Torretti, interpreters have shown how framing Poin-
caré’s view group-theoretically clarifies its scope and basis. Torretti thereby ex-
plains (1978, 336–37) why in Science and Hypothesis (specifically in the trans-
lation argument on 59–60, the appeal to the “world enclosed in a giant sphere”
on 75–78, and the discussion of experiment in chap. 5) Lobachevskian geom-
etry appears as the only alternative to Euclidean. Stump (1991, 644–46; 1996,
481–82) and Ben-Menahem (2006, 47) seem basically to agree that, while the
“dictionary” argument of chapter 3 is intended to show that the correct ge-
ometry cannot be determined a priori, Poincaré’s account of how the Euclidean
group is chosen over the Lobachevskian is directed against empiricism.20 Most
recently, Friedman has explained why Poincaré finds it so natural to treat axi-
omatized geometries as alternative descriptions of experience (as when he moves
from the intertranslatability of Euclidean and Lobachevskian geometry to their
empirical equivalence). Friedman observes that Poincaré “worked squarely
within the modern group-theoretic tradition of Felix Klein and Sophus Lie—
where . . . the general concept of group is characterized axiomatically (by the
associative law, the existence of an inverse [etc.]), but our primary focus is on
the concrete operations which instantiate this general concept” (2010, 642–
43). We are concerned, in particular, with operations (of transporting instru-
ments and observers) through which distances are measured.
But for all the clarity it brings, the view that geometry is based on the choice
of a group is not developed in Poincaré (1902/1929). The fourth and fifth

20. Stump formulates the arguments of Science and Hypothesis in terms of “arbitrary continuous
point-transformations” rather than groups because, in his view, there is “no doubt that Poincaré gave up
the notion of group as a primitive in geometry after reading Hilbert and that he took topology as basic”
(1996, 489). This appears odd in light of the quotation at the end of sec. 5. Stump may have in mind
Poincaré’s claim that Lie was constrained to regard space as a Zahlenmannig faltigkeit because he studied
the “common properties” possessed by “all the groups which one can imagine,” whereas Hilbert has
shown this restriction to be artificial. But Poincaré’s response seems to be to widen the group-theoretic
viewpoint beyond “continuous groups properly so called” (1902/1903, 21).

290

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

chapters of Science and Hypothesis both end by asserting that the concept of a
group in general “pre-exists” in our minds, while geometry is founded on the
choice of a particular group as the “standard to which to refer natural phenom-
ena” (1902/1929, 79 and 91). Poincaré’s account of how geometry is founded
on the notion of a group appears sufficient to establish at most the apriority
claim, leaving the conventionality thesis unsupported. This section will focus
on the argument for the apriority claim and show how the argument seems to
place the intuition that founds arithmetic (namely, of the power to repeat an
operation) at the basis of geometry. I will later claim that when the argument
for the conventionality thesis is taken into account, this understanding of
arithmetic’s relationship to geometry appears too simple.
Chapter 4 of Science and Hypothesis explains how we arrive at the idea of
space, on the assumption that for us to represent space is to “localize” objects at
points of space, by “represent[ing] to ourselves the movements it would be
necessary to make to reach” each object (Poincaré 1902/1929, 70).21 Poincaré’s
account is framed in terms of “aggregates of impressions,” which can be un-
derstood as multimodal (visual, tactual, etc.) experiential states associated with
perception of particular objects. “Qualitative identity” as it pertains to aggre-
gates is not the qualitative sameness of each aggregate’s constituent sensations
to one another, but qualitative sameness of entire states (experienced at dif-
ferent times) to one another. Poincaré first argues that changes in these ag-
gregates can come about either “involuntarily and without experiencing mus-
cular sensations” (71) or voluntarily and with such sensations; the first are
taken to represent changes (called “external”) pertaining to outer objects, and
the second are taken to represent movements of one’s own body. A “dis-
placement” is defined as a change of the first sort that can be “corrected,” that
is, reversed, by a change of the second sort. Central to Poincaré’s account is
that we “consider two external changes as the same displacement,” even if the
aggregates of impressions with which the first begins and ends differ from the
second’s beginning and end, if they “can be corrected by the same correlative
movement of our body” (identical in the muscular sensations accompanying
the correcting change, rather than the impressions that begin and end it; 73–
74). Thus, each displacement is what we would now call an “equivalence class”
of external changes (see Torretti 1978, 343; Johnson 1981, 92). Poincaré claims
that “a mind already familiar with geometry” would reason that for such a re-

21. In chap. 4 of Science and Hypothesis, Poincaré explicitly defends the assumption that for us to
“localize” an object at a point in space is to represent the movements required to reach the object (1902/
1929, 70). That we represent space only by locating objects (in this way) is argued in Poincaré (1913/
1963, 15). See also Poincaré (1898, 24–28) and Poincaré (1908/1929, 413–18).

291

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

versal to occur, “the external object, in the first change, must be displaced as is a
rigid solid, and so it must be with the whole of our body in the second change
which corrects the first.” Subjects who lack the notion of space, however, cannot
thus determine “a priori whether compensation is possible,” but must learn
from experience that it sometimes happens (Poincaré 1902/1929, 72).
Poincaré concludes that the “object of geometry” is “constituted by” the laws
of displacements, which “mathematicians sum up in a word by saying that dis-
placements form ‘a group.’” But the only law he mentions is that of “homoge-
neity,” which asserts that if more than one external change leads from a certain
qualitatively identical “totality of impressions” to another qualitatively identical
totality, then every movement that corrects one of these changes is “the same” in
the sense that it is accompanied by the same muscular sensations (1902/1929,
74–75).
Poincaré seems mainly concerned to show that the assertion of homogene-
ity is based on the same intuition that grounds arithmetical reasoning. He
restates the law as that “a movement which has once been produced [specifi-
cally, a movement correcting an external change from A to B] may be repeated
a second and a third time, and so on, without its properties varying.” Poincaré
then claims that “thanks to the law of homogeneity,” mathematical reasoning
“has a hold on the geometric facts” through the possibility of “repeating in-
definitely the same operation,” the importance of which was shown in chap-
ter 1 (1902/1929, 75). Since the assertion that displacements form a group is
thus seen to rest on the same intuition as the “rule of reasoning by recurrence,”
it can be accorded the same status: namely, a necessary “imposition” that only
“affirm[s] a property of the mind itself ” (40).
This argument appears subject to the same worry as Poincaré’s attempt to
base the concept of a continuum on the intuition of indefinite iteration. Poin-
caré holds that by representing (as a pattern of muscular sensation) the move-
ment required to restore an aggregate of impressions, we represent the point at
which the object corresponding to the impressions is located. But by indefi-
nitely iterating representations of such patterns, we can represent at most a
dense collection of points, not a truly continuous space.22 We have seen, how-
ever, that the existence claims implicit in the axioms of Euclidean geometry are

