Studies in History and Philosophy of Science 54 (2015) 80e89

Contents lists available at ScienceDirect

Studies in History and Philosophy of Science

journal homepage: www.elsevier.com/locate/shpsa

Conventionalism about what? Where Duhem and Poincaré part ways

Milena Ivanova
Munich Center for Mathematical Philosophy, Germany

a r t i c l e i n f o a b s t r a c t

Article history: This paper examines whether, and in what contexts, Duhem’s and Poincaré’s views can be regarded as
Received 28 February 2015 conventionalist or structural realist. After analysing the three different contexts in which convention-
Received in revised form alism is attributed to themdin the context of the aim of science, the underdetermination problem and
2 September 2015
the epistemological status of certain principlesdI show that neither Duhem’s nor Poincaré’s arguments
can be regarded as conventionalist. I argue that Duhem and Poincaré offer different solutions to the
Keywords:
problem of theory choice, differ in their stances towards scientific knowledge and the status of scientific
Conventionalism;
principles, making their epistemological claims substantially different.
neo-Kantianism;
Structural realism;
Ó 2015 Elsevier Ltd. All rights reserved.
Theory choice;
Underdetermination

When citing this paper, please use the full journal title Studies in History and Philosophy of Science

1. Introduction to their conventionalism. I distinguish three different types of

conventionalism associated with Duhem and Poincaré: (a) as a
Pierre Duhem’s and Henri Poincaré’s philosophy of science, and position in the scientific realism debate concerned with the aim of
especially their stance towards scientific theories, has caused sig- science, (b) as a position arising from the problem of under-
nificant debates in the philosophy of science literature. However, a determination of theories by the data, and (c) as an epistemological
unified and coherent understanding of their views is still to be thesis regarding the status of scientific principles. I examine these
given. Many regard Duhem and Poincaré as conventionalists (Ben- different understandings of conventionalism in the context of
Menahem, 2006; Popper, 1959; Reichenbach, 1958; Sklar, 1974), Duhem’s and Poincaré’s arguments and evaluate whether these
others regard them as instrumentalists (Laudan, 1981; Stanford, theses are rightly attributed to them. Last, I examine whether we
2006). More recently Duhem and Poincaré have been considered can attribute a structural realist reading of their views.
as the founders of structural realism (Giedymin, 1982; Gower, The structure of this paper is the following. In Section 2, I pre-
2000; Worrall, 1989). However, none of these interpretations pro- sent Duhem’s and Poincaré’s views on the aim of science and
vide a fully coherent understanding of the motivations Duhem and oppose the traditional reading that renders these arguments to
Poincaré have in developing their conventionalism and its rele- establish conventionalism understood as an anti-realist position
vance to their general philosophy of science and especially with with regard to scientific knowledge. I show that the arguments
their alleged structural realism. establish structuralism motivated by the history of science. In
This paper disentangles the different arguments of Duhem and Section 3, I discuss the problem of underdetermination. I show,
Poincaré, that are usually taken to motivate conventionalism or contrary to what is usually assumed, that Duhem is not led to
structural realism, in order to answer two questions: (1) what kind conventionalism about theory choice. Rather than being arbitrary,
of conventionalism is developed by Duhem and Poincaré, and (2) for Duhem the choice is settled by a rational faculty he calls ‘good
do they develop structural realism and how does this thesis relate sense’. I also show, contrary to the mainstream literature, that
Poincaré’s solution to the problem of underdetermination is driven
by the employment of aesthetic values, which he takes to play a
E-mail address: mail@milenaivanova.co.uk. regulative role in scientific practice. In Section 4, I examine

http://dx.doi.org/10.1016/j.shpsa.2015.09.004
0039-3681/Ó 2015 Elsevier Ltd. All rights reserved.
 M. Ivanova / Studies in History and Philosophy of Science 54 (2015) 80e89 81

Poincaré’s argument for the conventional and constitutive nature of 2.1.1. Arguments against scientific anti-realism
geometry and mechanics and show how he develops, contrary to Duhem explicitly articulates two objections against scientific
Duhem’s epistemological holism, a ‘hierarchy’ of science, which anti-realism, arguing that it cannot account for: (1) the predictive
presupposes the Kantian synthetic a priori. It is in this context, I success of science and (2) the unity in science.
argue, that Duhem and Poincaré significantly part ways and as a The first argument Duhem offers against scientific anti-realism
consequence their structuralisms differ in a crucial way. I argue that is that it cannot explain the novel predictive power of physical
according to Poincaré the structure of the world can only be theories. If theories did nothing but classify already known phe-
reached by conventional and mind-dependent elements, rendering nomena in some convenient way, how could we explain the fact
his structuralism an internalist one. I conclude with an analysis of that they manage to predict phenomena they were not constructed
Duhem’s and Poincaré’s understanding of structuralism and show to classify? Duhem argues that physical theories are not artificial
the limitations of the current literature regarding their alleged classifications but aim to be a ‘natural classification’ of the
structural realism. observable phenomena. According to Duhem, even though a
physical theory cannot reveal the unobservable reality, it can still
teach us something of the unobservable world, because “the more
2. The aim of science complete it becomes, the more we apprehend that the logical order
in which theory orders experimental laws is the reflection of an
Both Duhem and Poincaré are concerned with the question of ontological order, the more we suspect that the relations it estab-
what the aim of science is, what scientific theories can teach us lishes among the data of observation correspond to real relations
about the world, whether there is an unobservable reality beyond among things and the more we feel that theory tends to be a nat-
the appearances, and whether we can know it. Duhem often seems ural classification” (Duhem, 1954, 26). For Duhem, the strongest
to take an anti-realist approach to answering these questionsdhe indication of a theory being a natural classification is its ability to
holds that scientific theories should not search for any underlying make novel predictions (Duhem, 1954, 28).
ontology beyond what is observable, that theoretical terms such as Duhem does not believe scientific anti-realism can adequately
‘electrons’ and ‘genes’ should be eliminated from our theories, explain the ability of scientific theories to make novel predictions
because we could never observe their alleged referents, and that and thus suggests that a more optimistic stance towards successful
science in principle should not aim at providing explanations but scientific theories must be adopted. Duhem is nevertheless careful
only representations of observational laws. According to him: in articulating this optimism about scientific theories being natural
classifications by using terms like “we apprehend”, “we suspect”,
[T]he aim of physical theory is to become a natural classification,
“we feel”. His position is fallibilist, allowing that even theories that
to establish among diverse experimental laws a logical coordi-
make novel predictions might succumb and be abandoned. His
nation serving as a sort of image and reflection of the true order
position is also, as is made clear below, not full blown scientific
according to which the realities escaping us are organised.
realism.2
(Duhem, 1954, 31)
The second argument against instrumentalism is again based on
In a similar manner, Poincaré argues that the concept of natural classification and pertains to the ability of
theories to be unified with other theories. Maxwell’s theory of
The object of mathematical theories is not to reveal to us the real
electromagnetism unified electric and magnetic phenomena into a
nature of things; that would be unreasonable claim. Their only
single framework. Newton’s theory of universal gravitation
object is to coordinate the physical laws with which physical
accounted for two distinct sets of phenomena: terrestrial and ce-
experiments make us acquainted, the enunciation of which,
lestial. Duhem believes that this fact cannot be explained by the
without the aim of mathematics, would be unable to effect.
conventionalists who believe that distinct phenomena can be saved
(Poincaré, 1902, 117)
by different theories even if they are found to be in contradiction.
Duhem asks:
Despite appearances, Duhem and Poincaré are not content with
Why give up the ideal of a completely unified and perfectly
a purely anti-realist picture of science, they both develop several
logical theory, when the systems actually constructed have
objections to conventionalism and defend the continuity and unity
drawn closer and closer to this ideal from century to century?
of science.1
[.] physical theory through its successive advances tends to
arrange experimental laws in an order more and more analo-
2.1. Duhem on the continuity and unity of science gous to the transcendental order according to which the realities
are classified, that as a result physical theory advances gradually
Duhem argues that taking scientific theories to simply be
convenient tools for prediction does not capture our conviction that
scientific theories are continuous and make novel predictions,
2
It is this use of language that has motivated an alternative interpretation of
promoting in this way a more moderate position.
Duhem’s concept of natural classificationdneither as a selective realist position or
anti-realist position. First developed by Darling (2003) and articulated in more
depth by Bhakthavatsalam (2015), a motivational realist reading of Duhem’s
1
It is important to note a distinction here between conventionalism and anti- concept of ‘natural classification’ claims that his aim is to explain scientific practice
realism positions such as instrumentalism. Duhem and Poincaré are often taken rather than, as traditional realists, to offer an explanation for the success of science
to defend a ‘global conventionalism’, equated in the contemporary literature with or to give an account of theory’s approximate truth. This reading correctly points to
anti-realist positions such as instrumentalism. On this view, the aim of science is the language Duhem uses for justifying our conviction, it is a ‘feeling’ an ‘intuition’.
not to provide true descriptions of reality but, rather, to save the phenomena. On a However, I disagree that Duhem offers no empirical justification for this conviction.
‘localised’ understanding of conventionalism, the thesis regards the status of sci- Duhem gives us an empirical reason to justify our belief in a theory being natural
entific principles and does not make claims about whether scientific theories can classificationdwhen it makes novel predictions. Thus, we can project expectation
describe successfully the unobservable. As Giedymin (1982) and Ben-Menahem in the success of the theory by reflection on its past success at acting as a ‘prophet’
(2006) point out, Poincaré’s conventionalism is local because it concerns only the for us. Importantly, the selective-realist reading advocated here, and the motiva-
axioms of geometry and certain constitutive principles in mechanics, while the tional realist reading, need not be in conflict, as Bhakthavatsalam (2015) has
conventionalism of his contemporaries Abel Rey and Édouard LerRoy is generalized. insightfully noted.
82 M. Ivanova / Studies in History and Philosophy of Science 54 (2015) 80e89

