Contemporary Scientific Realism

Contemporary Scientific
Realism
The Challenge from the History of Science

Edited by

T I M O T H Y D. LYO N S A N D P E T E R V IC K E R S

1
 3
Oxford University Press is a department of the University of Oxford. It furthers
the University’s objective of excellence in research, scholarship, and education
by publishing worldwide. Oxford is a registered trade mark of Oxford University
Press in the UK and certain other countries.

Published in the United States of America by Oxford University Press

198 Madison Avenue, New York, NY 10016, United States of America.

© Oxford University Press 2021

All rights reserved. No part of this publication may be reproduced, stored in

a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by license, or under terms agreed with the appropriate reproduction
rights organization. Inquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above.

You must not circulate this work in any other form

and you must impose this same condition on any acquirer.

Library of Congress Control Number: 2021934069

ISBN 978–​0–​19–​094681–​4

DOI: 10.1093/​oso/​9780190946814.001.0001

1 3 5 7 9 8 6 4 2
Printed by Integrated Books International, United States of America
 Contents

1. History and the Contemporary Scientific Realism Debate  1

Timothy D. Lyons and Peter Vickers

PA RT I H I S T O R IC A L C A SE S F O R T H E D E BAT E

2. Theoretical Continuity, Approximate Truth, and the Pessimistic

Meta-​Induction: Revisiting the Miasma Theory  11
Dana Tulodziecki
3. What Can the Discovery of Boron Tell Us About the Scientific
Realism Debate?  33
Jonathon Hricko
4. No Miracle after All: The Thomson Brothers’ Novel Prediction that
Pressure Lowers the Freezing Point of Water  56
Keith Hutchison
5. From the Evidence of History to the History of Evidence: Descartes,
Newton, and Beyond  70
Stathis Psillos
6. How Was Nicholson’s Proto-​Element Theory Able to Yield
Explanatory as well as Predictive Success?  99
Eric R. Scerri
7. Selective Scientific Realism and Truth-​Transfer in Theories of
Molecular Structure  130
Amanda J. Nichols and Myron A. Penner
8. Realism, Physical Meaningfulness, and Molecular Spectroscopy  159
Teru Miyake and George E. Smith

PA RT I I C O N T E M P O R A RY S C I E N T I F IC R E A L I SM

9. The Historical Challenge to Realism and Essential Deployment  183

Mario Alai
10. Realism, Instrumentalism, Particularism: A Middle Path Forward
in the Scientific Realism Debate  216
P. Kyle Stanford
vi Contents

11. Structure not Selection  239

James Ladyman
12. The Case of the Consumption Function: Structural Realism
in Macroeconomics  257
Jennifer S. Jhun
13. We Think, They Thought: A Critique of the Pessimistic
Meta-​Meta-​Induction  284
Ludwig Fahrbach
14. The Paradox of Infinite Limits: A Realist Response  312
Patricia Palacios and Giovanni Valente
15. Realist Representations of Particles: The Standard Model, Top
Down and Bottom Up  350
Anjan Chakravartty

Index  375
 1
History and the Contemporary Scientific
Realism Debate
Timothy D. Lyons1 and Peter Vickers2

1Department of Philosophy

Indiana University–​Purdue University Indianapolis

[email protected]
2Department of Philosophy

Durham University
[email protected]

The scientific realism debate began to take shape in the 1970s, and, with the publica-
tion of two key early ’80s texts challenging realism, Bas van Fraassen’s The Scientific
Image (1980) and Larry Laudan’s “A Confutation of Convergent Realism” (1981),
the framework for that debate was in place. It has since been a defining debate of
philosophy of science. As originally conceived, the scientific realism debate is one
characterized by dichotomous opposition: “realists” think that many/​most of our
current best scientific theories reveal the truth about reality, including unobservable
reality (at least to a good approximation); and they tend to justify this view by the
“no miracles argument,” or by an inference to the best explanation, from the success
of scientific theories. Further, many realists claim that scientific realism is an empir-
ically testable position. “Antirealists,” by contrast, think that such a view is lacking in
epistemic care. In addition to discussions of the underdetermination of theories by
data and, less commonly, competing explanations for success, many antirealists—​in
the spirit of Thomas Kuhn, Mary Hesse, and Larry Laudan—​caution that the his-
tory of science teaches us that empirically successful theories, even the very best
scientific theories, of one age often do not stand up to the test of time.
The debate has come a long way since the 1970s and the solidification of its frame-
work in the ’80s. The noted dichotomy of “realism”/​“antirealism” is no longer a
given, and increasingly “middle ground” positions have been explored. Case studies
of relevant episodes in the history of science show us the specific ways in which a
realist view may be tempered, but without necessarily collapsing into a full-​blown
antirealist view. A central part of this is that self-​proclaimed “realists,” as well as
“antirealists” and “instrumentalists,” are exploring historical cases in order to learn

Timothy D. Lyons and Peter Vickers, History and the Contemporary Scientific Realism Debate In: Contemporary Scientific
Realism. Edited by: Timothy D. Lyons and Peter Vickers, Oxford University Press. © Oxford University Press 2021.
DOI: 10.1093/​oso/​9780190946814.003.0001
2 Contemporary Scientific Realism

the relevant lessons concerning precisely what has, and what hasn’t, been retained
across theory change. In recent decades one of the most important developments
has been the “divide et impera move,” introduced by Philip Kitcher (1993) and
Stathis Psillos (1999)—​possibly inspired by Worrall (1989)—​and increasingly
embraced by numerous other realists. According to this “selective” strategy, any re-
alist inclinations are directed toward only those theoretical elements that are really
doing the inferential work to generate the relevant successes (where those successes
are typically explanations and predictions of phenomena). Such a move is well
motivated in light of the realist call to explain specific empirical successes: those the-
oretical constituents that play no role, for instance, in the reasoning that led to the
successes, likewise play no role in explaining those successes. Beyond that, however,
and crucially, the divide et impera move allows for what appears to be a testable re-
alist position consistent with quite dramatic theory-​change: the working parts of a
rejected theory may be somehow retained within a successor theory, even if the two
theories differ very significantly in a great many respects. Even the working parts
may be retained in a new (possibly approximating) form, such that the retention
is not obvious upon an initial look at a theoretical system but instead takes consid-
erable work to identify. This brings us to a new realist position, consistent with the
thought that many of our current best theories may one day be replaced. Realism
and antirealism are no longer quite so far apart, and this is progress.
A major stimulus for, and result of, this progress has been a better under-
standing of the history of science. Each new case study brings something new to
the debate, new lessons concerning the ways in which false theoretical ideas have
sometimes led to success, or the ways in which old theoretical ideas “live on,” in
a different form, in a successor theory. At one time the debate focused almost
exclusively on three famous historical cases: the caloric theory of heat, the phlo-
giston theory of combustion, and the aether theory of light. An abundance of lit-
erature was generated, and with good reason: these are extremely rich historical
cases, and there is no simple story to tell of the successes these theories enjoyed,
and the reasons they managed to be successful despite being very significantly
misguided (in light of current theory). But the history of science is a big place,
and it was never plausible that all the important lessons for the debate could be
drawn from just three cases. This is especially obvious when one factors in a “par-
ticularist” turn in philosophy of science where focus is directed toward partic-
ular theoretical systems. These days, many philosophers are reluctant to embrace
grand generalizations about “science” once sought by philosophers, taking those
generalizations to be a dream too far. Science works in many different ways, in
different fields and in different contexts. It follows that the realism debate ought
to be informed by a rich diversity of historical cases.
It is with this in mind that the present volume is put forward. In recent years
a flood of new case studies has entered the debate, and is just now being worked
 History and the Contemporary Scientific Realism Debate 3

through. At the same time it is recognized that there are still many more cases
out there, waiting to be analyzed in a particular way sensitive to the concerns
of the realism debate. The present volume advertises this fact, introducing as it
does several new cases as bearing on the debate, or taking forward the discus-
sion of historical cases that have only very recently been introduced. At the same
time, the debate is hardly static, and as philosophical positions shift this affects
the very kind of historical cases that are likely to be relevant. Thus the work of
introducing and analyzing historical cases must proceed hand in hand with phil-
osophical analysis of the different positions and arguments in play. It is with this
in mind that we divide the present volume into two parts, covering “Historical
Cases for the Debate” and “Contemporary Scientific Realism,” each comprising
of seven chapters.
Many of the new historical cases are first and foremost challenges to the re-
alist position, in that they tell scientific stories that are apparently in tension with
even contemporary, nuanced realist claims. The volume kicks off in Chapter 2
with just such a case, courtesy of Dana Tulodziecki. For Tulodziecki, the miasma
theory of disease delivered very significant explanatory and predictive successes,
while being radically false by present lights. Further, even the parts of the theory
doing the work to bring about the successes were not at all retained in any suc-
cessor theory. Thus, Tulodziecki contends, the realist must accept that in this case
false theoretical ideas were instrumental in delivering successful explanations
and predictions of phenomena.
This theme continues in Chapter 3, with Jonathon Hricko’s study of the dis-
covery of boron. This time the theory in question is Lavoisier’s oxygen theory
of acidity. Hricko argues that the theory is not even approximately true, and yet
nevertheless enjoyed novel predictive success of the kind that has the power to
persuade. Just as Tulodziecki, Hricko argues that the realist’s divide et impera
strategy can’t help—​the constituents of the theory doing the work to generate the
prediction cannot be interpreted as approximately true, by present lights.
We meet with a different story in Chapter 4, however. Keith Hutchison provides
a new case from the history of thermodynamics, concerning the successful pre-
diction that pressure lowers the freezing point of water. Hutchison argues that al-
though, at first, it seems that false theoretical ideas were instrumental in bringing
about a novel predictive success, there is no “miracle” here. It is argued that the
older Carnot theory and the newer Clausius theory are related in intricate ways,
such that in some respects they differ greatly, while in “certain restricted situations”
their differences are largely insignificant. Thus the realist may find this case useful
in her bid to show how careful we need to be when we draw antirealist morals from
the fact that a significantly false theory enjoyed novel predictive success.
Chapter 5 takes a slightly different approach, moving away from the narrow
case study. Stathis Psillos surveys a broad sweep of history ranging from
4 Contemporary Scientific Realism

Descartes through Newton and Einstein. We have here a “double case study,”
considering the Descartes–​Newton relationship and the Newton–​Einstein rela-
tionship. Psillos argues against those who see here examples of dramatic theory-​
change, instead favoring a limited retentionism consistent with a modest realist
position. Crucially, Psillos argues, there are significant differences between the
Descartes–​Newton relationship and the Newton–​Einstein relationship, in line
with the restricted and contextual retention of theory predicted by a nuanced,
and epistemically modest, contemporary realist position.
Chapter 6—​courtesy of Eric Scerri—​introduces another new historical case,
that once again challenges the realist. This time the theory is John Nicholson’s
atomic theory of the early 20th century, which, Scerri argues, “was spectacu-
larly successful in accommodating as well as predicting some spectral lines in
the solar corona and in the nebula in Orion’s Belt.” The theory, however, is very
significantly false; as Scerri puts it, “almost everything that Nicholson proposed
was overturned.” Hence, this case is another useful lesson in the fact that quite
radically false scientific theories can achieve novel predictive success, and any
contemporary realist position needs to be sensitive to that.
Chapter 7 turns to theories of molecular structure at the turn of the 20th
century. Amanda Nichols and Myron Penner show how the “old” Blomstrand-​
Jørgensen chain theory was able to correctly predict the number of ions that will
be dissociated when a molecule undergoes a precipitation reaction. While prima
facie a challenge to scientific realism, it is argued that this is a case where the di-
vide et impera strategy succeeds: the success-​generating parts of the older theory
are retained within the successor, Werner’s coordination theory.
The final contribution to the “History” part of the volume—​Chapter 8—​
concerns molecular spectroscopy, focusing on developments in scientific
“knowledge” and understanding throughout the 20th century and right up to
the present day. Teru Miyake and George E. Smith take a different approach from
the kind of historical case study most commonly found in the realism debate.
Siding with van Fraassen on the view that Perrin’s determination of Avogadro’s
number, so commonly emphasized by realists, does not prove fruitful for re-
alism, and focusing on diatomic molecules, they emphasize instead the extraor-
dinary amount of evidence that has accumulated after Perrin and over the past
ninety years for various theoretical claims concerning such molecules. Taking
van Fraassen’s constructive empiricism as a foil, they indicate that, in this case,
so-​called realists and antirealists may really differ very little when it comes to this
area of “scientific knowledge.”
The second part of the volume turns to more general issues and philosoph-
ical questions concerning the contemporary scientific realist positions. This
part kicks off in Chapter 9 with Mario Alai’s analysis of the divide et impera
realist strategy: he argues that certain historical cases no longer constitute
 History and the Contemporary Scientific Realism Debate 5

counterexamples when hypotheses are essential in novel predictions. He

proposes refined criteria of essentiality, suggesting that while it may be impos-
sible to identify exactly which components are essential in current theories, rec-
ognizing in hindsight those which were not essential in past theories is enough to
make the case for deployment realism.
In Chapter 10, Kyle Stanford makes the case that so-​called realists and
instrumentalists are engaged in a project together. For Stanford, these tradition-
ally diametrically opposed protagonists are now working together to “actively
seek to identify, evaluate, and refine candidate indicators of epistemic security
for our scientific beliefs.” As philosophy of science, and the realism debate in par-
ticular, becomes increasingly “local,” Stanford also argues for “epistemic guid-
ance intermediate in generality” between the sweeping generalizations of 1970s
and ’80s realism, and a radical “particularism” where any realist claim should
always be specific to one particular theory, or theoretical claim.
Chapter 11 turns to the relationship between the realist’s divide et impera
strategy and the structural realist position. James Ladyman argues that struc-
tural realism is not a form of selective realism (or at least doesn’t have to be).
For Ladyman, structural realism represents a departure from standard scientific
realism, not a modification of it. He also argues that scientific realists face on-
tological questions (not only epistemic ones), and he defends a “real patterns”
approach to what he calls the “scale relativity of ontology.” This allows for
equally “realist” claims to be made at the level of fundamental physics and at the
macroscopic level.
In Chapter 12, Jennifer Jhun explores the realism debate in a different terri-
tory. In particular, she considers the possibility of taking a structural realist atti-
tude toward macroeconomic theory. Taking the consumption function as a case
study, she argues that a better take on macroeconomic theory involves a compro-
mise between structural realism and instrumentalism. For Jhun, when it comes
to economics (at least), “theories are instruments used to find out the truth.”
In Chapter 13, Ludwig Fahrbach considers a prominent antirealist argument
against the claim that realism can be defended against the pessimistic meta-​
induction (PMI) by invoking the exponential growth of scientific evidence. The
antirealist response to this defense depends on the claim that realists could have
said the same thing in the past. He introduces this antirealist response as the
“PMMI,” the pessimistic meta-​meta-​induction. Fahrbach’s challenge focuses on
a particular weak spot common to both the traditional PMI and the PMMI. Thus
realists unimpressed by the traditional PMI will not be moved by the new PMMI.
Chapter 14 turns to another important aspect of the modern realism de-
bate: the use of “radically false” theoretical assumptions such as infinite limits in
many contemporary, highly successful theories. Patricia Palacios and Giovanni
Valente note how “infinite idealizations” misrepresent the target system, and
6 Contemporary Scientific Realism

sometimes it appears that the introduction of such blatant falsity is necessary

to achieve empirical success. Focusing on various examples in physics, such as
classical and quantum phase transitions as well as thermodynamically reversible
processes, Palacios and Valente propose a realist response to such cases.
Last but not least, in Chapter 15 Anjan Chakravartty tackles the standard
model of particle physics, and in particular what the realist might say about
representations of fundamental particles. Introducing the realist “tightrope,”
Chakravartty discusses the trade-​off between committing to too little, and
committing to too much. But in the end, he argues, this is a tightrope that is not
too thin to walk.
These articles have been carefully collected for this volume over many years,
and in particular during the Lyons/​Vickers 2014–​18 Arts and Humanities
Research Council (AHRC) project “Contemporary Scientific Realism and the
Challenge from the History of Science.” The project enjoyed seven major events
over its lifetime, out of which the fourteen substantive chapters of this volume
ultimately grew. The seven events were:

(i) “The History of Chemistry and Scientific Realism,” a two-​day work-

shop held at Indiana University–​Purdue University Indianapolis, United
States, December 6–​7, 2014.
(ii) “The History of Thermodynamics and Scientific Realism,” a one-​day
workshop held at Durham University, UK, on May 12, 2015.
(iii) “Testing Philosophical Theories against the History of Science,” a one-​
day workshop held at the Oulu Centre for Theoretical and Philosophical
Studies of History, Oulu University, Finland, on September 21, 2015.
(iv) “Quo Vadis Selective Scientific Realism?”—​a symposium at the bien-
nial conference of the European Philosophy of Science Association,
Düsseldorf, Germany, September 23, 2015.
(v) “Contemporary Scientific Realism and the Challenge from the History
of Science,” a three-​day conference held at Indiana University–​Purdue
University, United States, February 19–​21, 2016.
(vi) “Quo Vadis Selective Scientific Realism?”—​a three-​day conference held
at Durham University, UK, August 5–​7, 2017.
(vii) “The Structure of Scientific Revolutions,” a two-​day workshop held at
Durham University, UK, October 30–​31, 2017.

The editors of this volume owe a great debt to the participants of all of these
events, with special thanks in particular to those who walked this path with us
a little further to produce the fourteen excellent chapters here presented. We are
also grateful to the unsung heroes, the many anonymous reviewers, who not
only helped us to select just which among the numerous papers submitted would
 History and the Contemporary Scientific Realism Debate 7

be included in the volume but also provided thorough feedback to the authors,
helping to make each of the chapters that did make the cut even stronger. The
volume has been a labor of love; we hope that comes across to the reader.

References
Kitcher, P. (1993). The Advancement of Science, Oxford: Oxford University Press.
Laudan, L. (1981). “A Confutation of Convergent Realism.” Philosophy of Science
48: 19–​49.
Psillos, S. (1999). Scientific Realism: How Science Tracks Truth, London: Routledge.
Van Fraassen, B. (1980). The Scientific Image, Oxford: Oxford University Press.
Worrall, J. (1989). “Structural Realism: The Best of Both Worlds?” Dialectica 43: 99–​124.
 PART I
HISTOR IC A L CASE S
F OR T HE DE BAT E
 2
Theoretical Continuity, Approximate
Truth, and the Pessimistic Meta-​Induction:
Revisiting the Miasma Theory
Dana Tulodziecki

Department of Philosophy
Purdue University
[email protected]

2.1 Introduction

The pessimistic meta-​induction (PMI) targets the realist’s claim that a theory’s
(approximate) truth is the best explanation for its success. It attempts to do so
by undercutting the alleged connection between truth and success by arguing
that highly successful, yet wildly false theories are typical of the history of sci-
ence. There have been a number of prominent realist responses to the PMI, most
notably those of Worrall (1989), Kitcher (1993), and Psillos (1999). All of these
responses try to rehabilitate the connection between a theory’s (approximate)
truth and its success by attempting to show that there is some kind of continuity
between earlier and later theories, structural in the case of Worrall and theo-
retical/​referential in the cases of Kitcher and Psillos, with other responses being
variations on one of these three basic themes.1
In this paper, I argue that the extant realist responses to the PMI are in-
adequate, since there are cases of theories that were both false and highly
successful (even by the realist’s own, more stringent criteria for success)
but that, nevertheless, do not exhibit any of the continuities that have been

1 It is worth pointing out that this is the case even for the (very) different structural realisms

that abound. In particular, even ontic structural realism, which differs substantially from
Worrall’s epistemic version, shares with Worrall’s approach an emphasis on mathematical con-
tinuities among successor theories (see, for example, Ladyman (1998) and French and Ladyman
(2011)).

Dana Tulodziecki, Theoretical Continuity, Approximate Truth, and the Pessimistic Meta- Induction: Revisiting the
Miasma Theory In: Contemporary Scientific Realism. Edited by: Timothy D. Lyons and Peter Vickers, Oxford University
Press. © Oxford University Press 2021. DOI: 10.1093/​oso/​9780190946814.003.0002
12 Historical Cases for the Debate

suggested by realists as possible candidates for preservation. I will make my

case through discussing an example of such a theory: the 19th century mi-
asma theory of disease. Specifically, I show that this theory made a number of
important and successful use-​novel predictions, despite the fact that its cen-
tral theoretical element—​miasma—​turned out not to exist. After showing
that miasma was crucially involved in virtually every successful prediction
the miasma theory made, I argue that not just is there no ontological con-
tinuity between the miasma theory and its successor, but neither can a case
be made for any other kind of continuity, be it in terms of structure, laws,
mechanisms, or kind-​constitutive properties. I conclude by arguing that
realists face problems regardless of whether the miasma theory is approxi-
mately true: if it is, the prospects for any kind of substantive realism are dim;
if it is not, the miasma case constitutes a strike against the “convergent” part
of convergent realism.
I will proceed as follows: In section 2.2, I briefly outline the PMI and the
responses by Worrall, Kitcher, and Psillos, including the more stringent notion
of success that Psillos argues is required in order for theories to be genuinely
successful. In section 2.3, I outline the most sophisticated version of the miasma
theory of disease and show that it ought to be considered a genuinely successful
theory, even on Psillos’s own terms. Section 2.4 is concerned with showing that
none of the extant realist responses to the PMI can account for the case of the
miasma theory. After discussing some objections in section 2.5, I conclude in
section 2.6.

2.2 The PMI and Realist Responses

The PMI received its most sophisticated and explicit formulation in Laudan
(1981, 1984). The argument’s main target is the Explanationist Defense of
Realism, according to which the best explanation for the success of science is
the (approximate) truth of our scientific theories (see, for example, Boyd (1981,
1984, 1990)). Anti-​realists employ the PMI to undercut this alleged connection
between truth and success by pointing to the (in their view) large number of
scientific theories that were discarded as false, yet regarded as highly successful.
Laudan, for example, provides a list of such theories that includes, among others,
the phlogiston theory of chemistry, the caloric theory of heat, and the theory
of circular inertia (1984: 121). However, since “our own scientific theories are
held to be as much subject to radical conceptual change as past theories” (Hesse
1976: 264), it follows, the anti-​realist argues, that the success of our current the-
ories cannot legitimize belief in unobservable entities and mechanisms: just as
 Revisiting the Miasma Theory 13

past theories ended up wrong about their postulates, we might well be wrong
about ours.2
Realist responses to the PMI typically come in several stages: first, realists try to
winnow down Laudan’s list by including only those theories that are genuinely suc-
cessful (Psillos 1999: ­chapter 5). While, according to Laudan, a theory is successful
“as long as it has worked reasonably well, that is, so long as it has functioned in a va-
riety of explanatory contexts, has led to several confirmed predictions, and has been
of broad explanatory scope” (1984: 110), Psillos argues that “the notion of empirical
success should be more rigorous than simply getting the facts right, or telling a story
that fits the facts. For any theory (and for that matter, any wild speculation) can be
made to fit the facts—​and hence to be successful—​by simply “writing” the right kind
of empirical consequences into it. The notion of empirical success that realists are
happy with is such that it includes the generation of novel predictions which are in
principle testable” (1999: 100). The specific type of novel prediction that Psillos has in
mind is so-​called use-​novel prediction: “the prediction P of a known fact is use-​novel
relative to a theory T, if no information about this phenomenon was used in the con-
struction of the theory which predicted it” (101). Once success is understood in these
more stringent terms, Psillos claims, Laudan’s list is significantly reduced.
With the list so shortened, the next step of the realist maneuver is to argue that
those theories that remain on Laudan’s list ought to be regarded as (approximately)
true, since they don’t, in fact, involve the radical discontinuity with later theories
that Laudan suggests. Rather than being discarded wholesale during theory-​change,
it is argued that important elements of discarded theories get retained: according to
Worrall (1989, 1994), theories’ mathematical structures are preserved, according to
Kitcher (1993) and Psillos (1996, 1999), those parts of past theories that were re-
sponsible for their success are. Because these components are preserved in later the-
ories, the argument goes, we ought to regard as approximately true those parts of
our current theories that are essentially involved in generating their successes, since
those are the parts that will carry over to future theories, just as essential elements
from earlier theories were carried over to our own. As I will show in section 2.4,
however, none of these strategies will work for the miasma theory: despite the fact
that the miasma theory made a number of use-​novel predictions, and so counts as
genuinely successful by realist standards, none of the elements or structures that
were involved in those successes were retained by its successor.3

2 Laudan’s argument is usually construed as a reductio (see Psillos (1996, 1999)). For some recent

discussions about how to properly interpret the argument, see Lange (2002), Lewis (2001), and Saatsi
(2005).
3 The miasma case is especially significant in view of the markedly different analysis that Saatsi

and Vickers (2011) provide of Kirchhoff ’s theory of diffraction. Saatsi and Vickers diagnose a specific
kind of underdetermination at work in the Kirchhoff case and argue that “it should not be implausible
14 Historical Cases for the Debate

2.3 The Miasma Theory and Its Successes

The most sophisticated version of the miasma theory saw its heyday in the
mid-​1800s. The situation with respect to the various accounts of disease at that
time was fairly complicated, however, and so it is in some sense misleading to
speak of the miasma theory of disease, since there was a whole cluster of related
views that went under this label, rather than one easily identifiable position
(see, for example, Baldwin (1999), Eyler (2001), Hamlin (2009), Pelling (1978),
and Worboys (2000)). However, since all members of that cluster shared basic
assumptions about the existence and nature of miasma, I will disregard this
complication here, and treat them as one. According to this basic miasma view,
diseases were caused and transmitted by a variety of toxic odors (“miasmas”) that
themselves were the result of rotting organic matter produced by putrefaction or
decomposition. The resulting bad air would then act on individual constitutions
to cause one of several diseases (cholera, yellow fever, typhus, etc.), depending
on a number of more specific factors. Some of these were thought to be extra-
neous, such as weather, climate, and humidity, and would affect the nature of
the miasmas themselves; others were related directly to the potential victims and
thought to render them more or less susceptible to disease, such as their general
constitution, moral sense, age, and so on. Lastly, there were a variety of local
conditions that could exacerbate the course and severity of the disease, such as
overcrowding and bad ventilation.
Although the miasma theory is sometimes contrasted with so-​ called
“contagionist” views of disease (the view that diseases could be transmitted di-
rectly from individual to individual), this opposition is also misleading, for two
reasons: first, because both contagionists and anti-​contagionists subscribed to
the basic miasmatic assumptions just described, with the debate not centering
on the existence or nature of miasmas, but, rather, on whether people them-
selves were capable of producing additional miasma-​like effects and through
these “exhalations” directly give the disease to others. Second, although some
diseases were generally accepted as contagious (smallpox, for example), and
some were generally held to be non-​contagious (malaria), most diseases (yellow
fever, cholera, typhoid fever, typhus, and so on) fell somewhere in between these

to anyone that given the enormous variation in the nature and methods of scientific theories across
the whole spectrum of ‘successful science’, some domains of enquiry can be more prone to this kind
of underdetermination than others” (44). In fact, they believe that “witnessing Kirchhoff ’s case, there
is every reason to expect that from a realist stance we can grasp the features of physics and mathe-
matics that contribute to such differences” (44). As a result, they take themselves to have “argued for
the prima facie plausibility” of a realist response that focuses on “showing how the field of theorizing
in question is idiosyncratic in relevant respects, so that Kirchhoff ’s curious case remains isolated and
doesn’t provide the anti-​realist with grounds for projectable pessimism” (44). The miasma case, how-
ever, is not prone to any of the idiosyncrasies Saatsi and Vickers identify (or any other idiosyncrasies,
as far as I can tell).
 Revisiting the Miasma Theory 15

two extremes. Pelling (1978: 9) appropriately terms these diseases the “doubtful
diseases”: instead of being straightforwardly contagionist or anti-​contagionist
about them, people espoused so called “contingent contagionism” with respect to
them, holding that they could manifest as either contagious or non-​contagious,
depending on the exact circumstances and, sometimes, even transform from one
into the other (see also Hamlin (2009)).
This version of the miasma theory was extremely successful with respect to a
number of different types of phenomena. Most famous are probably the sanitary
successes that it ultimately inspired, but it also managed to provide explanations
of disease phenomena that any theory of disease at the time had to accommo-
date. These included the fact that diseases were known to be highly seasonal, that
particular regions (especially marshy ones) were affected particularly harshly,
that specific locations (prisons, workhouses, etc.) were often known to suffer
worse than their immediate surroundings, that sickness and mortality rates in
urban centers were much worse than those in rural areas, and why particular
geographical regions/​countries were struck much worse by disease than others.
The miasma theory managed to explain all of these through its claims that de-
composition and putrefaction of organic material was responsible for producing
miasmas. Diseases peaked when conditions for putrefaction were particularly
favorable: this was the reason why certain diseases were particularly bad during
periods of high temperature and in certain geographical regions (for example,
the many fevers in Africa), why urban centers were much more affected than
rural areas, and why even specific locations in otherwise more or less healthy
areas could be struck (sewage, refuse, and general “filth” would sit around in
badly ventilated areas). It should also be noted that miasma theorists were not
just making vague or simple-​minded pronouncements about stenches produ-
cing toxic odors, but embraced very specific and often highly complex accounts
of how various materials and conditions gave rise to miasmas—​Farr, for ex-
ample, drew in some detail on Liebig’s chemical explanations (1852, lxxx-​lxxxiii;
see also Pelling (1978: ­chapters 3 and 4) and Tulodziecki (2016)). Moreover,
there were debates about exactly what sorts of materials were prime for potential
miasmas, such as debates about various sources of vegetable vs. animal matter,
and so on. Since, however, I cannot do justice to the details of these accounts and
their corresponding successes here, I will focus on a somewhat simpler, yet par-
ticularly striking, example of use-​novel success, while merely noting that others
could be given.
The example in question is that of William Farr’s (then famous) elevation law
(1852). Farr (1807–​83), although not himself a physician, was viewed as an au-
thority on infectious diseases in mid-​1800s Britain. Among the positions he
held were that of Statistical Superintendent of the General Register Office and
member of the Committee of Scientific Inquiries. Through his various writings,
16 Historical Cases for the Debate

especially those on the various big British cholera epidemics, he established his
credentials as a medical authority.4
As we have seen, according to the miasma theory, any decomposing organic
material could in principle give rise to miasmas. However, it was thought that
the soil at low elevations, especially around riverbanks, was a particularly good
source for producing highly concentrated miasma, since such soil held plenty of
organic material and the conditions for putrefaction were particularly favorable.
It thus followed directly from the miasma theory that, if miasmas were really
produced in the manner described and responsible for disease, the concentra-
tion of noxious odors ought to be higher closer to such sources and dissipate
with increasing distance. Correspondingly, it was to be expected that mortality
and sickness rates would be higher in close proximity to sources of miasma, de-
clining as one moved away. Farr confirmed that this was the case, in a number of
different ways.
First, he found that “nearly 80 percent of the 53,000 registered cholera deaths
in 1849 occurred among four-​tenth of the population living on one-​seventh of
the land area” (Eyler 1979: 114), and, moreover, that, “cholera was three times
more fatal on the coast than in the interior of the country” (Farr 1852: lii).
Furthermore, he found that those deaths that did occur inland were in either
seaport districts or close to rivers, noting that “[c]‌holera reigned wherever it
found a dense population on the low alluvial soils of rivers” and that it “was al-
most invariably most fatal in the port or district lying lowest down the river”
(lii). Concerning coastal deaths, he found that the “low-​lying towns on the coast
were all attacked by cholera,” while the high-​lying coast towns “enjoyed as much
immunity as the inland towns” (liv). Further, he noted that mortality increased
and decreased relative to the size of the port, with smaller ports having lower
mortality. The Welsh town of Merthyr-​Tydfil constituted an exception to this,
having a “naturally” favorable location, yet relatively high mortality. However,
Farr also noted that Merthyr-​Tydfil had a reputation, with the Health of Towns’
Commissioners’ Report noting that “[f]rom the poorer class of the inhabitants,
who constitute the mass of the population, throwing all slops and refuse into
the nearest open gutter before their houses, from the impeded courses of such
channels, and the scarcity of privies, some parts of town are complete networks of
filth emitting noxious exhalations” (liv). In addition, much of the refuse was being
carried to the local riverbeds, with the result that “the stench is almost intolerable
in many places” (lv). In one area, close to the river, an “open, stinking, and nearly
stagnant gutter, into which the house refuse is as usual flung, moves slowly before
the doors” (lvi).