22. Poincaré could be taken to acknowledge this problem when he claims that “the geometries of
Lie” (each of which “corresponds” to the “totality” of possible combinations of “a certain number of
transformations,” e.g., displacements) remain “subject to the forms of arithmetic and analysis.” But the
emphasis should probably be on the restriction to “forms of analysis,” for Poincaré here contrasts Lie,
who “confined himself to the study of continuous groups properly so called, to which the rules of the
ordinary infinitesimal analysis apply” (1902/1903, 21), with Hilbert, who “frees the theory of groups
from all appeal to the principles of the differential calculus” (23).

292

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

satisfied in a space every point of which is constructible by ruler and compass

operations. Friedman has suggested that understanding these constructive op-
erations in terms of a group (of motions) illuminates Poincaré’s view of the role
of arithmetical intuition in geometry. Specifically, the Euclidean group of mo-
tions can be understood as “the basis . . . of the procedure of construction of
straight edge and compass,” which “iteratively generates the domain of Eu-
clidean geometry.” Friedman stresses that insofar as “we are concerned . . . only
with constructive existence claims and the potential infinite,” this domain is
“precisely analogous . . . to the natural numbers” (2000, 205). Friedman does
not spell out the role he sees for intuition (of the possibility of indefinite
iteration), but presumably it serves to justify geometrical theory by validating
the theory’s existential commitments.
For his part, Friedman is most concerned to explain how the “connection
between groups of rigid motions and geometrical construction” can be said to
“subsist” in “all three classical cases of [spaces of ] constant curvature,” so that
the necessary “imposition” of the notion of a group does not restrict us to Eu-
clidean geometry. The key point is that in all three cases, “one can formulate
an elementary geometry where, in place of the Dedekind continuity axiom, one
simply has an axiom of intersection for straight lines and circles,” and which has
as its domain “precisely those points generated by straight-edge and compass
constructions in the sense of each of the geometries in question” (Friedman
2000, 206).23 The possibility of any such iteratively generated domain is ver-
ified by intuition (of the possibility of indefinite iteration). So long as we can
make out how the motions that generate (points in) each domain form a group
that could be chosen as “the standard to which we shall refer natural phenom-
ena,” we will have an attractive explanation of how intuition “imposes” the
“general group concept” on us, while leaving this choice open (Poincaré 1902/
1929, 79).

7. The Argument for the Conventionality of Geometry

in Science and Hypothesis
Unfortunately, Science and Hypothesis does not contain a sustained argument
that any other geometry (of constant curvature) could be chosen as the “stan-

23. The connection with groups of rigid motions is elaborated by representing each elementary
geometry’s domain analytically as “an appropriate space over a Euclidean subfield of the reals” (see n. 8
above). These analytic representations are “induced by corresponding analytic representations of pro-
jective geometry” when “all three cases are viewed in accordance with the Cayley-Klein program as
embedded within projective geometry, so that, in particular, the three different groups of motions
appear as subgroups of the projective group” (Friedman 2000, 206).

293

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

dard.” According to Poincaré, nothing “prevents us from imagining a series of

representations, similar in all points to our ordinary representations, but suc-
ceeding one another according to laws different from those to which we are
accustomed” (1902/1929, 75). But the example Poincaré gives for illustration
seems to fail of its purpose.
The “world enclosed in a great sphere” is introduced precisely to show how
beings could be “educated” in an environment in which representations suc-
ceed according to different laws, and thus be led to develop a non-Euclidean
geometry (Poincaré 1902/1929, 75). For these beings, the external changes sub-
ject to correction by internal changes are (considered from “the point of view
of our ordinary geometry”) deformations of solids, in “exact conformity” with
the law by which temperature varies with position.24 If the beings then formu-
late a geometry, Poincaré claims, “it will be the study of the changes of posi-
tion which they will thus have distinguished” (not, “as ours is, the study of the
movements of our rigid solids”), and will be non-Euclidean (77–78). Poincaré
maintains that if we (educated as we were “in our actual world”) “were sud-
denly transported into this new world, we should have no difficulty in referring
its phenomena to our Euclidean space,” and likewise that beings educated there
would, if transported into our environment, “relate our phenomena to non-
Euclidean space.” But he further claims that “with a little effort” (66) we, too,
could relate the phenomena of our world to non-Euclidean space, although “it
is certain we should find it more convenient not to change our habits” (80).
(Poincaré presumably also holds that it is possible, but inconvenient, for the be-
ings to relate the phenomena of their world to Euclidean space.) The example is
thus supposed to bear out Poincaré’s view that the choice of a convention for
spatial measurement, while “guided by experimental facts,” “remains free” (65),
specifically in the sense that both ways of measuring distance remain available
to a creature “educated” in either environment.
Poincaré does not, however, establish that the changes of position “distin-
guished” by the creatures comprise a group. Of course, he is thinking of these
transformations under an abstract mathematical description, on which it is
clear that they do comprise one.25 At issue, however, is whether the displace-
ments considered as measuring operations are subject to laws such that they

24. A sentient being’s “impressions will be modified by [such a change] of the object, but he [the
being] can reestablish them by moving in a suitable manner. It suffices if finally the aggregate of the
object and the sentient being, considered as forming a single body, has undergone [a change in
accordance with the law of thermal dilation]. This is possible if it be supposed that the limbs of these
beings dilate according to the same law as the other bodies of the world they inhabit” (Poincaré 1902/
1929, 77).
25. On the background to this argument in Poincaré’s mathematical work, see Gray (2013, 47).