towards its limiting form, namely, that of a natural classification, representation and not an explanation, Duhem argues that even at
and finally that logical unity is a characteristic without which the ‘ideal end of science’ a natural classification will not reveal the
physical theory cannot claim this rank of a natural classification. real causes of the phenomena. We can never “strip reality of the
(Duhem, 1954, 296-297) appearances covering it like a veil, in order to see the bare reality
itself” (Duhem, 1908, p. 7). But the more successful in their pre-
Against his contemporaries Abel Rey and Édouard LeRoy, who
dictions our theories are, the more they manage to uncover the
believe that physical theories are convenient tools for prediction,
relations that hold among these ‘realities’ or ‘things’. According to
suited for increasing empirical knowledge, Duhem argues:
Duhem, “physical theory through its successive advances tends to
But nature protests against this judgment; it declares that there arrange experimental laws in an order more and more analogous to
exists a universal and necessary truth, and that physical theory the transcendent order according to which the realities are classi-
through the steady progress which extends it continually while fied” (Duhem, 1908, p. 297).
rendering it still more unified gives us from day to day a more Duhem argues that only a ‘perfect theory’ could reveal the true
perfect insight into this truth, so that it constitutes a veritable ontological order hidden behind the appearances. A perfect theory
philosophy of the universe. (Duhem, 1954, 332-333) would be a ‘theory of everything’ that saves all the phenomena.
However, the scientific method can construct only an ‘imperfect’
Duhem’s most clear defense of unification is stated at the end of his
theory. Even if science was to reach an end, a ‘perfect theory’, a
To Save the Phenomena. After a detailed discussion of the method-
theory that explains all the phenomena, will not be reached.
ology of Kepler and Galileo, who claimed that their theories
described ‘real movements’ rather than, like Osiander and Bellar- [W]e do not possess this perfect theory, and mankind will never
mine held, simply saved the phenomena, Duhem argues that: possess it; what we possess and what mankind will always
possess is an imperfect and provisional theory, which by its
Despite Kepler and Galileo, we believe today, with Osiander and
innumerable groupings, hesitations and repentances proceeds
Bellarmine, that the hypotheses of physics are mere mathe-
slowly toward that ideal form which would be a natural classi-
matical contrivances devised for the purpose of saving the
fication. (Duhem, 1954, 302)
phenomena. But thanks to Kepler and Galileo, we now require
that they save all of the phenomena of the inanimate universe According to Duhem, science is not an explanation because “it
together. (Duhem, 1908, 117) cannot attain this degree of perfection [.] it cannot render
accessible to the senses the reality it proclaims as residing under-
Despite the fact Duhem disagrees with Kepler and Galileo’s scien-
neath those appearances” (1954, 8). Science is “a system of math-
tific realism, he also opposes the instrumentalism of Osiander and
ematical propositions, deduced from a small number of principles,
Bellarmine because Kepler and Galileo have managed to unify the
which aim to represent as simply, as completely, and as exactly as
celestial and terrestrial phenomena in a single mathematical
possible, a set of experimental laws” (Duhem, 1954, 19). Duhem is
scheme, a unification completed by Newton. This unification, for
also convinced that the evidence that science progresses and un-
Duhem, cannot be explained by instrumentalism, according to
covers the true relations among things:
which scientific theories are a ‘rack filled with tools’.
The above passages show that Duhem is not satisfied with an What is lasting and fruitful [in physical theories] is the logical
anti-realist view about scientific theories and wants to explain the work through which they have succeeded in classifying natu-
historical success of science. Like scientific realists, he does not rally a great number of laws by deducing from them a few
regard scientific theories as mere useful classifications; he believes principles; what is perishable and sterile is the labour under-
the unity and novel predictive power of theories should be taken to explain these principles in order to attach them to as-
accounted for as well as the continuity of science. However, like sumptions concerning the realities hiding underneath sensible
instrumentalists, he does not believe science can provide us with a appearances. (Duhem, 1954, 38)
‘perfect theory’, i.e. a complete metaphysical description of the
Worrall (1982, 1989) argues that Duhem’s position is best under-
world. Unlike scientific realists, Duhem argues that science cannot
stood as structural realist,4 accounting for both the argument from
uncover the ‘realities’ hidden beyond the observable phenomena.
theory change and the ‘no miracles’ argument. The idea is that at
However, he believes that a natural classification can teach us how
the level of ontology we should be anti-realists and believe that the
these realities are classified.3 The next subsection explores in more
entities postulated by our current theories will likely be aban-
detail the notion of ‘natural classification’ and its structuralist
doned, as was the case with the caloric, the ether and phlogiston.
interpretation.
But we can be realists at the level of structure, which is expressed in
the mathematical equations that survive theory change. According
2.1.2. ‘Natural classification’ to Worrall’s account the concept of natural classification can be
According to Duhem, physical theory cannot reveal the unob- understood as the structural content of the theory, content which is
servable ‘realities’ or ‘things’ that cause the phenomena we retained in theory change at least as a limiting case. This reading,
observe. However, the history of science reveals that despite the which will be examined in more detail in Section 4, captures
discontinuities occurring in theory change, we can reach some Duhem’s mediating position between anti-realism and scientific
knowledge of the unobservable realitydwe can learn how these realism.
‘realities’ are related. Duhem distinguishes between explanation
and representation and argues that the former belongs to meta-
physics while the latter to physics. Taking physical theory to be a