4 For more on Farr’s life and ideas, see Eyler (1979).

 Revisiting the Miasma Theory 17

All of these phenomena were exactly what was to be expected on the mi-
asma theory: wherever there was disease, one ought to be able to trace it back
to miasmatic conditions and, similarly, wherever conditions particularly favor-
able to decomposition were to be found, disease ought to be rampant. If there
were exceptions to the general rule, towns such as Merthyr-​Tydfil that were not
naturally vulnerable, one ought to be able to find alternative sources of miasma
without difficulty. Farr also proceeded to check these results against a variety of
data from different countries and concerning different diseases, and further con-
firmed what he had found. Moreover, the miasma theory did not merely accom-
modate these findings but, rather, all of the phenomena followed naturally from
the account.
In addition, while the miasma theory made predictions about what areas
ought to be affected by cholera and to what degrees, for example, none of these
predictions were confirmed until Farr analyzed the data from the General Board
of Health in the late 1840s (indeed, due to the fact that much of this data was not
collected until shortly before that time, it would have been impossible to confirm
these predictions until then). Thus, since it was not even clearly known to what
extent these predictions were borne out, they could not have played a role in
formulating the miasma theory in the first place and, so, ought to count as use-​
novel. One might object that, use-​novel or not, these predictions were simply too
vague to qualify a theory as genuinely successful in the realist sense. I will note,
however, that (i) the predictions were as specific as a theory of this type would
allow, (ii) the predictions were no more vague than Snow’s later predictions
about what ought to be expected on the (correct) assumption that cholera was
waterborne (see Snow 1855a, 1855b) and, so, if one regards Snow’s predictions as
successful, one ought to also regard the miasmatic predictions as successful, and
(iii) that miasmatists actually did a lot better than this (and, indeed, better than
Snow ever did) by providing some detailed and quantifiable results.
Farr’s elevation law is a particularly striking example of this and it is to this
that I will turn now. Farr’s law related cholera mortality to the elevation of the
soil. However, Farr did not just predict that there ought to be a relationship
between these two variables, but upon analyzing more than 300 pages of data,
he found that the “mortality from cholera is in the inverse ratio of the eleva-
tion” (Farr 1852: lxi). Farr grouped the various London districts by their alti-
tude above the Thames high water mark, in brackets of 20 feet (0–​20, 20–​40,
and so on) and was able to capture the exact relation between the decline of
cholera and increased soil elevation in the form of an equation. Specifically, he
found that:

The mortality from cholera on the ground under 20 feet high being represented
by 1, the relative mortality in each successive terrace [i.e. the terraces numbered
18 Historical Cases for the Debate

Figure 2.1 Farr, William. (1852). “Report on the Mortality of Cholera in England,
1848–49.” London: Printed by W. Clowes, for H.M.S.O., 1852, p. lxii.

“2,” “3,” etc.] is represented by ½ [for terrace 2], ⅓ [for terrace 3], ¼, ⅕, ⅙: or
the mortality on each successive elevation is ½, ⅔, ¾, ⅘, ⅚, &c. of the mortality
on the terrace immediately below it. (ibid.: lxiii; see Figure 2.1)

He then generalized this result: “Let e be any elevation within the observed
limits 0 and 350, and c be the average rate of mortality from cholera at that eleva-
tion; also let eʹ be any higher elevation, and cʹ the mortality at that higher eleva-
tion” (lxiii). Then, adding a as a constant, Farr found that the formula

c = c′ × (e′ + a)/​(e + a)

represented the decreasing series he obtained when considering the mor-

tality from cholera for districts with specific mean elevations. Farr then
calculated the expected series according to the formula, compared it to the
actual series recorded in London, and found remarkable agreement (see
Figure 2.2, lxiii).
Seeking further confirmation, Farr immediately proceeded to “submit the
principle to another test, by comparing the elevation and the mortality from
cholera of each sub-​district,” and found that this “entirely confirms the announced
law” (xv–​xvi).5 Trying to illustrate just how good Farr’s numbers were and how

5 The tables appear on pp. clxvi–​ix of Farr’s Report. (1852).

 Revisiting the Miasma Theory 19

Figure 2.2 Farr, William. (1852). “Report on the Mortality of Cholera in England,
1848–49.” London: Printed by W. Clowes, for H.M.S.O., 1852, p. lxiii

convincing they must have seemed, Langmuir (1961: 174) plotted Farr’s result
about cholera mortality in the various London subdistricts, grouped by elevation
(Figure 2.3). According to Langmuir, Farr had “found a confirmation that I be-
lieve would be impressive to any scientist at any time” (173).
Even more remarkably, it turned out that Farr’s predictions did not just hold
for the (sub-​)districts of London, but were also confirmed by others in different
regions. For example, “William Duncan, Medical Officer of Health for Liverpool,
wrote that when he grouped the districts of his city by elevation as Farr had
done, that cholera mortality in the last epidemic obeyed Farr’s elevation law for
Liverpool as well” (Eyler 1979: 228).
Now, these predictions of Farr’s were certainly use-​novel: Farr was predicting
new phenomena that were hitherto unknown and that were later borne out
by a variety of data from different regions, in different contexts, and from dif-
ferent times. Moreover, Farr’s law clearly could not have played a role in the
construction of the miasma theory since, first, it was obviously not even for-
mulated by then, but, second, even the data on which the law was based (the
statistics from the General Register Office, collected on Farr’s initiative) did not
exist and, indeed, in the case of Duncan in Liverpool, no one had even thought
about collecting the relevant information. In short: it followed from Farr’s law
that cholera mortality and soil elevation ought to exhibit a specific relation that
was then found to occur in the various sub-​districts of London and various other
parts of the country.
20 Historical Cases for the Debate

Figure 2.3 Correlation of cholera mortality and elevation above the Thames
River, London, 1849. Langmuir, Alexander D. (1961). “Epidemiology of Airborne
Infection.” Bacteriological Reviews, 25(3), p. 174.

2.4 Miasmas and Realists

As we have seen, the miasma theory made a number of use-​novel predictions

and so, by the realist’s own standard, ought to count as genuinely successful.6
However, the existence of such cases—​successful, yet false theories—​is not
enough, by itself, to spell trouble for realists. After all, realists themselves
admit that there are such examples. They also argue that we can explain
these cases by showing that crucial elements of those theories—​t hose that
were essentially involved in the theory’s success—​are preserved by later

6 Doppelt (2007) argues that requiring novel predictions is not sufficient for genuine success; in-

stead, we ought to also look for various explanatory virtues, such as consilience, simplicity, or uni-
fying power. To make a detailed case for this would take us too far afield here, but it can be shown
that the miasma theory also exhibits some of these more traditional virtues. Certainly one might
argue that independent strands of evidence (epidemiological and pathological, for example) pro-
duced a degree of consilience and that the miasma theory unified a number of different phenomena
(regarding the various diseases) in a simple and elegant way by providing essentially the same kind of
explanation for a number of different afflictions.
 Revisiting the Miasma Theory 21

theories. Kitcher (1993: 149) distinguishes working from presuppositional

posits and argues that cases support the PMI only if it is found that a theory’s
working posits (those elements of a theory essential for making predictions)
do not refer. Psillos resorts to what he calls the divide et impera move, and
claims that “it is enough to show that the theoretical laws and mechanisms
which generated the successes of past theories have been retained in our
current scientific image” (1999: 108). Worrall (1989: 11) also argues that
parts of past theories get retained, but instead of focusing on theoretical
content, Worrall’s suggestion is that there is retention at the structural level,
specifically in the form of mathematical content. This view does not commit
Worrall to the existence of unobservable entities and mechanisms, the way
the approaches of Kitcher and Psillos do, but it also goes beyond the merely
empirical.
Despite the fact that we have several options for what sorts of elements might
be retained in order to support some version of realism or another, it turns out
that the miasma theory does not exhibit any of the above realist continuities.
First and foremost, miasmas were crucially involved in virtually every pre-
diction the miasma theory made—​thus, some kind of miasmatic continuity
is essential to both Kitcher and Psillos: miasmas are working, not presupposi-
tional posits on Kitcher’s terms, and essential, not idle, components on Psillos’s.
Because miasmas were thought to be the cause of diseases and also the mech-
anism for transmission, all of the predictions—​such as where the incidence of
disease ought to be particularly high or low, what populations would be par-
ticular targets for diseases, what regions ought to suffer to what degrees, and
correlations between distance from concentrated sources of miasma and mor-
tality rates—​essentially depended on miasmas themselves: without reliance on
the concept of miasma and its rotting sources, there simply would not have been
any predictions whatsoever.7
Now, while it’s clear that there is no straightforward ontological continuity
between the miasma theory and the germ theory, perhaps some sort of refer-
ential stability can be salvaged in other ways. But there are no other viable
candidates: the mechanisms of disease transmission were not retained—​bad

7 Vickers’ discussion of possible refinements of the divide et impera move is worth mentioning

here, in particular his suggestion that, instead of the usual focus on working posits “it remains pos-
sible that we might develop a recipe for identifying certain idle posits” (2013: 209). Vickers develops
an account of how this might have gone in the case of Kirchhoff ’s theory of diffraction. However, it
is unclear how to extend Vickers’ conclusions from the Kirchhoff case to the miasma theory, and so
I will not discuss his views in any detail here. Note, however, that Vickers mentions further potential
cases (see also Lyons (2006), especially his discussion of Kepler’s Mysterium Cosmographicum).
22 Historical Cases for the Debate

smells don’t transmit disease—​and neither were any of the laws.8 Specifically, not
only is there no analogue of Farr’s law in any of the modern disease theories, but
not even the phenomenon associated with it—​the connection between cholera
mortality and soil elevation—​was retained.
Incidentally, this last example is also the best bet as a candidate for Worrall’s
structure. Since Worrall emphasizes mathematical components, and there is
virtually no mathematical structure to be had in the miasma theory, the only
candidates for this view were of the kind put forward by Farr. One might object
that there were other kinds of statistics that were retained: claims, for example,
about the peaks, courses, and durations of epidemics. Moreover, these didn’t in-
volve or rely on miasma or its properties, and so one might think that these are
prime candidates for preservation. The problem with these, however, is that while
these are good candidates for preservation they are not good candidates for re-
alism, since they are all observational. True, they don’t depend on any theoretical
components of the miasma theory (or any other disease theory, for that matter),
but they don’t depend on any other theoretical account either. Rather, this is
merely empirical data—​a constraint with which any viable theory of disease has
to work. Realists and anti-​realists agree on these data sets, and there’s absolutely
no debate here about approximate truth or realism about unobservables—​and
realism about observables was never the issue.
So, to sum up: the miasma theory made several use-​novel predictions and so
counts as genuinely successful on the realist’s own terms. Further, it does not
exhibit any kind of realist continuity, neither theoretical, nor referential, nor
structural. There are no miasmas, the laws the theory gave rise to have been
abandoned, its mechanisms of transmission turned out not to exist, and the
properties that were ascribed to miasmas are not now ascribed to any of its etio-
logical successors.

2.5 Objections

One objection realists might raise against Farr is that his prediction was not
really use-​novel, since it was based on empirical data and hence insufficiently
grounded in theoretical assumptions. Thus, the realist might argue, as Vickers
has done with respect to Meckel, that Farr “really reached his conclusion
not via his (false) theoretical ideas, but rather via his empirical knowledge”
(Vickers 2015).

8 One also cannot make a case for Psillos’s kind-​ constitutive properties (see Psillos
1999: ­chapter 12): miasma is akin in this respect to phlogiston, not the luminiferous ether, and an
analogous version of the argument that Psillos makes against phlogiston will also work for miasma.
 Revisiting the Miasma Theory 23

First, note that it is true that once all the data is in, in some sense the original
theory that prompted the prediction can be discarded. However, just because this
is possible does not mean that there was no theoretical basis for the prediction in
the first place. In Farr’s case, there were strong theoretical underpinnings: both
his predictions and the empirical data he used were dependent on his disease
theory. This data was not just readily available for him; instead, he needed to col-
lect and generate it, which, given the extremely limited contemporary means at
his disposal, amounted to a gargantuan effort. Moreover, he did not just have to
gather the raw data, which was already extremely work-​intensive but, in order
to make well-​formed predictions, he also needed to organize and interpret this
data appropriately. Even here, his work was not straightforwardly empirical but
based on a number of theoretically generated assumptions, for example about
how to construct appropriate mortality rates, and so on. None of these efforts
made sense unless one subscribes to a disease theory according to which such
relationships are to be expected. Without his disease theory, there would never
have been any reason for Farr to gather any of the data he did and, indeed, on
other disease theories—​such as Snow’s, say—​there would not have been any
point in collecting it. Without Farr, quite possibly the relation between elevation
and cholera mortality would have never been discovered; certainly, without the
accompanying theory, it would have been meaningless.
One might still object that Farr’s prediction amounts to mere curve-​fitting and
his data to “more of the same.”9 However, there is nothing “mere” about curve-​
fitting in science. Curve-​fitting is widespread, standard practice, and it is a mis-
take to think that it involves nothing besides empirical data. On the contrary,
most scientific curve-​fitting has important theoretical elements: most obviously,
it has to be decided what the relevant parameters of the curve are, what family of
curves is appropriate, and so on. Further, picking a curve comes with predictions
about future data points, and here different curves will of course make different
predictions. As a result, finding that future data fits one’s curve confirms the idea
that one has hit on the (or at least a) right relation. Moreover, this data is “more
of the same” only if one has, in fact, hit upon a curve that works.10 Importantly
for this discussion, all sorts of scientific laws involve curve-​fitting, so if realists
object to Farr’s law on these grounds, they ought to also object to other laws.11

9 An anonymous referee voiced this objection.

10 One might make this argument for other laws, too: once one has hit on a working law, any data it
predicts will be “more of the same,” but this is so only if the law already works.
11 Interestingly, curve-​fitting “was first proposed by Adrian Marie Legendre (1752–​1833) and
Carl Friedrich Gauss (1777–​1855) in the early 19th century as a way of inferring planetary trajecto-
ries from noisy data” (Forster 1999: 197). For some examples of other scientific theories that involve
curve-​fitting and for more details on the theoretical work that goes into successful curve-​fitting, see
Forster (2003).
24 Historical Cases for the Debate

Regardless of whether there is curve-​fitting, however, not just do seemingly

“merely” empirical laws come with theoretical components, it is also hardly ever
possible (if ever at all) to give entirely theoretical deductions of laws. Most, if not
all, scientific laws are based at least partially on data, and so usually at least some
data is necessary in order to come up with the right form of a law in the first place.
The fact that empirical data is used in the formulation of a law should not count
against it—​far from it, in fact. Science is, after all, empirical, and an overwhelm-
ingly large number of scientific laws seek to describe empirical phenomena. It
is a non-​surprising consequence of this that, once we have a certain law, we are
free to kick away the theoretical apparatus that generated it. As a result, the fact
that the theoretical assumptions that go into a prediction can at some point be
discarded is, once again, not an objection to Farr in particular, but to any theory
involving laws that seek to describe empirical phenomena. Thus, realists should
say about Farr’s law exactly what they say about other laws. And I take it that
most realists do, in fact, want to keep other laws since, if laws are taken away as
a source of confirmation or sign of a theory’s success, this would unduly limit
the class of theories that are even potential candidates for “genuine” success and,
hence, realist warrant.12
Lastly, it is important to note that whether Farr’s prediction was based on em-
pirical data is, at any rate, irrelevant in assessing use-​novelty: the only criterion
that matters for use-​novelty is whether the prediction’s content was used in the
construction of the miasma theory, which it was not. Even though the miasma
theory only received sophisticated treatment in the 19th century, some version
or other of it, but in particular those parts of the theory from which the elevation
prediction logically followed, had been around for a long time. The data, more-
over, as we have seen, did not exist until Farr generated it; yet the theory had been
around since long before his lifetime, regardless of whether one wants to trace
the basics back to Hippocrates, Galen, Sydenham, or someone else.
Another avenue of realist criticism is to grant that Farr’s prediction was use-​
novel but to argue that we should not on that basis infer that his theory was
successful in a way that supports the success-​to-​truth inference. As we have al-
ready seen, realists place great importance on the kind of success involved in
this inference. The standard line is that use-​novel predictive success is what is
required for genuine success but it is, of course, open to realists to argue that
this notion needs to be refined further. On this view, what the Farr case shows
is that use-​novel predictive success is insufficient for genuine success and that

12 Note that this is compatible with the view that in certain domains the theoretical plays a heavier

role than in others, where the empirical is a more common starting point. In fact, this does not strike
me as implausible. The general point about the interplay between the theoretical and the empirical in
the generation of scientific laws is a question of degree, not kind, and so I do not see how this could be
an objection to Farr without also being an objection to other laws.
 Revisiting the Miasma Theory 25

further work needs to be done in order to capture what truly realism-​warranting

success looks like. In other words, Farr’s law is seen as a counterexample to the
view that use-​novel predictions are signs of genuine success. But, what might this
more refined notion of genuine success look like? One suggestion comes from
Vickers’ discussion of Meckel’s prediction of gill slits in human embryos (Vickers
2015). In 1811, the German anatomist J. F. Meckel predicted that the human em-
bryo ought to have gill slits, a prediction that was found to be correct by fellow
anatomist M. H. Rathke in 1827. Since this prediction was temporally (and use-​)
novel, it counts as an instance of novel predictive success for Meckel’s theory. For
the realist to preserve the success-​to-​truth inference, it now needs to be the case
that those parts of Meckel’s theory that were responsible for this success were
approximately true and were retained in successor theories. However, Vickers
argues, it looks like this is not the case here: those of Meckel’s assumptions re-
sponsible for the gill slit prediction cannot be regarded as approximately true
and so it seems realists have a problem. The case of Farr, they might suggest, falls
into the same category. What Vickers points out is that while Meckel might have
made a novel prediction, “the prediction isn’t ‘risky’ in the way the prediction of
the Poisson white spot was for Fresnel’s theory of light” and that this lack of riski-
ness “reduces the significance of the prediction” (2015). Further, Vickers invokes
Bayes’ theorem to argue that in cases of “remarkable novel predictive success,”
the prior probability of the evidence is small and, thus, such a prediction—​if
found to be true—​ought to raise our degree of belief in the theory much more
than a true prediction whose prior probability was high (2015). In Meckel’s case,
Vickers suggests, the prediction was not surprising and had a high prior proba-
bility, since “with a little bit of imagination” Meckel could have come to the gill
slit conclusion given the other empirical knowledge of the day. The same re-
sponse might be given to Farr.13
Let’s start with this last point: Could Farr have come to his conclusions about
cholera mortality and elevation without the miasma theory, given the existing
empirical knowledge and with a little bit of imagination? Presumably we don’t
want to speculate about Farr’s counterfactual psychology, so I take it what is at
issue is whether it is plausible to think that someone would or could have come to
the same conclusions as Farr, absent the miasma theory, and given the empirical
knowledge of the day. It is not. The given empirical knowledge concerned data
about the seasonality of cholera, the fact that it was worse in certain localities
than others, such as swamps and areas rife with sewage, and so on. Making an
inference from this sort of data to a relationship between mortality and elevation
only makes sense on the assumption that there is a connection between disease

13 In personal communication, Vickers has stated that this is, in fact, his preferred response to the

Farr case.
26 Historical Cases for the Debate

and decomposition and that disease can be transmitted through the air, both
central assumptions of the miasma theory.14 Without these assumptions, the
prediction about elevation and cholera mortality just disappears. Further, since
without these core theoretical assumptions, the prediction did not even make
sense, it is hard to see how Farr could somehow have made the elevation predic-
tion merely from other empirical knowledge. Both assumptions do theoretical
work for Farr and on taking out either one—​the connection to decomposition or
beliefs about aerial transmission—​there is no reason to expect the elevation rela-
tion. As a result, it is extremely hard to see how someone could have predicted it
without either of these components.15
It is perhaps noteworthy here that John Snow, famous for his view that cholera
was water-​, not air-​borne and on whose view there wasn’t a central connection
between cholera and putrefaction, did not predict the elevation relation. In fact,
not just did Snow not predict it, he did not even accommodate it, which was one
of the biggest criticisms voiced by his contemporaries (see, for example, Parkes
(1855)).16 Thus, not everybody came to Farr’s conclusion; notably, those who
had a different disease theory did not. Further, the other well-​known empirical
knowledge had been around for a long time, yet it took Farr to predict a form
of the elevation hypothesis. Presumably, if the elevation relation had somehow
followed straightforwardly from this existing empirical knowledge, other people
would have predicted it long before Farr.17 So, to sum up: it is implausible to
think that Farr could have come to his conclusions about elevation and cholera
without the central theoretical assumptions of the miasma theory.
What about the other part of the objection, that Farr’s prediction was not as
risky as that of Poisson and, therefore, less significant? The first thing to note is

14 There are, of course, as with any other theory, some auxiliary assumptions involved, such as

assumptions about air dilution (that the air is cleaner the further away it is from miasmatic sources
and so on). An anonymous referee has suggested that different auxiliaries are needed for different
diseases and that the “miasma theory of cholera” is different from, say, the “miasma theory of ty-
phoid fever.” This, however, is to misunderstand the context of mid-​19th century disease discussions.
Diseases, at the time, were not thought of as entities that would attack the body from the outside and
make it sick; instead, the prevalent view was a more physiological notion of disease, according to
which, in some sense, diseases originated in the victim. The ontological conception became influen-
tial later in the century, with the idea that different diseases were different species and with the advent
of early germ theories. For more detail, see Hamlin (1998: ­chapter 2).
15 I have previously argued that both assumptions ought to count as essential for realists. For more

detail, see Tulodziecki (2017).

16 We can now partially explain the elevation law in terms of the water-​hypothesis. Snow, however,

did not do so.

17 On the flipside, we might also have the following worry: assume we could somehow make an

argument for the claim that Farr could have come to the elevation conclusion just from existing em-
pirical knowledge, without miasmatic assumptions. Given the assumption that he somehow could
have done so, even if it is hard to see how, we might now worry that we are able to make this same ar-
gument in many other instances and, in particular, in instances that realists want to keep as examples
of genuine successes. Given enough ingenuity, perhaps we can always construct such a hypothetical
argument (see also Greg Frost-​Arnold’s comment in response to Vickers (Frost-​Arnold 2015)).
 Revisiting the Miasma Theory 27

that even granting that Farr’s prediction was not as risky or significant as Poisson’s
does not mean that Farr’s prediction was not good. I have no qualms with the
claim that Poisson’s prediction was superior to Farr’s; however, it is not clear how
typical or frequent predictions like Poisson’s actually are and what percentage
of scientific predictions are Poisson-​like.18 Indeed, it is entirely possible, if not
plausible, that such predictions, while exemplars of excellent predictions, are
somewhat atypical. Just as most scientists—​even very good ones—​are not Curies
or Einsteins, most predictions—​even very good ones—​are not like Poisson’s.
Thus, Farr’s prediction might have been very good, even if it was less good than
Poisson’s. And, thus, the mere fact that his prediction was less significant than
Poisson’s is not a reason to think that Farr’s prediction should not be viewed as an
instance of genuine success. But what about the claim that Farr’s prediction was
less risky? Even without a specific notion of riskiness it is clear that it would have
been bad for Farr if it had turned out that there was no relationship between ele-
vation and cholera mortality since this clearly followed from the miasma theory.
If no relation had been found, there would have been a need for another mias-
matic explanation (such as more sewage at high altitudes, for example) and, ab-
sent any such explanation, the absence of a relationship would have constituted a
strong argument against the theory.19,20
Still, a realist might argue, what matters is that what followed deductively from
the miasma theory was only the general fact that there ought to be some relation
between elevation and cholera mortality, but not the exact form of this relation
or even the fact that it was possible to capture it through a mathematical law.
This is true; however, once again, the fact that what followed logically from the
theory was only a general, not a specific prediction should not be held against
it, since this is all that could follow from such a theory, even under the very best
circumstances. In physical theories, laws and predictions, even if heavily based
on data, eventually usually have to be integrated into the theory mathematically,

18 It actually strikes me as an interesting project to look into the different types of prediction one

might find in different domains, what categories they might fall into, how frequent they are, and so
on. I would not find it surprising if different kinds of predictions were prevalent in different fields.
19 The same was true for other localities where high mortality was expected but not found, and for

localities where low mortality was expected, but it turned out to be high. Indeed, many of the con-
temporary discussions of the miasma theory and its competitors centered around just such issues.
20 What about the thought that Farr’s prediction did not have a low prior probability, but Poisson’s

did? I honestly do not know how to assign the priors in this case. It makes sense to assign Poisson’s
prediction a low prior probability but, in this case, the prediction was against the prevalent theory.
In Farr’s case, however, the prediction followed from a theory that was already the dominant di-
sease theory. It thus seems reasonable to think that its predictions, especially plausible ones like the
elevation relation, would not have been surprising to anyone. But what would the prior probability
have been if the miasma theory had been new? Certainly, given the humoral theory, according to
which diseases were an imbalance of individual humors, the probability of Farr’s prediction would
have been extremely low. What would the prior have been on Snow’s view, or some of the other
alternatives? Since we do not want to make the priors contingent on historical circumstances, it is
unclear to me how to determine what Vickers considers the crucial term, p(E/​–T ​ ).
28 Historical Cases for the Debate

in a way that gives rise to various theoretical deductive relationships. It is true

that this is not the case here, but this is not surprising, given that the miasma
theory, and indeed theories of disease etiology in general, are not mathematised
in the same way in which physical theories are. The point here is that Farr is
making the best kind of prediction that can be expected from this kind of theory.
Predictions about causal factors involved in a given disease are not of the math-
ematical kind, and the best predictions possible involve likely disease incidence,
mortality, and so on. However, these predictions are, by their nature, statistical
at best (if that) and, as a result, their exact numerical form will never be logically
entailed by a particular disease etiology. The most that can ever follow deduc-
tively from such a theory are general population-​level claims.
It is also notable in this context that our current theories do not, in kind, do
much better on this front than Farr did with respect to elevation, despite the fact
that we have vastly improved resources. Indeed, it is precisely because theories
about disease etiology do not usually make numerically precise predictions in
the form of laws that Farr’s prediction seems so remarkable and was seen by
many as such a triumph. Lastly, it is noteworthy that Snow also did not make
better predictions than Farr; in fact, Snow’s predictions were worse, since they
never approached any sort of numerical precision. What followed deductively
from Snow’s claim that cholera was waterborne were predictions of the same
type as Farr’s: population-​level predictions about disease incidence and mor-
tality, such as “disease incidence ought to be higher among people who drink
from contaminated water sources” (since, of course, not everyone who drank
contaminated water became sick). Thus, no theory of this kind could make
predictions which were both mathematically precise and logically entailed by the
theory in question.
This, in turn, gives rise to the following problem for realists: If Farr made the
best kind of prediction that this type of theory (especially given the limited means
available) allowed for, and realists do not consider this good enough for genuine
success, they end up with a view according to which no theory of this kind can,
even in principle, make predictions that would render it genuinely successful.
As a result, we would never, in any such case, be licensed to make the success-​to-​
truth inference, and thus never, in any such case, be warranted in being realists
about such theories, at least not on the basis of their predictions, no matter how
good.21 Thus, if the elevation prediction is not good enough for genuine success,

21 This is not to say, of course, that Farr’s prediction, on its own, made the miasma theory suc-

cessful, or that a single prediction of this type could warrant realist commitment in the (approximate)
truth of a theory. Obviously, a theory’s success depends on many factors. The point is, rather, that the
best type of prediction a theory can make should play a large role in assessing that theory’s success
and that, if such predictions are discounted as a source of success for some class of theories, it is hard
to see how theories in that class could ever count as genuinely successful, even in principle.
 Revisiting the Miasma Theory 29

realists will have to pay a cost that goes much beyond the specific example of Farr
and that now also includes, as we have seen, Snow’s water-​hypothesis and con-
temporary disease etiologies. In fact, the cost might be even greater, since there
are plenty of theories whose predictions are not predominantly mathematical in
nature or concern population-​levels. Realists might be willing to bite this bullet,
but the result would be a much narrower, perhaps domain-​specific realism.

2.6 Conclusion

It follows from earlier sections that the case of the miasma theory supports the
PMI: it does so by supporting the view that there is no connection between suc-
cess and approximate truth.22 Regardless, realists need some kind of account of
what is going on in this case. Perhaps the most obvious option is to identify some
other kind of continuity at work in the miasma case and add it to the possibilities
presented earlier. In that event, perhaps the realist intuition that the (working
parts of the) predecessors to our current theories are approximately true can be
preserved.
However, this strategy is problematic: first, because it is not clear what pos-
sible candidates for retention are left after we have excluded theoretical entities,
mechanisms, laws, and structures. Second, even if a candidate could be found,
it’s unlikely that the resulting proposal could be generalized to the cases in the
already existing literature, and which already have other realist accounts. One
might not be worried by this and instead be tempted by a kind of pluralism about
retention: different types of preservation apply to different types of theories
under different circumstances. I am not unsympathetic to this proposal, but it
comes with one big worry for the realist, namely whether the resulting position
has any realism left in it. After all, if the proposal is that theories get better over
time, and increasing success brings with it increasing approximate truth, yet this
approximate truth consists of such different things in different cases, it’s not clear
what we should be realist about. We might still be committed to some vague idea
of progress, but we won’t be able to tell what that progress involves. In particular,
we cannot be realist about any particular element: we cannot be realist about the
parts that were involved in predictions, since, sometimes, those are not retained;
we cannot be realists about structures, since, sometimes, those are not retained;
we cannot be realists about theoretical entities or mechanisms, since, sometimes,
they are not retained. The same goes for a theory’s laws and, indeed, if realists
want to argue that the miasma theory is approximately true, they’ll have to keep

22 Note that of course it is not enough to establish the PMI. In order to achieve the latter, it also

needs to be shown that these kinds of cases are pervasive (see also Kitcher (1993: 138–​139)).
30 Historical Cases for the Debate

adding to this list. But this means that, from the perspective of our current the-
ories, we simply have no idea which of their parts we ought to be realists about,
beyond a vague commitment to the view that something about them is probably
right, although we have not the slightest inkling what.
The alternative, of course, is to bite the bullet in the miasma case and accept
that this theory, although genuinely successful, was simply not one that was
approximately true. However, this is also a somewhat uncomfortable posi-
tion for the realist, quite independently of the fact that this case now supports
the PMI. The source of this discomfort is that we now have a case in which
we went from a not even approximately true theory directly to a successor
(the germ theory) that we do take be true—​and not just approximately true,
but true simpliciter. This is uncomfortable, because the whole realist idea is
that there is a relation between increasing success over time and increasing
truth-​content, and if it now turns out that discontinuities between theories
might sometimes be completely radical, then this does shed some doubt on
the “converging” part of convergent realism. Of course just because approxi-
mate truth might not be necessary for truth simpliciter does not mean that it
is not sufficient, and it doesn’t follow from the fact that (some) true theories
had no approximately true predecessors that approximate truth is not a reli-
able guide to truth as such.23 The point is rather that, if the convergent realist’s
story is right—​if the right way to view the history and development of science
is as a succession of increasingly true theories that get better and better over
time—​then we ought to expect the predecessors of theories that we now know
to be true to be approximately true. Finding true theories that do not conform
to this expectation does not invalidate the story—​one case does not make a
pattern—​but it is a strike against the plausibility of the general story that con-
vergent realists tell.