294

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

form a group. In his most extensive discussion of the group-theoretic basis of

geometry (“On the Foundations of Geometry,” a Monist article of 1898 not
reprinted in Science and Hypothesis), Poincaré explains that if displacements
form a group, it follows that “space must be homogeneous,” meaning that all its
points must be “capable of playing the same part.” For the points to “play the
same part” is, more precisely, for one’s body and other objects to remain “ca-
pable of the same movements as before” after being “transported from one
point to another” (Poincaré 1898, 12). But it need not be granted that a trans-
ported body remains capable of the same movements. To be sure, its move-
ments may remain subject to correction by internal changes that are “the same”
in Poincaré’s sense, that is, accompanied by identical muscular sensations; but
the movements need not be considered “the same” in the sense that they mea-
sure the same distance. (For a concrete illustration, consider the motion of
sliding a ruler a distance equal to its length, as when “laying off” feet, meters, or
yards.) To make it easy for us to “refer the phenomena” of the sphere-enclosed
world to Euclidean space, Poincaré describes them “from the point of view of
our ordinary geometry,” namely, as involving deformation (of solids). But once
this point of view is assumed, how can the reader be compelled to abandon
it?26 That is, should we not just consider it the beings’ mistake to suppose that
in their world, a body (such as the ruler) occupies the same volume of space
after being moved? Then the movements they count as displacements would
not satisfy the law of homogeneity and so would not form a group, by our lights.
Poincaré’s argument would then fail to establish that an alternative conven-
tion for (spatial) measurement is available to us.
Of course, to deny that a body occupies the same volume of space after
transport, we must presuppose some other standard by which to compare the
quantity of spatial regions. Poincaré’s view that space does not “possess geo-
metric properties independent of the instruments used to measure it,” but is
rather “amorphous” (1913/1963, 17), gives him grounds to deny the legiti-
macy of such a comparison. Indeed, his argument has been taken to turn on
this point.27 But this claim does not appear in Science and Hypothesis,28 and it
appears to express the sort of metaphysical thesis—about the “things them-
selves”—that Poincaré there declares beyond our ken (in, e.g., the introduction

26. The question is particularly pressing given Poincaré’s claim that he has “describe[d] the fantastic
worlds above imagined without ceasing to employ the language of ordinary geometry. And in fact, we
should not have to change it if transported thither” (1902/1929, 80).
27. Some interpreters who hold this view are Grünbaum (1973, 27), O’Gorman (1977, 305), and
Stump (1991, 647).
28. It does appear in a 1903 Revue de métaphysique et de morale article, “L’espace et ses trois
dimensions.” The relevant portion is reprinted as “The Notion of Space,” chap. 3 of The Value of Science.

295

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

and chap. 10).29 On the other hand, the claim could be taken epistemologically,
as denying only that metric properties can be attributed to space “independent
of the instruments used to measure it,” not that it has them. The claim then has
a basis in Poincaré’s argument that experiments bear on only “the relations of
bodies to one another,” not “the relations of bodies to space” or “mutual rela-
tions of different parts of space” (1902/1929, 86). However, it appears to hold
only for the imaginary beings within the sphere-enclosed world; we, on the
other hand, come to know the metric properties of its space through Poincaré’s
description. So it again seems that the beings’ conventions for measurement are
not available to us.

8. Poincaré’s Full Argument for the Conventionality of Geometry

As it appears in Poincaré (1902/1929), the argument that the displacements
studied in our geometry form a group is too brief for us to see how it applies to
the movements “distinguished by” the imaginary beings. Therefore, it does not
seem adequate to establish the thesis that geometry is founded on a choice
among groups. The fuller version of the argument in the 1898 Monist article
extends more readily. I contend, however, that this way of defending the view
that geometry involves the choice of a group leaves little room to argue that ge-
ometry depends on intuition in the same way as arithmetic.
The argument of Science and Hypothesis begins with the observation that
some external changes are in fact corrected by internal changes. Poincaré claims
that it is on the basis of this “experimental fact” that we begin to classify
external changes into “changes of position,” the ones capable of being so cor-
rected, and “changes of state” (1902/1929, 72). He also seems to take as basic
that the correction is exact, in the sense that the relevant internal changes issue
in an exactly similar aggregate of impressions. For in his view, we notice that
some bodies typically, and others “only exceptionally,” “undergo displacements
susceptible of being corrected by a correlative movement of our own body,”
and work out that the reason we (usually) cannot “bring back our sense-organs
into the same relative situation with regard to” bodies of the latter kind is that
they change both position and shape. We can then learn to “decompose” such
“bodies of variable form” into smaller elements and on this basis distinguish

29. One might argue that Poincaré’s denial that space has metric properties is not a claim about
“things themselves,” since he also claims that “space is only a word that we have believed a thing” (so its
concept can be analyzed without “sacrificing reality to I know not what phantom”; 1902/1929, 5). But
then the notorious problem with Kant’s view would extend to Poincaré’s, namely, how it can be
maintained that space is not among the things-in-themselves, while denying all knowledge of such
things.