4
The term ‘structural realism’ was first introduced by Grover Maxwell in his
3
Lugg (1990) and McMullin (1990) offer lucid presentations of Duhem’s concept (1962), referring to the thesis developed by Bertrand Russell in The Analysis of
of natural classification, but their conclusions are in opposition. While Lugg argues Matter (1927). According to this view, our knowledge of the unobservable world is
that Duhem’s view is a convergent realism, McMullin argues that Duhem’s view is a limited to second-order relations between the structure of our stimuli and the
middle way between conventionalism and scientific realism. However, neither of structure of their causes. Structural realism, as a thesis motivated by the problem of
these readings explains how the two seemingly opposing motivations for realism theory change, was first introduced by Worrall (1989), who traces the position back
and instrumentalism can be reconciled into a coherent view. to Duhem and Poincaré.
 M. Ivanova / Studies in History and Philosophy of Science 54 (2015) 80e89 83

2.2. Poincaré on the continuity and unity of science Using the transition from Fresnel’s elastic ether theory to
Maxwell’s theory of the electromagnetic field, Worrall argues that
Duhem’s opposition to anti-realism was inspired by the debate there is continuity between the two theories but this continuity is
between Poincaré and LeRoy. LeRoy generalizes Poincaré’s argu- expressed in the mathematical equations rather than their onto-
ment for the conventional status of geometry and mechanics to logical interpretation. Fresnel and Maxwell attributed different
defend a generalised conventionalism, termed at the time ‘nomi- ‘nature’ to light, but they identified correctly its ‘structure’. The idea
nalism’. Poincaré opposes LeRoy’s conventionalism because he of structural continuity according to Worrall, originates in both
believes that such an anti-realist position has undesired conse- Duhem and Poincaré. Zahar (2001) shares Worrall’s conviction that
quences, it does not account for the continuity and unity of scien- Poincaré is a structural realist “according to Poincaré, only the
tific theories. In his chapter ‘Is Science Artificial’ from The Value of universals, and more particularly the relations occurring in a uni-
Science, Poincaré responds to LeRoy by arguing that if science has a fied and empirically successful theory, mirror the ontological order
value, because of its ability to make successful predictions, then this of things. As for the nature of the relata, it will forever remain
value must be due to the fact that theories are more than just hidden from us” (Zahar, 2001, 37). Furthermore, Gower argues that
practical recipes (Poincaré, 2001, 320). Like Duhem, Poincaré sees “for both Poincaré and Duhem [.] a defensible scientific realism
science as “a classification, a manner of bringing together facts must be structural in the sense that it attributes reality to the
which appearances separate, though they were bound together by relational structure of a scientific theory” (Gower, 2000, 86).
some natural and hidden kinship. Science, in other words, is a According to this understanding of structural realism science is
system of relations. Now we have just said, it is in the relations cumulative; in theory change there are elements of the old theory
alone that objectivity must be sought.” (Poincaré, 2001, 347) that are retained in the new one. This preservation is structural, not
Poincaré’s main concern is to account for ‘the bankruptcy of ontological, and is expressed in the mathematical equations of
science’, a problem widely discussed during his time. empirically successful theories. In the transition from Fresnel’s
theory to Maxwell’s theory of electromagnetism, the equations of
The ephemeral nature of scientific theories takes by surprise the
the former theory are completely preserved in the latter theory,
man of the world. Their brief period of prosperity ended, he sees
receiving a different metaphysical interpretation.
them abandoned one after another; he sees ruins piled upon
In this section I have highlighted Duhem’s and Poincaré’s main
ruins; he predicts that the theories in fashion to-day will in a
arguments against conventionalism understood as an anti-realist
short time succumb in their turn, and he concludes that they are
position with respect to the aim of science. While both Duhem
absolutely in vain. This is what he calls the bankruptcy of sci-
and Poincaré argue that science can at best produce approximate,
ence. (Poincaré, 2001, 122)
and not true, theories, they do not want to adopt conventionalism.
However, Poincaré is not convinced that this argument from theory They believe that a more moderate position needs to be adopted if
change, if properly understood, shows that there is no progress in one wants to account for the novel predictive success of theories,
science and leads to anti-realism about scientific theories. Poincaré their unifying power and the continuity in theory change. Duhem
continues his argument by defending the structural continuity in and Poincaré develop an epistemological view according to which
theory change, by examining the transition from Fresnel’s wave our knowledge of the world is limited to its structure, the ‘reali-
theory of light to Maxwell’s. Poincaré argues that while Fresnel ties’dthe unobservable entities that form this structuredremain
understood light in a different way, he nevertheless identified the epistemically inaccessible. The tenability of this reading will be
correct relations between optical phenomena and the equations further evaluated in Section 5.
that express these relations are preserved in Maxwell’s theory. He
argues that “[t]he true relations between these real objects are the 3. The problem of underdetermination
only reality we can attain, and the sole condition is that the same
relations shall exist between these objects as between the images It is commonly believed that Duhem and Poincaré are lead to
we are forced to put in their place. If the relations are known to us, conventionalism because of the problem of underdetermination
what does it matter if we think it convenient to replace one image (Ben-Menahem, 2006). In this section I show that the focus of the
by another?” (Poincaré, 2001, 123) underdetermination argument for Duhem and Poincaré is different
Poincaré argues that science cannot teach us the nature of and they develop different responses to it. While Duhem focuses
‘things’ but it can teach us the true relations among things. The primarily on articulating a solution to the problem of theory choice,
evidence for this claim is the history of science itself. “When we ask Poincaré’s main concern regards the epistemological implications
what the objective value of science is, that does not mean: Does of the problem of underdetermination for the status of geometry. I
science teach us the true nature of things? but it means; Does it present Duhem’s articulation of the problem of under-
teach us the true relations of things?” (Poincaré, 2001, 347-349).5 determination and his solution to theory choice and contrasts it
with Poincaré’s articulation of the argument for the under-
determination of geometry and his own understanding of theory
5
choice and epistemic virtues. I show that neither Duhem nor
Several authors have argued that Poincaré clearly opposes anti-realism, espe-
cially at the end of his career when he endorsed the existence of the atom after
Poincaré can be associated with conventionalism in the context of
initially having opposed it. David Stump (1989) argues that “[Poincaré] does not the problem of underdetermination.
hold that theoretical objects are unknowable in principle, rather, he thinks that one
must look at the state of experimental research at the time to see whether or not 3.1. Underdetermination, theory virtues and good sense
belief in the existence of a particular theoretical entity is justified” (Stump, 1989,
339). Stump’s interpretation would challenge our current understanding of Poin-
caré’s structuralism, which would not allow for the commitment to the existence The problem of underdetermination of theory by the evidence,
and causal detection of unobservable entities. Krips (1986), on the other hand, for Duhem, stems from the holistic nature of confirmation. Given
argues that Poincaré’s (1913)’s acceptance of the atom did not indicate a shift from that scientific hypotheses are always tested as a conjunction with
instrumentalism to scientific realism. Rather, the shift concerns the status of the auxiliary hypotheses and assumptions, a negative piece of evidence
atomic hypothesisdfrom ‘indifferent’ it becomes ‘empirical’. Ivanova (2013) argues
that Poincaré’s acceptance of the atom is not in tension with his structuralism
does not refute the testing hypothesis but the whole conjunction.
because the significance of Poincaré’s argument lies in the fact that he argues As a consequence, neither deductive logic nor experience can
against atoms being fundamental entities. isolate which of the tested hypotheses or assumptions are at fault.
84 M. Ivanova / Studies in History and Philosophy of Science 54 (2015) 80e89