Acknowledgments

For helpful comments, many thanks to Tim Lyons, Marshall Porterfield, and Jan
Sprenger. Special thanks are due to Peter Vickers for numerous exchanges and
conversations about this case, and for generous written comments on a different
but related paper.

23 For that, we would need cases of approximately true predecessors that led to false successors,

and it is not clear how we would find these, since we also need true successors in order to assess these
claims in the first place.
 Revisiting the Miasma Theory 31

References
Baldwin, P. (1999). Contagion and the State in Europe, 1830–​1930. Cambridge: Cambridge
University Press.
Boyd, R. (1981). “Scientific Realism and Naturalistic Epistemology,” in P.D. Asquith and
T. Nickles (eds) PSA 1980, Vol. 2, East Lansing, MI: Philosophy of Science Association,
pp. 613–​662.
Boyd, R. (1984). “The Current Status of the Realism Debate,” in J. Leplin (ed.) Scientific
Realism, Berkeley: University of California Press, pp. 41–​82.
Boyd, R. (1990). “Realism, Approximate Truth and Philosophical Method,” in C.W.
Savage (ed.) Scientific Theories, Minnesota Studies in the Philosophy of Science, Vol. 14,
Minneapolis: University of Minnesota Press, pp. 355–​391.
Doppelt, G. (2007). “Reconstructing Scientific Realism to Rebut the Pessimistic Meta-​
Induction.” Philosophy of Science 74 (1): 96–​118.
Eyler, J. M. (1979). Victorian Social Medicine: The Ideas and Methods of William Farr.
Baltimore/​London: Johns Hopkins University Press.
Eyler, J. M. (2001). “The Changing Assessments of John Snow’s and William Farr’s Cholera
Studies.” Sozial und Präventivmedizin 46: 225–​232.
Farr, W. (1852). Report on the Mortality of Cholera in England, 1848–​49. London: W.
Clowes.
French, S., and Ladyman, J. (2011). “In Defence of Ontic Structural Realism,” in A.
Bokulich and P. Bokulich (eds.) Scientific Structuralism (Boston Studies in the Philosophy
of Science: Volume 281), Dordrecht: Springer, pp. 25–​42.
Forster, M. R. (1999). “Curve-​fitting problem.” In R. Audi (ed.) The Cambridge
Dictionary of Philosophy, Second Edition, Cambridge: Cambridge University Press,
pp. 197–​198.
Forster, M.R. (2003). “Philosophy of the Quantitative Sciences: Unification, Curve Fitting,
and Cross Validation.” Manuscript. Available at http://​philosophy.wisc.edu/​forster/​pa-
pers/​Part1.pdf
Frost-​Arnold, G. (2015). “Comment on Vickers (2015).” https://​thebjps.typepad.com/​
my-​blog/​2015/​06/​srpetervickers.html, last accessed 24 June 2018.
Hamlin, C. (1998). Public Health and Social Justice in the Age of Chadwick: Britain, 1800–​
1854.Cambridge: Cambridge University Press.
Hamlin, C. (2009). Cholera: The Biography. New York: Oxford University Press.
Hesse, M. (1976). “Truth and the Growth of Scientific Knowledge.” Proceedings of the
Biennial Meeting of the Philosophy of Science Association 2: 261–​280.
Kitcher, P. (1993). The Advancement of Science. New York: Oxford University Press
Ladyman, J. (1998). “What is Structural Realism?” Studies in History and Philosophy of
Science Part A 29 (3): 409–​424.
Lange, M. (2002). “Baseball, Pessimistic Inductions and the Turnover Fallacy.” Analysis
62: 281–​285.
Langmuir, A.D. (1961). “Epidemiology of airborne infection.” Bacteriological Reviews 25
(3): 173–​181.
Langmuir, Alexander D. (1961). “Epidemiology of Airborne Infection.” Bacteriological
Reviews, 25(3): 174
Laudan, L. (1981). “A Confutation of Convergent Realism.” Philosophy of Science 48
(1):19–​49
32 Historical Cases for the Debate

Laudan, L. (1984). Science and Values. Berkeley and Los Angeles: University of
California Press
Lewis, P. (2001). “Why the Pessimistic Induction Is a Fallacy.” Synthese 129: 371–​380.
Lyons, T.D. (2006). “Scientific Realism and the Stratagema de Divide et Impera.” British
Journal for the Philosophy of Science 57 (3): 537–​560.
Parkes E.A. (1855). “Mode of Communication of Cholera. By John Snow, M.D. Second
Edition.” British and Foreign Medico-​Chirurgical Review 15: 456.
Pelling, M. (1978). Cholera, Fever and English Medicine: 1825–​1865. Oxford: Oxford
University Press.
Psillos, S. (1996). “Scientific Realism and the ‘Pessimistic Induction.’” Philosophy of
Science 63 (3): 314.
Psillos, S. (1999). Scientific Realism: How Science Tracks Truth. London: Routledge
Saatsi, J. (2005). “On the Pessimistic Induction and Two Fallacies.” Philosophy of Science
72 (5): 1088–​1098.
Saatsi, J., Vickers, P. (2011). “Miraculous Success? Inconsistency and Untruth in
Kirchhoff ’s Diffraction Theory.” British Journal for the Philosophy of Science 62
(1): 29–​46.
Snow, J. (1855a). “On the comparative mortality of large towns and rural districts, and the
causes by which it is influenced.” Journal of Public Health, and Sanitary Review 1: 16–​24.
Snow, J. (1855b). On the Mode of Communication of Cholera (second much enlarged ed.).
London: J. Churchill.
Tulodziecki, D. (2016). “From Zymes to Germs: Discarding the Realist/​Anti-​Realist
Framework,” in T. Sauer and R. Scholl (eds.) The Philosophy of Historical Case-​Studies,
Boston Studies in Philosophy of Science, Switzerland: Springer, 265–​284.
Tulodziecki, D. (2017). “Against Selective Realism(s).” Philosophy of Science 84
(5): 996–​1007.
Vickers, P. (2013). “A Confrontation of Convergent Realism.” Philosophy of Science 80
(2): 189–​211.
Vickers, P. (2015). “Contemporary Scientific Realism and the 1811 Gill Slit Prediction,”
Auxiliary Hypotheses: Blog of the British Journal for the Philosophy of Science, http://​
thebjps.typepad.com/​my-​blog/​2015/​06/​srpetervickers.html, last accessed 24
June 2018.
Worboys, M. (2000). Spreading Germs: Diseases, Theories, and Medical Practice in Britain,
1865–​1900. Cambridge: Cambridge University Press.
Worrall, J. (1989). “Structural realism: The best of both worlds?” Dialectica 43
(1–​2): 99–​124.
Worrall, J. (1994). “How to Remain (Reasonably) Optimistic: Scientific Realism and
the” Luminiferous Ether.” PSA: Proceedings of the Biennial Meeting of the Philosophy of
Science Association 1994, 1: 334–​342.
 3
What Can the Discovery of Boron Tell Us
About the Scientific Realism Debate?
Jonathon Hricko

Education Center for Humanities and Social Sciences

National Yang-​Ming University
[email protected]

3.1 Introduction

In much of the recent work on the scientific realism debate,1 realists and their
anti-​realist opponents have focused on theories with two significant features.
First of all, those theories purport to describe some kind of unobservable reality,
for example, unobservable entities like electrons or genes. Second, those theories
exhibit novel predictive success, i.e., they generate true predictions that scientists
did not use when constructing those theories. The clearest cases of such success
involve temporal novelty, where a prediction of some phenomenon is temporally
novel if that phenomenon was not known at the time the theory was constructed.
Other cases involve only use-​novelty, where a prediction is use-​novel, but not
temporally novel, if it predicts a phenomenon that was known at the time the
theory was constructed, but knowledge of that phenomenon was not used in the
construction of the theory.2 In this chapter, I’ll focus on an example of a tempo-
rally novel predictive success.
One of the central issues in the realism debate concerns whether the novel
predictive success of a theory constitutes a sufficient reason for thinking that (at
least some of) that theory’s claims regarding unobservables are (at least approxi-
mately) true. Realists of various stripes argue that some form of realism or other
provides the best explanation of why theories exhibit novel predictive success
(Musgrave 1988; Psillos 1999). This argument is the so-​called no-​miracles ar-
gument, the name of which comes from the realist’s contention that, without a

1 For a comprehensive, up-​to-​date survey of the debate, see Alai (2017).

2 See Psillos (1999, 105–​107) for a good discussion of novelty.

Jonathon Hricko, What Can the Discovery of Boron Tell Us about the Scientific Realism Debate? In: Contemporary
Scientific Realism. Edited by: Timothy D. Lyons and Peter Vickers, Oxford University Press. © Oxford University Press
2021. DOI: 10.1093/​oso/​9780190946814.003.0003
34 Historical Cases for the Debate

realist explanation, novel predictive success would be miraculous. Anti-​realists

argue that the success of a theory does not warrant or require any sort of realist
attitude toward theoretical claims regarding unobservables (van Fraassen 1980;
Stanford 2006). Although realists and anti-​realists disagree about this issue, it’s
worth emphasizing that they actually have much else in common. Their assess-
ment of a theory’s claims regarding observables need not differ at all. And when
it comes to theories that have not exhibited novel predictive success, today’s
realists and anti-​realists may be equally skeptical of such a theory’s claims re-
garding unobservables.
Theories that, by present lights, are not even approximately true and yet
exhibited novel predictive success pose one of the strongest challenges to re-
alism.3 Recent work has uncovered many examples of such false-​but-​successful
theories (Lyons 2002, 2006, 2016, 2017; Vickers 2013). The challenge for the re-
alist is to clearly articulate what sort of realism is warranted in such cases. Many
realists have argued that the most viable option is some form of selective realism,
which involves a commitment, not to the (approximate) truth of entire theories,
but only to certain parts of theories. Selective realism comes in a number of dif-
ferent varieties.4 To take one example, deployment realists (Kitcher 1993; Psillos
1999) argue that we ought to commit only to the parts of theories deployed in
deriving successful novel predictions. Their strategy is to explain the novel pre-
dictive success of false theories in terms of the (approximately) true parts of
those theories.
Insofar as this strategy of providing such a realist explanation is successful,
selective realists can accommodate these examples of false-​but-​successful theo-
ries. However, even if there were just one example of novel predictive success that
realists failed to explain (i.e., one miracle), that example would pose a challenge
to realism insofar as it would undermine the claim that realism is required to ex-
plain such success.5 After all, if a non-​realist explanation is required in one case,
why stop there? Nothing compels us to endorse a realist explanation once we see
that a non-​realist explanation is going to be required regardless of what realists
have to say about all of the other examples. Realists might retreat to the claim
that miracles are permitted provided that they are rare. Additional examples of
novel predictive success that don’t yield to realist explanations cast doubt on this

3 Throughout the chapter, when I claim that past theories are/​are not approximately true, these

claims should be understood as abbreviations for the conditional claim that, if our current theories
are approximately true, then these past theories are/​are not approximately true. Otherwise, I might
be accused of presupposing realism about our current theories, and of thereby begging some central
questions in the realism debate.
4 See Lyons (2017, 3215) for a useful list of selective realist positions.
5 This point is especially clear within the context of Lyons’s pessimistic meta-​ modus tollens
(2002, 67).
 Scientific Realism Debate 35

further claim. The many examples of false-​but-​successful theories thus pose a

challenge to the viability of such a realist retreat.
My goal in this chapter is to present an example of a false-​but-​successful
theory that has not yet received sufficient attention within the realism debate and
use that example to pose a challenge to selective realism. The false-​but-​successful
theory is the oxygen theory of acidity due to Antoine-​Laurent Lavoisier (1743–​
94). The temporally novel prediction that it made was that boracic acid consists
of oxygen combined with a hypothetical, combustible substance that Lavoisier
called the boracic radical. And that prediction was subsequently confirmed. In
1808, the British chemist Sir Humphry Davy (1778–​1829) used potassium to
extract oxygen from boracic acid and thereby discovered that boron is the bo-
racic radical, while the French chemists Joseph Louis Gay-​Lussac (1778–​1850)
and Louis Jacques Thénard (1777–​1857), working together but independently of
Davy, did the same. This example of novel predictive success poses a strong chal-
lenge to selective realism because the parts of Lavoisier’s theory responsible for
its success are not even approximately true.
I proceed as follows. In section 3.2, I present Lavoisier’s oxygen theory of acidity
and show why it is, from our present perspective, not even approximately true. In
section 3.3, I identify four novel predictions that Lavoisier made regarding bo-
racic acid. In section 3.4, I demonstrate the success of these predictions by exam-
ining the work by Davy, Gay-​Lussac, and Thénard that constituted the discovery
of boron. In section 3.5, I show how the derivation of these predictions made use
of claims from Lavoisier’s theory that are not even approximately true. I go on to
consider some ways in which selective realists might try to accommodate this
example of novel predictive success, and I argue that these attempts are unsuc-
cessful. Sections 3.2 through 3.5 present the main argument of the chapter in a
relatively straightforward way, without anticipating and responding to some rel-
evant objections and without exploring aspects of the history of this episode that,
while relevant, are not essential to understanding the main argument. I flag these
issues in footnotes as they arise and direct the reader to the relevant subsections
of section 3.6, in which I address these issues more fully.

3.2 The False Theory

The false theory at the center of this example of novel predictive success is
Lavoisier’s oxygen theory of acidity. According to that theory, all acids contain
oxygen and a non-​oxygen component that Lavoisier called the “acidifiable base”
or “radical” of the acid, two terms that he regarded as synonyms (1790 [1965],
65). Lavoisier was led to this theory by the fact that the acids that were most well
understood at the time were all shown to contain oxygen. Lavoisier holds that
36 Historical Cases for the Debate

different kinds of acids differ from one another in one of two ways. They may
contain different radicals, as phosphoric acid and sulfuric acid do—​the radicals
in these acids are phosphorus and sulfur, respectively (66). And even if they con-
tain the same radical, they may still differ from one another by virtue of the fact
that they contain different amounts of oxygen, as sulfuric acid and sulfurous acid
do (66–​68). Moreover, Lavoisier emphasizes that the base or radical of an acid
can be either a simple substance or a compound (115–​116, 176–​177). For ex-
ample, he claims that many vegetable acids have compound radicals composed
of carbon and hydrogen.
Oxygen is the substance that plays the most central role in Lavoisier’s theory of
acidity. It is the element that all acids share, and which “constitutes their acidity”
(65). Oxygen, for Lavoisier, is what gives acids their acidic properties, for ex-
ample, their sour taste. Lavoisier labels oxygen “the acidifiing [sic] principle” (65)
and in fact the very name “oxygen” that Lavoisier proposed for this substance
comes from the Greek words for “acid-​generator” (51; Chang 2012b, 9). Though
to be sure, as Le Grand (1972, 11–​12) has emphasized, Lavoisier admitted that
many compounds that contain oxygen are not acids, and his point was that ox-
ygen is a necessary condition for acidity, not a sufficient one.
Lavoisier’s theory is also a theory of the formation of acids. After recounting
a series of experiments in which acids are formed by the combustion of the
acidifiable bases phosphorus, sulfur, and carbon, Lavoisier writes: “I might
multiply these experiments, and show by a numerous succession of facts, that
all acids are formed by the combustion of certain substances” (1790 [1965],
64). Lavoisier’s view, then, is that an acidifiable base must be a combustible sub-
stance. In order to understand Lavoisier’s view regarding the formation of acids
by means of combustion, we must start with his claim that oxygen gas is a com-
pound of caloric and the base of oxygen gas (“oxygen base” for short, following
Chang (2011, 415)); and it is oxygen base which is the true acidifying prin-
ciple (Lavoisier 1790 [1965], 51–​52). Lavoisier illustrates the combustion of an
acidifiable base and the subsequent formation of an acid in terms of the example
of phosphorus, the acidifiable base of phosphoric acid:

at a certain degree of temperature, oxygen possesses a stronger elective attrac-

tion, or affinity, for phosphorus than for caloric; . . . in consequence of this, the
phosphorus attracts the base of oxygen gas from the caloric. (57)

In other words, phosphorus effects the decomposition of oxygen gas into caloric
and oxygen base; oxygen base then combines with phosphorus to form an acid.
More generally, “[b]‌efore combustion can take place, it is necessary that the base
of oxygen gas should have greater affinity to the combustible body than it has to
caloric” (414–​415). Hence, Lavoisier’s theory of acidity makes use of his theories
 Scientific Realism Debate 37

of combustion and caloric as well. Lavoisier goes on to propose the term “oxy-
genation” to name the process by which combustible substances combine with
oxygen, in which case acidifiable bases are converted to acids by oxygenating
them (61–​62).
As a whole, Lavoisier’s theory of acidity isn’t even approximately true.6 By the
early years of the 19th century, chemists already knew of two counterexamples to
Lavoisier’s theory, namely, prussic acid (hydrocyanic acid, HCN) and muriatic
acid (hydrochloric acid, HCl), neither of which contains oxygen, Lavoisier’s acid-
ifying principle. Additionally, by positing the muriatic radical as the non-​oxygen
component of muriatic acid, Lavoisier posited a non-​existent entity.7 Moreover,
if one or both of our current conceptions of acidity are approximately true, then
Lavoisier’s theory cannot be. According to the Brønsted-​Lowry concept, acids
are proton donors. And according to the Lewis concept, acids are electron pair
acceptors. While these concepts diverge from one another in interesting ways,8
neither gives any sense to the claim that oxygen is the acidifying principle, which
is the central claim of Lavoisier’s theory.
It’s worth pressing this point regarding the falsity of Lavoisier’s theory a bit fur-
ther by considering the theories of combustion and caloric that Lavoisier made
use of in the context of his theory of acidity. One might think that Lavoisier’s
theory of combustion, at least, is approximately true. But it’s not obvious that it is.
Lavoisier’s theory requires the presence of oxygen gas for combustion. But by the
early 19th century, chemists knew of cases of combustion that occurred without
the presence of oxygen gas. Moreover, combustion, for Lavoisier, requires the de-
composition of oxygen gas into caloric and oxygen base (1790 [1965], 414–​415).
The caloric theory of heat has been a problem case for realists at least since it
appeared on Laudan’s well-​known list of false-​but-​successful theories (1981, 33).
Even if we confine our attention to the period in which Lavoisier was working,
this theory exhibited a number of successful explanations. As Chang makes clear,
these included explanations of:

the flow of heat toward equilibrium, the expansion of matter by heating, latent
heat in changes of state, the elasticity of gases and the fluidity of liquids, the heat
released and absorbed in chemical reactions, [and] combustion. (2003, 907)

Moreover, these explanations made use of a number of assumptions about the

nature of caloric which are not even approximately true. These included the

6 See Chang (2012b, 8–​10) for a good discussion of the problems with Lavoisier’s theories of

acidity, combustion, and caloric, which has informed the discussion in the remainder of this section.
7 See Hricko (2018) for a detailed discussion of the muriatic radical within the context of the re-

alism debate.
8 See Chang (2012a) for a good discussion of the ways in which these concepts diverge.
38 Historical Cases for the Debate

assumption “that heat was a ‘self-​repulsive’ (or ‘elastic,’ or ‘expansive’) substance,

while it was attracted to ordinary matter”; “the postulation that caloric existed
in two different states: sensible and latent”; and the identifications of caloric as a
chemical element, of sensible heat with free or uncombined caloric, and of latent
heat with caloric that was chemically combined with matter (907–​908). The ap-
pearance of caloric in Lavoisier’s theory of acidity thus provides some additional
support for the claim that this theory was not even approximately true.

3.3 The Novel Predictions

The novel predictions on which I’ll focus concern the composition of what
Lavoisier called boracic acid. Lavoisier writes:

The boracic radical is hitherto unknown; no experiments having as yet been

able to decompose the [boracic] acid; we conclude, from analogy with the other
acids, that oxygen exists in its composition as the acidifying principle. (1790
[1965], 245)

As we’ll see shortly, there is a bit more to the predictions than what Lavoisier
says here.
Lavoisier’s novel predictions regarding boracic acid belong to a larger group
of similar novel predictions that he made on the basis of his theory of acidity. At
the time, there were a number of acids that chemists could neither produce from
simple substances nor decompose into simple substances. These acids included
boracic acid, muriatic acid, and fluoric acid (hydrofluoric acid, HF). Lavoisier
hypothesized that these acids are composed of oxygen and the boracic, muriatic,
and fluoric radicals, respectively. Regarding these radicals, Lavoisier writes:

the combinations of these substances [the radicals], either with each other, or
with the other combustible bodies, are hitherto entirely unknown . . . We only
know that these radicals are susceptible of oxygenation, and of forming the mu-
riatic, fluoric, and boracic acids. (209–​210)

None of these radicals had been isolated, and in this passage, Lavoisier predicts
that they are combustible substances that form acids by means of oxygenation.
This prediction is a straightforward consequence of his theory of acidity, ac-
cording to which acids contain oxygen and a combustible radical. It’s also worth
emphasizing that Lavoisier’s theory predicts that the only components of these
three acids are oxygen and their respective radicals. The radical, for Lavoisier, is
the acid’s non-​oxygen component, and cases in which an acid has more than one
 Scientific Realism Debate 39

non-​oxygen component are cases in which the radical is a compound as opposed

to a simple substance.
In order to gain a better understanding of these predictions, it’s worth going
into a bit more detail regarding how Lavoisier understood the hypothetical
substances that he referred to as the boracic, muriatic, and fluoric radicals. The
terms boracic radical, etc., are best understood as descriptions of the role that
these substances were hypothesized to play within their respective acids, i.e., the
role of the combustible component substance that forms an acid by oxygena-
tion. These terms should not be understood as names of chemical substances,
because they were temporary placeholders for the names of whatever might be
discovered to play this role. Lavoisier included all three radicals in his table of
simple substances (175).9 By doing so, he did not mean to imply that they would
be discovered to be previously unknown substances, elementary or otherwise.
Regarding the muriatic radical, Lavoisier recognized that it might be “discovered
to be a known substance, though now unknown in that capacity” (72), and this
point applies to the other two radicals as well. If, say, the boracic radical turned
out to be a known substance or a compound of known substances, then Lavoisier
would have simply eliminated the term boracic radical from his table of simple
substances. If, however, the boracic radical turned out to be a previously undis-
covered simple substance, then Lavoisier would have replaced the term boracic
radical with a name for this new substance. The important point here is that
Lavoisier’s theory of acidity did not entail that the three radicals are three hith-
erto undiscovered chemical substances.
At this point, we can distinguish four predictions that Lavoisier made re-
garding the composition and formation of boracic acid:

• Boracic acid contains oxygen.

• Boracic acid contains a combustible substance (the boracic radical).
• The only components of boracic acid are oxygen and the boracic radical
(though the boracic radical itself might be a compound radical).
• Boracic acid is formed by the combustion of the boracic radical with oxygen.

Importantly, these predictions are novel predictions. In fact, the novelty of these
predictions is temporal novelty, which arises from the fact that chemists at the
time had not yet succeeded in decomposing boracic acid into simpler substances
or producing it by means of simpler substances. Hence, the composition of

9 He thereby indicated his view that they are elements. In section 3.6.2, I discuss the prediction

that the boracic radical is an element and conclude that it is not relevant to the issue concerning
whether selective realists can accommodate false-​but-​successful theories.
40 Historical Cases for the Debate

boracic acid and the way in which it is formed were unknown and so couldn’t
have been written into the oxygen theory of acidity at the time Lavoisier
constructed it.

3.4 The Predictive Success

In order to determine whether Lavoisier’s predictions were successful, we must

first identify what substance chemists at the time referred to when they used
the term boracic acid. In fact, they primarily had in mind the substance that we
know as boron trioxide (B2O3).10 Boron trioxide is composed of oxygen and a
combustible substance, namely, boron; and it can be formed by the combustion
of boron with oxygen. Moreover, it turns out that, unlike the muriatic radical, the
boracic radical exists—​it is what we now call boron.11 Hence, Lavoisier’s oxygen
theory of acidity, although not even approximately true, exhibited novel predic-
tive success. I’ll now examine the work in the early 19th century that confirmed
Lavoisier’s predictions regarding boracic acid.
Three chemists are credited with decomposing boracic acid and isolating
boron: Davy in England, and Gay-​Lussac and Thénard in France (Lowry 1915,
288; Fontani, Costa, and Orna 2015, 30). Because they suspected boracic acid
to contain oxygen, they attempted to decompose it with potassium, a substance
known to have a strong affinity for oxygen. Gay-​Lussac and Thénard isolated
boron by heating boracic acid and potassium in a copper tube and washing
the mixture that they obtained with water to separate boron from the other
substances (1808, 169–​171). They report that they thereby produced a greenish-​
brown substance (171). Davy first attempted the decomposition by acting on
moistened boracic acid with a battery, but without much success (1808, 43; 1809,
75–​76). Davy (1809, 76–​78) eventually succeeded using more or less the same
methods as Gay-​Lussac and Thénard. He reports that he produced a dark, olive-​
colored powder (1809, 78). Today, we would say that these chemists produced
relatively impure samples of amorphous boron.12
By producing these impure samples of boron, these chemists thought that
they had isolated the base or radical of boracic acid. Gay-​Lussac and Thénard
label the substance that they had isolated the “radical boracique,” for which
they propose a new name: “bore” (1808, 171, 173). Davy, however, initially

10 I discuss the details regarding the identification of boracic acid with boron trioxide in

section 3.6.1.
11 In section 3.6.4, I consider and respond to an objection to the identification of boron with the

boracic radical.
12 Given the impurity of the samples, one might question whether these chemists really confirmed

Lavoisier’s predictions. I consider this issue in section 3.6.3.

 Scientific Realism Debate 41

suspected that the substance he had isolated was a compound of oxygen and
“the true basis of the boracic acid,” which he conjectured was a metal and for
which he proposed the name “boracium” (1809, 84–​85). Several years later,
he concluded that this conjecture had not been vindicated, and in light of
the analogy between the substance that he had isolated and carbon, he pro-
posed another name for what he calls “the basis of the boracic acid”: “boron”
(1812, 178). The terminology that these chemists used does not necessarily
imply a commitment to Lavoisier’s theory of acidity. In fact, at the time they
discovered boron, Davy was skeptical of Lavoisier’s theory of acidity while
Gay-​Lussac and Thénard were committed to it.13 And by 1812, although
Davy was still using the term “basis” to refer to the non-​oxygen component
of boracic acid, he had more decisive reasons for rejecting Lavoisier’s theory
since, in 1810, he had argued that the components of muriatic acid are hy-
drogen and chlorine (Davy 1811). However, Davy did previously claim that
“the combustible matter obtained from boracic acid, bears the same rela-
tion to that substance, as sulphur and phosphorus do to the sulphuric and
phosphoric acids” (1809, 82), which basically amounts to an identification
of boron with the boracic radical. Importantly, regardless of their attitude
toward Lavoisier’s theory of acidity, these three chemists isolated the sub-
stance that Lavoisier had hypothesized years earlier; in other words, boron
is the boracic radical.
Once they had isolated boron, Davy, Gay-​Lussac, and Thénard also deter-
mined some of its properties, most notably, that it is a combustible substance
that combines with oxygen to form boracic acid. Both Davy (1809, 79, 82; 1812,
179) and Gay-​Lussac and Thénard (1808, 172–​173) describe the experiments
by which they converted the newly discovered substance into boracic acid by
means of combustion with oxygen. And by doing so, they also showed that the
components of boracic acid are boron and oxygen.
At this point, we can conclude that Lavoisier’s four novel predictions, which
I listed in section 3.3, were successful. The work of Davy, Gay-​Lussac, and
Thénard showed that boracic acid, which we know as boron trioxide, contains
both oxygen and a combustible substance, namely, boron. Moreover, these
chemists confirmed Lavoisier’s prediction regarding the formation of boracic
acid, namely, that it forms by the combustion of the boracic radical (boron)
with oxygen. And by both producing and decomposing boracic acid, they
provided a convincing demonstration that its sole components are boron and
oxygen.

13 See Brooke (1980, 124, 158) for Davy’s skeptical attitude toward Lavoisier’s theory of acidity and

Crosland (1978, 45) for Gay-​Lussac’s acceptance of that theory.

42 Historical Cases for the Debate

3.5 The Challenge for Selective Realism

At this point, the upshot is that we can add Lavoisier’s oxygen theory of acidity to
the increasingly long list of theories that, although not even approximately true,
exhibited novel predictive success. As I discussed in section 3.1, such theories
pose a challenge to realism. My goal in this section is to consider two ways in
which a selective realist might respond to this challenge and argue that neither of
these ways is successful.
Before doing so, it’s worth going into a bit more detail regarding the chal-
lenge that the selective realist faces. The derivation of Lavoisier’s four successful
predictions involved a number of claims from his oxygen theory of acidity that,
by present lights, are not even approximately true. These claims include:

(1) Oxygen gas is a compound of caloric and oxygen base.

(2) All acids contain oxygen base, which is the acidifying principle, i.e., the
substance in virtue of which acids have their acidic properties.
(3) All acids contain an acidifiable base or radical, which is the non-​oxygen
component of the acid, and which may be a simple substance or a
compound.
(4) Acids form by the combustion of an acidifiable base, which requires that
oxygen base has a stronger affinity for the acidifiable base than it does
for caloric; oxygen base then combines with the acidifiable base to form
an acid.