296

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

“deformations” (in which “each element undergoes a mere change of posi-

tion, . . . but the modification undergone by the aggregate is more profound
and is no longer susceptible of correction by a correlative movement”) from
other “changes of state.” But Poincaré emphasizes that the notion of such a
partially correctible change requires sophistication and “could not have arisen
if the observation of solid bodies had not already taught us to distinguish
changes in position” (73). Now, given this observational knowledge of “changes
of position” and our ability to exactly “correct” them, we have no discretion as
to whether two (or more) such changes constitute the same displacement: they
do if and only if each is exactly corrected by the same voluntary movement
(same, that is, with respect to the muscular sensations accompanying it).
In contrast, in the Monist article Poincaré stresses that the partition (as we
might call it) of external changes into equivalence classes (each of which con-
stitutes a displacement) is “not a crude datum of experience, because the afore-
mentioned compensation of the two changes, the one internal and the one
external, is never exactly realized” (1898, 9). Poincaré grants that it is “due to
experience,” namely, to observation “that the compensation has approximately
been effected,” that the mind has “occasion to perform” an “operation [of ]
identifying two changes because they possess a common character,” but it does
so “in spite of their not possessing it exactly” (9). Specifically, the mind adopts
the convention of considering a change that “obeys [the laws of displacements]
only approximately” as “the resultant of two other component changes.” The
first component change is “a displacement rigorously satisfying the laws,” which
would be the motion of a perfectly rigid solid body, and the second component
is a “qualitative alteration” of the body, such as a warming, cooling, or warp-
ing, that accounts for the failure (of the internal change) to restore the original
situation (11). (It is presumably because “natural solids” in our environment
undergo such alterations that compensation is “never exactly realized.”)30
According to this view, whether an external change is corrected by an in-
ternal change, and thus whether two external changes compose the same dis-
placement, is decided by a convention. To say (“from the point of view of our
ordinary geometry”) that in the sphere-enclosed world bodies do not pre-
serve their dimensions when moved assumes our resolution of the question;

30. In chap. 4 of Science and Hypothesis, Poincaré asserts that “in our world,” “natural solids”
doubtless “undergo variations of form and volume due to heating and cooling.” He then notes that we
“neglect these variations in laying the foundations of geometry, because besides their being very slight,
they are irregular and consequently seem to us accidental”; but in the sphere-enclosed world, the
variations “follow regular and very simple laws” and thus have a greater claim on geometers’ attention
(1902/1929, 76). The crucial difference (from the 1898 treatment) is that Poincaré does not make
explicit that whether to neglect or consider the variations is a matter of convention.

297

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

but since this is just our convention, other resolutions remain available to
us. Poincaré makes explicit that the convention by which external changes are
identified with each other, as the same displacement, “further brings us to rec-
ognize that two displacements are identical,” so that the same displacement can
be repeated, which now “permits measurement where formerly pure quality
alone held sway” (1898, 9). Thus, the lacuna in the argument of Poincaré (1902/
1929) is filled.
In the 1898 article, Poincaré emphasizes the spontaneous character of the
“operation” by which changes are identified, describing it as an “endeavor” to
“insert the crude results of experience into a pre-existing form, or category” (9).
Poincaré’s use of Kantian terminology (see Heinzmann 2009, 185–86) indi-
cates that this operation should be subsumed under his general account of how
science is shaped by the demands of our intellect. Indeed, as I will now show,
the notion of “group” serves as a prime example of a formative contribution
of ours. The problem this will raise is that Poincaré seems to have no way to
disallow similar contributions in the foundations of arithmetic, which would
make arithmetic’s basis logical.

9. The Importance of Linguistic Innovation for Science

The Value of Science fills out Poincaré’s account, which Science and Hypothesis
only adumbrates, of our active role in the generation of scientific facts. So while
in Science and Hypothesis Poincaré stresses the importance of generalization for
science (1902/1929, 117–20 and 127–29), in The Value of Science he makes
explicit that science is “before all a classification, a matter of bringing together
facts” that appear as separate but are “bound together by some natural and
hidden kinship” (1905/1929, 349). Poincaré claims to have shown this by dis-
tinguishing between scientific facts and “crude” or “rough” ones.31 So long as a
fact is “completely in the rough,” it remains “individual,” that is, “completely
distinct from all other facts.” Facts become scientific by being “enunciated,”
where the enunciation “would suit an infinity of other facts” (327). Poincaré
explains that what “the scientist creates in a fact is the language in which he
enunciates it,” and that “the scientist intervenes actively in choosing the facts
worth observing”—in particular, the facts whose “verification is the confir-
mation of a law,” because they were predicted (332). These forms of creative
activity are two sides of the same coin, because by “enunciating,” the scientist

31. Poincaré claims to “have explained this” in chap. 10 of The Value of Science, where he contrasts
his views with Édouard Le Roy’s, in part by explaining how crude and scientific facts should be
distinguished (1905/1929, 325–33).

298

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

selects out the myriad facts whose verification confirms the enunciation (or a
law presupposed by it).32 They are vital for science, since in Poincaré’s view,
merely to “inventory” isolated facts without connecting them “would have
no value for us”: it would neither enable foresight nor satisfy “our craving for
order and harmony” (1902/1929, 119; see also Poincaré 1905/1929, 332; Poin-
caré 1908/1929, 535).
Poincaré sums up his view with the claim that a scientific fact is “the crude
fact translated into a convenient language” (1905/1929, 330). In chapter 5
(“Analysis and Physics”) of The Value of Science, he uses the example “elec-
tricity” to illustrate how language highlights the “kinship” among facts: who-
ever invented the word “had the good fortune to implicitly endow physics with
a new law, that of the conservation of electricity, which, by a pure chance, has
been found exact, at least until now” (281). This is part of Poincaré’s argument
for the relevance of mathematics to physics, which—in contrast to his treat-
ment of this theme in Poincaré (1902/1929)—stresses that such felicitous
conventions are mathematical in origin.33 Poincaré argues that in the generaliza-
tions required “to get laws from experiment,” we need something to steer our
choice from among the innumerable ways in which any “particular truth” can
be “extended.” Our choice must be based on analogy, but analogies may be either
banal (“those which strike the senses,” e.g., “of colors or sounds”) or profound
(such as “likening light to radiant heat”; Poincaré 1905/1929, 281). According
to Poincaré, we grasp the “true, profound analogies” by means of the “mathe-
matical spirit [esprit],” which “disdains matter to cling only to mere form” and
so teaches us “to give the same name to things differing only in material” (282).
That language aids in finding analogies is seen already from Poincaré’s
“simple example that comes first to mind” of how “the possibility of ” gener-
alizations is perceived. The example is “an algebraic formula which gives us the
solution to a type of problem when finally we replace the letters by numbers”
(Poincaré 1908/1929, 372). Poincaré goes on to assert that mathematics “is