The holistic nature of confirmation allows, in light of a negative Since logic does not determine with strict precision the time
outcome of an experiment, for the modification of either the when an inadequate hypothesis should give way to a more
auxiliary assumptions used to derive the prediction or the testing fruitful assumption, and since recognizing this moment belongs
theory. For Duhem, scientists faced with the problem of under- to good sense, physicists may hasten this judgement and in-
determination have two choices. They could either modify the crease the rapidity of scientific progress by trying consciously to
auxiliary assumptions to accommodate the new evidence or make good sense within themselves more lucid and more
formulate a new hypothesis. However, there is no “absolute prin- vigilant. (Duhem, 1954, 218)
ciple” which dictates this decision.
There is an ongoing debate regarding the proper interpretation of
[T]he rules of syllogistic logic are not adequate. They must be Duhem’s notion of good sense.8 Here I am not interested in eval-
assisted by a certain sense of soundness that is one of the forms uating these accounts or defending the tenability of Duhem’s so-
of good sense [.] good sense will intervene at the moment at lution.9 What suffices for the purposes of this section is pointing at
which one realizes that the consequences of a preconceived idea the fact that Duhem’s concept of good sense is an attempt to
are either contradicted or confirmed by the experiment. [.] articulate a non rule-governed solution to theory choice. Recog-
What a delicate task, concerning which no precise rule can nizing the involvement of values and judgements in theory choice
guide the mind! It is essentially a matter of insight and in- and attempting to provide an account that explains how these
genuity! (Ben-Menahem, 2006, 23-25) values and judgements lead to consensus in the scientific com-
munity, shows that Duhem is not content with a conventionalist
One traditional way for privileging a theory over its empirically
response to the problem of underdetermination. Duhem claims
equivalent rival is to employ theory virtues and choose the
that the notion of good sense can explain the history of science,
simplest, most unified, elegant or fertile theory. Duhem, however,
where Duhem sees rational resolutions of theory choices despite
does not believe that theory virtues are always sufficient for the
the lack of empirical evidence or conclusive theory virtues.
resolution of theory choice. This is because different scientists can
Whether Duhem’s solution is feasible is a subject of ongoing
understand these virtues in different ways, leading to a ‘state of
investigation. However, this solution shows that conventionalism
indecision’ in the community. Even if each individual scientist has
about theory choice was not a position Duhem accepted and he
made up their mind about their preferred theory on the basis of
believes there is space for rational solution to the problem of
theory virtues, the community might have failed to agree upon a
underdetermination.
theory.
No doubt the physicist will choose between these logically
equivalent theories, but the motives which will dictate his 3.2. Geometric underdetermination, conventions and epistemic
choice will be considerations of elegance, simplicity, and con- virtues
venience, and grounds of suitability which are essentially sub-
jective, contingent, and variable with time, with schools, and Understanding Poincaré’s solution to the problem of under-
with persons; as serious as these motives may be in certain determination requires a distinction between his argument for the
cases, they will never be of a nature that necessitates adhering underdetermination of geometry from experience and the general
to one of the two theories and rejecting the other, for only the argument for underdetermination of theory by the data. Poincaré’s
discovery of a fact that would be represented by one of the alleged conventionalism is rooted in his articulation of the argu-
theories, and not by the other, would result in a forced opinion. ment from underdetermination in the context of geometry. In
(Duhem, 1954, 288) chapter 4 of Science and Hypothesis, Poincaré presents his famous
argument for the empirical equivalence of Euclidean and non-
What, according to Duhem, guides the scientist’s choice in weigh- Euclidean geometries. He argues (2001, 55-56) that the appear-
ing theory virtues and making the optimal or rational decision is a ances can be described either by employing a non-Euclidean ge-
faculty he calls ‘good sense’.6 Duhem argues that every scientist ometry of constant negative curvature or by employing Euclidean
possesses good sense, however, some exemplify it more than oth- geometry and modifying the laws of physics (adding heat forces
ersdthus the disagreements occurring in science (Duhem, 1954, that distort the measurements). There is, thus, a choice to be made
218). Also, he claims that good sense is closely connected to moral as to which geometry to choose in order to describe the
virtues, such as impartiality, sobriety, lack of self-interest. He claims phenomena.
that “in order to estimate correctly the agreement of a physical According to Poincaré, one could cultivate an alternative,
theory with the facts, it is not enough to be a good mathematician observationally indistinguishable theory, by changing the geometry
and skillful experimenter; one must also be an impartial and and modifying the physical laws. Experience, then, is compatible
faithful judge” (Duhem, 1954). He condemns personal interest, with alternative geometries, leading to the problem of choosing
stresses the importance of impartiality in science and suggests that which geometry to be employed. However, this choice is not arbi-
developing one’s impartiality would lead to scientific progress. “[N] trary and is limited to only three alternative geometries of constant
othing will delay the decision which should determine a fortunate curvature (Friedman, 1999). Many have suggested that this argu-
reform in a physical theory more than the vanity which makes a ment is a special case of Duhem’s argument for under-
physicist too indulgent towards his own system and too severe determination of theory by experiment and leads to
towards the system of another” (Duhem, 1954). Duhem claims that conventionalism about scientific theories.10 However, a more
good sense is cultivated with scientific practice and scientists
should aim to develop their good sense in order to accelerate the
progress of science7: 8
Stump (2007) and Fairweather (2012) argue that because of his appeal to moral
virtues, Duhem’s notion can be understood as a form of virtue epistemology.
Ivanova (2010) develops objections to these readings and offers an alternative ac-
count that she takes to fit better with Duhem’s general epistemology of science.
6 9
For a detailed presentation of Duhem’s notion of good sense and its properties Ivanova (2010) develops objections to Duhem’s notion by arguing that good
in different contexts of scientific inquiry, see Ivanova & Paternotte (2013). sense can justify a preference for a theory only retrospective, after new evidence
7
The accelerating property of good sense has been the focus of Ivanova & has supported the theory, which makes the notion redundant.
10
Paternotte’s (2013) interpretation of ‘good sense’. See Schlick (1915), Carnap (1966) and Torretti (1978).
 M. Ivanova / Studies in History and Philosophy of Science 54 (2015) 80e89 85