Lavoisier’s prediction that boracic acid contains oxygen made use of (1), (2),
and (4). His prediction that it contains a combustible substance made use of (1),
(3), and (4). His prediction that it contains only these two substances made use
of all of these claims, as did his prediction regarding the formation of boracic
acid in terms of the combustion of an acidifiable base with oxygen. If our cur-
rent theories are correct, then none of these four claims is even approximately
true. (1) is false in light of the rejection of the caloric theory of heat. Moreover,
we now consider oxygen gas to be an elementary substance, not a compound.
(2) is false, not just because there are acids that do not contain oxygen, but also
because we’ve rejected the idea that a chemical substance can be an acidifying
principle in Lavoisier’s sense. (3) is false because Lavoisier characterized the rad-
ical of an acid as its non-​oxygen component, and so acids that lack oxygen also
lack radicals in Lavoisier’s sense. And (4) is false, not just because it involves the
caloric theory of heat, but also because there are acids that do not form by com-
bustion with oxygen because they do not contain oxygen.
The first selective realist response I’ll consider is to show, contrary to first
appearances, that the theoretical claims required for deriving Lavoisier’s
 Scientific Realism Debate 43

successful predictions are in fact approximately true. One way the selective realist
might attempt to do so makes use of Vickers’s (2013, 199) distinction between
derivation external posits (DEPs) and derivation internal posits (DIPs), along
with his notion of the working part of a DIP.14 Realist commitment does not ex-
tend to DEPs, which merely inspire scientists to consider ideas; and it may not
even extend to all parts of a DIP. Realist commitment extends only to the working
part of a DIP, i.e., the part that “actually contributes to a given derivational step,”
and not to the other parts, which Vickers labels “surplus content” (201).
In order to evaluate this response, I’ll focus on one of Lavoisier’s predictions,
namely, the prediction that boracic acid contains oxygen. While this prediction
was in fact motivated by (1), (2), and (4), for the sake of argument, let’s grant
that (1) and (4) are DEPs, and only (2) is a DIP. It’s plausible that (2) has some
surplus content and that the working part of (2) is a more minimal claim that
suffices for deriving the prediction without mentioning oxygen base or an acidi-
fying principle. What might this more minimal claim be? The claim that all acids
contain oxygen is more minimal, and it suffices for deriving the prediction; but
it is not even approximately true since not all acids contain oxygen. The claim
that boracic acid contains oxygen is even more minimal, and also true; but it
doesn’t provide an explanation of this instance of novel predictive success since it
is identical to the prediction itself, which makes the derivation trivial. The claim
that many acids contain oxygen is more minimal, and also true; but it doesn’t
entail the prediction, and on its own, it doesn’t even make the prediction likely
since it’s consistent with the claim that many acids do not contain oxygen. The
challenge for the selective realist is to identify a working part that has two prop-
erties: it must be at least approximately true, and it must be sufficient for deriving
the prediction in a nontrivial way. The problem for the selective realist is that the
claims that have one property lack the other property.
Another consideration that makes this kind of selective realist response prob-
lematic concerns Lavoisier’s predictions regarding the other two undecomposed
acids: muriatic (hydrochloric) acid and fluoric (hydrofluoric) acid. The selective
realist must explain why Lavoisier’s prediction regarding the presence of oxygen
in boracic acid succeeded while his predictions regarding the presence of ox-
ygen in the muriatic and fluoric acids failed. And this explanation must be con-
sistent with the fact that Lavoisier’s predictions regarding the presence of oxygen
in these three acids were derived in exactly the same way, as I discussed in section
3.3. These predictions were motivated solely by Lavoisier’s theory of acidity and
by the fact that these three substances are acids. Explaining why one prediction
succeeded while the other two failed requires adopting a view that is historically

14 Vickers uses the term posit “as essentially synonymous with ‘proposition’ or ‘hypothesis’

(broadly construed)” (2013, 198, fn 7).

44 Historical Cases for the Debate

inaccurate. This is the view that Lavoisier’s derivations of these three predictions
differ in some significant way, say, because he identified some property of boracic
acid that muriatic acid and fluoric acid lack and that truly indicates the presence
of oxygen in a compound. In short, for any candidate for the working part of a
DIP, the selective realist will have to explain, in a historically plausible way, why it
works when it comes to boracic acid and doesn’t work when it comes to the other
two acids. It’s worth emphasizing that the issue here is not simply about failed
predictions. It’s about whether, in light of those failed predictions (which were
derived in exactly the same way as the successful predictions), the working parts
in the derivation of the successful predictions can be regarded as approximately
true. My claim is that they cannot.
The second selective realist response is that the novel predictive success of
Lavoisier’s theory is not impressive enough to pose a challenge to realism. Vickers
(2013, 195–​198) argues that successful novel predictions need to be sufficiently
impressive in order to pose a challenge to realism. For Vickers, the distinction
between impressive and unimpressive novel predictive successes is a matter of
degree. One way in which they differ in degree concerns “the degree to which the
prediction could be true just by luck (perhaps corresponding to the Bayesian’s
‘prior probability’)” (196). If the probability that Lavoisier’s predictions could be
true just by luck is sufficiently high, then we have a group of lucky guesses that
the realist can dismiss, as opposed to “miracles” that require a realist explanation.
Given the state of knowledge regarding acids in the years leading up to Davy’s
and Gay-​Lussac and Thénard’s work on boracic acid, what can we say about the
prior probability of Lavoisier’s predictions being true? Many acids (e.g., sulfuric
acid and phosphoric acid) were known to contain oxygen and a combustible sub-
stance and to form via the combustion of that substance with oxygen. However,
there were problem cases. In 1787, Claude Louis Berthollet (1748–​1822) dem-
onstrated that prussic acid (hydrocyanic acid, HCN) contains hydrogen, carbon,
and nitrogen, but not oxygen (1789, 38). Thomas Thomson (1773–​ 1852)
mentions another problem case: “Sulphurated hydrogen [i.e., hydrogen sul-
fide, H2S], for instance, possesses all the characters of an acid, yet it contains no
oxygen” (1802, 4). In light of these problem cases, assigning a very high prior
probability to Lavoisier’s predictions regarding boracic acid would have been
unwarranted. In that case, Lavoisier’s predictions were, at least to some degree,
impressive.
Vickers (2013, 196) discusses two other senses in which predictions can be
unimpressive; but neither of them applies to Lavoisier’s predictions regarding
boracic acid. First of all, Vickers argues that vague predictions are unimpres-
sive. His example is the prediction that the planet Venus is hot, which is suc-
cessful so long as Venus has a temperature greater than, say, 50°C. The problem
here seems to relate, not just to the vagueness of the term hot, but also to its
 Scientific Realism Debate 45

imprecision, since there are many ways in which Venus could be hot. Lavoisier’s
predictions regarding boracic acid are, however, both less vague and more pre-
cise than this prediction. Secondly, Vickers argues that predictions may be un-
impressive if the match between the prediction and the experimental results is
not sufficiently close. But Davy’s and Gay-​Lussac and Thénard’s results match
Lavoisier’s predictions quite well. It’s also worth mentioning that, for Vickers,
qualitative predictions (e.g., the Poisson white spot) can be very impressive, and
so the qualitative nature of Lavoisier’s predictions is no reason to dismiss them
as unimpressive.
I conclude that the novel predictive success of Lavoisier’s oxygen theory of
acidity poses a strong challenge to selective realism. The working parts of
Lavoisier’s theory are not even approximately true. And the predictive success
of that theory is sufficiently impressive that the selective realist cannot dismiss
Lavoisier’s predictions as likely to be true just by luck.

3.6 Complications, Clarifications, Objections, and Replies

At this point, I’ve presented the main argument of the chapter. In the course
of doing so, I’ve also flagged and so far ignored a number of relevant issues, to
which I now turn.

3.6.1 What Did Boracic Acid Refer To?

In section 3.4, I claimed that when chemists working in the late 18th and early
19th centuries used the term boracic acid, they primarily had in mind the sub-
stance that we call boron trioxide (B2O3). Boron trioxide is the anhydride of boric
acid (H3BO3), and one might wonder why boracic acid didn’t refer solely to the
substance that we call boric acid. If it did, then some of Lavoisier’s predictions
(e.g., the prediction that oxygen and a combustible substance are the only
components of boracic acid) are false. Moreover, in order to conclude that the
work of Davy, Gay-​Lussac, and Thénard demonstrated the success of Lavoisier’s
predictions, we need to be sure that these three chemists were experimenting
with the substance that was the subject of Lavoisier’s predictions. Since my ar-
gument depends on the claim that chemists used the term boracic acid to refer
primarily to boron trioxide, it’s important for me to defend this claim.
The use of the term boracic acid to refer primarily to boron trioxide was a
particular instance of a more general trend. A number of commentators have
observed that chemists at the time used the term acid to refer to two kinds of
substances that we distinguish today, namely, acids and their anhydrides (Laurent
46 Historical Cases for the Debate

1855, 44–​45; Miller 1871, 314; Lowry 1915, 250). Knight (1992, 83) explains that
when chemists at the time used the term acid, they had in mind primarily the
anhydride, though they recognized that the presence of water was required for
anhydrides to manifest their acidic properties. Crosland (1973, 307) points out
that this water “was considered as no more an essential part of the acid than, say,
water of crystallization is a part of certain salts.” To take an example, Lavoisier
(1790 [1965], 69) uses the term carbonic acid to refer to what we could call
carbon dioxide (CO2), the anhydride of carbonic acid. And when he claims that
the components of carbonic acid are oxygen and carbon, he’s not neglecting hy-
drogen as a component, since he had in mind the anhydride, and since he recog-
nized that carbonic acid mixed with water does contain hydrogen.
Chemists at the time took the same attitude toward boric acid and its an-
hydride, boron trioxide. For instance, in his Dictionary of Chemistry, William
Nicholson (1753–​1815) writes:

In a moderate heat this concrete acid melts with less intumescence than borax
itself, and runs into a clear glass, which is not volatile unless water be present.
This glass does not differ from the original acid, except in having lost its water
of crystallization. (1795, 18)

The concrete acid is what we would now call boric acid, while the clear glass
is its anhydride, which we would call boron trioxide. Lavoisier (1790 [1965]
244) distinguishes the combinations of boracic acid with other substances “in
the humid way” and “via sicca,” i.e., in the dry way, which corresponds to the dis-
tinction we draw between boric acid and boron trioxide today. Years later, Davy
treats the acid and the anhydride as two forms of the same substance:

Boracic acid, in its common form, is in combination with water; it then appears
as a series of thin white hexagonal scales; . . . By a long continuous white heat
the water is driven off from it, and a part of the acid sublimes; the remaining
acid is a transparent fixed glass, which rapidly attracts moisture from the air.
(1812, 180)

The white hexagonal scales are what we would call boric acid, while the trans-
parent fixed glass is what we would call boron trioxide.
Importantly, Lavoisier’s predictions regarding boracic acid are primarily
predictions regarding what we would call boron trioxide. Lavoisier was prima-
rily concerned with the composition of boracic acid without water (i.e., boron
trioxide); and he predicted that it is composed of oxygen and a combustible sub-
stance, namely, the boracic radical. He was also perfectly aware that boracic acid
with water (i.e., boric acid) contains hydrogen since it contains water.
 Scientific Realism Debate 47

It is also the case that Davy, Gay-​Lussac, and Thénard decomposed boron tri-
oxide as opposed to boric acid. To be sure, there is some disagreement about
this point in the literature, with some commentators claiming that it was boric
acid (e.g., Fontani, Costa, and Orna 2015, 30) and others claiming that it was
boron trioxide (e.g., Crosland 1978, 78). There are two good reasons for con-
cluding that these chemists decomposed boron trioxide. First of all, Gay-​Lussac
and Thénard (1808, 170) describe the boracic acid they were working with as
“very pure and vitreous,” which looks like a description of boron trioxide as op-
posed to boric acid. Second, decomposing boron trioxide by means of alkaline
metals like potassium is a recognized method of producing impure samples of
boron. Chemists in the mid-​20th century did attempt to substitute boric acid for
boron trioxide in such reactions, but these attempts were eventually abandoned
because of the risk of producing an explosive mixture (Zhigach and Stasinevich
1977, 216–​217).
Since boracic acid referred primarily to boron trioxide, we can conclude that
Lavoisier’s predictions were successful and that the work of Davy, Gay-​Lussac,
and Thénard confirmed those predictions.

3.6.2 Lavoisier’s Prediction That the Boracic Radical Is

an Element

In section 3.3, I briefly noted another prediction regarding the boracic, muri-
atic, and fluoric radicals that one finds in Lavoisier’s work, namely, that they are
elements. Lavoisier included the three radicals in his table of simple substances,
thereby indicating his view that they should be considered elements (1790
[1965], 175).
For Lavoisier, the terms element and simple substance are synonyms, and he
understands elements as substances that chemists cannot decompose into sim-
pler component substances (xxiv). As Scerri (2005, 129) puts it, elements, for
Lavoisier, are “observable simple substances that can be isolated,” i.e., from the
other substances that they combine with to form compounds, which requires
chemical decomposition. Lavoisier (1790 [1965], xxiv, 177) emphasizes that
“simple” should be understood relative to our present state of knowledge, so that
a substance that we regard as simple today may be shown to be a compound to-
morrow if chemists succeed in decomposing it. Until chemists are able to decom-
pose a substance, Lavoisier states that we should regard it as an element.
However, there’s a puzzle regarding the inclusion of the three radicals on
the table of simple substances. As Hendry (2005, 41–​42) has noted, Lavoisier
applies his notion of element in an inconsistent way—​since the boracic, muri-
atic, and fluoric acids had not yet been decomposed, they should appear on the
48 Historical Cases for the Debate

table of simple substances. So why do their radicals appear instead? Lavoisier’s

theory of acidity and his notion of element give us the resources to solve this
puzzle. The undecomposed acids do not appear on the table because, according
to Lavoisier’s theory of acidity, acids in general are compounds, not simple
substances. Lavoisier doesn’t explicitly justify the inclusion of the radicals on his
table of simple substances; and since his theory allows for compound radicals,
it’s not immediately obvious why he includes them. The most plausible reason is
that, since the radicals had not yet been isolated, they had not been shown to be
compounds, in which case they belong on the table of simple substances. To be
sure, Hendry is correct about Lavoisier’s inconsistent application of his notion of
element. My point is just that he had his reasons for being inconsistent.
The upshot of this way of solving the puzzle is that, although it turned out to
be true that the boracic radical is an element, namely, boron, the predictive suc-
cess in this case is not the success of one of Lavoisier’s theories. Lavoisier’s theory
of acidity is what led him to omit the three undecomposed acids from his table
of simple substances. It did not lead him to include the three radicals on that
table since his theory allowed for the possibility of compound radicals. It was
his notion of element that led him to include the three radicals, and this notion
does not amount to a theory of the nature of elements in any deep sense. It basi-
cally amounts to a norm regarding the application of the term element: Until a
substance is demonstrated to be a compound, we should call it an element. As
Lavoisier (1790 [1965], xxiv) emphasizes, this notion does not contain within
it a view regarding “those simple and indivisible atoms of which matter is com-
posed.” It was precisely this kind of metaphysical view that Lavoisier attempted to
avoid by proposing his operational notion of element. If the boracic radical were
isolated and subsequently shown to be a compound, that result would be con-
sistent with both Lavoisier’s theory of acidity and his notion of element. It would
merely necessitate an update to the table of simple substances. Hence, although
this instance of novel predictive success may be interesting for other reasons, it is
not relevant to the issue of whether selective realists can accommodate cases of
false-​but-​successful theories because it is not a success of a theory.

3.6.3 The Purity of the Samples

In section 3.4, I noted that Davy, Gay-​Lussac, and Thénard isolated relatively
impure samples of amorphous boron.15 Given the impurity of the samples that
these three chemists isolated, did they really confirm Lavoisier’s predictions?

15 Methods for producing pure samples of boron were not developed until the 20th century

(Naslain 1977, 167).

 Scientific Realism Debate 49

In short, the samples that they isolated were pure enough to confirm the novel
predictions from section 3.3, because those predictions don’t require that the
boracic radical is an absolutely pure chemical substance. Confirming the pre-
diction that the boracic radical is an element might require isolating a sample
with a greater degree of purity. But as I argued in section 3.6.2, this prediction is
not relevant to the issue at hand. That said, it’s worth going into a bit more detail
regarding the ways in which Davy, Gay-​Lussac, and Thénard isolated as pure a
sample as they could, and how they thereby confirmed Lavoisier’s predictions. In
order to do so, I’ll make use of Chen’s (2016) account of experimental individu-
ation and go into a bit more detail regarding the experimental work on boron in
the early 19th century.
According to Chen, scientists experimentally individuate an entity or a sample
of an entity when their experiments satisfy three conditions: (1) they separate the
entity or sample from its surrounding environment, (2) they maintain the struc-
tural unity of the entity or sample, and (3) they manipulate the entity or sample
so as to investigate other aspects of nature.
Conditions (1) and (2) are the conditions most relevant to the isolation of
samples of chemical substances, and therefore to the confirmation of Lavoisier’s
prediction that boracic acid contains a combustible substance. According to
condition (1), chemists must separate the substances that they produce “from
their environments” (Chen 2016, 348), and “from the experimental instruments
that may have helped produce [them]” (365). And according to condition (2),
chemists must maintain the structural unity of these substances in the process.
In general, structural unity is the idea that “the components of an individual are
structured into a whole in some specific manner” (358). When it comes to chem-
ical substances, maintaining structural unity includes such things as ensuring
that a highly reactive substance that has been isolated doesn’t react with other
substances in its environment. These two conditions collectively yield an under-
standing of what it means to isolate a sample of a chemical substance. Moreover,
both of these conditions relate to the quality of the sample that chemists manage
to produce. We must ask whether chemists managed to produce and preserve
a sample that is both large enough and pure enough for them to determine the
properties of the substance.
To a large degree, Davy’s and Gay-​Lussac and Thénard’s experiments satisfied
these two conditions. Davy’s initial attempts to obtain the substance by means of
decomposing boracic acid with a battery were unsatisfactory precisely because
they didn’t satisfy these conditions. Davy (1809, 76) writes that, via this method,
he “was never able to obtain it, except in very thin films,” and so “[i]‌t was not
possible to examine its properties minutely, or to determine its precise nature,
or whether it was the pure boracic basis.” By heating boron trioxide in a metal
tube with potassium, these chemists were able to produce more of the substance
50 Historical Cases for the Debate

in question. While Gay-​Lussac and Thénard report using a copper tube for the
experiment (1808, 170), Davy reports repeating the experiment with a number
of different metal tubes, including tubes made of copper, brass, gold, and iron
(1809, 76–​77). According to Davy, “[i]n all cases, the acid was decomposed, and
the products were scarcely different” (77). By showing that the results were the
same regardless of what kind of metal was used, Davy goes some length toward
showing that the substance had been separated from the instruments used to
produce it. Both Davy (1809, 78) and Gay-​Lussac and Thénard (1808, 171) also
used water and muriatic acid to wash the mixture that resulted from this process
and thereby separate the boron they had produced from other substances like
potash and borate of potash. By doing so, they go some length toward showing
that the substance had been separated from its surrounding environment. In
sum, they were able to produce samples of the substance that were large enough
and pure enough for them to determine some of its properties (e.g., combusti-
bility) and conclude that it is a component of boracic acid.
Condition (3) is relevant to the confirmation of Lavoisier’s prediction re-
garding the formation of boracic acid. This condition concerns the “instru-
mental use” of an entity or sample of an entity “to investigate other phenomena
of nature” (Chen 2016, 358). Davy (1809, 79, 82; 1812, 179) and Gay-​Lussac
and Thénard (1808, 172–​173) used the samples of boron they had isolated to
form boracic acid by means of combustion with oxygen. Importantly, their
samples of boron were pure enough to confirm Lavoisier’s prediction. And in
the course of confirming this prediction, these three chemists used boron as
an instrument. However, it’s not clear that they were using boron to investigate
other phenomena of nature as opposed to continuing an investigation into the
same phenomena.
Experimental individuation is a matter of degree. Samples of chemical
substances can be more or less separated from their environments and from
the instruments that produced them, they can be more or less pure, and their
structural unity can be more or less maintained. The important point is that the
samples of boron that chemists produced in the early 19th century were pure
enough to confirm Lavoisier’s predictions.

3.6.4 Is Boron the Boracic Radical?

In section 3.4, I claimed that boron is the boracic radical. This claim is crucial
to my argument since, if boron is not the boracic radical, we can’t conclude that
Lavoisier’s predictions regarding boracic acid are successful. Hence, this claim is
in need of some defense, especially since Chang (2012b, 54) has labeled all three
 Scientific Realism Debate 51

of Lavoisier’s hypothetical radicals (the muriatic, fluoric, and boracic radicals) as

“non-​existent.” If the boracic radical doesn’t exist, then there is no way to identify
it with boron.
Chang’s attitude toward the three hypothetical radicals is presumably moti-
vated by some observations he makes regarding Lavoisier’s system of chemistry.
For Lavoisier, the element oxygen is, among other things, what one obtains by
taking an acid and removing its radical, or by taking oxygen gas and removing
caloric. More schematically, “oxygen = acid − radical” and “oxygen = oxygen
gas − caloric.” Chang holds that these are “meaningless formulations in modern
chemistry, with the empty set as the extension in each case” (2011, 417). For
Chang, it makes no sense, from our current standpoint, to ask what the radical
of an acid is, since the radical, for Lavoisier, is the acid’s non-​oxygen compo-
nent, and since we’ve long ago given up the idea that all acids contain oxygen.
Moreover, to say that the radical is the acid’s non-​oxygen component is to say
that it is the component of the acid that is not oxygen base, i.e., the substance that
results when oxygen gas loses its caloric. Chang’s claim that the three hypothet-
ical radicals don’t exist is presumably motivated by the fact that the very notion
of an acid’s radical is tied so closely to Lavoisier’s theory of acidity, which is not
even approximately true if our current theories are correct.
Chang’s view is quite reasonable when it comes to the muriatic and fluoric
radicals. Since the muriatic and fluoric acids contain no oxygen, there is no
way to identify the non-​oxygen component. In section 3.3, I discussed how the
terms muriatic radical and fluoric radical are best understood as temporary
placeholders for the names of whatever substances play the role of the non-​
oxygen components within the muriatic and fluoric acids, respectively. Once
chemists understood the composition of these two acids, this role no longer
made sense, and so the most reasonable conclusion to draw is that these two
radicals do not exist.
While Chang’s view only concerns Lavoisier’s three hypothetical
radicals, it’s worth emphasizing that it would be implausible to claim that
all of Lavoisier’s radicals are non-​existent. Consider, for example, the sul-
furic, phosphoric, and carbonic radicals, i.e., the radicals of sulfuric, phos-
phoric, and carbonic acid, respectively. If we were to deny that these three
radicals exist, that would amount to denying that they should be identified
with sulfur, phosphorus, and carbon, respectively. But in the case of these
three acids, the role descriptions of the radicals make some sense. The role
descriptions are approximately true in these cases, though it’s fair to note
that sulfur, etc., do not combine with oxygen base to form acids. Moreover, in
order to understand Lavoisier’s theory of acidity, one has to be able to see cer-
tain substances as radicals of acids. Without entering into the mindset that
52 Historical Cases for the Debate

sulfur is the radical of sulfuric acid, it’s difficult to comprehend Lavoisier’s

theory. So while we have good reasons to conclude that the muriatic and
fluoric radicals do not exist, these reasons do not extend to the sulfuric,
phosphoric, and carbonic radicals, and we have good reason to identify
them with sulfur, phosphorus, and carbon, respectively. Importantly, these
identifications don’t require a commitment to Lavoisier’s theory of acidity.
We can of course admit that his theory as a whole was not even approxi-
mately true. These identifications merely require an attempt to see the world
through Lavoisier’s theory insofar as it’s possible to do so.
The boracic radical is much more like the sulfuric, phosphoric, and car-
bonic radicals than the muriatic and fluoric radicals, and so it’s better to
conclude that it exists, and to identify it with boron. The role description in
this case is approximately true—​t he boracic radical is what combines with
oxygen to form boracic acid, and that is exactly what boron does. Moreover,
identifying the boracic radical with boron helps us to understand the his-
tory of the discovery of boron. For example, it helps us to understand why
two of the chemists credited with discovering boron, namely, Gay-​Lussac
and Thénard, label the substance they isolated the “radical boracique”
(1808, 171). Importantly, one can label something as a base or radical while
rejecting Lavoisier’s theory, and this seems to be exactly what Davy himself
did. As I noted in section 3.4, Davy (1812, 178) was still referring to boron
as “the basis of the boracic acid” two years after he had argued that muriatic
acid contains no oxygen. The primary reason for concluding that the muri-
atic and fluoric radicals do not exist is that neither muriatic acid nor fluoric
acid contains oxygen; and that reason does not extend to the boracic radical.
In short, insofar as there are good reasons for identifying, say, sulfur with
the sulfuric radical, there are good reasons for identifying boron with the
boracic radical.

3.7 Conclusion

In this chapter, I’ve attempted to gain a better understanding of the work

in chemistry that led to the discovery of boron and the significance of this
work for the scientific realism debate. I’ve added Lavoisier’s oxygen theory
of acidity to the list of theories that, although not even approximately true,
exhibited novel predictive success. I’ve considered some ways in which the
selective realist may attempt to accommodate this instance of novel predic-
tive success. And I’ve argued that these attempts are unsuccessful. In that
case, Lavoisier’s oxygen theory of acidity poses a strong challenge to selective
realism.
 Scientific Realism Debate 53

Acknowledgments

This chapter branched off from a co-​authored project with Ruey-​Lin Chen,
and I’d like to single him out for special thanks. This chapter also benefited
from very helpful suggestions from Timothy Lyons. Thanks to the organizers
and participants of the Quo Vadis Selective Scientific Realism? conference at
Durham University in August 2017, especially Hasok Chang, Ludwig Fahrbach,
Amanda J. Nichols, Myron Penner, Jan Potters, Juha Saatsi, Yafeng Shan, and
Peter Vickers. Thanks to three anonymous referees for their helpful comments
and suggestions. Finally, thanks to the Ministry of Science and Technology in
Taiwan for supporting this work (MOST 106-​2410-​H-​010-​001).

References
Alai, Mario. 2017. “The Debates on Scientific Realism Today: Knowledge and Objectivity
in Science.” In Varieties of Scientific Realism: Objectivity and Truth in Science, edited by
Evandro Agazzi, 19–​47. Cham: Springer.
Berthollet, Claude Louis. 1789. “Extrait d’un mémoire sur l’acide prussique.” Annales de
Chimie 1: 30–​39.
Brooke, John Hedley. 1980. “Davy’s Chemical Outlook: The Acid Test.” In Science and
the Sons of Genius: Studies on Humphry Davy, edited by Sophie Forgan, 121–​175.
London: Science Reviews.
Chang, Hasok. 2003. “Preservative Realism and Its Discontents: Revisiting Caloric.”
Philosophy of Science 70 (5): 902–​912.
Chang, Hasok. 2011. “The Persistence of Epistemic Objects through Scientific Change.”
Erkenntnis 75 (3): 413–​429.
Chang, Hasok. 2012a. “Acidity: The Persistence of the Everyday in the Scientific.”
Philosophy of Science 79 (5): 690–​700.
Chang, Hasok. 2012b. Is Water H2O? Evidence, Realism and Pluralism. Dordrecht: Springer.
Chen, Ruey-​ Lin. 2016. “Experimental Realization of Individuality.” In Individuals
across the Sciences, edited by Alexandre Guay and Thomas Pradeu, 348–​ 370.
New York: Oxford University Press.
Crosland, Maurice P. 1973. “Lavoisier’s Theory of Acidity.” Isis 64 (3): 306–​325.
Crosland, Maurice P. 1978. Gay-​Lussac: Scientist and Bourgeois. Cambridge: Cambridge
University Press.
Davy, Humphry. 1808. “The Bakerian Lecture [for 1807]: On Some New Phenomena
of Chemical Changes Produced by Electricity, Particularly the Decomposition of the
Fixed Alkalies, and the Exhibition of the New Substances Which Constitute Their
Bases; and on the General Nature of Alkaline Bodies.” Philosophical Transactions of the
Royal Society of London 98: 1–​44.
Davy, Humphry. 1809. “The Bakerian Lecture [for 1808]: An Account of Some New
Analytical Researches on the Nature of Certain Bodies, Particularly the Alkalies,
Phosphorus, Sulphur, Carbonaceous Matter, and the Acids Hitherto Undecompounded;
with Some General Observations on Chemical Theory.” Philosophical Transactions of
the Royal Society of London 99: 39–​104.
54 Historical Cases for the Debate

Davy, Humphry. 1811. “The Bakerian Lecture [for 1810]: On Some of the Combinations
of Oxymuriatic Gas and Oxygene, and on the Chemical Relations of These Principles,
to Inflammable Bodies.” Philosophical Transactions of the Royal Society of London
101: 1–​35.
Davy, Humphry. 1812. Elements of Chemical Philosophy. Philadelphia: Bradford and
Inskeep.
Fontani, Marco, Mariagrazia Costa, and Mary Virginia Orna. 2015. The Lost Elements: The
Periodic Table’s Shadow Side. Oxford: Oxford University Press.
Gay-​Lussac, Joseph Louis, and Louis Jacques Thénard. 1808. “Sur la décomposition et la
recomposition de l’acide boracique.” Annales de Chimie 68: 169–​174.
Hendry, Robin Findlay. 2005. “Lavoisier and Mendeleev on the Elements.” Foundations of
Chemistry 7 (1): 31–​48.
Hricko, Jonathon. 2018. “Retail Realism, the Individuation of Theoretical Entities, and the
Case of the Muriatic Radical. In Individuation, Process, and Scientific Practices, edited
by Otávio Bueno, Ruey-​Lin Chen, and Melinda B. Fagan, 259–​278. New York: Oxford
University Press.
Kitcher, Philip. 1993. The Advancement of Science: Science without Legend, Objectivity
without Illusions. New York: Oxford University Press.
Knight, David M. 1992. Humphry Davy: Science and Power. Cambridge, MA: Blackwell.
Laudan, Larry. 1981. “A Confutation of Convergent Realism.” Philosophy of Science 48
(1): 19–​49.
Laurent, Auguste. 1855. Chemical Method, Notation, Classification, and Nomenclature.
London: Harrison and Sons, St. Martin’s Lane.
Lavoisier, Antoine Laurent. 1790 [1965]. Elements of Chemistry. 1st ed. Translated by
Robert Kerr. New York: Dover. Translation of Traité élémentaire de chimie, 1789.
Le Grand, H. E. 1972. “Lavoisier’s Oxygen Theory of Acidity.” Annals of Science 29
(1): 1–​18.
Lowry, Thomas Martin. 1915. Historical Introduction to Chemistry. London: Macmillan.
Lyons, Timothy D. 2002. “Scientific Realism and the Pessimistic Meta-​Modus Tollens.” In
Recent Themes in the Philosophy of Science: Scientific Realism and Common Sense, ed-
ited by Steve Clarke and Timothy D. Lyons, 63–​90. Dordecht: Kluwer.
Lyons, Timothy D. 2006. “Scientific Realism and the Stratagema de Divide et Impera.” The
British Journal for the Philosophy of Science 57 (3): 537–​560.
Lyons, Timothy D. 2016. “Structural Realism versus Deployment Realism: A Comparative
Evaluation.” Studies in History and Philosophy of Science 59: 95–​105.
Lyons, Timothy D. 2017. “Epistemic Selectivity, Historical Threats, and the Non-​Epistemic
Tenets of Scientific Realism.” Synthese 194 (9): 3203–​3219.
Miller, William Allen. 1871. Elements of Chemistry: Theoretical and Practical. Part II.
Inorganic Chemistry. 3rd London ed. New York: John Wiley & Son.
Musgrave, Alan. 1988. “The Ultimate Argument for Scientific Realism.” In Relativism and
Realism in Science, edited by Robert Nola, 229–​252. Dordrecht: Kluwer.
Naslain, R. 1977. “Crystal Chemistry of Boron and of Some Boron-​ Rich Phases;
Preparation of Boron Modifications. In Boron and Refractory Borides, edited by Vlado
I. Matkovich, 139–​202. Berlin: Springer-​Verlag.
Nicholson, William. 1795. A Dictionary of Chemistry. Vol. 1. London: G. G. and
J. Robinson, Paternoster Row.
Psillos, Stathis. 1999. Scientific Realism: How Science Tracks Truth. London: Routledge.
 Scientific Realism Debate 55

Scerri, Eric R. 2005. “Some Aspects of the Metaphysics of Chemistry and the Nature of
the Elements.” HYLE—​International Journal for Philosophy of Chemistry 11: 127–​145.
Stanford, P. Kyle. 2006. Exceeding Our Grasp: Science, History, and the Problem of
Unconceived Alternatives. New York: Oxford University Press.
Thomson, Thomas. 1802. A System of Chemistry. Vol. 2. Edinburgh: Bell & Bradfute, and
E. Balfour.
van Fraassen, Bas C. 1980. The Scientific Image. Oxford: Clarendon Press.
Vickers, Peter. 2013. “A Confrontation of Convergent Realism.” Philosophy of Science 80
(2): 189–​211.
Zhigach, A. F., and D. C. Stasinevich. 1977. “Methods of Preparation of Amorphous
Boron.” In Boron and Refractory Borides, edited by Vlado I. Matkovich, 214–​226.
Berlin: Springer-​Verlag.
 4
No Miracle after All
The Thomson Brothers’ Novel Prediction that Pressure
Lowers the Freezing Point of Water
Keith Hutchison

School of Historical and Philosophical Studies

University of Melbourne
[email protected]

4.1 A Problem and Its Context

Defenders of scientific realism sometimes suggest that it would be “miraculous”

for false theories to generate successful predictions. But it is fairly common for
theories to make good predictions. So it must be rare for such theories to be false.
The predictive success of science, then, seems good evidence for its truth.
The early history of thermodynamics, however, poses a couple of pressing
challenges to simplistic understandings of this no-​miracle argument. For it seems
to provide clear examples of false theories producing successful predictions.
Rankine’s thermodynamics of around 1850, for example, anticipated a string of
experimental effects, even though it was based on a model of heat (as a non-​
random vortex motion in the interior of atoms) that is now thoroughly ignored.
Indeed it was the surprising success of one of these predictions that prompted
Rankine to start publishing his investigations.1 About the same time he predicted
that saturated steam would have a negative specific heat, a phenomenon that
soon seemed to be equally confirmed by observation, albeit via ambiguous
data.2 More momentous however was Rankine’s use of the same model of heat
to predict (around 1854) the existence of the entropy function, a function that

1 Rankine, ‘On an equation’ (1849) = Misc. Sci. Pap., pp.1–​12. The prediction here was of the

form of the relationship between the temperature and saturated vapour pressure of a liquid/​vapour
mixture, without a specification of precise numerical values, and this may well place it outside the
intended scope of the no-​miracle argument.
2 For an overview, see: Hutchison, ‘Miracle or mystery,’ pp.106–​7 and ‘Mayer’s hypothesis,’ esp.

pp.288–​294.