32. In Poincaré’s example, the statement “there is running in this circuit a current of so many
amperes” means both that “I shall see [a] spot come to the division a” if “I adapt to this circuit” a certain
galvanometer and that I shall see a spot “go to the division b” if I adapt an electrodynamometer to the
circuit, and can mean “still many other things” because the current can also “manifest itself ” by thermal
or luminous effects. The statement can “suit” the different facts that it conveniently packages together
because it presupposes “a law according to which whenever such a mechanical effect shall happen, such a
chemical effect will happen also,” which is confirmed by verifying any of the facts (Poincaré 1905/1929,
329).
33. Poincaré’s account of the “service rendered” by mathematical to experimental physics in
Poincaré (1902/1929, 127–30) relies on a vague notion of “direct[ing] generalization.” According to it,
language is a source of “preconceived ideas,” but Poincaré does not make clear whether these are
“consciously introduced” ideas or the pernicious ones “counterbalanced” by them (129).

299

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

the art of giving the same name to different things,” that is, that its distinc-
tive achievement is to highlight commonalities through language. Well-chosen
words enable us to see the applicability of “all the proofs made for a certain
object” to more objects, and also “do away with the exceptions from which the
rules stated in the old way suffer,” exceptions that “are pernicious because they
hide the laws.” As examples of such exception-eliminating innovations, Poin-
caré gives “negative quantities, imaginaries, points at infinity,” and, from phys-
ics, “energy,” which “also made the law by eliminating the exceptions, since it
gave the same name to things differing in matter and like in form” (375).34
Poincaré puts “group” among the “words that have had the most fortunate
influence.” We now know, he says, that “in a group the matter is of little in-
terest, the form alone counts, and that when we know a group we thus know all
the isomorphic groups.” The words “isomorphism” and “group” “condense in a
few syllables this subtle rule and quickly make it familiar to all minds” (Poin-
caré 1908/1929, 375–76), and thereby cue us to extend arguments and results
from a given group to all those isomorphic with it.
Seeing how the use of the word “group” exemplifies the activity by which
scientists “create facts” helps, I think, to explain why it should be up to our dis-
cretion to consider the motions of measuring instruments (in particular, in the
sphere-enclosed world) as a group.

10. Poincaré’s Argument against Definitions “by Postulate”

in Set Theory
The tension with Poincaré’s account of arithmetic arises because he seems there
to deny us the same prerogative.
Indeed, there Poincaré seems to flatly rule out that felicitous expression
could be a source of novel knowledge. If “the conclusion is nothing but the
premises translated into another language,” then, he claims, the reasoning is
“sterile” (1902/1929, 33). In contrast, in “The Future of Mathematics” Poin-
caré says that the word “energy” has been “prodigiously fecund [ fécond ]” (1908/
1929, 375).
The conflict runs much deeper than these opposing metaphors, as can be
seen from Poincaré’s criticism of logicism in arithmetic. Poincaré’s argument
in Science and Hypothesis that arithmetical notions (in particular, “reasoning
by recurrence”) cannot be “reduced” to logical ones seems to underestimate the
logicist’s resources. For Poincaré supposes that a logical foundation must pro-

34. This is perhaps a better illustration than “electricity” of how the introduction of a name makes
salient the conservation of a quantity and so leads to a law.

300

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

ceed by substitution from explicit definitions. As Michael Detlefsen observes,

this gives his argument force against only a “Leibnizian form of logicism,” on
which the truths of mathematics are derivable from logical identities by such
substitutions. The argument would not apply to a view (such as Frege’s or Rus-
sell’s) on which the derivation of consequences from a sentence S is not “just
like a definitional transform of S,” but may instead result in a sentence express-
ing a proposition different from that expressed by S (Detlefsen 1992, 359). For
our purpose—which is to see the difference between Poincaré’s accounts of
arithmetic and geometry—Poincaré’s treatment of definitions is more relevant
(than of rules of inference). In a 1912 essay (reprinted as “Mathematics and
Logic” in Last Essays), Poincaré considers a form of definition that, if legiti-
mate, would have the “creative virtue” by which mathematics transcends tau-
tology. He expounds a view (which is, however, not necessarily his own) that
denies its legitimacy.35 Our problem is that according to this view, it would also
seem illegitimate to claim that a collection (such as the class of displacements)
forms a group.
In the 1912 essay, Poincaré contrasts “direct” with implicit definition (which
he calls “definition by postulates”). Each typically gives the genus to which the
object belongs and the marks or differentiae distinctive to it. But in the former
the “specific difference” describes the object in (finitely many) words,36 while
in the latter the “specific difference” is expressed by means of “a ‘postulate’
which the object being defined must satisfy” (Poincaré 1913/1963, 69).
If all definitions were direct, Poincaré argues, we could substitute a defi-
nition for each term “in any proposition whatsoever.” When the substitution
was completed, the proposition would reduce to an identity “and then would
be merely a more or less cleverly disguised tautology” (or else it would not re-
duce to an identity, which would show that our reasoning was not “purely log-
ical” after all). Thus, “the impotence of pure logic could not be contested”
(Poincaré 1913/1963, 69).
Definition by postulate offers a way out of the impasse. In general, though,
it is subject to the objection raised against the attempt to define the natural

35. In “Mathematics and Logic” Poincaré claims to present the opposing views “from a purely
objective point of view, just as if we ourselves were not a member of these schools” (1913/1963, 65).
Although much in his writings (particularly the 1909 essay “The Logic of Infinity”) accords with the
view he attributes to the “pragmatists” or “idealists,” here he distances himself from it, as McLarty
observes (1997, 109–10).
36. Poincaré allows that direct definitions may be “incomplete” in the sense that “they do not define
a particular thing but rather an entire genus.” But on the restrictive view he is expositing, “it is necessary
to understand therein the set of the particular objects which satisfy the definition and which could
finally be defined in a finite number of words” (Poincaré 1913/1963, 69). He thus appears to single out
description in a finite number of words, as essential for direct definition.