convincing interpretation is to take the argument to establish that simplicity has an important regulative role for Poincaré insofar as it
geometry by itself makes no empirical predictions, that geometrical is linked to the goal of unification.13
knowledge is not empirical.11 The conventional epistemological To sum up, both Duhem and Poincaré articulate solutions to the
status of geometry is established in the chapter 6 of Science and problem of theory choice. For Duhem, the problem of theory choice
Hypothesis, where Poincaré argues that geometry is neither syn- emerges from the holistic nature of confirmation that results in
thetic a priori nor empirical, rendering it conventional. I discuss this underdetermination of theories by the data. This problem, which
argument in detail in Section 4.2. Duhem believes often occurs in scientific practice, does not lead to
The argument for underdetermination of geometry by experi- conventionalism. Despite the fact that epistemic virtues cannot al-
ence leads to the problem of choosing which geometry to employ. ways guarantee the resolution of the underdetermination, the choice
Contrary to Duhem, who points at the ambiguity problem for is not arbitrary but is grounded in the scientist’s intuitions that guide
aesthetic values, Poincaré argues that simplicity leads to a resolu- them to the optimal choices. Poincaré, on the other hand, develops a
tion of theory choice. On several occasions Poincaré claims that we different solution to the problem. He limits the choice of geometries
resolve this choice by employing some ‘conventions’, which sug- to three alternatives, but does not regard it as arbitrary. He argues
gests that he does not attribute any epistemic significance to these that simplicity is a regulative rule that picks Euclidean geometry as
values but treats them merely as pragmatic devices for choice. In the ultimate option. While not being taken to have epistemic import,
the case of geometry, Poincaré believes that we have good reasons simplicity is regulative because it promotes the goal of science. As a
to choose one of the alternative geometries and this choice is consequence, neither Duhem nor Poincaré is content with regarding
guided by considerations of simplicity: theory choice to be settled on arbitrary grounds. They both believe a
rational choice can be made between competing rivals, disagreeing
Euclidean geometry is, and will remain, the most convenient:
on the role of theory virtues in that choice.
first, because it is the simplest, and it is not so only because of
our mental habits or because of the kind of direct intuition that
4. Conventions, constitutivity and structuralism
we have of Euclidean space; it is the simplest in itself. (Poincaré,
2001, 45)
While both Duhem and Poincaré advance the holistic nature of
Poincaré argues that we choose between different geometries on theory testing, Poincaré opposes the epistemological holism asso-
grounds of simplicity and convenience, making Euclidean geometry ciated with Duhem. His view is motivated by his neo-Kantianism
always the most convenient option. Since Poincaré renders the and conventionalism about the status of constitutive principles.
choice between Euclidean and non-Euclidean geometries a con- In this section I present how Poincaré responds to Duhem’s
ventional one, it is important to clarify whether simplicity is also epistemic holism by introducing a ‘layered’ account of sciences. I
considered a pragmatic value when employed in choices between show the substantial differences between Duhem’s and Poincaré’s
competing physical theories or whether it has any epistemic epistemologies and the structuralisms that stem from them.
import. Poincaré claims that the question of whether nature is
simple or not is irrelevant to our practices. 4.1. Duhem’s epistemological holism
[T]hose who do not believe that natural laws must be simple are
still often obliged to act as if they did believe it. They cannot Duhem famously claims that theories are always tested holis-
entirely dispense with the necessity without making all gener- tically, and, in light of negative evidence, deductive logic cannot
alisations, and therefore all science, impossible. It is clear that determine which element (or elements) needs to be revised; there
every law can be generalised in a number of ways, and it is a is a certain freedom with respect to the revisions we can make to
question of choice. The choice can only be guided by consider- our theories so that they can accommodate the evidence. In this
ations of simplicity [.] every law is held to be simple until the sense, any statement in a theory is equally subject to revision and
contrary is proved (Poincaré, 2001, 113). no sentence bears a special epistemic status. Duhem rejects Poin-
caré’s distinction between constitutive conventions and empirical
According to Poincaré, simplicity should be employed as a regula- statements and claims that “hypotheses that by themselves have no
tive rule in the selection of fact, forming of hypotheses and com- physical meaning undergo experimental testing in exactly the same
parison of theories. This rule, he claims, is part and parcel of our manner as other hypotheses” (1954, 216). Thus, any seemingly a
intellectual capacities: “in formulating a general, simple, and priori or conventional statement could in principle be revised on
formal law, based on a comparatively small number of not alto- the grounds of experience.14
gether consistent experiments, we have only obeyed a necessity
from which the human mind cannot free itself” (Poincaré, 2001,
100). Poincaré gives an interesting justification for using simplicity 13
It is interesting to note that both Duhem and Poincaré do not only link
as a regulative rule; he claims simplicity is linked to the ultimate simplicity and unity of a scientific theory to its utility, but also to its beauty. Duhem
aim of science, that of reaching a highly unified theory of every- claims that “[o]rder whenever it reigns, brings beauty with it. Theory not only
renders the group of physical laws it represents easier to handle, more convenient,
thing (Poincaré, 2001, 112).
and more useful, but also more beautiful” (1954, 24). Similarly, Poincaré claims that
Poincaré believes physical theories can reveal real12 relations “the scientist does not study nature because it is useful to do so. He studies it
and are not simply convenient systems for prediction. The con- because he takes pleasure in it, and he takes pleasure in it because it is beautiful.
siderations we employ to prioritise a certain theory over its [.] I am not speaking, of course, of the beauty which strikes the senses, of the
empirically equivalent rivals might be pragmatic considerations beauty of qualities and appearances. I am far from despising this, but it has nothing
to do with science. What I mean is that more intimate beauty which comes from
leading us to choose the most convenient (simple) rival, but
the harmonious order of its parts, and which pure intelligence can grasp” (Poincaré,
2001, 368).
14
Quine (1951) argues that some statements are more ‘entrenched’ than others
and are thus less likely to undergo modification or abandonment. The notion of
entrenchment, however, bears no epistemic weight. Our thinking that some
11
See Stump (1989), Giedymin (1982), Friedman (1999), Zahar (1997) and Ben- statements have a special or fundamental epistemic status, in comparison to others,
Menahem (2001, 2006). is “nothing more than a loose association reflecting the relative likelihood, in
12
In section 4.3 I clarify that for Poincaré what is real is ultimately what is practice, of our choosing one statement rather than another for revision in the
commonly accepted by rational beings (his account is not an externalist one). event of recalcitrant experience” (Poincaré, 2001, 43).
86 M. Ivanova / Studies in History and Philosophy of Science 54 (2015) 80e89