Keith Hutchison, No Miracle after All In: Contemporary Scientific Realism. Edited by: Timothy D. Lyons and Peter Vickers,
Oxford University Press. © Oxford University Press 2021. DOI: 10.1093/​oso/​9780190946814.003.0004
 No Miracle After All 57

consequently became familiar to physicists a decade before Clausius renamed

it in his famous 1865 declaration of the connection between entropy and the
arrow of time.3 I surveyed these predictions in a paper published some 20 years
ago, though I then left unresolved the ultimate impact of Rankine’s successes on
the defence of realism, for I underestimated the obstacles to showing that his
predictions really did depend on the discredited theory.4 Even now, that task re-
mains beyond me, if only because of the logical maze that I glimpse within it.
So I wish to analyse here a quite different example of a successful prediction
from the early history of thermodynamics, one that was (beyond any doubt)
made using another abandoned theory, the 1824 proto-​thermodynamics of Sadi
Carnot.5 This magnificent analysis of the efficiency of steam engines was bal-
anced upon a postulate that its author personally doubted, the conservation of
heat. Accordingly, it allowed no inter-​conversions between heat and other forms
of what would today be called energy, and for this reason was unequivocally
abandoned around 1850, as the first formulations of modern thermodynamics
emerged, with its so-​called first law replacing Carnot’s postulate.6
But just before this happened, William Thomson (the later Lord Kelvin) was
able to confirm the magnitude of an “entirely novel physical phaenomenon”
predicted “in anticipation of any direct experiments on the subject” by his
brother James on the basis of Carnot’s theory: a lowering of the freezing point of
water, produced by an increase in the external pressure.7 The Thomson brothers
rightly saw the success of this prediction as good evidence for a theory that was
attractive for a lot of other reasons. But at the same time, James Joule was pro-
ducing strong evidence that Carnot was wrong, and William was much puz-
zled, acknowledging the persuasiveness of Joule’s investigations. He concluded
that a resolution of the dilemma required further empirical data—​but this was
wrong. For in 1850, Clausius managed to harmonise the two understandings by
purely intellectual (and relatively harmless) modifications to the Carnot theory.

3 Rankine called it the “thermodynamic function”, and denied (e.g. Misc. Sci. Pap., p.233) that it

had any asymmetric tendency to increase. For its first explicit identification, see pp.351–​2 of his Misc.
Sci. Pap. For Clausius’s renaming, see the English trans. on pp.364–​5 of Mechanical theory. A predic-
tion like this, that a particular state-​function exists, is definitely subject to empirical verification, but
it is certainly atypical, so could perhaps be outside the scope of the no-​miracle argument.
4 Hutchison, ‘Miracle or Mystery’ (2002). In that study, I did consider the dependence question,

but only briefly (on pp.109–​11), but that discussion now feels too thin. As I recall, those remarks were
inserted at the request of a referee.
5 Carnot, Réflexions sur la puissance motrice du feu.
6 For a survey, see e.g., Cardwell, Watt to Clausius, esp. pp.186–​260.
7 The whole process took place between 1847 and 1850. Though the prediction came out under

James’s name, William was heavily involved and may even have been the principal investigator. For
the chronology etc., see: Smith & Wise, Energy and empire, pp.294–​99. For the original prediction,
see: J. Thomson, ‘Theoretical considerations’ and W. Thomson ‘The effect of pressure,’ reprinted
together on pp.156–​69 of W. Thomson, Math. Phys. Pap., v.1. My quotations are from pp.165–​6 of
William’s paper. Given that the sensitivity of the boiling point of water to pressure had long been
known, it is odd that the sensitivity of the freezing point was (apparently) quite unsuspected in 1848.
58 Historical Cases for the Debate

To preserve Carnot’s central claim, however, Clausius did need to invoke a new
premise—​a second law—​and with these decisive steps, modern thermody-
namics emerged. Thomson soon accepted Clausius’s understandings.
My concern here is with that slightly earlier prediction however. What was the
thinking that led James to conclude that external pressure lowered the freezing
point? How did he defend this conclusion? How did he estimate the magnitude
of the lowering? Did he really use the theory that was about to be discarded?
By answering these historical questions, I shall demystify the prediction, by
showing that though James’s argument uses suppositions rejected in modern
thermodynamics, and applies those falsehoods to a situation where the two the-
ories differ, the quantitative difference between the two accounts of James’s phe-
nomenon is only marginal. One says that heat is conserved here, the other says it
is not, but agrees that in this particular situation heat is nearly conserved. James’s
argument furthermore needed a principle that was central to both theories, but
at the time he wrote that principle was only supplied by the Carnot theory. So his
prediction did indeed need the precarious theory. Yet the critical inference did
not itself directly use those portions of the Carnot theory that were soon aban-
doned. The false components did however give credence to the true portions—​
so mattered. But there is (seemingly) little miracle in using the truish sections
of a false theory to make a prediction. To demolish the no-​miracle argument,
a bigger miracle is needed, one where a successful prediction hinges on bigger
deviations from the truth.
In discussing this episode, I deploy a very naive distinction between “true”
and “false” statements. This greatly simplifies my narrative, for it avoids compli-
cated hedging, circumlocutions, etc. The associated over-​simplifications seem to
me to do no harm, because the nature of “scientific truth” is not the issue here
and does not need to be confronted. But my language might at times seem an-
noying to the fastidious reader.

4.2 The Carnot Theory of 1824

The half-​century beginning with 1775 saw a ten-​fold increase in the effi-
ciency of steam engines, an improvement so extraordinary that the question
began to circulate whether there was any limit to this process.8 Attempts to
answer this question were generally of limited scope, focusing too closely
on familiar devices, but one was quite exceptional. In a masterpiece of ele-
gant analysis, Sadi Carnot introduced (in 1824) the notion of a “heat-​engine,”
one characterised by the single fact that it produces mechanical work by

8 See, e.g. Fox, ‘Introduction etc.,’ pp.2–​12 and Cardwell, Watt to Clausius, pp.153–​81.
 No Miracle After All 59

exchanging heat with its environment. The recent prevalence of the notion
that heat was a special fluid, caloric, suggested that there was, accordingly,
some sort of analogy between heat-​engines and water-​engines, given that the
latter operated through an exchange of water with their environment. This
analogy lies at the heart of Carnot’s discussion, but it also incorporates the
flaw that eventually led to the analysis being abandoned, the presumption that
heat (like water) was conserved. Yet it provided Carnot with his key insight,
an insight that survived the later demise of the 1824 analysis: just as a water-​
engine requires two “reservoirs,” a high one to supply water and a low one to
receive “waste,” a heat-​engine requires two heat-​reservoirs, one at a high tem-
perature to supply heat, and one at a low temperature to receive heat. (Watt’s
separate condenser had made it abundantly clear that steam engines produced
waste-​heat, and that, accordingly, their coal needed to be supplemented by a
source of cold, simply for the engine to function.)
The same analogy also provided Carnot with the idea of a heat-​pump, and
that of a reversible engine, and he was then able to make a simple adjustment
to an argument that had already been applied to water-​engines, to show that
if it was accepted that perpetual motion was impossible, no heat-​engine could
be more efficient than one which was perfectly reversible (for given source
and sink levels). This is Carnot’s answer to the question challenging his con-
temporaries, and to enhance its bona fides, Carnot showed that his theory
was very fertile indeed. He produced some rough calculations of the maximal
efficiencies he had identified, rough only because too much of the requisite
data was lacking; and he produced a series of novel lemmata about the proper-
ties of fluids, typical of those deductions which are a standard feature of later
thermodynamics.9

4.3 Thomson Revives the Carnot Theory

But Carnot’s bold theory attracted little attention—​until the late 1840s, after a
copy of his memoir had been discovered by William Thomson, then working
in Regnault’s laboratory in Paris (where the sort of data needed to complete
Carnot’s calculations was being systematically generated). Thomson was en-
thusiastic about the analysis. He soon published his own rehearsal of the theory,
with some of the core calculations systematized (via Regnault’s new data). But
before this came out, Thomson also showed that Carnot’s understanding pro-
vided a rationale for absolute thermometry, a temperature scale independent of

9 For the efficiency estimates, see Carnot, Réflexions, pp.94–​ 101. For the lemmata, see: ibid.,
pp.78–​90; Fox, ‘Introduction etc.,’ pp.132–​45; Cardwell, Watt to Clausius, pp.201–​4.
60 Historical Cases for the Debate

the idiosyncrasies of any particular chemical substance. And he noticed that the
Stirling air-​engine vividly illustrated Carnot’s claims: it used two heat reservoirs
and could be run in two different directions: as a heat-​engine; or as a heat-​
pump—​where the transfer of heat was palpable.10

4.4 The Rationale Behind James Thomson’s Prediction

When a model Stirling engine was run as a pump, shifting heat between two
reservoirs at the same temperature, virtually no work was required to operate
the pump (just as Carnot’s theory would predict).11 It was this fact that led to
James’s prediction, because it meant that heat could easily be pumped from
beneath the piston of a cylinder containing a water-​ice mixture at 0℃ to
(say) an icy lake at the same temperature (where the heat received would melt
some ice without changing the lake’s temperature). Yet if this was done, some
of the water in the cylinder would freeze, and in doing so it would expand
(for water colder than 4℃ expands when cooled). And it was well known that
the forces generated by freezing water are very great, for they often caused
serious damage. So quite a high force could be applied to the piston, without
preventing the expansion, and combining this force with the movement of
the piston would yield a supply of external work that exceeded (by far) the
negligible quantity of work consumed in pumping the heat to the lake. So a
perpetual motion would seem to be on offer. Or rather there had to be some
flaw in the reasoning here: some phenomenon, hitherto overlooked, must be
sabotaging the proposal.
Reflection quickly shows one such flaw: the argument that threatens a per-
petual motion presumes the temperature in the high-​pressure cylinder to be
the same as that in the lake, i.e. that the temperatures of the two different
ice-​water mixtures are identical, despite the different pressures at which the
phases change. (For it is this identity that enables the more or less work-​f ree
pumping of the heat from the cylinder.) So if the temperature at which water
freezes changes with the pressure the argument collapses. Indeed, if the tem-
perature of the high-​pressure cylinder were lower than that of the lake (acted
on by normal atmospheric pressure) work would be consumed in pumping
the heat “uphill” into the lake, and this work could suffice to prevent a per-
petual motion.

10 W. Thomson, ‘Notice of Stirling’s air engine.’

11 So this minor episode might provide another example of a false theory successfully predicting
an unknown effect.
 No Miracle After All 61

4.5 A More Polished Analysis

Such were the thoughts that led to James’s discovery, though his published an-
nouncement of the discovery includes a more systematic analysis, one extended
to include a quantitative estimate of the magnitude of the newly appreciated ef-
fect.12 Let us now sketch that argument.
We begin by observing that to understand the operation of any engine, we
obviously need to make a clear distinction between the fuel (materials which
take on permanently new forms in the course of the operation of the device)
and the engine proper (that which suffers no permanent change, the “working
substance” as it is often called, plus its mechanical container, etc.). Carnot
accommodates this fundamental distinction by having his engines operate in
cycles, their materials passing through a sequence of changes that bring them
back to their original configuration. To avoid wastage, these changes need (as
mentioned earlier) to be reversible; and in a thermal context, the simplest such
engine would seem to be one with a single working substance undergoing just
four changes, two to exchange heat with the two different reservoirs (each at the
same temperature as the engine materials13), and two to switch the temperature
of those materials with no exchange of heat. The materials need to be brought,
back and forth, between the pair of temperatures asserted by Carnot to be essen-
tial to a heat-​engine.14
Adapting this idea of a “Carnot cycle” to his context, James considers an en-
gine composed of a cylinder containing a mixture of water and ice beneath a
piston. As the engine operates, the quantities of water and ice here vary, as ice
sometimes melts, and as water sometimes freezes, in association with the ener-
getic exchanges between the engine and its environment.
To see the details, imagine the cylinder to be thoroughly insulated, with a rel-
atively large weight on the piston, so that the water-​ice mixture is at a highish
pressure, and in internal equilibrium at whatever temperature happens to be
the melting point of water for that pressure. Suppose now (as a preliminary) the

12 For the more informal discussion, see J. Thomson, ‘Theoretical considerations,’ pp.157–​9; for

the quantitative version, see pp.160–​4.

13 It might (very reasonably) be objected that no heat will flow into the reservoir if it is at the same

temperature as the ice-​water mixture. Like Carnot before him (e.g. Réflexions, p.68n), James recog-
nized this problem, and avoided it. I, by contrast, create it here—​by using a circumlocution that is
quite common in thermodynamics, which serves to abbreviate the wording, and which I presume
will be familiar to any concerned reader. Since he was writing at the very beginning of thermody-
namic analysis, James could not presume such familiarity and explains the logic more fully, overtly
imagining a small difference in temperature, but one that could be made arbitrarily small, perhaps
infinitesimal (etc.) Some such artifice lies behind many references to isothermal change in thermo-
dynamic argument, and is routinely deployed in descriptions of a Carnot cycle.
14 Thermocouples however operate without undergoing any change themselves, but electrical

engines were a rarity at the time of the Thomson brothers’ discovery, so I just ignore any minor chal-
lenge they might pose to my narrative.
62 Historical Cases for the Debate

insulation is removed, and the cylinder brought into thermal contact with a heat
reservoir at that same temperature, and that heat flows out of the cylinder and
into the reservoir, so some of the water in the cylinder freezes, and the resulting
expansion causes the piston to lift the weight, doing mechanical work on its
environment.
After a period of expansion, one chosen simply for pragmatic reasons, the
cylinder is insulated again, and the weight on the piston is slowly reduced. As
the pressure is reduced, the mixture in the cylinder expands, and work is done
on the environment, while water freezes, or ice melts, to bring the cylinder to a
new temperature, that which is the melting point of water at the new, reduced,
pressure. This is to be the first stage of our cycle. For stage 2, the insulation is
removed, and the cylinder is brought into contact with a second heat-​reservoir,
one at the same temperature as the low-​pressure cylinder. Heat is now injected
into the cylinder, melting some of the ice, so that the piston descends (as the
volume of its contents decreases) and work is done on the system. After a con-
venient time, the cylinder is insulated again, and stage 3 begins: the weight on the
piston is slowly increased, further compressing the mixture in the cylinder, with
yet more work being done by the environment. The stage terminates when the
weight on the piston reaches its original value. Along the way, some water will
freeze, or some ice will melt, and the cylinder will return to its original tempera-
ture. The final, fourth, stage then mimics the preliminary process described ear-
lier, beginning as thermal contact with the first heat-​reservoir is reestablished.
Expansion resumes and heat flows to the reservoir, until the initial volume of
stage 1 (= the final volume of the preliminary stage) is reached. The cycle is now
complete, and can be repeated if desired etc.
This cycle constitutes a heat-​engine. For it exchanges nothing but heat and
work with its environment. And it produces mechanical work: for each change
of volume takes place in two directions, and the compression that reverses an ex-
pansion takes place at a lower pressure than the corresponding expansion. So if
Carnot is right, heat must be flowing from a high-​temperature reservoir to a low-​
temperature one. But we know that in this case heat is flowing into the reservoir
during stage 4 and out of the reservoir during stage 2 (while no heat is flowing
during stages 1 and 3). So the high-​pressure stage 4 must take place at a lower
temperature than the low-​pressure stage 2. In other words, an increase in pres-
sure reduces the freezing point.

4.6 No Miracle Involved in This Prediction

There is no doubt that James’s prediction here uses the Carnot theory, for it is
this that tells James heat must flow from a high temperature source to a low
 No Miracle After All 63

temperature sink. But Carnot does more for James than simply assert this neces-
sity, it embeds the requirement in a sustained analysis which enhances its plau-
sibility. The fact that the Carnot theory requires such a thermal descent is not,
however, what makes it false. The theory is false because of something quite dif-
ferent, the fact that it declares the quantity of heat released to the sink identical to
that acquired from the source. Carnot’s discussion made much use of the heat-​
conservation that this alleged identity illustrates,15 and the rationale for insisting
upon a descent of heat presumes conservation. But once that descent is accepted,
the prediction requires no further appeal to conservation. Indeed, James’s deri-
vation works just as well in post-​1850 thermodynamics where the heat released
to the sink differs from that taken from the source. Cosmetic adjustments do, of
course, need to be made to articulations of the story; and a different context (like
that provided by Clausius), is needed to make the necessity of a low tempera-
ture sink plausible. So James may be thought of as taking a trustworthy deriva-
tion and placing it in bad company (via the alternative rationale for its central
premise). But no-​one is challenged by a trustworthy argument making a suc-
cessful prediction. In other words, James’s prediction requires no miracle, and
the episode shows that a false theory really can—​sometimes—​make a successful
prediction. But it is a weak counter example to the no-​miracle argument, for the
prediction proceeds via the true components of a false theory. To get a better
counter-​example, it would be necessary to show something stronger: that James’s
deduction would collapse if denied some false component of the Carnot theory,
heat-​conservation in particular. But it does not collapse. All that happens is that
its critical premise is left without a rationale. The Rankine story is far more pow-
erful, then, and does seem to be a real counter-​example—​if I got that story right.

4.7 Some Methodological Hesitation

Before moving on to James’s quantitative discussion, where he estimates the

magnitude of the effect predicted, thoroughness requires some navel-​gazing, in
which we briefly consider the relationship between my summary of James’s ar-
gument and his own wording. Needless to say the two differ, though the discrep-
ancies are no more severe than those standard in the history of science, where
historians routinely paraphrase their sources. (Indeed, I find I cannot trust a his-
torian whose articulation of an argument too closely mirrors that of the original
version: it suggests the argument has not been mastered.) Accordingly, we need

15 The argument given by Carnot (at p.69 of Réflexions) to show that no engine can be more effi-

cient than a reversible uses the conservation of heat. See also: Carnot, Réflexions, pp.65, 76n, 100–​
101; Fox, ‘Introduction etc.,’ pp.131, 150.
64 Historical Cases for the Debate

to consider how much this matters in the context of the no-​miracle argument—​
arguing that the differences are of no moment.
Like me, James gives two versions of the argument (see n. 12), one informal
(and focused on that key insight supplied by the Stirling engine) and the
other phrased, more systematically, in terms of a Carnot cycle. James however
introduces the Carnot cycle as a means of making his quantitative estimate, and
regards the effect itself as established by the more informal argument; while I use
that cycle to perform both these tasks (with the quantitative evaluation delayed
until later). Otherwise, our discussions of the cycle are very close, though that of
James seems more verbose, aimed at an audience with little exposure to thermo-
dynamic analysis (as, e.g. indicated in n. 13).
Our informal arguments however differ markedly, mainly because I cannot
follow the original argument in detail. There may (accordingly) be some serious
blemish in James’s prediction, a blemish hidden by my rephrasing.16 Our concern
here, however, is with the plausibility of the prediction, whether the conclusions
are reasonably justified by the premises (and their context), not whether a par-
ticular string of words adequately displays that connection. So a blemish in some
wording of the argument is of no moment, so long as some minor redrafting
sustains the conclusion—​without introducing new premises. James may have
written his argument down hastily, and not done full justice to his thoughts, or
he may have gone to an opposite extreme, and penned numerous drafts, none of
which is canonical. In either extreme, the published argument does not tell the
full story, and the real argument for the prediction is that elaboration of the overt
wording that each of us privately prepares, as we absorb the message conveyed
to us by the written summary. So there are (I suggest) no grounds for concern
in the mere existence of small discrepancies between James’s words and my own
phrasing of the argument that supplies the prediction.

4.8 The Quantitative Calculation

To estimate the magnitude of the effect he had discovered, i.e. how large a re-
duction (“−t”) in the freezing point is produced by a given increase (“[+]‌p”) in
pressure, James examines a concrete example of the Carnot cycle we introduced
to set out the qualitative argument.17 To reveal the relationship between p and t,

16 James does not, for example, evaluate the work done on the environment during the adiabatic

stages of his cycle, just treating the work done in these stages as negligibly small. I, by contrast, ex-
panded my description of the cycle slightly, in such a manner as would (hopefully) reduce any un-
certainty here. James’s brevity probably does not matter, but I am not sure. The issue hardly seems to
matter, since it is so easy to rephrase James’s argument.
17 James Thomson, loc. cit., n.12 = ‘Theoretical considerations,’ pp.160–​4.
 No Miracle After All 65

he uses three “known” quantities: (1) how much expansion is produced by the
freezing of water; (2) the latent heat required to melt ice at 0℃; and (3) the effi-
ciency of a Carnot engine that operates by absorbing heat at temperature 0℃ and
releasing heat at temperature −t℃. At that time, there was no novelty in the first
two of these data, while the third had recently been calculated by William, in his
systematic revision of Carnot’s fragmentary calculations.
The first of these items gives James the work done by his engine, as a multiple
of p—​viz. “0.087 × p.” (James ignores the work done during the changes in tem-
perature, since the adiabatic stages of the cycle are so small, “infinitesimal” so
to speak.) The other two quantities give another, quite different, expression for
this same work, now as a multiple of t—​viz. “4925 × 4.97 × t,” just the product
of William’s figure for the efficiency (“4.97”), the fuel consumed (“4925”) and
the magnitude of temperature drop (“t”). Equating these two figures, gives James
his final estimate, “t = 0.00000355p,” for the change in melting point produced
by an additional pressure p. Without specifying James’s units (viz. “cubic feet”;
“foot-​pounds”; “pounds on a square foot”; “the quantity of heat required to raise
a pound of water from 0 to 1 degree centigrade”) these results are verging on the
meaningless, but our aim here is not to reveal the actual magnitude of the effect
he discovered, just to check that James’s calculation requires no appeal (direct or
indirect) to the conservation of heat.
We have clearly not established such an independence yet, for though the first
two items of data make no discernible appeal to conservation, the earlier cal-
culation of efficiency was made in the context of the Carnot theory, and used
data that had been prepared in a laboratory that did not embrace the possibility
that heat was created or destroyed. So application of this calculation to James’s
cycle could well have required some identification, overt or hidden, of the heat
supplied to some engine with that it released. And some such comparison had
indeed been required in the very calculations used by James, those performed
by William to indicate that his first “absolute” scale of temperatures was in good
agreement with familiar laboratory scales. When the Carnot theory was aban-
doned, William realised his calculations were suspect, so checked the agreement
he had claimed for the earlier scale, and found it no longer applied. He decided to
rescue the convenience it had promised, by developing a rather different absolute
scale, effectively the one in use today.18
We thus need to scrutinise James’s use of William’s earlier calculations. But
before doing so, it is worth observing that James’s method of calculating the
magnitude of the effect he had discovered is effectively a standard component
of modern thermodynamics textbooks, used to generate the so-​called Clausius-​
Clapeyron equation, relating the variation in the boiling-​point of a liquid to

18 Hutchison, ‘Mayer’s hypothesis,’ pp.294–​300; Chang, Inventing temperature, pp.182–​6.

66 Historical Cases for the Debate

the external pressure. A version of this equation had been published by Émile
Clapeyron, in his important 1834 injection of Carnot’s discussion into a far more
mathematical, and especially graphical, idiom.19 The later survival of the calcula-
tion suggests that the result was not essentially tied to heat-​conservation, but we
need here to confirm that suggestion—​for on a strictly literal reading of James’s
account, a dependence of heat-​conservation is evident.

4.9 William Thomson’s Figure for the Efficiency

of James’s Engine

James tells us that his estimate of the efficiency of his engine comes from “tables
deduced . . . from the experiments of Regnault” and published in his brother’s
major paper of 1849—​to which James’s own discussion is attached as an appendix
(in the 1882 reprint). Strictly speaking this is not correct, as the coldest engine
listed in William’s table is one that operates just above 0℃, while James’s engine
operates just below that temperature. But William’s figure for the efficiency of a
perfect engine taking heat from 1℃ to 0℃ is 4.96, and James has presumably got
his 4.97 by extrapolation (something perfectly reasonable), or by access to data
not used in the final version of William’s table.
William had published those efficiency figures as the ones applicable to all
Carnot engines, so James was entitled to apply the figure to his own ice/​water de-
vice, and that is the logic used by James here.20 Yet an appeal to heat-​conservation
is lurking beneath the surface here, for (as observed in n. 15) the argument that
the figure at issue applied to all engines used the problematical presumption. But
a validation of this figure does not require a universal conservation of heat. It is
enough that heat be conserved in two particular engines, that used by William
to compute the efficiency and that used by James to evaluate the change in the
freezing point. And both these engines operate over a very small temperature
range, where the difference between the fuel heat and the waste heat is tiny. So the
claim that heat is conserved is effectively true in these particular circumstances,
and the suspect evaluation of the “infinitesimal” engine’s efficiency is in fact reli-
able, as William later checked.21
But there is also a risk, far less obvious, of a further appeal to heat-​conservation
in William’s use of Regnault’s experimental data. For Regnault could well have

19 See the three equations on p.90 of Clapeyron, ‘Motive power.’ For modern versions, see

Pippard, Classical thermodynamics, p.54 (eq. 5.12) and Fermi, Thermodynamics, pp.63–​69. Clausius
points out that James’s prediction is easily transferred to the new theory: see his 1850 remarks on
pp.80–​3 of Mech. Theory.
20 The critical passage is on p.163 of Math. Phys Papers, v.1. William’s table is on p.140.
21 See n. 22.
 No Miracle After All 67

presumed that heat was conserved in making his measurements, and this back-
ground assumption could have tainted his data. When, soon after these events,
William accepted that Joule had refuted the conservation of heat and endorsed
Clausius’s modification of the Carnot theory, William published a review of the
data behind James’s calculation and concluded that it had not been compromised
by any such presumption by Regnault.22
I see no reason to doubt William’s judgment here, all the more so because he
was certainly happy to admit error. For at the very same time, William reviewed
his earlier estimates of the efficiency of Carnot engines, concluding that these
were seriously in error, and publishing revised estimates (effectively those in use
today).23 This meant that the table cited by James was vitiated by the presumption
of heat-​conservation made in its compilation, but it turned out that the errors got
smaller and smaller as the range of temperatures over which the engine operated
shrank. So the one figure used by James, the 4.97, was quite acceptable. Once
again, there is no dependence on the false components of Carnot’s account.

4.10 Concluding Review: What Can We Take Home?

At the risk of some repetition, let us summarise my conclusions. The no-​miracle

argument presumes it highly unlikely that a substantially false theory will make
a precise, non-​trivial, and correct observational prediction, if that prediction
makes important use of the false components of the theory at issue. At first sight,
the Thomson episode seems to provide one of these very rare events. It thus
challenges the no-​miracle argument by making it seem these predictions may
not be so rare after all.
But closer scrutiny shows the damage is minor, and indeed gives us a glimpse
of how a false theory will sometimes manage a surprising prediction that does
not seem mysterious. The brothers’ argument definitely uses a false theory, but
what makes that theory false is its presumption that heat is conserved. This
matters to the prediction for two quite different reasons. Firstly, because it gets
involved in calculating the magnitude of the effect discovered. If we overlook
these calculations for the moment, we see a second, quite different role for the
Carnot falsehood. It provides the context for the core premise of the argument,
the insistence that heat needs to descend for mechanical work to be produced.

22 See W. Thomson, Math. Phys. Papers, v.1, pp.193–​5. Chang, Inventing temperature, pp.76–​

77, 86–​7, 99–​100 notes that Regnault was extremely cautious about allowing theory to taint his
measurements and (pp.174–​5) that William Thomson found this caution a bit excessive. But Chang
notes too (77, n.48) that some of Regnault’s calorimetric measurements did require heat to be con-
served. Thomson had, of course, worked with Regnault’s laboratory, so his assessment here would be
especially reliable.
23 W. Thomson, Math. Phys. Papers, v.1, pp.189–​91, 197–​8.
68 Historical Cases for the Debate

Without the context supplied by Carnot’s false interpretation of heat-​engines, the

premise of the Thomson argument would be ill-​supported. But “ill-​supported” is
not the same as “false.” So heat-​conservation was needed to make the prediction
plausible, but it was not needed simply to make the prediction.
This last remark is perhaps in slight error though, for I made it under the
tentative presumption that we could ignore the associated calculations. Heat-​
conservation was used in the calculations twice, in both cases to declare that
the quantity of heat involved in one process was the same as that used in a dif-
ferent one. The Clausius theory insists these declarations are false, but also allows
that in the particular processes at issue, the difference between the two quanti-
ties is tiny. So the errors in the quantitative estimation are of no moment. The
Carnot theory has a big flaw in it, but that does not mean that every calculation
that invokes that flaw will make big errors. A big error can sometimes have tiny
consequences.
For the purpose of this discussion, it is granted that scientists sometimes come
up with theories that are deemed true and make successful predictions. Given
this, it is reasonable to acknowledge that some of the false theories devised by
scientists are approximately true. And it is barely miraculous for an approxi-
mately true theory to make—​sometimes—​much the same successful prediction
as a true theory.
This is not what happens with the Thomson prediction, though, for the Carnot
theory is not approximately true. But there are certain restricted situations in
which the two theories, Carnot and Clausius, do approximate each other closely.
Here, they effectively overlap and make extremely similar predictions about
observable phenomena. In these circumstances, a prediction made by the false
theory is no more surprising than the same prediction being made by the true
theory. James’s prediction is an example of this.
It is not a case of a false theory making the very same predictions as a true one.
The predictions differ, but get very close together in special circumstances, and
indeed (in our example) they get too close to be separated observationally. If a
false theory were to make exactly the same significant predictions as some genu-
inely different true theory, that would be far closer to a miracle.