301

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

numbers as satisfying the Peano postulates: the definition must be proved con-
sistent, either by finding an example (model for the postulates) or by induction
on their deductive consequences (see sec. 5 above).37 Moreover, Poincaré ob-
serves that although we cannot substitute the definition for the defined term in
a proposition, we can “eliminate this term between the proposition and the
postulate that serves as its definition.” (If we thereby obtain an identity, the
proposition was merely tautologous; if not, then the proposition “cannot be
proved by means of pure logic.”) But Poincaré’s focus is the special case in
which “the postulate . . . is a relation between the object to be defined and all
the individual objects of a genus of which the object defined is itself supposed
to be a member.” According to “Cantorians” who propound such definitions,
“the genus . . . is already given, consequently we know all its members,” and
the definition serves only “to distinguish from among these members the one
that has the expressed relation” to the others (Poincaré 1913/1963, 70). In this
way, the “example” that proves the definition’s consistency is already at hand.
And (as we are about to see in more detail) the interplay between the defined
term and the defining conditions precludes “eliminating” the former in favor of
the latter. Thus, according to Poincaré, “if this type of definition were per-
mitted, logic would no longer be sterile.” He continues, “we are amazed at the
power which a word can possess. Here is an object from which nothing could
be derived until it had been christened; all it needed was to be given a name and
it worked wonders” (71).
Now, we have seen that according to Poincaré, a “christening” can make it
possible to derive novel consequences (as in the cases of “energy” and “group”).
But this form of definition is not legitimate from the point of view he adopts in
the 1912 essay. Poincaré explains that contradiction can arise unless the genus
is “posited ne varietur as the Cantorians do.” Specifically, if the genus is re-
garded as “a collection capable of being increased indefinitely, whenever new
members are formed which possess appropriate characteristics,” then when the
genus is “completed by including the [defined object],” the object picked out
by the definition within the “genus thus completed” may not be the object
picked out by the definition on its first application (Poincaré 1913/1963, 71–
72). In a precursor essay, “The Logic of Infinity” (first published in 1909,
reprinted in Last Essays), Poincaré explains why, in general, such specifications
“will never be fixed”:38

37. Specifically, Couturat’s attempt (Poincaré 1908/1929, 463). On this background, see Goldfarb
(1988, 65).
38. Poincaré begins “Mathematics and Logic” with a reference to “The Logic of Infinity” and the
dispute it provoked (1913/1963, 65n1).

302

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

Let us suppose that it is desired to classify the elements of [an infinite

collection] and that the principle of the classification rests on some re-
lation of the element to be classified with the entire collection. . . . There
is no actual infinity, and when we speak of an infinite collection, we un-
derstand a collection to which we can add new elements unceasingly
(similar to a subscription list which would never end . . . ). For the
classification could not properly be completed except when the list was
ended; every time that new elements are added to the collection, this
collection is modified; it is therefore possible to modify the relation of
this collection with the elements already classified; and since it is in ac-
cordance with this relation that these elements have been arranged in
this or that drawer, it can happen that, once this relation is modified,
these elements will no longer be in the correct drawer and it will be nec-
essary to shift them. (1913/1963, 47)

Since “it will never happen that there will not be new elements to be intro-
duced” (47), we are constantly under threat of having to revise our classifi-
cation. For this reason, such classifications are unacceptable to those “who
believe that the only objects about which one may reason are those which can
be defined in a finite number of words.” Poincaré says he will defend this
restriction (48). Indeed, we should expect him to reject the “Cantorian” as-
sumption that the “drawers” for objects exist “independent of the curators
charged with the task of arranging the objects” (68), given the importance he
puts on our practices of naming and classifying.

11. Comparing the Group-Theoretic Foundation of Geometry

with the Set-Theoretic Foundation of Arithmetic
The problem we now face is that the criticism seems to apply equally well to
linguistic innovations that Poincaré sanctions. For the assertion that a collec-
tion composes a group has the same logical structure as the definitions he
rejects: a group is characterized by the existence of an “identity element” that
bears a certain specified relation to every object in the group.39
To be sure, in Poincaré’s application of this notion to geometry the relevant
collection is in a sense “given,” specifically as the totality of displacements. One
might suppose that insofar as this application of the notion involves only se-

39. Namely, the identity element e satisfies the equation a · e = e · a = a for every a, where “·” denotes
the operation by which the group is defined.