4.2. The hierarchy of science: Poincaré’s rejection of Duhem’s Having established the synthetic a priori status of arithmetic,
holism Poincaré turns to the status of mathematical magnitude (2001, 24).
He argues that we create the mathematical continuum by applying
Poincaré rejects Duhem’s epistemological holism and advocates the law of non-contradiction using as a raw material the idea of the
the idea that different parts of scientific theories have different physical continuum, which we build from our sensory experience.
epistemic statusdsome are more epistemically fundamental than Because the law of non-contradiction plays an essential part in
others. He distinguishes between three different types of scientific creating the notion of mathematical continuum, Poincaré con-
hypothesesdindifferent, conventional and empirical. Indifferent cludes, as in the case of the synthetic a priori status of arithmetic,
hypotheses are metaphysical hypotheses that cannot be observa- that “this notion has been created entirely by the mind, but it is
tionally confirmed. Poincaré’s famous example of an indifferent experience that has provided the opportunity.” (Poincaré, 2001)
hypothesis is the atomic hypothesis, which he originally regarded Poincaré’s next step is to introduce a way to measure magnitudes
as purely fictional and after Perrin’s experiments admitted that its on the continuum and he does this by introducing the addition
status changed from an indifferent to an empirical hypothesis operation (Poincaré, 2001, 27). Friedman (2001) argues that even
(Poincaré, 1913). Empirical hypotheses are testable hypotheses, e.g. though Poincaré regards the theory of mathematical magnitude as
the law of universal gravitation and the laws of electromagnetism. a synthetic a priori, the addition operator required to make the
Conventional statements are statements that are neither apodicti- continuum measurable requires the introduction of purely arbi-
cally nor empirically true but are fixed by convention and play a trary elements, which shows the first introduction of a convention
constitutive role in a scientific theory; they are, to use Poincaré’s into the hierarchy. On the next level of Poincaré’s hierarchy of sci-
term, ‘definitions in disguise’. ence he places geometrydhe takes the one-dimensional mathe-
matical continuum, applies it to many dimensions and adds a
We also see that there are various kinds of hypotheses; that
metric.
some are verifiable and, when once confirmed by experiment,
become truths of great fertility; that others, without being able Space is another framework which we impose on the world.
to lead us into error, become useful to us in fixing our ideas, and Whence are the first principles of geometry derived? Are they
that the others, finally, are hypotheses in appearance only and imposed on us by logic? Lobatschewsky, by inventing non-
reduce to definitions or conventions in disguise. (Poincaré, 1913, Euclidean geometries, has shown that this is not the case. Is
28) space revealed to us by our senses? No, for the space revealed to
us by our senses is absolutely different from the space of ge-
According to Poincaré geometry and the fundamental laws of me-
ometry. Is geometry derived from experience? Careful discus-
chanics have a conventional nature, since they lack properly
sion will give an answer e no! We therefore conclude that the
empirical content. They are, however, constitutive of empirical
principles of geometry are only conventions; but these con-
science,
ventions are not arbitrary (Poincaré, 2001, 6)
we now arrive at the physical sciences properly speaking. Here
The laws of Newtonian mechanics are found on the next level of
the science changes: we meet with hypotheses of another kind,
Poincaré’s hierarchy bearing the same epistemological status as
and we recognize their great fertility. No doubt at first sight our
geometrydthat of a convention.16 The three laws of motion, ac-
theories appear fragile, and the history of science shows us how
cording to Poincaré are not empirical but conventional because
ephemeral they are; but they do not entirely perish, and from
they serve as implicit definitions of the concepts involved in
each of them something remains. It is this something that it is
themdtime, motion, mass, force, and without these definitions the
necessary to discover, because it is this, and this alone, that is the
laws have no empirical applicability.
true reality. (Poincaré, 1913, 29-30)
Having fixed the laws of mechanics, the next level of Poincaré’s
Poincaré’s idea is based on what he calls ‘the series of sciences’, hierarchy contains all properly empirical sciences. Here, laws are
which Friedman (1999) calls a ‘hierarchy’, according to which sci- allowed to be modified in light of experiment, since the constitutive
entific knowledge is made possible by constitutive principles of framework has been fixed so all their central terms have been
different epistemic status. While Poincaré regards the most defined. As Poincaré explains:
fundamental layer of arithmetic as synthetic a priori, at the end of
We now come to the physical sciences, properly so called, and
the hierarchy we have the purely empirical laws.
here the scene changes. We meet with hypotheses of another
Let us describe how Poincaré understands the status and func-
kind, and we fully grasp how fruitful they are. No doubt at the
tion of different elements in the hierarchy of science.15 Poincaré
outset theories seem unsound, and the history of science shows
starts his description of the ‘series of sciences’ by examining the
us how ephemeral they are; but they do not entirely perish, and
epistemological status of arithmetic. Arithmetic has synthetic a
of each of them some traces still remain (Poincaré, 2001, 6)
priori status because arithmetical knowledge is based on our
reasoning by recurrence, what he calls ‘mathematical induction’. The hierarchical picture of scientific theories shows that some
This reasoning allows us to show that a given theorem does not principles bare different status. As Friedman (2001) stresses, this
hold only for a finite amount of cases, but can be taken to hold of all, view strictly opposes the DuhemeQuine holism, according to
infinite, cases. As he suggests, “reasoning by recurrence [.] is the which all sentences in a scientific theory have the same (empirical)
only instrument which enables us to pass from the finite to the status. On this framework, the likelihood of revision is established
infinite” (Poincaré, 2001, 16). For Poincaré arithmetic is synthetic a not on epistemological grounds but on purely intentional
priori: whereas it enriches our knowledge, since it takes us from the
finite to the infinite (and is thus non-tautological), it is based on our
intuitive capacity to represent infinite instances of the very same
16
operation. Interestingly, while Poincaré regards both the laws of mechanics and the ax-
ioms of geometry as constitutive conventions, their origin is different. While ge-
ometry is ultimately constructed by our mind by the a priori concept of a group
(Poincaré, 2001, 59), the laws of mechanics are idealized from our experience
(Poincaré, 2001, 105; Poincaré, 1913, 123-124). For Poincaré’s two-type conven-
15
Poincaré’s hierarchy of science is developed in Friedman (1999), chapter 3. tionalism, see Stump (2015), chapter 3.
 M. Ivanova / Studies in History and Philosophy of Science 54 (2015) 80e89 87