References
Cardwell, Donald S. L. From Watt to Clausius: The rise of thermodynamics in the early
industrial age. (London: Heinemann, 1971).
Carnot, Sadi. Réflexions sur la puissance motrice du feu. [1824]. Critical edition, ed.
Robert Fox. (Paris: Vrin, 1978). My pages reference are however, to Fox’s English trans-
lation, Reflexions on the motive power of fire, (Manchester: Manchester University
Press, 1986).
 No Miracle After All 69

Chang, Hasok. Inventing temperature: measurement and scientific progress. (Oxford:

University Press, 2004)
Clapeyron, Emile. “Memoir on the motive power of heat,” [1834]. Trans. Eric Mendoza
on pp.71–​105 of: Mendoza, ed. Reflections.
Clausius, Rudolf J. E. The mechanical theory of heat, with its applications to the steam-​
engine and to the physical properties of bodies. Trans. J. Hirst. (From several German
texts). (London: Van Voorst, 1867).
Fermi, Enrico. Thermodynamics. (New York: Dover, 1956). Originally 1936.
Fox, Robert. “Introduction” [on pp.1-​57] and annotations etc. to Carnot, Reflexions.
Hutchison, Keith. ‘Mayer’s hypothesis: a study of the early years of thermodynamics.’
Centaurus. 20 (1976): 279–​304.
Hutchison, Keith. “Miracle or mystery? Hypotheses and predictions in Rankine’s
thermodynamics.” On pp. 91–​ 120 of: Clarke, Steve & Timothy Lyons, (ed).
Recent themes in the philosophy of science. Scientific realism and commonsense.
(Dordrecht: Kluwer, 2002).
Mendoza, Eric, ed. etc. Reflections on the motive power of fire. English trans. of Carnot,
Réflexions and papers by Clapeyron and Clausius. (New York: Dover, 1960).
Pippard, Alfred B. Elements of classical thermodynamics for advanced students of physics.
(Cambridge: Cambridge University Press, 1966).
Rankine, William J. M. Miscellaneous scientific papers. With a memoir of the author by P.
G. Tait. Ed. W. J. Millar. (London, 1881).
Rankine, William J. M. “On an equation between the temperature and the maximum
elasticity of steam and other vapours.” Edinb. New Phil. J. 47 (1949): 29–​42. Reprinted
on pp.1–​12 of Rankine, Misc. Sci. Pap.
Smith, Crosbie & M. Norton Wise, Energy and empire: a biographical study of Lord Kelvin.
(Cambridge: Cambridge University Press, 1989.
Thomson, James. “Theoretical considerations on the effects of pressure in lowering the
freezing point of water.” Read 2 Jan 1849. Trans. R. Soc. Edinb. 16 (1849): 575–​80.
Reprinted on pp.156–​64 of William Thomson, Math. Phys. Papers, v.1. Page references
are to this reprint.
Thomson, William. “The effect of pressure in lowering the freezing point of water, ex-
perimentally demonstrated.” Read 21 Jan 1850. Proc. R. Soc. Edinb. 2 (1851): 267–​71.
Reprinted on pp.165–​9 of William Thomson, Math. Phys. Papers, v.1.
Thomson, William. “Notice of Stirling’s air engine.” Read 21 Apr. 1847. Proc. Phil. Soc.
Glasg. 2 (1847): 169–​70. Reprinted on pp.38–​39 of William Thomson, Math. Phys.
Papers, v.5. Appears to be a summary, written by someone else.
Thomson, William. Mathematical and physical papers. Vol. 1. (Cambridge: University
Press, 1882). Vol. 5, ed. J. Larmor. (Cambridge: University Press, 1911).
 5
From the Evidence of History to
the History of Evidence
Descartes, Newton, and Beyond
Stathis Psillos

Department of History and Philosophy of Science

University of Athens
[email protected]

5.1 Introduction

The pessimistic induction (PI) takes it that the evidence of the history of sci-
ence is an indictment for scientific realism: if past theories have been invariably
false, then it will be very unlikely that current theories will turn out to be true,
unless a sort of epistemic privilege is granted to current theories. But, the argu-
ment might go, what could possibly be the ground for this privilege? Assuming a
substantial difference between current theories and past (and abandoned) ones
would require that the methods used by current scientists are substantially dif-
ferent than those of the past scientists. But methodological shifts, the argument
would conclude, are rare, if they occur at all.
Yet, standards of evidence as well as conceptions of explanation have changed
as science has grown. If we look at these changes, we may draw different
conclusions regarding the bearing of the history of science on scientific realism.
Theories are not (nor have they been) invariably worthless because false. Some
have been outright falsehoods, while others hold in their domains under various
conditions of approximation. More importantly, some are such that important
parts of them have been retained in subsequent theories. Some, but not others,
have been licensed by more rigorous standards of success. Some, but not others,
have initiated research programs, which result in successive approximations to
truth. Hence, the history of science carries a mixed message, which needs to be
read with care.

Stathis Psillos, From the Evidence of History to the History of Evidence In: Contemporary Scientific Realism.
Edited by: Timothy D. Lyons and Peter Vickers, Oxford University Press. © Oxford University Press 2021.
DOI: 10.1093/​oso/​9780190946814.003.0005
 Descartes, Newton, and Beyond 71

In this chapter, I will look into two episodes in the history of science: the tran-
sition from the Cartesian natural philosophy to the Newtonian one, and then to
the Einsteinian science. These are complex episodes, and I can only offer sketches
here. But the point I want to drive home should still be visible: though the shift
from Descartes’s theory to Newtonian mechanics amounted to a wholesale re-
jection of Descartes’s theory (including his account of vortices as well as his
methodology), in the second shift, a great deal was retained; Newton’s theory
of universal gravitation gave rise to a research program which informed and
constrained Einstein’s theory. Newton’s theory was a lot more supported by the
evidence than Descartes’s and this made it imperative for its successor theory
to accommodate within it as much as possible of Newton’s theory: evidence for
Newton’s theory became evidence for Einstein’s. Besides, a point that these two
successive cases of theory-​change will highlight is that the very judgment that
some theories are more supported by the evidence than others, and the concom-
itant judgement that some standards of evidence and explanation are better than
others, need not be made in a post hoc way by current philosophers of science.
Rather, they are made by the protagonists of these episodes themselves.
This double case study will motivate a rebranding of the divide et impera
strategy against PI that I introduced in my (1996) and developed in my (1999).
This rebranding urges shifting attention from the (crude) evidence of the his-
tory of science to the (refined) history of evidence for scientific theories. What’s
the connection between the two? As I will show in some detail before the
conclusions, to paraphrase Clausewitz, “moving from the evidence of history to
the history of evidence” is the continuation of divide et impera by other means.
The reader might wonder at the outset: are the lessons drawn from this double case
generalizable? Similar studies in optics and the theories of heat that I have performed
in the past (see my 1999, c­ hapter 6) suggest they are. But all these are theories of
physics and the reader might still wonder whether these lessons can be extended to
other areas of science. I will leave it open whether they can or cannot. The key point is
that it is sufficient to neutralize the force of the PI to show that some typical historical
cases, which have been used to feed epistemic pessimism about current science, do
not license this pessimism. On the contrary, they suggest that a more fine-​tuned ac-
count of theory-​change in science gives us reasons for epistemic optimism.
Here is the road-​map. In section 5.2, I examine in some detail Descartes’s
account of method and explanation. I focus my attention on the role of
hypotheses in Cartesian science and the circumstances under which they
might be taken to be true. In section 5.3, I discuss Newton’s more stringent
account of scientific explanation and his critique of Descartes’s vortex theory
of planetary motions. I focus my attention on the fact that Newton’s initiated
a research program which aimed at the discovery of physical causes of the
72 Historical Cases for the Debate

systematic discrepancies between the predictions of the theory and the ac-
tual orbits of the planets; a program which put Newton’s theory to a series of
rigorous tests. In section 5.4, I discuss the retentionist pattern that Newton’s
theory of gravity set in motion, as this was exemplified in Einstein’s general
theory of relativity (GTR) and in particular in the explanation of the anoma-
lous perihelion of Mercury. In section 5.5, I compare the strategy suggested in
this paper with the earlier divide et impera strategy and show their substantial
similarity. Finally, I draw some general conclusions concerning the need to
shift our attention from the evidence of the history of science to the history of
evidence for scientific theories.

5.2 Descartes on Mechanical Hypotheses

Descartes was the first to put forward a comprehensive theory of the world
based on entities invisible to the naked eye. All worldly phenomena, all mac-
roscopic objects and their behavior, are accounted for by the motion(s),
notably collisions, of invisible corpuscles, subject to general and universal
laws of nature. Within this framework, to offer a scientific explanation of a
worldly phenomenon X was to provide a particular configuration Y of matter
in motion, subject to laws, such that Y could cause X. Elsewhere, I have
explained the status and role of laws of nature in Cartesian natural philos-
ophy (cf. 2018b; forthcoming).1 The key relevant idea is that in Cartesian
physics, the possible empirical models of the world are restricted from above
by a priori principles, which capture the fundamental laws of motion, and
from below by experience. Between these two levels there are various me-
chanical hypotheses, which refer to configurations of matter in motion. As
Descartes explains in (III, 46) of the Principia, since it is a priori possible that
there are countless configurations of matter in motion that can underlie the
various natural phenomena, “unaided reason” is not able to figure out the
right configuration. Mechanical hypotheses are necessary, but experience
should be appealed to, in order to pick out the one in conformity with the
phenomena:

[W]‌e are now at liberty to assume anything we please [about the mechanical
configuration], provided that everything we shall deduce from it is {entirely} in
conformity with experience. (III, 46; 1982, 106)

1 There has been considerable debate about the status and role of laws of nature in Descartes. Some

notable contributions include Ott (2009), Hattab (2000), and Wilson (1999, c­ hapter 28).
 Descartes, Newton, and Beyond 73

There are two sorts of hypotheses in Cartesian physics and his concomitant
hypothetico-​deductive method: vertical and horizontal. “Vertical” I call the
hypotheses which posit invisible entities and their primary modes (shape and
size, i.e., bulk, and motion). These kinds of entities, being invisible, have to be
hypothesized. But there are “horizontal” hypotheses too. These concern mac-
roscopic (that is visible) entities. Being visible, these entities do not have to be
hypothesized; yet their structural relations have to be hypothesized because dif-
ferent structures yield the same appearances. The prime example is the “system
of the world”—​the solar system. Here, there are competing, but empirically
equivalent, possible structures of the planetary motions.
In Principia (IV, 201; 1982, 283), Descartes states that sensible bodies are com-
posed of insensible particles. So insensible particles (of various bulks and speeds)
are everywhere: they fill (literally) space. A number of problems crop up, at this
juncture. How can there be epistemic access to these invisible entities? How is
knowledge of them possible? How can hypotheses be justified and accepted?
How can one of them be more preferable than another, given that there are more
than one which tally with the phenomena? All these are familiar questions. They
are at the heart of the current scientific realism debate. Let us see how Descartes
answers them.

5.2.1 The Bridge Principle

Descartes’s ingenious attempt to answer them is based on a bridge principle,

that is a principle which bridges the gap between the visible macrocosm and the
invisible microcosm. According to this principle, the properties of the minute
particles should be modeled on the properties of macro-​bodies. Here is how
Descartes put it:

Nor do I think that anyone who is using his reason will be prepared to deny
that it is far better to judge of things which occur in tiny bodies (which escape
our senses solely because of their smallness) on the model of those which our
senses perceive occurring in large bodies, than it is to devise I know not what
new things, having no similarity with those things which are observed, in order
to give an account of those things [in tiny bodies]. {E.g., prime matter, substan-
tial forms, and all that great array of qualities which many are accustomed to
assuming; each of which is more difficult to know than the things men claim to
explain by their means}. (IV, 201; 1982, 284)

In this passage Descartes advances a continuity thesis: it is simpler and con-

sonant with what our senses reveal to us to assume that the properties of
74 Historical Cases for the Debate

micro-​objects are the same as the properties of macro-​objects. This continuity

thesis licenses certain kinds of explanations: those that endow matter in general,
and hence the unobservable parts of matter, with the properties of the perceived
bits of matter. It therefore licenses as explanatory certain kinds of unobservable
configurations of matter; viz., those that resemble perceived configurations of
matter. The key point here is that the bridge principle makes “visibility” irrele-
vant ontically, though epistemically important. It allows transferring knowledge
gained by experience (and the fundamental laws) to the invisible particles. There
is not much justification for this principle offered by Descartes. A principle such
as this is assumed as required for gaining knowledge of the invisible based on
experience.
Let’s see how this bridge principle works in practice in the (in)famous vortex
hypothesis, according to which the planets are carried by vortices around the
sun. A vortex is a specific configuration of matter in motion—​subtle matter re-
volving around a center. The underlying mechanism of the planetary system
then is a system of vortices:

[T]‌he matter of the heaven, in which the Planets are situated, unceasingly
revolves, like a vortex having the Sun as its center, and . . . those of its parts which
are close to the Sun move more quickly than those further away; and . . . all the
Planets (among which we {shall from now on} include the Earth) always remain
suspended among the same parts of this heavenly matter. (III, 30; 1982, 196)

The very idea of this kind of configuration is suggested by experience, and

by means of the bridge principle it is transferred to the subtle matter of the
heavens. Hence, invisibility doesn’t matter. The bridge principle transfers the
explanatory mechanism from visible bodies to invisible ones. More specif-
ically, the specific continuity thesis used is the motion of “some straws {or
other light bodies}. . . floating in the eddy of a river where the water doubles
back on itself and forms a vortex as it swirls.” In this kind of motion, we can
see that the vortex carries the straws “along and makes them move in circles
with it.” We also see that

some of these straws rotate about their own centers, and that those which
are closer to the center of the vortex which contains them complete their
circle more rapidly than those which are further away from it. (III, 30;
1982, 196)

Given the continuity thesis, we may transfer this mechanical model to the motion
of the planets and “imagine that all the same things happen to the Planets; and
this is all we need to explain all their remaining phenomena” (III, 30; 1982, 196).
 Descartes, Newton, and Beyond 75

5.2.2 The System of the World

How about horizontal hypotheses? Here the chief rivalry was between the three
systems of the world: the Ptolemaic, the Tychonic, and the Copernican. The
bridge principle cannot be of help. But experience can help refute one of them,
viz., the Ptolemaic. As Descartes notes it is ruled out because it does not con-
form with observations, notably the phases of Venus (cf. III, 16; 1982, 89–​90).
But when it comes to Copernicus vs. Tycho, Descartes notes that they “do not
differ if considered only as hypotheses”—​they “explain all phenomena equally
well,” though Copernicus’s hypothesis is “simpler and clearer.” Hence, qua poten-
tial (hypothetical) explanations these two systems do not differ.
Descartes advocated a version of the Copernican theory because it’s “the sim-
plest and most useful of all; both for understanding the phenomena and for in-
quiring into their natural causes” (III, 19; 1982, 91). In his own version, he denied
the motion of the earth “more carefully than Copernicus and more truthfully
than Tycho” (III, 19; 1982, 91). To understand what goes on, we need to make a
short digression into Descartes’s theory of motion.
Notoriously, Descartes defines motion twice over. On the one hand, there is
the ordinary account of motion as “the action by which some body travels from one
place to another.” On the other hand, there is the “true” account of motion: “the
transference of one part of matter or of one body, from the vicinity of those bodies
immediately contiguous to it and considered as at rest, into the vicinity off some]
others” (Principia II, 24, 25; 1982, 50–​51). According to this account, motion is
change of position relative to contiguous bodies that are considered at rest. But
the earth is carried over by a vortex on which it is, so to speak, pinned; hence, it
does not move (it is at rest) relative to its contiguous and at-​rest bodies. Nor do
the other planets. The observed changes in their positions are wholly due to the
motion of the vortex which contains them.2
It is noteworthy that Tycho’s account is in conflict with Descartes’s theory of
motion: the Earth appears immobile, but it is not at rest. The reason for this is
based on a) the relativity of motion; and b) on Descartes’s claim that a body is
in motion only if it moves as a whole. As he observes in Principia (III, 38; 1982,
101–​102), the relativity of motion implies that it could be either that the earth
rotates and the heavens with the stars are rest, or the other way around. But, on
the Tychonic hypothesis, condition b) implies that it is the earth which is actu-
ally in motion, since the earth moves as a whole body relative to the heavens (its
separation from the contiguous heavens taking place along the whole of the sur-
face of the earth), whereas the heavens with the stars in them are at rest, since the
separation takes place over a small fraction of their surface, viz., this part which

2 For an excellent account of Descartes’s theory of vortices, cf. Aiton (1957).

76 Historical Cases for the Debate

is contiguous with the earth. In fact, there is a deeper reason why Descartes
dismissed Tycho’s model of the System of the World. Given that the planets
(apart from the earth) revolve around the sun, and the sun revolves around the
earth, it is hard to see how this combination of motions could be modeled by
vortices. Hence, Descartes concluded that Tycho “asserted only verbally that the
Earth was at rest and in fact granted it more motion than had his predecessor”
(III, 18; 1982, 91).3
Despite his heliocentrism, Descartes noted that he accepted the Copernican
account not as true, “but only as a hypothesis,” which may be false. Hypotheses
need not be true to be useful. But can they be true? Or better, under what
circumstances, if any, should they be taken to be true?

5.2.3 Was Descartes an Instrumentalist?

Descartes seems to endorse an instrumentalist account of hypotheses. He says

quite explicitly in the third part of the Principia (article 44) that though his
hypotheses may be false, it is enough that “all the things which are deduced from
them are entirely in conformity with the phenomena” (III, 44; 1982, 105). He
further says, (IV, 204; 1982, 286), that his causal story enables us to understand
a possible way the world might have been (in its invisible structure), but that we
“shouldn’t conclude that” it was actually “made in that way.” To illustrate the dif-
ference, he envisages a watchmaker who could have produced “two equally reli-
able clocks that looked completely alike from the outside but had utterly different
mechanisms inside.” God, being the “supreme maker of everything,” could have
produced all we see in “in many different ways.” In light of this, Descartes says, it
is enough that his hypothesis “corresponds accurately with all the phenomena of
nature.” For after all, this is all that is necessary for all the practical applications of
science, which are directed “towards items that are sense-​perceptible.”
Yet, to read Descartes as an instrumentalist would be far too quick. He cer-
tainly thinks that there is an alternative theory concerning the creation of the
Solar System (and of everything else), viz., the story told in the Bible (cf. IV, 1;
1982, 181). However, he is clear that his own corpuscularian hypothesis offers
a better explanation than a) no explanation at all and b) the biblical story. The
reason for this is that his own hypothesis identifies the natural causes of things
and is based on simple and “easy to know” principles. Descartes occasionally
uses the expression “as if ”: “my hypothesis will be as useful to life as if it were
true” or “as if they were [the] causes [of these effects].” But the context suggests
that this use is not the same as the instrumentalist’s. The key difference is that

3 For a more detailed account of Descartes’s theory of motion, cf. Garber (1992).
 Descartes, Newton, and Beyond 77

for Descartes “as if true” seems to mean something like “true for all practical
purposes” (or, morally certain, as Descartes would put it). In the French transla-
tion of the Principia he added right after the claim that his hypothesis “will be as
useful to life as if it were true”: “because we will be able to use it in the same way
to dispose natural causes to produce the effects which we desire” (III, 44; 1982,
105). And again: “And I do not think it possible to devise any simpler, more intel-
ligible or more probable principles than these” (III, 47; 1982, 107).
Simplicity, intelligibility, and probability are marks of truth. Descartes adds
novel predictions, too; it is further support of the truth of a hypothesis, that is of
its being right about the causes of the phenomena, when it explains “not only the
effects that we were initially trying to explain but all these other phenomena that
we hadn’t even been thinking about” (III, 42; 1982, 104). On various occasions,
Descartes stresses that unification is a mark of truth too. In (III, 43; 1982, 104), he
notes emphatically that it “can scarcely be possible that the causes from which all
phenomena are clearly deduced are false.” He grounds this on the further claim
that if the principles are evident and the deductions of observations from them
“are in exact agreement with all natural phenomena,” it would “be an injustice
to God to believe that the causes of the effects which are in nature and which we
have thus discovered are false.”
The best explanation is known with moral certainty, i.e., certainty for all prac-
tical purposes. What is known with moral certainty could be false in the sense
that it would be in the absolute power of God to make things otherwise than
our best theory says they are. Descartes uses the metaphor of decoding a secret
message to make an analogy with the explanatory power of theories. The more
complex the message is the more unlikely it is that the code that was devised to
successfully decode it is a matter of coincidence. Similarly, the more a theory
reveals about the “fabric of the entire World,” the more unlikely it is that it is false
(cf. IV 205; 1982, 287).

5.2.4 The Fate of Vortices

The picture that emerges from this account of Descartes’s methodology is

nuanced and intricate. Empirical adequacy, so to speak, is a necessary condi-
tion for the truth of a hypothesis. But it being insufficient, what should possibly
be added to render a hypothesis true? Descartes seems clear that the theoretical
virtues of simplicity, intelligibility, unification, and novel predictive success are
enough to render a hypothesis the best explanation of the evidence, and hence
true, though known with moral certainty. Descartes clearly thought that his own
vortex theory met these conditions. He was wrong about this. Descartes did try
to show how planets move in non-​circular orbits—​by assuming that the solar
78 Historical Cases for the Debate

vortex is somewhat distorted because the neighboring vortices exert unequal

pressures on it and then arguing that subtle matter at the wider part of the vortex
moves more slowly than at the narrower parts thus making the planet recede
further from the sun (cf. Principia II, 141; 1982, 169). But he otherwise paid no
attention to Kepler’s laws; nor did he try to accommodate them within his theory.
Nor did he account for the parallax of the equinoxes. In sum, his theory was em-
pirically inadequate and shown to be false by Newton, whose theory superseded
Descartes’s.
As we are about to see, Newton’s theory offered not only a better explanation
of gravitational phenomena than Descartes’s but also a better account of explana-
tion as well as a better research strategy for devising such explanations. In doing
all this, Newton changed the standards of evidence.

5.3 Newton’s Deduction from the Phenomena

As is well known, Newton subjected Cartesian physics to trenchant criticism.

In the unpublished De Gravitatione, he ventured to dispose of Descartes’s
“fictions” (2004, 14). One of his key points was that Descartes’s theory of mo-
tion leads to absurdities since, if we take it seriously, it follows that a moving
body has neither determinate velocity nor trajectory of motion; hence, it does
not move!4 Newton was particularly opposed to Descartes’s theory of vortices.
He produced a number of arguments against it. Toward the end of Book II of the
Principia, Newton demonstrated that “the hypothesis of vortices can in no way
be reconciled with astronomical phenomena and serves less to clarify the celes-
tial motions than to obscure them” (1729 [2016], 790). In particular, Newton
showed that the vortex hypothesis could recover neither Kepler’s area law nor the
harmonic law. In one key argument, Newton constructed a simple model of the
solar system in which planets are carried by vortices and showed that planets will
move more swiftly in the aphelion and more slowly in the perihelion, contrary to
Kepler’s area law.5 In the second edition of the Principia in 1713, Newton started
his famous General Scholium by stressing that “hypothesis of vortices is beset
with many difficulties” (1729 [2016], 939) and added that the second and third
laws of Kepler cannot be mutually satisfied by the layers of vortex and that the
motions of comets are incompatible with the existence of vortices.

4 For a thorough account of Newton’s criticism of Descartes’s theory of motion and of body, see

Katherine Brading (2012). Brading argues that both Descartes and Newton shared a law-​constitutive
account of bodies but Newton developed it more systematically. See also Jalobeanu (2013).
5 For a detailed account of Newton’s criticism of Cartesian vortices, see Snow (1924) and Aiton

(1958a).
 Descartes, Newton, and Beyond 79

In fact, after Newton’s accommodation of Kepler’s laws within his theory,

Cartesian and non-​Cartesian advocates of the vortex theory aimed to show how
their favorite versions of the vortex theory could accommodate Kepler’s laws.6
But all attempts resulted in failures (cf. Aiton 1996).

5.3.1 What was the Problem with Vortices?

But the problem with vortices appeared to be epistemological too: they were hy-
pothetical entities. In the preface to the second edition of the Principia, Roger
Cotes equated vortices with “occult causes,” being “utterly fictitious and com-
pletely imperceptible to the senses” (1729 [2016], 392).
But is the problem really that vortices, if they exist, are imperceptible? It seems
it is not. To see why it is not let us make a short digression to Huygens and Leibniz.

5.3.1.1 Huygens vs. Leibniz

Christiaan Huygens came to doubt the vortex theory “which formerly appeared
very likely” to him (1690 [1997], 32). He didn’t thereby abandon the key tenets of
mechanical philosophy. For Huygens too the causal explanation of a natural phe-
nomenon had to be mechanical. He said, referring to Descartes:

Mr Descartes has recognized, better than those that preceded him, that nothing
will be ever understood in physics except what can be made to depend on prin-
ciples that do not exceed the reach of our spirit, such as those that depend on
bodies, deprived of qualities, and their motions. (1–​2)

Huygens posited a fluid matter that consists of very small parts in rapid motion
in all directions, and which fills the spherical space that includes all heavenly
bodies. Since there is no empty space, this fluid matter is more easily moved in
circular motion around the center, but not all parts of it move in the same di-
rection. As Huygens put it “it is not difficult now to explain how gravity is pro-
duced by this motion.” When the parts of the fluid matter encounter some bigger
bodies, like the planets: “these bodies [the planets] will necessarily be pushed
towards the center of motion, since they do not follow the rapid motion of the
aforementioned matter” (16). And he added:

This then is in all likelihood what the gravity of bodies truly consists of: we can
say that this is the endeavor that causes the fluid matter, which turns circularly

6 As Aiton (1958b) argues, it was only after the publication of Newton’s Principia that Cartesians

started to try to account for Kepler’s laws.

80 Historical Cases for the Debate

around the center of the Earth in all directions, to move away from the center
and to push in its place bodies that do not follow this motion. (16)

In fact, Huygens devised an experiment with bits of beeswax to show how this
movement toward the center can take place. At the same time, however, he had
no difficulty in granting that Newton’s law of gravity was essentially correct when
it comes to accounting for the planetary system, and especially Kepler’s laws. As
he put it:

I have nothing against Vis Centripeta, as Mr. Newton calls it, which causes the
planets to weigh (or gravitate) toward the Sun, and the Moon toward the Earth,
but here I remain in agreement without difficulty because not only do we know
through experience that there is such a manner of attraction or impulse in na-
ture, but also that it is explained by the laws of motion, as we have seen in what
I wrote above on gravity. (31)

Explaining the fact that gravity depends on the masses and diminishes with dis-
tance “in inverse proportion to the squares of the distances from the center” was,
for Huygens, a clear achievement of Newton’s theory, despite the fact that the me-
chanical cause of gravity remained unidentified. His major complaint was that he
did not agree with Newton that the law of gravity applies to the smallest parts of
bodies, because he thought that “the cause of such an attraction is not explicable
either by any principle of mechanics or by the laws of motion” (31).7
Leibniz in his Tentamen de motuum coelestium causis (1689), adopted a ver-
sion of the vortex theory despite his overall disagreement with Cartesian physics
and metaphysics. On this theory, a planet is carried by a “harmonic vortex,” the
layers of which moved with speeds inversely proportional to the distance from
the center of circulation, thereby satisfying Kepler’s area law. But the planet is
also subject to a radial motion (motus paracentricus), which is due to two forces
acting on the planet: the gravity exerted by the sun on the planet and the cen-
trifugal force arising from the circulation of the planet with the vortex. Given
all this, Leibniz was able to show that an elliptical orbit requires that the gravity
(which was conceived by Leibniz along the lines of the magnetic action) is in-
versely proportional to the distance, and then to deduce from this Kepler’s har-
monic law. In effect, Leibniz had posited two independent but superimposed
vortices for each planet, the vortex that causes gravity and the harmonic vortex
(cf. Aiton 1996, 12).

7 For further discussion of Huygen’s theory of gravity and his reaction to Newton’s, see Martins

(1994).
 Descartes, Newton, and Beyond 81

Huygens, in correspondence, (Huygens to Leibniz 11 July 1692) challenged

Leibniz on this, arguing that the harmonic vortex was not necessary since
Newton had shown that gravity was enough for Kepler’s laws to hold. Leibniz
replied that he could not abandon this “deferent matter” (the harmonic vortex)
because it explains why “all the planets [as well as the satellites of Jupiter and of
Saturn] move somewhat in the same direction”; a thing that Newton’s theory just
takes for granted (Leibniz 1989, 415). But, as was pointed out by Newton’s dis-
ciple David Gregory in his Astronomiae physicae et geometricae elementa (1702),
Kepler’s third law was inconsistent with the idea that planets are carried by har-
monic vortices. Besides, Leibniz’s theory cannot account for the motion of the
comets: comets, on Leibniz’s theory, should also be drawn by harmonic vortices,
since their motions satisfy Kepler’s area law, but their orbits are inclined at large
angles to the plane of circulation of the planetary vortices. How can it be that the
comets’ vortices and the planetary vortices do not interact with each other and
each of them remains intact with no modification of its state of motion?8 Newton
himself took issue with Leibniz’s account arguing that Leibniz had the wrong
measure of centrifugal force, which was for Newton equal and opposite to the
force of attraction and that Leibniz’s mathematical reasoning was unsound (cf.
Aiton 1962).9
Leibniz wrote to Newton on March 7, 1692/​3. In this letter, he praised Newton
for having made “the astonishing discovery that Kepler’s ellipses result simply
from the conception of attraction or gravitation and passage in a planet” (cf.
Newton 2004, 106). He added, however, that he thought that Newton’s account
is incomplete, since it leaves out the causes of attraction, which Leibniz took to
be “the motion of a fluid medium.” Leibniz, as we have seen, had developed his
own version of the vortex theory. But the real purpose of the letter was to solicit
Newton’s opinion regarding Huygens’s vortex theory.
Newton’s reply (October 16, 1693) was very illuminating. He objected to
vortex theory on the grounds that the vortices would disturb the motions of
the planets and the comets. He made the further point that the laws of gravity
he discovered were enough to account for the motion of the heavenly bodies.
Moreover, as he put it, “since nature is very simple, I have myself concluded that
all other causes are to be rejected and that the heavens are to be stripped as far as
may be of all matter, lest the motions of planets and comets be hindered or ren-
dered irregular” (2004, 109). The point here is double: on the one hand, Newton’s
explanation of planetary motions is simpler than the various accounts based on
vortices; on the other hand, the explanations based on vortices are inadequate
since they were at odds with the regular motions of the heavenly bodies. The

8 For an insightful discussion, see Bussotti (2015, section 4.2).

9 For a detailed account of Leibniz’s theory, see Aiton (1958a; 1962; 1996) and Bussotti (2015).
82 Historical Cases for the Debate

inferential strategy is guided by a theoretical virtue (simplicity) plus the available

evidence against the rival theories.
But Newton went on to add something seemingly puzzling: “But if, mean-
while, someone explains gravity along with all its laws by the action of some
subtle matter, and shows that the motion of planets and comets will not be dis-
turbed by this matter, I shall be far from objecting” (2004, 109). This suggests that
Newton was not in principle against a “deeper” explanation of gravity by reference
to the action of some subtle matter. Newton’s problem with vortex theory was not
that it posited the action of some subtle matter. Hence his problem was not that
the vortex theory referred to insensible particles and their motion. Rather his
problem was that the very vortex theory was inadequate as a potential explana-
tion of the heavenly orbits. Invisibility does not matter; explanation does!