303

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

lection from within an antecedently given collection, it is less prone to incon-

sistency. But in general, Poincaré sets little store in the “limitation of size” ap-
proach to avoiding paradox.40 More importantly, the classification of changes
as displacements involves the convention (as we have seen) that they are to be
regarded as satisfying the group axioms. So why couldn’t a “Cantorian” like-
wise stipulate that the incorporation of new elements into the genus must be
held to preserve relations between its members?
It might be thought that what Poincaré objects to in the Cantorians’ pro-
cedure is their treating an infinite series of operations (of, specifically, “adding
elements to the collection”) as completed, whereas in the group case the ex-
tension into the infinite has a basis, in the appeal to intuition that grounds the
assertion of “homogeneity.” Then the Cantorian procedure would be legitimate
if we had intuition of a power to continue adding elements. But I think that
in Poincaré’s view, intuition would be needed already to justify earlier stages
of the set-theoretic development of arithmetic, and once its role is acknowl-
edged, definitions of the sort he criticizes remain without basis. A standard way
of developing arithmetic in set theory is to define the natural numbers as the
“minimal closure” of the “inductive sets,” that is, the smallest set containing
everything common to those sets that contain 0 and contain the successor of x
if they contain x.41 It seems clear that on Poincaré’s view, we need intuition to
be justified in conceiving and affirming the existence of inductive sets; grant-
ing that we can, it is a further question whether we may quantify over all in-
ductive sets in the definition of one of them. In the same way, the quantifi-
cation over all elements of the group in the designation of the identity element
will remain in need of justification.
Now the notion of infinite set had been shown to give rise to contradiction—
according to Poincaré, precisely through the “mutability” of such classifications
(1913/1963, 45)—and (as far as I know) the notion of group had not. So it may
seem obvious that the former had to be handled more cautiously, and there-
fore frivolous to ask why its application should not be as warranted as that of

40. See, e.g., Poincaré’s criticism of Zermelo’s view that “by positing beforehand a Menge M” (from
which elements satisfying a certain condition are to be selected), “he has erected an enclosing wall which
keeps out the intruders who could come from without”: “But he does not query whether there could be
intruders from within whom he has enclosed inside his wall. If the Menge M possesses an infinite
number of elements, this means . . . that it is possible for new ones to arise constantly; they will arise
inside the wall instead of outside” (1913/1963, 59–60).
41. This has one clear advantage over Couturat’s definition of the natural numbers. This definition
blocks the possibility that the set of natural numbers contains something that cannot be obtained by
iterated application of the successor operation (starting from 0), in which case Couturat’s definition fails
(because the set does not satisfy the principle of induction).

304

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

the notion of group. Yet the notion of Lie groups—particularly as formulated

by Lie—was regarded with some suspicion, notably in Berlin, where Karl Wei-
erstrass spearheaded the drive for rigor. In 1906 Johannes Knoblauch, an assis-
tant professor in Berlin, objected specifically that Lie’s method based on groups
of infinitesimal transformations was not legitimate “since its application, at the
first step, destroys the system of equations, which is the basis for the true sub-
ject of the investigation” (quoted in Hawkins 2000, 189). (Knoblauch’s point
appears to be that certain higher-order infinitesimals are ignored in determining
the equations; see Hawkins 2000, 189.). In general, where Lie presented his
work analytically he did not take care to show that the relevant series converge
(see Gray 2013, 489–95), and to Weierstrass the work seemed so lacking in
rigor “that it would have to be redone from the ground up” (Hawkins 2000,
188). At the beginning of the twentieth century, then, the notion of group may
not have appeared so much safer than the notion of infinite set.
These considerations would make it difficult for Poincaré to base his asym-
metrical treatment of the notions on his view that the notion of group “pre-
exists in our minds” (see Poincaré 1902/1929, 79 and 91). Poincaré cannot
mean that the notion of group is innate in the sense that it is automatically
and universally utilized (by normal human reasoners). Rather, it became avail-
able once our reasoning practices attained a certain level of sophistication, and
serves to represent their basis. According to Poincaré’s view, the notion of group
has the following necessary role in organizing spatial experience: Every point
in space is associated with a certain subgroup of the group formed by the dis-
placements of a rigid body. That its displacements form a group means that
the condition of free mobility is satisfied, so that the body can be applied as a
measure everywhere in space, which is required for geometry as Poincaré con-
ceives it. For the group concept to have the mathematical importance Poin-
caré attributes to it, these must be conditions on the possibility of geometry
itself, not merely on our capacity to think geometrically. But in that case, it
seems open to the logicist to argue that it is just as much a condition of arith-
metic’s possibility that certain (infinite) collections constitute sets—regardless
of whether we now acknowledge it. The notion of set would then have a com-
parable role in organizing experience. It would not count against the notion’s
innateness, thus understood, that its formulation was at first unsatisfactory,
since the notion of group itself was rigorously formulated only after it was con-
ceived—by Helmholtz and Lie, as Poincaré claims (1898, 40)—to have a foun-
dational role.
We began our examination of Poincaré’s treatment of groups with the view
that intuition justifies the use of the group notion by warranting the assump-
tion of an iteratively generated domain (sec. 6). I argued (in secs. 7–8) that

305

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

Poincaré’s arguments in Science and Hypothesis do not suffice to establish his

view that it is conventional which of the candidate groups is employed to orga-
nize our spatial experience. His full argument for the conventionality of the
group appears to undermine his case for the necessity of intuition in arithmetic.
In part 2 of this article, I will show by another example how Poincaré’s account
of the role of linguistic innovation seems to conflict with his view he defends in
chapter 1 of Science and Hypothesis. This persisting tension shows that Poincaré’s
views of arithmetic and geometry cannot be reconciled by supposing geome-
try to be founded on arithmetical operations or principles. I will then explain
how intuition can be seen to found both geometry and arithmetic, without the
former depending on the latter.