groundsdby how likely it is that the scientific community will Poincaré eventually accepted the atomic theory, showing that
want to revise a given statement. This is a very important conclu- experimental evidence can shift the status of hypotheses con-
sion; it shows where Duhem and Poincaré part waysdwhile for cerning the unobservable, he did not defend an entity-type realism.
Duhem science is seen in a holistic way, where all sentences have However, the question at stake here is whether Duhem and Poin-
the same epistemic status, for Poincaré science is a hierarchy where caré claim that science gives us epistemic access to a mind-
some elements have an asymmetric (constitutive) role and cannot independent world.19 Poincaré’s opposition to such realism is
be revised unless the whole theoretical framework is modified. For clear. When talking about the possibility of knowing a mind-
Poincaré the synthetic a priori and conventional elements serve a independent reality, he argues that:
different function within the theory and are not on equal epistemic
a reality completely independent of the mind which conceives
footing.
it, see it or feels it, is an impossibility. A world as exterior as that,
even if it existed, would for us be forever inaccessible. But what
we call objective reality is, in the last analysis, what is common
4.3. Where Duhem and Poincaré part ways
to many thinking beings, and could be common to all; this
common part, we shall see, can only be the harmony expressed
One of the most important tasks in evaluating Duhem’s and
by mathematical laws. It is this harmony then which is the sole
Poincaré’s place in the scientific realism debate regards whether
objective reality, the only truth we can attain. (Poincaré, 2001,
the terms ‘structuralism’ and ‘realism’ can be attributed to their
14)
positions. From the passages I discuss in Section 2, we can establish
that Poincaré and Duhem defend some kind of cumulative struc- In another place in The Value of Science, Poincaré argues that:
turalism. They believe that the content of a scientific theory
It will be said that science is only a classification and that a
amounts to its structural claimsdclaims about the relations among
classification cannot be true, but convenient. But it is true that it
‘things’ and not about the ‘nature’ of those thingsdand that those
is convenient, it is true that it is so not only for me, but for all
claims tends to survive theory change. Poincaré explicitly states
men; it is true that it will remain convenient for our de-
that:
scendants; it is true finally that this cannot be by chance. In sum,
The aim of science is not the things in themselves, as the dog- the sole objective reality consists in the relations of thing
matists in their simplicity imagine, but the relations between whence results the universal harmony. Doubtless these re-
things; outside those relations these is no reality knowable. lations cannot be conceived outside of a mind which conceives
(1902, xxiv) them. But they are nevertheless objective because they are, will
become, or will remain, common to all thinking beings.
Both Duhem and Poincaré believe that the mathematical content of
(Poincaré, 2001, 350)
the theory is the only content that captures true relations and hold
that these relations survive theory change. The structuralist idea at
first glance appears compatible with conventionalism, in particular With the help of the discussion of Poincaré’s hierarchy of science
the idea of conventions as implicit definitions. Implicit definitions we can now evaluate where Duhem and Poincaré stand in the
focus on the relational content between the elements that satisfy scientific realism debate. According to scientific and structural re-
them just like the structural realist focus on the mathematical re- alism, reality has a mind-independent structure and our best the-
lations between the elements that satisfy them and not on the ories provide true descriptions of this reality. These claims clash
‘nature’ of these elements. Two developments in mathematics with the above-mentioned claims that a mind-independent reality
particularly influenced Poincaré’s articulation of a structuralist cannot be known. Given that Poincaré’s thesis is based on the
position: Hilbert’s axiomatisation of geometry and Dedekind’s Kantian premise that we cannot discover facts about the world that
interpretation of arithmetic. According to Hilbert, geometrical are independent of our cognitive apparatus, it directly clashes with
concepts, such as lines and points, are defined only implicitly as the realist claim that that there is a mind-independent reality that
whatever satisfies the axioms of geometry. The same idea is artic- scientific theories discover. According to Poincaré’s epistemology,
ulated in arithmetic, where numbers are implicitly defined as ele- there is no direct access to the nature or structure of the world
ments that stand in a structure.17 However, contrary to the account because scientific knowledge is conditioned on the conventions we
of implicit definitions defended by Dedekind and Hilbert, Poin- decide to employ.20
caré’s conventions are offered as a response to the question of how Despite the fact that Poincaré articulates the argument from
we link mathematical structure to the appearances. continuity against the ‘bankruptcy of science’, an argument that
While it is clear that both Duhem and Poincaré develop a Worrall (1989) famously takes to support structural realism, it is
structuralist account of theories, it is controversial whether their difficult to see how Poincaré’s overall view on the structure of
position is ‘realist’. The question at stake here is not about justifi- scientific theories can be seen in a realist way. Whilst in a broad
cation of belief in unobservable entities. Duhem clearly opposed sense divorced from the synthetic a priori in the context of geom-
such realism with his lifelong rejection of atomism.18 While etry, Poincaré’s view has Kantian elements. Poincaré still relies on
the synthetic a priori in arithmetic and claims that the construction
of geometry, while guided by our experience, still had mind-
17
According to Dedekind (1888), structuralism in arithmetic allows for the dependent foundation, since the concept of a group from which
elimination of numbers from one’s ontology. Numbers are regarded as ‘free crea- we construct spatial geometry ‘preexists in our mind’. He makes
tions of the mind’ that have no properties over and above the properties they
knowledge of the world conditioned on our choice of geometry and
receive as elements in the system of arithmetic. The mathematical structuralism
articulated by Dedekind is eliminative (see Gower (2000)). constitutive principles.
18
Duhem refused to accept the atom on methodological grounds due to atoms
being unobservable entities. “The school of the neo-atomism [.] have taken up
again with supreme confidence the method we refuse to follow. This school thinks
its hypotheses attain at last the inner structure of matter, that they make us see the
19
elements as if some extraordinary ultra-microscope were to enlarge them until For a presentation of the different epistemic and metaphysical presuppositions
they were made perceivable to us.” (Duhem, 1996 [1913], 238). An account of this of scientific realism, see Chakravartty (2007).
20
rejection is developed in Achinstein (2007), Ivanova (2013) and Coco (2015). I explore this problem further in Ivanova (2015).
88 M. Ivanova / Studies in History and Philosophy of Science 54 (2015) 80e89