5.3.2 How to Do Natural Philosophy

This point, viz., that the problem with vortex theory was that it does not offer
a good explanation of gravity, is echoed by Roger Cotes: “For even if these
philosophers [the Cartesians] could account for the phenomena with the greatest
exactness on the basis of their [vortices] hypotheses, still they cannot be said to
have given us a true philosophy and to have found the true causes of the celestial
motions until they have demonstrated either that these causes really do exist or
at least that others do not exist” (Newton 1729 [2016], 393). But, as Newton has
shown, not only does the vortex hypothesis fail to account for the gravitational
phenomena; there is a rival theory—​Newton’s own—​that accounts for them.
In fact, Newton stressed repeatedly that gravity is a non-​mechanical force;
hence it cannot be modeled by action within a mechanical medium. In the
General Scholium he noted that gravity does not operate “in proportion to the
quantity of the surfaces of the particles on which it acts (as mechanical causes are
wont to do) but in proportion to the quantity of solid matter” (1729 [2016], 943).
It transpires then that Newton’s theory is superior to Descartes’s as an ex-
planation. As Aiton (1958a, 1958b) has shown, most Cartesians succumbed to
Newton’s theory, and by the 1740s, the vortex theory had been fully abandoned.
But even before this, most Cartesians who advocated the vortex theory tried to
make it consistent with Newton’s.10 But there is more to it. Newton goes beyond

10 One of the last attempts to defend Descartes’s theory over Newton was by the Swiss mathema-

tician Johann Bernoulli who won the 1730 Académie Royale des Sciences prize for his essay on the
causes of elliptical orbits in a Cartesian vortex. He disagreed with two “bold suppositions” made by
Newton, which “shock the spirits accustomed to receive in Physics only incontestable and obvious
principles.” The first of these suppositions, according to Bernoulli, was “to attribute to the body a
virtue or attractive faculty” (1730, 6), while the second is the perfect void. But Bernoulli only gestured
 Descartes, Newton, and Beyond 83

the hypothetico-​deductive account of explanation that was favored by Descartes

and places more stringent demands on explanation. Let us pursue this point
further.
What counts as a good explanation, according to Newton? As Cotes notes, it
is an explanation that shows that the posited causes exist or “at least that others
do not exist” (Newton 1729 [2016], 393). Hence for an explanation to be good it
should be shown that other potential explanations are excluded; in effect, that it
is the unique explanation of the phenomena. An explanation need not be full or
complete to be good. Newton considered it enough that he identified the law that
gravity obeys. The fact, if it is a fact, that there is a hitherto unknown “cause” of
gravity does not detract from the fact that he identified gravity as “the force by
which the moon is kept in its orbit.” His agnosticism as to the “physical seat” of
gravity, as Andrew Janiak (2008, 56–​57) has nicely put it, neither renders the law
a mere calculating device nor prevents Newton from stressing that gravity “really
exists.”
Newton’s “hypotheses non fingo” has become emblematic. Here is the full
quotation:

I have not as yet been able to deduce from phenomena the reason for these
properties of gravity, and I do not “feign” hypotheses. For whatever is not
deduced from the phenomena must be called a hypothesis; and hypotheses,
whether metaphysical or physical, or based on occult qualities, or mechanical,
have no place in experimental philosophy. In this experimental philosophy,
propositions are deduced from the phenomena and are made general by in-
duction. The impenetrability, mobility, and impetus of bodies, and the laws of
motion and the law of gravity have been found by this method. And it is enough
that gravity really exists and acts according to the laws that we have set forth
and is sufficient to explain all the motions of the heavenly bodies and of our sea.
(1729 [2016], 943)

In this, hypotheses of any sort are contrasted to propositions which “are deduced
from the phenomena and are made general by induction.” We shall see in the se-
quel how this claim may be understood. For the time being, the relevant point is
that Newton criticizes a certain account of explanation. But on what grounds? This
is explained by Cotes in his preface to the second edition of the Principia. Cotes
aims to highlight Newton’s methodological novelty, and he contrasts it with both
the Aristotelian way and the Cartesian way. He does not doubt that Cartesian me-
chanical philosophy is an improvement over the Aristotelian science. Nor does he

as to how Kepler’s laws could be accommodated in his system of vortices. For a detailed discussion,
see Aiton (1996, 17–​18). See also Iltis (1973).
84 Historical Cases for the Debate

doubt that the Cartesians were right in aiming to explain the perceptible in terms
of the imperceptible. Indeed, Cotes praises the Cartesians for trying to show that
the “the variety of forms that is discerned in bodies all arises from certain very
simple and easily comprehensible attributes of the component particles” (Newton
1729 [2016], 385). Where did they go wrong then? In relying on speculative and
unfounded hypotheses about the properties of the invisible corpuscles.
If Cartesians are right on demonstration based on laws, but wrong to rely on
speculative and unfounded hypotheses, what is the alternative? It is to use the
phenomena as premises such that, together with the general laws of motion, fur-
ther laws are derived on their basis. What is thereby derived is not speculative
and unfounded. Indeed, that’s how Cotes describes Newton’s method:

Although they [the Newtonians] too hold that the causes of all things are to be
derived from the simplest possible principles, they assume nothing as a prin-
ciple that has not yet been thoroughly proved from phenomena. They do not
contrive hypotheses, nor do they admit them into natural science otherwise
than as questions whose truth may be discussed. (1729 [2016], 385)

The aim of explanation then changes: it is to use the general laws of motion
and phenomena in order to deduce further special laws, which have been estab-
lished in such a way that they cannot be denied while affirming the general laws
and the phenomena. In this sense, the special laws have been deduced from the
phenomena. They are then at least as certain as the phenomena and general laws
of motion. That’s how the law of gravity has been found.

5.3.3 Newton’s Account of Explanation

We can then say that Newton’s experimental philosophy reverses the order of ex-
planation in science. Schematically put:

Descartes
General Laws & Hypotheses ➔ phenomena

Newton
General Laws & phenomena ➔ special laws

What, of course, Newton called “phenomena” in book 3 of the Principia were

not data or observations but empirical structures that exemplify a certain pat-
tern, captured by an empirical law (Kepler’s area law, Kepler’s harmonic law).
So, in a certain sense, they exhibit robustness and have been built inferentially
 Descartes, Newton, and Beyond 85

from data and observations. Hence, they are amenable to mathematical analysis.
For instance, phenomenon one is that the satellites of Jupiter (the circumjovial
planets), by radii drawn to the center of Jupiter, satisfy Kepler’s area law and
Kepler’s harmonic law. And phenomenon six is that the moon, by a radius drawn
to the center of the earth, satisfies Kepler’s area law.
It is then straightforward for Newton to apply Propositions 1, 2, and 4 (corol-
laries 6 and 7) of Book 1 of the Principia, which are proved for bodies qua math-
ematical points and forces qua mathematical functions, to the phenomena. As is
well known, Newton demonstrates the following two equivalences:
A body A moving in an ellipse (a conic section) around another body B which
is in a focus of the ellipse describes equal areas in equal times (that is, it satisfies
Kepler’s area law) if and only if it is urged by a centripetal force that is directed
toward body B.
A body A moving in an ellipse (a conic section) around another body B which
is in a focus of the ellipse has its periodic time T as the 3/​2 power of its radius r
(that is, it satisfies Kepler’s harmonic law) if and only if the centripetal force that
urges it toward B is inversely proportional to the square of the distance r between
the two bodies.11
What is special about Newton’s account of explanation, as William Harper
and George Smith point out, is that he explains a phenomenon by relying on
equivalences such as those just described. These equivalences “make the phe-
nomenon measure a parameter of the theory which specifies its cause” (Harper
and Smith 1995, 147).12 In Book 3, the mathematically described centripetal
force of the foregoing equivalences is identified with gravity. In the scholium to
Proposition 5, Newton says: “Hitherto we have called ‘centripetal’ that force by
which celestial bodies are kept in their orbits. It is now established that this force
is gravity, and therefore we shall call it gravity from now on” (1729 [2016], 806).
Gravity is rendered a universal force by means of Rule III of Philosophy, which
is a rule of induction.13 This rule licenses the generalization of a quality, which is
found to all bodies on the basis of experiments, to all bodies whatsoever. Or, as
Newton put it:

11 In fact, Newton added a further proof of the inverse square law, one that as he says proves its

existence “with the greatest exactness” (1729 [2016], 802). The proof is based on the empirical fact
that “the aphelia [of the planets] are at rest.” As Newton argued, invoking proposition 45 of Book 1,
the slightest departure from the inverse square of the distance would “necessarily result in a notice-
able motion of the apsides in a single revolution and an immense such motion in many revolutions.”
This “systematic dependence,” as William Harper has put it, between the inverse square ratio and the
immobility of the aphelia is something that makes Newton’s account of gravity robust. Besides, as
Harper (2011, 42ff) has argued, looking for systematic dependencies between theoretical parameters
and the phenomena renders Newton’s method distinct from a simple hypothetico-​deductive method.
12 For an elaboration of the idea of theory-​mediated measurement, see Harper (2011).
13 For a useful discussion of Newton’s account of induction vis-​à-​vis Locke’s and Hume’s, see de

Pierris (2015, c­ hapter 3).

86 Historical Cases for the Debate

Finally, if it is universally established by experiments and astronomical

observations that all bodies on or near the earth gravitate [lit. are heavy] to-
ward the earth, and do so in proportion to the quantity of matter in each body,
and that the moon gravitates [is heavy] toward the earth in proportion to the
quantity of its matter, and that our sea in turn gravitates [is heavy] toward the
moon, and that all planets gravitate [are heavy] toward one another, and that
there is a similar gravity [heaviness] of comets toward the sun, it will have to be
concluded by this third rule that all bodies gravitate toward one another. (1729
[2016], 796)

Unlike Descartes’s bridge principle, Newton’s Rule III, is a more rig-

orous bridge principle: not all observable properties of matter are trans-
ferable to all bits of matter (no matter how small); but only those shown
to be possessed by experiments. It should be noted that this bridge prin-
ciple is both horizontal (it extends to all observable bodies) and vertical
(it extends to all unobservable particles of matter). As such, it is a vehicle
for unification. In justifying rule III, which yields the unification, Newton
appeals to nature’s being “always simple and ever consonant to itself.”
Newton’s theory of gravity is offered as the best explanation of the gravi-
tational phenomena and is licensed as such by the theoretical virtues and
(rigorous) evidence.

5.3.4 Quam Proxime

We know that Newton’s theory of gravity was superseded by Einstein’s GTR.

But even if this hadn’t happened, Newton’s theory would still have been, at
best, an approximation to the phenomena to be accounted for. In what Smith
(2008) has called the Copernican Scholium of the “De Motu” tracts that pre-
ceded the Principia, Newton stated the great complexity of the actual plane-
tary motions:

By reason of the deviation of the Sun from the center of gravity, the centrip-
etal force does not always tend to that immobile center, and hence the planets
neither move exactly in ellipses nor revolve twice in the same orbit. Each time
a planet revolves it traces a fresh orbit, as in the motion of the Moon, and each
orbit depends on the combined motions of all the planets, not to mention
the actions of all these on each other. . . . But to consider simultaneously all
these causes of motion and to define these motions by exact laws admitting of
easy calculation exceeds, if I am not mistaken, the force of any human mind.
(Herival 1965, 301)
 Descartes, Newton, and Beyond 87

Kepler’s laws hold only approximately of the planets and their satellites mainly
because the actual motions are lot more complicated than the theory could cap-
ture. Was this something to give Newton pause? To cut a long story short, it was
certainly not. Not only Kepler’s laws hold quam proxime, but Newton himself
took pains to show that the equivalences he proved in Book I of the Principia
hold quam proxime too. This phrase (which occurs 139 times in the Principia)
literally means “most nearly to the highest degree possible”; it is probably best
translated “very, very nearly” or, as is more customary, “very nearly.”14
After proving Propositions 1 and 2 of Book I (which, as we have seen, assert
the equivalence: the centripetal force by which a body is drawn to an immo-
bile center of force is directed to this center iff that body satisfies Kepler’s area
law), Newton went on, in corollaries 2 and 3 of Proposition 3, to show that the
centripetal force is directed to the center of force quam proxime iff Kepler’s area
law is satisfied quam proxime. More generally, the theoretical equivalences that
Newton proved in Book I hold quam proxime too: p quam proxime iff q qua
proxime. Hence, given the laws of motion, Newton’s law of gravity is true at least
quam proxime (cf. Smith 2002, 156).
Still, Newton also showed the conditions under which the theoretical
equivalences would hold, not simply quam proxime, but exactly of the phe-
nomena. As he characteristically put it in Proposition 13 of Book III: “if the
sun were at rest and the remaining planets did not act upon one another, their
orbits would be elliptical, having the sun in their common focus, and they would
describe areas proportional to the times” (1729 [2016], 817–​818). It follows,
as Smith put it, that “the true motions would be in exact accord with the phe-
nomena were it not for specific complicating factors” (2002, 157).
This interplay between exact, idealized motions and quam proxime ac-
tual motions renders Newton’s account of explanation substantially different
from Descartes’s. As Ducheyne put it, Newton’s target is not merely to explain
Keplerian motions but rather to show that, given the laws of motion, there is a
unique explanation of Keplerian motions (cf. 2012, 104). Given that Keplerian
motions are idealized motions, Newton’s target is to explain the actual deviations
from these motions.
As Smith has noted, Newton’s theory makes possible the following: On the as-
sumption that the theory is exact (that is on the basis of idealizations embodied
in the theory) there are systematic discrepancies between the predictions of the
theory and the observations. These discrepancies constitute “second-​order phe-
nomena” which the theory, in the first instance, should explain by attributing
them, as far as possible, to physical causes. Hence, the theory makes possible

14 Ducheyne (2012, 82) offers a more literal translation as “as most closely as possible” or “utter-

most closely.”
88 Historical Cases for the Debate

further explanations of the discrepancies between predicted values and observed

values of various parameters. The theory is being further elaborated and tested
by finding the causes of discrepancies. This is achieved by “asking in a sequence
of successive approximations, what further forces or density variations are af-
fecting the actual situation?” (2014, 277).
The discovery of physical causes of these systematic discrepancies is a rigorous
test of the physical theory because it shows

a) the circumstances under which the theory holds exactly;

b) the sense in which it can be taken to be an approximation of real worldly
phenomena; and
c) the ways the theory can be improved upon by taking into account further
causal factors in the world.15

The discovery of Neptune by Le Verrier in 1846 is a case in point. The idea

that the theory holds quam proxime of worldly phenomena is tied to the con-
tinuous development and further testing of the theory, which leaves open the
possibility that some phenomena may well be genuine exceptions to the theory;
that is, that some phenomena might not be such that they can be accommodated
within the theory by the method of successive approximations. One reason why
Smith has called this kind of phenomena “second-​order” is the fact that their
being acknowledged as phenomena worthy of accommodation presupposes the
theory that makes them available. That is, it presupposes that the theory offers
the best explanation of the first-​order phenomena. In the case of the planetary
motions, these second-​order phenomena, which call for a physical explanation,
presuppose Newton’s theory of gravity and in particular that it holds exactly.
They are made available qua phenomena as discrepancies between the idealized
predictions of Newton’s theory (which hold on the assumption that it is an exact
theory) and observations. As Smith put it: “they are what you get by subtracting
observations from idealized calculated results.” And he adds: “They are second-​
order because they categorically presuppose the theory of gravity, taken as
holding exactly. They are phenomena because they are systematic and hence
constitute regularities that cannot initially be identified because more dominant
regularities mask them” (2014, 278). The method of successive approximations
turn Newton’s theory into a research program for its own development and fur-
ther testing (cf. Smith 2002, 159). Instead of relying on ad hoc forces to save the
law of gravity, the systematic discrepancies between the predicted theoretical

15 For a detailed discussion of the method of successive approximations and the differences it

introduces vis-​à-​vis the simple HD method, see Ducheyne (2012, 161ff).

 Descartes, Newton, and Beyond 89

planetary motions and the actual ones call for an explanation in terms of specific
gravitational forces.
This kind of attitude toward the theory is embodied in Newton’s Rule IV.

In experimental philosophy, propositions gathered from phenomena by in-

duction should be considered either exactly or very nearly [quam proxime]
true notwithstanding any contrary hypotheses, until yet other phenomena
make such propositions either more exact or liable to exceptions. (1729
[2016], 796)

Being already an idealized description of the phenomena under investigation,

Newton’s theory is false of the actual world. But this kind of falsity is cheap and
fully consistent with the theory being, in some sense, a good approximation of
worldly phenomena. What mandates theory change is not falsity per se but rather
the fact that the theory systematically fails to find robust physical explanations
(framed with the conceptual resources of the theory) of further (including some
second-​order) phenomena.

5.4 Einstein’s Retentionism

Unlike the transition from Descartes’s theory to Newton’s, the transition to

Einstein’s GTR highlights a pattern in theory-​change which capitalizes on the
successes (and I would add, the truth-​content) of the predecessor theory to ad-
vance a better explanation of the phenomena. In the case we are about to dis-
cuss, the cause of falsity, as it were, was that spacetime is curved and that this
contributes substantially to the explanation of the discrepancies found between
prediction and observation. By the same token, however, identifying the causes
of falsity presupposed treating Newton’s theory as partially correct.
In a survey article on gravitation in 1903, the US astronomer Jonathan
Zenneck presented the situation regarding Newton’s gravitational theory as
follows:

Two independent fields insure that Newton’s law, even if not absolutely accurate,
represents real conditions with a far-​reaching accuracy unmatched by hardly
any other law. In the astronomical domain, this law yields planetary motions
not only to the first approximation (Kepler’s laws); but even to the second ap-
proximation, the deviations of planetary motion due to perturbations by other
planets follow from Newton’s law so accurately that the observed perturbations
led to the prediction of the orbit and relative mass of a hitherto unknown planet
(Neptune). (1903, 86)
90 Historical Cases for the Debate

Yet, there are “astronomical observations that show deviations compared to

calculations based on Newton’s law.” The most significant one that was cited was
the “ca. 40″ per century in the perihelion motion of Mercury.” As it was noted
by Zenneck, the case of Mercury, unlike the case of Neptune, presented a real
difficulty for Newton’s theory of gravity. In fact, already in 1894, there had been
attempts to revise Newton’s law of gravity in order to account for the perihelion
of Mercury—​by Asaph Hall. As Zenneck noted:

Indeed, M. Hall proved that the law previously examined by G. Green, which
replaces 1⁄r2 with 1⁄  r2+λ where λ stands for a small number, is sufficient to ex-
plain the anomalous perihelion motion of Mercury, if λ = 16 ⋅10–​8. This figure
for λ would also give the right result for the observed anomalous perihelion
motion of Mars, though for Venus and Earth the consequence would be some-
what too large a perihelion motion. (1903, 92)

But this revision led to conflicts with other phenomena, such as the position of
the lunar apogee. Fixing the exponent of the inverse square law in such a way that
it accounted for the various phenomena was proved to be more difficult than
anticipated. An adequate account of the anomalous perihelion of Mercury had to
wait until a new theory emerged: Einstein’s GTR.
In November 18, 1915, Einstein published a paper in which, as he put it, he
found “an important confirmation of this most fundamental theory of relativity,
showing that it explains qualitatively and quantitatively the secular rotation of
the orbit of Mercury (in the sense of the orbital motion itself), which was dis-
covered by Leverrier and which amounts to 45 sec of arc per century” (quoted
in, Kennedy 2012, 170). He was, of course, referring to the GTR, which he had
presented a week earlier, in November 1915, in the same journal.16
Einstein took for granted that Newton’s theory (aided by standard perturba-
tion theory) successfully accounted for 531 arc-​seconds per century of the total
575.1 arc-​seconds per century precession of the perihelion of Mercury. In fact,
taking seriously the 43.4 arc-​seconds per century as the observed discrepancy
that constituted an anomaly for Newton’s theory presupposed that the remaining
531 arc-​seconds were accounted for by the gravitational effects on Mercury of all
other planets, as explained by Newton’s theory. This suggests that the evidence
for Newton’s theory was solid enough to allow taking this theory as the bench-
mark for the explanation of the 531 arc-​seconds per century.
In fact, Einstein never doubted that he had to recover Newton’s theory of
gravity as a limiting case of his own theory. In his letter to Hilbert (November 18,

16 For a thorough and detailed study of Einstein’s account of the anomalous perihelion of Mercury

see Earman and Janssen (1993).

 Descartes, Newton, and Beyond 91

1915) he emphasized the role of the Newtonian limit in his formulation of the
theory of relativity. Stressing that he arrived at the field equations first, he noted
that three years ago

it was hard to recognize that these equations form a generalisation, and indeed
a simple and natural generalisation, of Newton’s law. It has just been in the last
few weeks that I succeeded in this (I sent you my first communication), whereas
3 years ago with my friend Grossmann I had already taken into consideration
the only possible generally covariant equations, which have now been shown to
be the correct ones. We had only heavy-​heartedly distanced ourselves from it,
because it seemed to me that the physical discussion had shown their incom-
patibility with Newton’s law. (quoted in Renn et al. 2007, 908).

In his 1916 systematic presentation of GTR, Einstein (1916, 145) made clear
that starting from the field equations of gravitation “give us, in combination with
the equations of motion . . . to a first approximation Newton’s law of attraction,
and to a second approximation the explanation of the motion of the perihelion of
the planet Mercury . . . These facts must, in my opinion, be taken as convincing
proof of the correctness of the theory”.
How can this be, if Newton’s law is characteristically false? In other words,
how can recovering Newton’s law of gravity as an approximation be “convincing
proof ” for Einstein’s theory, if Newton’s theory has had no substantial truth-​
content that could (and should) be recovered by (and retained in) Einstein’s
theory? What’s really interesting in Einstein’s case is that Einstein’s chief con-
ceptual innovation, the curvature of spacetime, provided evidence for Newton’s
account of gravity being in some sense a correct approximation to the actual grav-
itational phenomena. Conversely, as Smith has insisted, Newton’s law of gravity
provided evidence for Einstein’s own theory (and hence Einstein’s own account
of gravity) being more correct than Newton’s. This is because Einstein’s ac-
count identified “non-​Newtonian sources of discrepancies, like the curvature of
space,” which provide “evidence for the previously identified Newtonian sources
presupposed by the discrepancies” (2014, 328). In other words, the evidence for
Newton’s account of gravity was solid enough to allow taking this account as the
benchmark for the explanation of the 531 arc-​seconds per century.
In sum, some key theoretical components of Newton’s theory of gravity—​the
law of attraction and the claim that the gravitational effects from the planets on
each other were a significant cause of the deviations from their predicted orbits—​
were taken by Einstein to be broadly correct and explanatory (of at least part) of
the phenomena to be explained.
An objection to this retentionist argument could be that there is indeed reten-
tion in the transition from Newton’s theory of gravity to Einstein’s, but this is only
92 Historical Cases for the Debate

at the mathematical level.17 Indeed, Stanford has capitalized on this objection in

order to point out that “Newtonian mechanics is just plain false, and radically
so.” For him, the claim that “Newtonian mechanics itself is “approximately true,”
can “only mean that its empirical predictions approximate those of its successors
across a wide range of contexts. It cannot mean that it is approximately correct as
a fundamental description of the physical world” (2006, 9).
Falsity, we noted already, is cheap. Newton’s theory is false, and so is Descartes’s.
But this judgement obscures an important difference: Newton’s theory is still
used to model various phenomena while Descartes’s is not. Newton’s account
of gravity is not like phlogiston or vortices. And this is so because Newton’s ac-
count of gravity, though false, is still a good approximation of the gravitational
phenomena. This feature is not accidental (nor a matter of convenience only). It
relies on the fact that Newton’s account of gravity identified features of gravity
that were retained in Einstein’s theory and form part of the current scientific
image of the world. Einstein himself stressed that his field equations, together
with the laws of motion, yield “to a first approximation Newton’s law of attrac-
tion.” As is well known, Einstein’s chief point was that in a quasi-​static and weak
gravitational field, he could recover Newton’s law of attraction, in the form of the
gravitational potential. Besides, Newton correctly identified masses as causally
relevant factors in the stability of the solar system and he accounted for gravity
on their basis. For a realist, it should be enough that Newton’s account of gravity
was world-​involving in a way that Descartes’s was not.18
Here is a characteristic way to put the relation between GTR and Newton’s
theory of gravity:

Our modern acceptance of general relativity is not only based on experiments

or observations related to some of its special predictions but also on the fact
that it incorporates the entire Newtonian knowledge on gravitation, including
its relation to other physical interactions, that has been accumulated over a
long period of time in classical physics and in the special theory of relativity.
This knowledge embraces, among other aspects, Newton’s law of gravitation
including its implications for the conservation of energy and momentum, the
relation between gravitation and inertia, the understanding that no physical
action can propagate with a speed greater than that of light, which was first
achieved by the field theoretic tradition of classical physics and then finally
established with the formulation of special relativity, and, more generally, the
local properties of space and time, also formulated in special relativity. (Renn
and Sauer, 303)

17 Many thanks to Peter Vickers for pressing me on this point.

18 For a more detailed account of the “world-​involving” explanation, see my (2017).
 Descartes, Newton, and Beyond 93

5.5 Divide et Impera by Other Means

Before some conclusions are drawn, let me discuss the connection of the strategy
followed in the foregoing double case study with the move I introduced many
years ago (cf. 1996) and called the divide et impera strategy to address PI. I al-
ready noted in the introduction that the “moving from the evidence of history to
the history of evidence” strategy is, to paraphrase Clausewitz, the continuation of
divide et impera by other means. To see this let us recall the core motivation and
the core idea behind the divide-​and-​conquer strategy.
The key motivation was to refute PI by showing that the pessimistic con-
clusion that current theories are likely to be false because most past theo-
ries were false was too crude to capture the dynamics of theory change in
science. The problem with the conclusion of the PI is not that it is false. As
noted already, falsity is cheap; current theories are indeed likely to be false,
strictly speaking, if only because they employ idealizations and abstractions.
The problem with the conclusion of PI is that it obscures important senses in
which current theories are approximately true (better put: they have truth-​
like components). More specifically, the conclusion of PI obscures the sub-
stantial continuity there is in theory-​change. Now, the way I read PI was as
a reductio of the realist thesis that currently successful theories are approx-
imately true. PI, I argued, capitalizes on the claim that current theories are
inconsistent with past ones. So, according to PI, if we were to assume, for the
sake of the argument, that current theories are true, we would have to assume
that past theories were false. But given that those false past theories were
empirically successful (recall Larry Laudan’s list of past successful-​yet-​false
theories), truth cannot be the best explanation of the empirical success of
theories, as realists assume.
The key idea behind the divide et impera gambit was that the premise of PI
that needed rebutting was that past theories cannot be deemed truth-​like be-
cause they are inconsistent with current theories. And the novelty of divide et
impera, if I may say so, was that it refocused the debate on individual past the-
ories and their specific empirical successes. If we focus on specific past theory-​
led successes (e.g., the empirical successes of the caloric theory of heat, or of
Newton’s theory of gravity), the following questions suggest themselves: How
were these particular successes brought about? In particular, which theoretical
constituents of the theory essentially contributed to them? An important as-
sumption of the divide et impera strategy was that it is not, generally, the case that
no theoretical constituents contribute to a theory’s successes. By the same token,
it is not, generally, the case that all theoretical constituents contribute (or, con-
tribute equally) to the empirical successes of a theory. Theoretical constituents
that essentially contribute to successes are those that have an indispensable role
94 Historical Cases for the Debate

in their generation. They are those which really fuel the derivation of the predic-
tion. Hence, the theoretical task was two-​fold:

(i) to identify the theoretical constituents of past genuine successful theories

that essentially contributed to their successes; and
(ii) to show that these constituents, far from being characteristically false,
have been retained in subsequent theories of the same domain.

If both aspects of this two-​fold task were fulfilled, the conclusion would be that
the fact that our current best theories may well be replaced by others does not,
necessarily, undermine scientific realism. Clearly, we cannot get at the truth all at
once. But this does not imply that, as science grows, there is no accumulation of
significant truths about the deep structure of the world. The lesson of the divide
et impera strategy was that judgements from empirical support to approximate
truth should be more refined and cautious in that they should only commit us
to the theoretical constituents that do enjoy evidential support and contribute
to the empirical successes of the theory. Moreover, it was argued that these re-
fined judgements are made by the historical protagonists of the various cases.
Scientists themselves tend to identify the constituents that they think responsible
for the successes of their theories, and this is reflected in their attitude toward
their own theories.
I recently reviewed the debate that the divide et impera strategy has generated
(cf. Psillos 2018a). Hence, there is no point in repeating it here too. But given
the outline just presented, it transpires that the key thought behind the divide
et impera strategy was precisely that the evidence from the history of science
was too crude, undifferentiated, and misleading to indict realism. Besides, this
evidence portrayed theory-​change in science as an all or nothing approach. The
divide et impera move had in it the thesis that the proper target was looking at
the history of evidential relations between the various successes of the theory
and the parts of the theory that fueled them. In this sense, the “moving from the
evidence of history to the history of evidence” strategy is a continuation of the
divide et impera strategy.
But what are the “other means” suggested by paraphrasing Clausewitz? These
are suggested by the dynamics of the theory-​change we have discussed in the
double case study. What makes Newton’s strategy distinctive was a) the idea that
a law (or a theory, for that matter) holds quam proxime; and b) the idea that the
various discrepancies between the predictions of the theory and the observed
facts can become the source of a strategy for the development of theory and for
testing it more rigorously, until such time that there is reason to abandon the
theory. Hence, as for instance Harper and Smith (1995) have repeatedly claimed,
Newton significantly altered the criteria of rigorous testing of a theory. Now, a
 Descartes, Newton, and Beyond 95

rigorously successful theory is never really abandoned, as Einstein made sure to

stress. Not only because it holds quam proxime; but also (and mainly) because
it identifies and reveals some elements of reality which, precisely because they
enjoy evidential support, are retained in subsequent theories. Descartes’s theory,
on the other hand, was really abandoned. His theory was not fully unsuccessful.
(Actually, Descartes’s theory, as well as Leibniz’s, offered accounts of the fact that
all planets rotate around the sun in the same direction). Still, Descartes’s theory
failed to identify elements of reality which could be retained in subsequent the-
ories; hence it failed to transfer whatever evidence there was for it to Newton’s
theory. And it was for this reasons that it was discarded. (The whole account of
vortices was one such important failure.)
Hence, the other means that the “history of evidence” strategy adds to the di-
vide et impera strategy is retentionism: those components of a false theory are
deemed retainable by the historical actors for which there is strong evidence that
identify and capture elements of reality, and which can therefore become evi-
dence for the successor theory (which retains parts of the predecessor). Whereas
Newton found little retainable in Descartes’s theory, Einstein found much retain-
able in Newton’s.

5.6 Conclusions

In the preface to the Treatise on Light (1690), Huygens summed up the Cartesian
approach to hypotheses as follows:

One finds in this subject a kind of demonstration which does not carry with it
so high a degree of certainty as that employed in geometry; and which differs
distinctly from the method employed by geometers in that they prove their
propositions by well-​established and incontrovertible principles, while here
principles are tested by the inferences which are derivable from them. The na-
ture of the subject permits of no other treatment. It is possible, however, in this
way to establish a probability which is little short of certainty. This is the case
when the consequences of the assumed principles are in perfect accord with the
observed phenomena, and especially when these verifications are numerous;
but above all when one employs the hypothesis to predict new phenomena and
finds his expectations realized. (quoted in Harper 2011, 373)

We have seen that Newton replaced this conception of method with a more elab-
orate account of explanation, which, without neglecting the role of theoretical
virtues, took it that the explanation offered by the theory should be taken to be
quam proxime and should be further put to the test, while the theory is developed
96 Historical Cases for the Debate

to account for more first-​and second-​order phenomena. Both Descartes and

Newton developed their theories as the best explanation of the evidence, though
they differed as to what counts as the best explanation.
Both Descartes’s and Newton’s theories are false. Both can be used to feed the
premises of the PI. Yet, there is a substantial difference between them: Newton’s
theory has given rise to a research program that led to, and put constraints on,
the subsequent theory.19 Some key elements of Newton’s account of gravity have
become part of the current scientific image of the world. Newton’s account of
gravity has latched onto causal features of the world in a way that Descartes’s has
not. Newton’s account of gravity holds quam proxime, and because of this it led
to a retentionist pattern in theory-​change. The evidence for Newton’s theory be-
came evidence for its successor GTR.
Looking grossly at the evidence from the history of science obscures a more
fine-​tuned look at the history of evidence for the truth—​quam proxime—​of sci-
entific theories (past and present). In thinking about the bearing of the evidence
on theories we should shift our attention from the evidence that comes from his-
tory (of science) to the history of the evidence for theories and in particular for
the key theoretical claims they make.