REFERENCES
Ballue, E. 1894. “Le nombre entire considéré comme fondement de l’analyse mathéma-
tique.” Revue de métaphysique et de morale 2:317–28.
Ben-Menahem, Yemima. 2006. Conventionalism. Cambridge: Cambridge University Press.
Black, Max. 1942. “Conventionalism in Geometry and the Interpretation of Necessary
Statements.” Philosophy of Science 9:335–49.
Brading, Katherine, and Elise Crull. 2013. “Epistemic Structural Realism and Poincaré’s
Philosophy of Science.” Preprint. http://philsci-archive.pitt.edu/10079/.
Couturat, Louis. 1905/1979. Les Principes des Mathématiques. Repr. Hildesheim: Olms.
Da Silva, Jairo José. 1997. “Husserl’s Phenomenology and Weyl’s Predicativism.” Synthese
110:277–96.
Detlefsen, Michael. 1992. “Poincaré against the Logicians.” Synthese 90:349–78.
DiSalle, Robert. 2006. Understanding Space-Time. Cambridge: Cambridge University Press.
Folina, Janet. 1994. “Poincaré on Mathematics, Intuition, and the Foundations of Science.”
PSA: Proceedings of the Biennial Meeting of the Philosophy of Science Association, 1994,
217–26.
———. 2006. “Jules Henri Poincaré.” In The Philosophy of Science: An Encyclopedia, ed.
S. Sarkar and J. Pfeifer, 568–71. New York: Routledge.
———. 2008. “Intuition between the Analytic-Continental Divide.” Philosophia Mathe-
matica, ser. 3, 16:25–55.
Friedman, Michael. 1995–96/1999. “Poincaré’s Conventionalism and the Logical Positiv-
ists.” In Reconsidering Logical Positivism, 71–88. Cambridge: Cambridge University
Press. Originally published in Foundations of Science 1:299–314.
———. 2000. “Geometry, Construction and Intuition in Kant and His Successors.” In
Between Logic and Intuition, ed. G. Sher and R. Tieszen, 186–218. Cambridge: Cam-
bridge University Press.
———. 2010. “Synthetic History Reconsidered.” In Discourse on a New Method, ed. M.
Domski and M. Dickson, 571–814. Chicago: Open Court.

306

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Dunlop | FA L L 2 0 1 6

Goldfarb, Warren. 1988. “Poincaré against the Logicists.” In History and Philosophy of
Modern Mathematics, vol. 11 of Minnesota Studies in the Philosophy of Science, ed. W.
Aspray and P. Kitcher, 61–81. Minneapolis: University of Minnesota Press.
Gray, Jeremy. 2013. Henri Poincaré: A Scientific Biography. Princeton, NJ: Princeton Uni-
versity Press.
Grünbaum, Adolf. 1973. Philosophical Problems of Space and Time. Vol. 12 of Boston Studies
in the Philosophy of Science. Dordrecht: Reidel.
Hawkins, Thomas. 2000. Emergence of the Theory of Lie Groups. Berlin: Springer.
Heinzmann, Gerhard. 2009. “Hypotheses and Conventions in Poincaré.” In The Signifi-
cance of the Hypothetical in the Natural Sciences, ed. M. Heidelberger and G. Schiemann,
169–92. Berlin: De Gruyter.
Heinzmann, Gerhard, and David Stump. 2013. “Henri Poincaré.” The Stanford Encyclo-
pedia of Philosophy, ed. Edward N. Zalta. Stanford, CA: Stanford University. http://plato
.stanford.edu/archives/win2013/entries/Poincare.
Johnson, Dale M. 1981. “The Problem of the Invariance of Dimension in the Growth of
Modern Topology, Part II.” Archive for History of Exact Sciences 25:85–226.
Klein, Felix. 1908/1932. Elementary Mathematics from an Advanced Standpoint: Arithmetic,
Algebra, Analysis. Trans. E. R. Hedrick and C. A. Noble. New York: Macmillan. Orig-
inally published as Elementarmathematik vom höheren Standpunkte aus, Band 1 (Berlin:
Springer).
McLarty, Colin. 1997. “Poincaré: Mathematics & Logic & Intuition.” Philosophia Mathe-
matica, ser. 3, 5:97–115.
O’Gorman, F. P. 1977. “Poincaré’s Conventionalism of Applied Geometry.” Studies in His-
tory and Philosophy of Science 8:303–40.
Poincaré, Henri. 1889. “La logique et l’intuition dans la science mathématique et dans
l’enseignement.” L’Enseignement Mathématique 1:157–62.
———. 1894. “Sur la nature du raisonnement mathématique.” Revue de métaphysique et de
morale 2:371–84.
———. 1898. “On the Foundations of Geometry.” Monist 9:1–43.
———. 1902/1903. “Poincaré’s Review of Hilbert’s ‘Foundations of Geometry.’” Bulletin
of the American Mathematical Society 10 (1): 1–23. Originally published as “Les fonde-
ments de la géométrie,” Bulletin des Sciences Mathématiques, 2nd ser., 26:249–72.
———. 1902/1929. Science and Hypothesis. In The Foundations of Science. New York: Sci-
ence. Originally published as La science et l’hypothése (Paris: Flammarion).
———. 1905/1929. The Value of Science. In The Foundations of Science. New York: Science.
Originally published as La valeur de la science (Paris: Flammarion).
———. 1908/1929. Science and Method. In The Foundations of Science. New York: Science.
Originally published as Science et méthode (Paris: Flammarion).
———. 1913/1963. Mathematics and Science: Last Essays. Trans. J. W. Bolduc. New York:
Dover. Originally published as Derniéres Pensées (Paris: Flammarion).
Stump, David J. 1989. “Henri Poincaré’s Philosophy of Science.” Studies in History and
Philosophy of Science 20:335–63.

307

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 HOPOS | Forum: Poincaré Reconsidered

———. 1991. “Poincaré’s Thesis of the Translatability of Euclidean and Non-Euclidean

Geometries.” Noûs 25:639–57.
———. 1996. “Poincaré’s Curious Role in the Formalization of Mathematics.” In Henri
Poincaré: science et philosophie-Science and Philosophy-Wissenschaft und Philosophie, ed.
J.-L. Greffe, G. Heinzmann, and K. Lorenz, 481–92. Berlin: Akademie.
Torretti, Roberto. 1978. The Philosophy of Geometry from Riemann to Poincaré. Dordrecht:
Reidel.
Wilder, R. L. 1967. “The Role of Intuition.” Science 156:605–10.
Zahar, Elie. 1997. “Poincaré’s Philosophy of Geometry, or Does Geometric Convention-
alism Deserve Its Name?” Studies in History and Philosophy of Modern Physics 28:183–
218.

308

This content downloaded from 129.100.058.076 on December 24, 2019 08:11:12 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).