This discussion necessarily leads us to consider the role and constituted by synthetic a priori and conventional principles. While
status of mathematics in Duhem’s and Poincaré’s epistemologies both express a form of structuralism, Duhem’s and Poincaré’s
and this investigation requires another paper. Here I note that little different understandings of the status of scientific principles makes
work has been done in explicating Duhem’s approach to mathe- their claims regarding structural knowledge substantially different.
matical knowledge. While Poincaré emphasizes the constitutive
and a priori nature of mathematics, for Duhem “[m]athematics is Acknowledgements
the instrument necessary to construct all physical theory. But it is
only a means, not an end” (1996, 25). Mathematics, for Duhem, I am grateful to two anonymous referees for this journal for their
helps us to classify phenomena in a concise way, to represent the constructive comments.
phenomena in an elegant and convenient manner. While arith-
metic and geometry are revisable on Duhem’s framework, for
References
Poincaré they are synthetic a priori or conventional. In German
Science, Duhem argues that mathematical axioms are ‘self-evident’ Achinstein, P. (2007). Atom’s empirical eve: Methodological disputes and how to
(Duhem, 1991, 6-11)), in his (1906) he rejects Poincaré’s geometric evaluate them. Perspectives on Science, 15(Part 3), 359-390.
conventionalism and hierarchical approach to sciences by arguing Ben-Menahem, Y. (2001). Convention: Poincaré and some of his critics. British
Journal for Philosophy of Science, 52(3), 471-513.
that any element is revisable in light of experience, even geometry Ben-Menahem, Y. (2006). Conventionalism: From Poincaré to Quine. Cambridge:
and arithmetic. The important lesson to draw is that while Duhem Cambridge University Press.
takes mathematics as a representational too, for Poincaré it plays a Bhakthavatsalam, S. (2015). The rationale behind Pierre Duhem’s natural classifi-
cation. Studies in the History and Philosophy of Science, 51, 11-21.
constitutive role in our experience and access to the world. Bonnet, C., & de Calan, R. (2009). Moritz Schlick: Between synthetic a priori judg-
To sum up, despite the fact the structural realist interpretation ment and conventionalism. In M. Bitbol, P. Kerszberg, & J. Petitot (Eds.),
captures correctly Duhem’s and Poincaré’s aims to account for the Constituting objectivity: Transcendental perspectives on modern physics (pp. 117-
126). Berlin and New York: Springer.
argument from theory change and the argument for the novel Carnap, R. (1966). Philosophical foundations of physics, M. Gardner (Ed.). Basic
predictive and unifying power of scientific theories, the meta- Books: New York.
physical and epistemological claims of structural realism do not Chakravartty, A. (2007). A metaphysics for scientific realism: Knowing the unobserv-
able. Cambridge: Cambridge University Press.
represent Poincaré’s view. By arguing that scientific knowledge is Coco, K. (2015). Epistemology of a believing historian: Making sense of
built on conventional and synthetic a priori principles, Poincaré Duhem’s anti-atomism. Studies in the History and Philosophy of Science
opposes the possibility of knowing the mind-independent struc- Part A, 50, 71-82.
Darling, K. M. (2003). Motivational realism: The natural classification for Pierre
ture of the world. Duhem, on the other hand, just as Quine after
Duhem. Philosophy of Science, 70(5), 1125-1136.
him, renders all statement in a conjunction of beliefs empirical and Dedekind, R. (1996 [1888]). Was sind und was sollen die Zahlen? (W. W. Beman,
revisable in light of new evidence. As a consequence, despite the Trans.), pp. 790-833, From Kant to Hilbert: A source book in the foundations of
fact Duhem and Poincaré seem to both advance a structuralist mathematics, Vol. 2. W. B. Ewald (Ed.). Oxford, Clarendon Press.
Duhem, P. (1954 [1906]). The aim and structure of physical theory, Princeton:
position, their views substantially differ with respect to how Princeton University Press.
knowledge of the structure of the world is reached.21 Duhem, P. (1969 [1908]). To save the phenomena (E. Doland, & C. Mascher, Trans.).
Chicago: The University of Chicago Press.
Duhem, P. (1991 [1915]). German science: Some reflections on German science:
5. Conclusion German science and German virtues (J. Lyon, Trans.). La Salle, IL: Open Court.
Duhem, P. (1996 [1913]). Logical examination of physical theory. In R. Ariew, & P.
Barker (Eds.), Pierre Duhem: Essays in the History and Philosophy of Science
This paper analyses whether and in what contexts Duhem’s and
(pp. 29 50). Indianapolis: Hackett (First published 1913).
Poincaré’s views can be seen as conventionalist or structural realist Duhem, P. (1996). Essays in the history and philosophy of science. Hackett Publishing
and gives a novel, and hopefully more coherent, presentation of Company.
their views whilst also highlighting the essential differences be- Fairweather, A. (2012). The epistemic value of good sense. Studies in the History and
Philosophy of Science, 43, 139-146.
tween them. While it has been generally noted that Duhem and Friedman, M. (1999). Reconsidering logical positivism. Cambridge: Cambridge Uni-
Poincaré are conventionalists because they do not believe that versity Press.
theories can reveal the underlying nature of the world, on a closer Friedman, M. (2001). Dynamics of reason. CSLI Publications.
Giedymin, J. (1982). Science and convention: Essays on Henri Poincare’s philosophy of
look their aim is to balance the pessimistic meta-induction with the science and the conventionalist tradition. Oxford: Pergamon Press.
no-miracles argument and show that we can have knowledge of Gower, B. S. (2000). Cassirer, Schlick and ‘Structural’ realism: The Philosophy of the
the structure of the world. With regard to the problem of under- exact sciences in the background to early logical empiricism. British Journal for
the History of Philosophy, 8, 71-106.
determination, their views have also been misconstrued. Duhem’s Ivanova, M. (2010). Pierre Duhem’s good sense as a guide to theory choice. Studies in
solution to underdetermination is not a conventionalist one; he the History and Philosophy of Science, 41, 58-64.
develops an account of rational theory choice despite his belief that Ivanova, M. (2013). Did Perrin’s experiments convert Poincaré to scientific realism?
HOPOS: The Journal of the International Society for the History of Philosophy of
aesthetic values are not always sufficient for choice. For Poincaré,
Science, 3(1), 1-19.
on the other hand, choices are led by simplicity which promotes the Ivanova, M. (2015). Conventionalism, structuralism and neo-Kantianism in Poin-
ultimate goal of science, that of unification. Most importantly, when caré‫׳‬s philosophy of science. Studies in History and Philosophy of Modern Physics.
http://dx.doi.org/10.1016/j.shpsb.2015.03.003 (online first)
it comes to their epistemology, Duhem and Poincaré adopt very
Ivanova, M., & Paternotte, C. (2013). Theory choice, good sense and social consensus.
different views. Duhem’s epistemological holism renders all Erkenntnis, 78(5), 1109-1132.
knowledge empirical and does not assign special epistemic status Krips, H. (1986). Atomism, Poincaré and Plank. Studies in History and Philosophy of
to any theoretical postulates. Poincaré, on the other hand, develops Science, 17(1), 43-63.
Laudan, L. (1981). A confutation of convergent realism. Philosophy of Science, 48, 19-
a hierarchical approach to science, where empirical theories are 49.
Lugg, A. (1990). Pierre Duhem’s conception of natural classification. Synthese, 83,
409-420.
21 Maxwell, G. (1962). The ontological status of theoretical entities. In H. Feigl, &
The tension of equating Duhem’s and Poincaré’s positions is noted by Bonnet &
G. Maxwell (Eds.), Minnesota studies in the philosophy of science (Vol. 3, pp. 3-14)
de Calan (2009) who claims that: “even by any stretch of the imagination, Minneapolis: University of Minnesota Press.
comparing Duhem and Poincaré seems rather far-fetched given their radical McMullin, E. (1990). Comment: Duhem’s middle way. Synthese, 83, 421-430.
disagreement as far as the tenets on which the foundation of physical theory is Poincaré, H. (1963 [1913]). Science and mathematics: Last essays. Dover Publications.
based are concerned, that is to say the nature of the mathematical reasoning, the Poincaré, H. (2001 [1902, 1905, 1908]). The value of science: Essential writings of
role of definitions, the existence of a continuity or not between the mathematical Henri Poincaré, S. Gould (Ed.), New York: The Modern Library.
and physical methods, and the status of hypotheses in physics” (2009, 122). Popper, K. (1959). The logic of scientific discovery. London: Hutchinson.
 M. Ivanova / Studies in History and Philosophy of Science 54 (2015) 80e89 89

Quine, W. (1951). Two dogmas of empiricism. Philosophical Review, 60, 20-43. Stump, D. (2015). Conceptual change and the philosophy of science: Alternative in-
Reichenbach, H. (1958) Philosophy of space and time. M. Reichenbach (Trans. and terpretations of the A Priori. Rautledge.
Ed.), New York: Dover. Torretti, R. (1978). Philosophy of geometry from Riemann lo Poincare. D. Reidel
Russell, B. (1927). The analysis of matter. London: George Allen & Unwin. Publishing Company.
Schlick, M. (1978 [1915]) The philosophical significance of the principle of relativity. Worrall, J. (1996 [1989]) Structural realism: The best of both worlds? In D. Papineau
In Mulder & van de Velde-Schlick, pp. 153-189. (Ed.), The philosophy of science, pp. 139-166, Oxford: Oxford University Press.
Sklar, L. (1974). Space, time, and spacetime. Berkeley: University of California Press. Worrall, J. (1982). Scientific realism and scientific change. The Philosophical Quaterly,
Stanford, K. (2006). Exceeding our Grasp: Science, history, and the problem of 32(128), 201-231.
unconceived alternatives. Oxford: Oxford University Press. Zahar, E. (1997). Poincaré’s philosophy of geometry, or does geometric conven-
Stump, D. (1989). Henri Poincare’s philosophy of science. Studies in the History and tionalism deserve its name? Studies in the History and Philosophy of Modern
Philosophy of Science, 20, 335-363. Physics, 28, 183-218.
Stump, D. (2007). Pierre Duhem’s virtue epistemology. Studies in History and Phi- Zahar, E. (2001). Poincaré’s philosophy: From conventionalism to phenomenology.
losophy of Science, 38, 149-159. Open Court Publishing Company.