Acknowledgments

Many thanks to Timothy Lyons and Peter Vickers for their encouragement and
useful comments. I am also grateful to an anonymous reader for insightful criti-
cism. Earlier versions of this paper were presented at the History of Science and
Contemporary Scientific Realism Conference, IUPUI, Indianapolis, in February
2016; the philosophy seminar of the University of Hannover, in May 2016; and
at the inaugural conference of the East European Network for Philosophy of
Science (EENPS), Sofia, June 2016. I would like to thank all those who asked
questions and made comments on these occasions. This paper would not have
been written without the generous intellectual help of Robert DiSalle, William
Harper, and Stavros Ioannidis.

References
Aiton, E. J. 1957. “The Vortex Theory of the Planetary Motions—​I.” Annals of Science
13: 249–​264.

19 For an elaborate account of how Descartes’s and Newton’s theories differ in the ways they repre-

sent the world mathematically, see Domski (2013).

 Descartes, Newton, and Beyond 97

Aiton, E. J. 1958a. “The Vortex Theory of the Planetary Motions—​II.” Annals of Science
14: 132–​147.
Aiton, E. J. 1958b. “The Vortex Theory of the Planetary Motions—​III.” Annals of Science
14: 157–​172.
Aiton, E. J. 1962. “The Celestial Mechanics of Leibniz in the Light of Newtonian Criticism.”
Annals of Science 18: 31–​41.
Aiton, E. J. 1996. “The Vortex Theory in Competition with Newtonian Celestial Dynamics.”
In R. Taton and C. Wilson (eds.) The General History of Astronomy: Planetary
Astronomy from the Renaissance to the Rise of Astrophysics; vol 2B, The Eighteenth and
Nineteenth Centuries. Cambridge: Cambridge University Press, pp. 3–​21.
Bernoulli. J. 1730. Nouvelles Pensees vers le Systeme de M. Descartes, et la Maniere d’en
Deduire les Orbites et les Aphelies des Planetes. Paris: Rue S. Jacques.
Brading, K. 2012. “Newton’s Law-​Constitutive Approach to Bodies: A Response to
Descartes.” In A. Janiak and E. Schliesser (eds.) Interpreting Newton: Critical Essays.
Cambridge: Cambridge University Press, pp. 13–​32.
Bussotti, P. 2015. The Complex Itinerary of Leibniz’s Planetary Theory. Dordrecht: Springer.
De Pierris, G. 2015. Ideas, Evidence, and Method: Hume’s Skepticism and Naturalism con-
cerning Knowledge and Causation. Oxford: Oxford University Press.
Descartes, R. 1982/​1644. Principles of Philosophy. Trans. Valentine Rodger Miller and
Reese P. Miller. Dordrecht: D. Reidel Publishing Company.
Domski, M. 2013. “Mediating between Past and Present: Descartes, Newton, and
Contemporary Structural Realism.” In Mogens Laerke, Justin E. H. Smith, and Eric
Schliesser (eds.) Philosophy and Its History: Aims and Methods in the Study of Early
Modern of Philosophy. Oxford: Oxford University Press, pp. 278–​300.
Ducheyne, S. 2012. The Main Business of Natural Philosophy: Isaac Newton's Natural-​
Philosophical Methodology. Dordrecht: Springer.
Earman, J., & Janssen M. 1993. “Einstein’s Explanation of the Motion of Mercury’s
Perihelion.” In John Earman, Michel Janssen, and John D. Norton (eds.) The Attraction
of Gravitation: New Studies in the History of General Relativity. Boston: Birkhaeuser, pp.
129–​172.
Einstein, A. 1916. “The Foundation of the General Theory of Relativity” reprinted in
A. Einstein The Principle of Relativity. Dover 1932.
Garber, D. 1992. Descartes’ Metaphysical Physics. Chicago: University of Chicago Press.
Harper, W. 2011. Isaac Newton’s Scientific Methodology. New York: Oxford
University Press.
Harper, W. & Smith, G. E.1995. “Newton’s New Way of Inquiry.” In Jarrett Leplin (ed.) The
Creation of Ideas in Physics. Dordrecht: Kluwer, pp. 113–​166.
Hattab, H. 2000. “The Problem of Secondary Causation in Descartes: A Response to
DesChene.” Perspectives on Science 8: 93–​118.
Herival, J. 1965. The Background to Newton’s Principia. Oxford: Clarendon Press.
Huygens, C. 1690/​ 1997. Discourse on the Cause of Gravity. Karen Bailey (trans.).
Mimeograph.
Iltis, C. 1973. “The Decline of Cartesianism in Mechanics: The Leibnizian-​Cartesian
Debates.” Isis 64: 356–​373.
Jalobeanu, D. 2013. “The Nature of Body.” In Peter R. Anstey (ed.) The Oxford Handbook
of British Philosophy in the Seventeenth Century. Oxford: Oxford University Press, pp.
213–​239.
Janiak, A. 2008. Newton as a Philosopher. Cambridge: Cambridge University Press.
98 Historical Cases for the Debate

Kennedy, R. 2012. A Student’s Guide to Einstein’s Major Papers. Oxford: Oxford

University Press.
Leibniz, G. W. 1989. Philosophical Papers and Letters. L. E. Loemker (ed.) 2nd Edition.
Dordrecht: Springer.
Martins, Roberto De. 1994. “Huygens’s Reaction to Newton’s Gravitational Theory.” In J. V.
Field and Frank A. J. L. James (eds.) Renaissance and Revolution. Cambridge: Cambridge
University Press, pp. 203–​213.
Newton, I. 1729/​2016. Mathematical Principles of Natural Philosophy. I. Bernard Cohen,
Anne Whitman, Julia Budenz (trans.), Oakland, CA: University of California Press.
Newton, I. 2004. Philosophical Writings. Cambridge Texts in the History of Philosophy.
Andrew Janiak (ed.) Cambridge: Cambridge University Press.
Ott, W. 2009. Causation and Laws of Nature in Early Modern Philosophy.
Oxford: Clarendon Press.
Psillos, S. 1996. “Scientific Realism and the ‘Pessimistic Induction.’ ” Philosophy of Science
63: S306–​14.
Psillos, S. 1999. Scientific Realism: How Science Tracks Truth. London; New York: Routledge.
Psillos, S. 2017. “World-​Involving Scientific Understanding.” Balkan Journal of Philosophy
9: 5–​18
Psillos, S. 2018a. “Realism and Theory Change in Science.” In The Stanford Encyclopedia
of Philosophy, Summer 2018 Edition, Edward N. Zalta (ed.). URL = <https://​plato.stan-
ford.edu/​archives/​sum2018/​entries/​realism-​theory-​change/​>.
Psillos, S. 2018b. “Laws and Powers in the Frame of Nature.” In Lydia Patton and Walter R.
Ott (eds.) Laws of Nature. Oxford: Oxford University Press, pp. 80–​107.
Psillos, S. Forthcoming. “From Natures to Laws of Nature.”
Renn, J. et al (eds.). 2007. The Genesis of General Relativity, 4 volume-​set. (Boston Studies
in the Philosophy of Science, vol. 250.) Dordrecht: Springer.
Renn, J. & Suer, T. 2007. “Pathways out of Classical Physics”. In Jürgen Renn et al. (eds.)
The Genesis of General Relativity. Dordrecht: Springer.
Smith, G. E. 2002. “The Methodology of the Principia.” In I. B. Cohen and G. E. Smith
(eds.) The Cambridge Companion to Newton. Cambridge: Cambridge University Press,
pp. 138–​73.
Smith, G.E. 2008. “Newton's Philosophiae Naturalis Principia Mathematica.” In Edward
N. Zalta (ed.) The Stanford Encyclopedia of Philosophy (Winter 2008 Edition), https://​
plato.stanford.edu/​entries/​newton-​principia/​.
Smith, G. E. 2014. “Closing the Loop: Testing Newtonian Gravity, Then and Now.” In Zvi
Beiner and Eric Schliesser (eds.) Newton and Empiricism, Oxford: Oxford University
Press, pp. 262–​351.
Snow, A. J. 1924. “Newton’s Objections to Descartes’s Astronomy.” The Monist 34: 543–​557.
Stanford, P. K. 2006. Exceeding our Grasp. New York: Oxford University Press.
Wilson, M. D. 1999. Ideas and Mechanism: Essays on Early Modern Philosophy.
Princeton: Princeton University Press.
Zenneck, J. 1903. “Gravitation.” In Arnold Sommerfeld (ed.) Encyklopädie der
mathematischen Wissenschaften, Vol. 5 (Physics). Leipzig: Teubner, 25–​67. Reprinted
in Renn, Jürgen et al (eds) 2007. The Genesis of General Relativity, 4 volume-​set.
(Boston Studies in the Philosophy of Science, vol. 250.)
 6
How Was Nicholson’s Proto-​Element
Theory Able to Yield Explanatory as well
as Predictive Success?
Eric R. Scerri

Department of Chemistry & Biochemistry

UCLA
[email protected]

6.1 Introduction

Let me begin by saying how grateful I am to Peter Vickers and Tim Lyons for
having invited me to two of their conferences.1 I can honestly say that I have
seldom participated in such interesting interdisciplinary settings and from
which I learned so much. The following article is based on a lecture that I gave
at one of these meetings on the work of the English mathematical physicist John
Nicholson. This particular episode in the history of physics has received little at-
tention in the scientific realism debate even though Nicholson’s theory had con-
siderable explanatory and predictive success. What makes the case all the more
remarkable is that by most standards almost everything that Nicholson proposed
was overturned.
My study is most closely connected with the research interests of Peter Vickers
who has published on the subject of inconsistent theories.2 To quote from the
brief directive that was given to speakers by the conference organizers: “The
following is among the key historical questions to be asked: To what extent did
theoretical constituents that are now rejected lead to significant predictive or
explanatory successes?” Many authors, some of whom are represented in this

1 I presented papers at the History of Chemistry and Scientific Realism meeting held in

Indianapolis between December 6 and 7, 2014, as well as the meeting Contemporary Scientific
Realism and the Challenge from the History of Science that also took place in Indianapolis.
2 P. Vickers, Understanding Inconsistent Science, Oxford, Oxford University Press, 2013.

Eric R. Scerri, How Was Nicholson’s Proto-​Element Theory Able to Yield Explanatory as well as Predictive Success?
In: Contemporary Scientific Realism. Edited by: Timothy D. Lyons and Peter Vickers, Oxford University Press. © Oxford
University Press 2021. DOI: 10.1093/​oso/​9780190946814.003.0006
100 Historical Cases for the Debate

volume, appear to be somewhat puzzled by the fact that some inconsistent the-
ories were able to attract a good deal of success. However, in the approach that
I will be proposing, such features are seen to arise in a far more natural manner.
I will consider the work of Peter Vickers on quantum theory and quantum
mechanics as an example of the current research that has focused on the sci-
entific realism debate and the challenge from the history of science. In his
recent book Vickers considers two well-​trodden examples from the history
of quantum theory. The first one consists of Bohr’s calculation of the spec-
trum of He+, a feat that brought admiration from no less a person than Albert
Einstein when it was first published. Several authors have claimed that Bohr’s
theory was in several respects inconsistent to the point of containing in-
ternal contradictions.3 And yet Bohr succeeded not only in calculating ex-
actly the energy of the hydrogen atom but in also assigning the electronic
configurations of many atoms in the periodic table. Even more dramatically,
Bohr gave a highly accurate calculation for the energy of the helium +1 ion.
The second episode began when Fowler criticized Bohr’s initially published
theory because it predicted an energy of precisely four times that of the hy-
drogen atom, or 4 Rydbergs, when in fact experiments revealed the energy of
He+ to be 4.00163 Rydbergs. In response, Bohr pointed out that he would take
account of the reduced mass of the electron in order to carry out a more accu-
rate treatment of the problem.4 On doing so he obtained an energy of 4.00160
Rydbergs, which is what led Einstein to say, “This is an enormous achieve-
ment. The theory of Bohr must then be right.”5
Vickers recounts a similar story concerning Arnold Sommerfeld.6 It emerges
that precisely the same formula is featured in Sommerfeld’s semi-​classical
atomic theory as it is in the fully quantum mechanical theory that came later.
This equivalence occurs in spite of the fact that the existence of electron spin
and the wave nature of electrons, which play a prominent role in the fully
quantum mechanical theory, were completely unknown when Sommerfeld
published his account. Many commentators wonder how Sommerfeld could
have arrived at the correct formula without knowing about these aspects of
the more mature quantum mechanics, as opposed to the old quantum theory
within which he worked. For example, while Sommerfeld assumed definite

3 T. Bartelborth, “Is Bohr's Model of the Atom Inconsistent?” in P Weingartner and G. Schurz,

eds., Philosophy of the Natural Sciences, Proceedings of the 13th International Wittgenstein Symposium
: H, 1989, pp. 220–​223.
4 Bohr’s revised calculation essentially consisted of considering the reduced mass of the hydrogen

atom rather than assuming that the nucleus was infinitely heavier than the electron.
5 http:// ​ g alileo.phys.virginia.edu/ ​ c lasses/ ​ 2 52/ ​ B ohr_​ t o_​ Waves/​ B ohr_​ t o_​ Waves.

html#Mysterious%20Spectral%20Lines.
6 Vickers’s more recent views on the Sommerfeld question appear in P. Vickers, Disarming the

Ultimate Historical Challenge to Scientific Realism, British Journal for the Philosophy of Science, 2018.
 Nicholson’s Proto-Element Theory 101

trajectories or orbits for electrons, the quantum mechanical account denies the
existence of orbits and prefers to speak of orbitals in which definite particle tra-
jectories are abandoned.
The response from somebody wishing to maintain a form of realism might
be to claim that something about Sommerfeld’s theory may have been correct,
and it is that part of the theory which accounts for the fact that the two math-
ematical formulas in question were identical. Those wanting to defend such a
position might claim that the core ideas were correct and were responsible for
Sommerfeld success, in spite of his lack of knowledge of electron spin, the wave
nature of electrons, and any other such later developments. Such a preservative
realist would presumably claim that some aspect of Sommerfeld’s theory was
latching on to “the truth.” There has been much discussion of this issue in the
philosophy of science literature and especially in connection with the realism
and anti-​realism debate. Realists have tended to appeal to the slogan of divide
and conquer (divide et impera) that was first proposed, as far as I am aware, by
my old friend from graduate school in London, Stathis Psillos. Such an approach,
although perhaps not precisely under the same banner, has also been promoted
by Philip Kitcher and Larry Laudan.7 Meanwhile Carrier, Chang, Chakravartty,
and Lyons among others have expressed strong reservations about this strategy.8
It should be noted that these debates have usually been waged over what were
once very successful theories and scientific entities such as caloric, phlogiston,
or the ether.
As I will endeavor to show, it becomes more difficult to argue for some
form of “preservative realism” in cases like the one I am about to present,
concerning the mathematical physicist John Nicholson who was a contem-
porary of Niels Bohr at the time of the old quantum theory. In fact I shall
be arguing that it is rather difficult to find anything about Nicholson’s view
that has been preserved, except perhaps for the notion of the quantization of
angular momentum, which ironically did not play a prominent role in any
of the success that Nicholson’s theories achieved at the time when they were
first presented. What the defenders of preservative realism recommend in

7 P. Kitcher, The advancement of science, Science without legend, objectivity without illusions,

Oxford, Oxford University Press, 1993; L. Laudan, “A Confutation of Convergent Realism, Philosophy
of Science, 48, 19–​48, (1981).
8 M. Carrier, “Experimental Success and the Revelation of Reality: The Miracle Argument

or Scientific Realism.” In Knowledge and the World: Challenges beyond the Science Wars, ed.
Martin Carrier, Johannes Roggenhofer, Günter Küppers, and Philippe Blanchard, 137–​ 61.
Berlin: Springer (2004); A. Chakravartty, A Metaphysics for Scientific Realism: Knowing the
Unobservable. Cambridge: Cambridge University Press (2007); H. Chang, “Preservative Realism
and Its Discontents: Revisiting Caloric.” Philosophy of Science 70, (5), 902–​12 (2003); T.D. Lyons,
“Scientific Realism and the Pessimistic Meta-​modus Tollens.” In Recent Themes in the Philosophy
of Science: Scientific Realism and Commonsense, ed. Steve Clarke and Timothy D. Lyons, 63–​90.
Dordrecht: Kluwer, (2002).
102 Historical Cases for the Debate

the case of theories that featured the caloric or phlogiston, is that some parts
of the theories tracked the truth while others did not. However, this strategy
cannot be deployed in the case of Nicholson. Another strategy, that has been
recommended by John Worrall, has been the notion of structural realism.
Briefly put, this is the view that although the entities postulated by succes-
sive theories might change as history unfolds, the underlying mathematical
structure is seen to persist and to display continuity. I believe that this form
of strategy too will also fail in the case that I will be examining, all of which
leads me to say that my historical case has a good deal to contribute to the
general debate addressed in the present volume. I will now give a review of
the scientific work in atomic physics of John Nicholson before turning back
to the wider questions as to what one should make of the question of realism
and the development of scientific theories.

6.2 Introduction to John Nicholson and His Early Work

John Nicholson was born in Darlington in County Durham in 1881. He

attended Middlesbrough High School and then the University of Manchester,
where he studied mathematics and physical sciences. He continued his edu-
cation at Trinity College, Cambridge, where he took the mathematical tripos
exams in 1904.9 Nicholson won a number of prizes at Cambridge including
the Isaac Newton Scholar Prize for 1906 and was a Smith Prizeman in 1907,
as well as an Adams Prizeman in 1913 and again in 1917. His first position
was as lecturer at the Cavendish Laboratory in Cambridge, followed by a
similar position at Queen’s University in Belfast. In 1912 Nicholson was ap-
pointed professor of mathematics at King’s College, London. In 1921 he was
named fellow and director of studies at Balliol College, Oxford, before re-
tiring in 1930 due a recurring problem with alcoholism. He died in Oxford
in 1955.
Nicholson proposed a planetary model of the atom in 1911 that had certain
features in common with those of Jean Perrin, Hantaro Nagaoka, and Ernest
Rutherford,10 in that he placed the nucleus at the center of the atom. However, it

9 The mathematical tripos was a distinctive written examination of undergraduate students of the

University of Cambridge, consisting of a series of examination papers taken over a period of eight
days in Nicholson’s time. The examinations survive to this day although they have been reformed
in various ways. A. Warwick, Masters of Theory: Cambridge and the Rise of Mathematical Physics.
Chicago: The University of Chicago Press, 2003.
10 E.R. Scerri, The Periodic Table, Its Story and Its Significance, New York, Oxford University

Press, 2020.
 Nicholson’s Proto-Element Theory 103

Figure 6.1 John Nicholson

must be emphasized that Nicholson arrived at this conclusion independently of

Rutherford and the other physicists just mentioned.
In fact, Nicholson’s model had more in common with that of Thomson,
which regarded the electrons as being embedded in the positive charge that
filled the whole of the volume of the atom. Thomson’s later models envisaged
electrons as circulating in rings, but still within the main body of the atom.
More specifically, the way in which Nicholson’s model resembled those of
Thomson lies in the mathematical analysis and the concern for the mechanical
stability of the system.
Where Nicholson’s model differed from all previous ones, regardless of whether
planetary or not, was in his emphasis on astronomical data. He postulated a series
of proto-​atoms, as he called them, that would combine to form the familiar terres-
trial elements. Nicholson believed that the proto-​atoms, and the corresponding
proto-​elements, existed only in the stellar regions and not on the earth. This way of
thinking was part of a British tradition that included William Crookes and Norman
Lockyer, each of whom believed in the evolution of the terrestrial elements from
matter present in the solar corona and the astronomical nebulae. Like Crookes
104 Historical Cases for the Debate

Figure 6.2 Nicholson’s proto-​elements.

Note: hydrogen* does not represent terrestrial hydrogen but hydrogen the proto-​element.

and Lockyer, Nicholson was an early proponent of the study of spectra for gaining
a deeper understanding of the physics of stars as well as the nature of terrestrial
elements.
The particular details of Nicholson’s proto-​atoms were entirely original and
are represented in the form of a table (Figure 6.2). The first feature to notice is
a conspicuous absence of any one-​electron atom.11 This is because Nicholson
believed that such a system would be unstable according to his electromagnetic
analysis.12
For Nicholson, the identity of any particular atom was governed by the
number of positive charges in the nucleus, regardless of the number of orbiting
electrons present in the atom. Nicholson may thus be said to have anticipated
the notion of atomic number that was later elaborated by van den Broek and
Moseley. Nicholson argued that a one-​electron system could not be stable since
he believed this would produce a resultant acceleration towards the nucleus.
By contrast, Nicholson assumed that two or more electrons adopted equidis-
tant positions along a ring so that the vector sum of the central accelerations of
the orbiting electrons would be zero. The smallest atom therefore had to have at
least two electrons in a single ring around a doubly positive nucleus.13
By using his proto-​atoms, Nicholson set himself the onerous task of calcu-
lating the atomic weights of all the elements and of explaining the unidenti-
fied spectral lines in some astronomical objects such as the Orion nebula and
the solar corona. It is one of the distinctive features of Nicholson’s work that

11 Nicholson rejected a one-​electron atom because he believed that at least one more electron

was needed to balance the central acceleration of a lone electron. See p. 163 of McCormmach, “The
Atomic Theory of John William Nicholson,” Archives for History of Exact Sciences, 3, 160–​184, 1966,
for a fuller account.
12 Nicholson’s list of proto-​elements was extended to include two further members in 1914 when

he added proto-​hydrogen with a single electron and archonium with six orbiting electrons.
13 By the year 1914, he had accepted the possibility of a one-​electron atom. See H. Kragh, “Resisting

the Bohr Atom,” Perspectives in Physics, 13, 4–​35, (2011).

 Nicholson’s Proto-Element Theory 105

his interests ranged across physics, chemistry, and astrophysics and that he
placed great emphasis on astrophysical data above all other data forms.

6.3 Accounting for Atomic Weights of the Elements

Of the four proto-​atoms that Nicholson originally considered, he believed that

coronium did not occur terrestrially.14 He therefore set out to accommodate the
atomic weights of all the elements in terms of the three remaining proto-​atoms,
namely his special kind of hydrogen, nebulium, and proto-​fluorine. It is impor-
tant to consider the relative weights that Nicholson attributed to the proto-​atoms
and to delve a little further into Nicholson’s theory.
Although Rutherford’s planetary model had recently been proposed,
Nicholson’s work was much more indebted to the earlier Thomson model. As
is well known, Thomson regarded the atom as consisting of a diffuse positive
charge in which electrons were embedded as “plums in a pudding.”15 In a later
development the electrons were seen as circulating in concentric rings but still
within the main body of the positive charge.
According to Thomson, the orbital radius of any electron had to be less
than the size of the atom as a whole. However, Nicholson rejected this no-
tion for reasons that were quite independent of the arguments that were
being published by Rutherford at about the same time. There is a sense in
which Nicholson’s atom can be said to have been intermediate between that
of Thomson and the later one due to Rutherford. Nicholson retained much
of the mathematical apparatus that Thomson had used to argue for the me-
chanical stability of the atom but required that the positive nucleus should
shrink down to a size much smaller than the radius of the electrons. As a con-
sequence, Nicholson could no longer use estimates of the size of the atom to
fix the radius of the electron orbits. On the other hand, Nicholson could use
his atom to give what seems to have been an excellent accommodation of the
atomic weights of all the elements and some astronomical spectral lines as
will be discussed later.

14 Some of the proto-​elements postulated by Nicolson had been invoked earlier by other authors.

For example, Mendeleev had predicted the existence of an element called coronium and Emmerson
had featured proto-​fluorine in one of his periodic tables. B.K. Emmerson, “Helix Chemica, A Study
of the Periodic Elements,” American Chemical Journal, 45, 160–​210, (1911).
15 It turns out that the name “plum pudding” was never used by Thomson nor any of his con-

temporaries. A.A. Martinez, Science Secrets, Pittsburgh, PA, University of Pittsburgh Press, 2011.
Because of the currency of the term I will continue to refer to it as such.
106 Historical Cases for the Debate

In order to see precisely how Nicholson envisaged his atom we consider his
expression for the mass of an atom, which he published between 1910 and 1911
in a series of articles16 on a theory of electrons in metals.17

2 e 
2

m=  2 
3  rc 
In this expression m is the mass of an atom, e the charge on the nucleus, r the
radius of the electron’s orbit, and c the velocity of light. This expression can be
simplified to read
m ∝ e2 /r (i)

given the constancy of the velocity of light. Nicholson also assumed the positive
charge, ne, for any particular nucleus with n electrons would be proportional to
the volume of the atom.

ne ∝ V

Next by substituting e = ne into (i) he obtained

m ∝ n2 e2 /r (ii)

He also assumed that the positive charge would be uniformly distributed

throughout a sphere of volume V so that

ne ∝ V

or ne ∝ r 3  (since V ∝ r 3 ),

and so   r ∝ n1/3

Substituting into (i) the nuclear mass would take the form

n2
m∝
n1/3
or
m ∝ n5/3

16 Nicholson’s theory of metals appears in, J.W. Nicholson, “On the Number of Electrons

Concerned in Metallic Conduction,” Philosophical Magazine, series 6, 22, 245–​266, (1911).

17 Interestingly, Niels Bohr’s academic career also began in earnest with his development of a

theory of electrons in metals.

 Nicholson’s Proto-Element Theory 107

Figure 6.3 Relative weights of Nicholson’s proto-​atoms.

Figure 6.4 Nicholson’s calculations and observed weights of the noble gases.
Note: Slightly modified table based on a report of Nicholson’s presentation.
Source: Nature, 87, 2189-​501-​501, 1911.

At this point Nicholson assigned the mass of 1.008 to his proto-​atom of hy-
drogen,18 which allowed him to estimate the relative masses of the other proto-​
atoms as shown in Figure 6.3.
From here Nicholson combined different numbers of these three particles
(omitting Cn) to try to obtain the weights of the known terrestrial elements
(Figure 6.4). For example, terrestrial helium could be expressed as,

He = Nu + Pf = 3.9896,

a value that compares very well with the weight of helium that was known at the
time, namely 3.99.19

18 This step seems a little odd given Nicholson’s statements to the effect that hydrogen the proto-​

atom is not necessarily the same as terrestrial hydrogen. In using a mass of 1.008 he surely seems to be
equating the two.
19 The error amounts to approximately 0.3 of one percent. Moreover, Nicholson takes account of

the much smaller weight of electrons in his atoms. After making a correction for this effect he revises
the weight of helium to 3.9881 (or, to three significant figures, 3.99) in apparent perfect agreement
with the experimental value. Such was the staggering early success of Nicholson’s calculations. See
J.W. Nicholson, “A Structural Theory of the Chemical Elements,” Philosophical Magazine series 6, 22,
864–​889, 871–​872, (1911).
108 Historical Cases for the Debate

Figure 6.5 Nicholson’s composite atoms for the first 11 elements in the
periodic table.

Nicholson’s calculations of atomic weights were not confined to just the first
few elements as shown in the Figure 6.3. He was able to extend his accommo-
dation to all the elements up to and including the heaviest known at the time,
namely uranium, and to a very high degree of accuracy. For example, Figure
6.4 shows his calculations as well as the observed atomic weights for the noble
gases.20 Meanwhile Figure 6.5 shows the calculated and observed weights for the
first eleven elements in the periodic table.
Nicholson’s contemporaries initially reacted to his work in a positive but cau-
tious manner. For example, one commentator wrote,

The coincidence between the calculated and observed values is great, but the
general attitude of those present seemed to be one of judicial pause pending
the fuller presentation of the paper, stress being laid on the fact that any true
scheme must ultimately give a satisfactory account of the spectra.21

Nicholson promptly rose to this challenge and provided precisely such an ac-
count of the spectra of some astronomical bodies in his next publication.
This contribution involved the hypothetical proto-​element nebulium, which
Nicholson took to have four electrons orbiting on a single ring around a cen-
tral positive nucleus with four positive charges. Like his other proto-​elements,

20 Figure 6.4 did not appear in Nicholson’s own papers but in a 1911 article in Nature magazine as

part of a report on the annual conference of the British Association for the Advancement of Science
meeting at which Nicholson had presented some of his findings.
21 Anonymous, Nature, 501, (October 12, 1911).
 Nicholson’s Proto-Element Theory 109

Nicholson did not believe that this element existed on the earth but only in the
nebulae that had long ago been discovered by astronomers, such as the one in the
constellation of Orion. Following a series of intricate mathematical arguments
building on Thomson’s model of the atom, Nicholson found that he could ac-
count for many of the lines in the nebular spectrum that had not been explained
by others who had only invoked lines associated with terrestrial hydrogen or
helium.
Nicholson’s feat could well be regarded as a numerological trick, given that it is
always possible to explain a set of known data points given enough doctoring of
any theory. In fact when it was first publicly proposed at a meeting of the British
Society for the Advancement of Science the reaction was indeed one of further
caution. A report that appeared in the magazine Nature stated that,

Dr. J.W. Nicholson contributed a paper on the atomic structure of elements,

with theoretical determinations of their atomic weights, in which an at-
tempt was made to build up all the elementary atoms out of four prolytes
containing respectively 2, 3, 4 and 5 electrons in a volume distribution of
positive electricity. Representing the prolytes by the symbols Cn (coronium),
H (hydrogen), Nu (nebulium), Pf (protofluorine), the accompanying table
indicates the deductions of the author with regard to the composition of sev-
eral elements, allowance being made for the masses of both positive and neg-
ative electrons.22

Scientists usually demand that a good theory should also make successful
predictions so as to avoid any suspicion that a theory may have been deliber-
ately rigged in order to agree with the experimental data.23 Surprisingly enough,
Nicholson’s theory was also able to make some genuine predictions. In addition
to providing a quantitative accommodation of many spectral lines that had not
previously been identified, Nicholson predicted several experimental facts that
were confirmed soon afterward.
Nicholson assumed that each spectral frequency could be identified with
the frequency of vibration of an electron in any particular ring of electrons.
Furthermore, he believed that these vibrations took place in a direction that
was perpendicular to the direction of circulation of the electrons around the
nucleus (Figure 6.6). The model that was eventually developed by Bohr in 1913
differed fundamentally, in that spectral frequencies are regarded as resulting

22 Anonymous, Nature, 501 (October 12, 1911).

23 There is nevertheless a long-​standing discussion in the philosophy of science regarding the rel-
ative worth of temporal predictions as opposed to accommodations, or retro-​dictions as they are
sometimes termed. See, S.G. Brush, Making 20th Century Science, New York, Oxford University
Press, 2015.
110 Historical Cases for the Debate

Figure 6.6 Nicholson’s atomic model.

Note: Figure created by the present author; no diagram was published by Nicholson. As the electron
orbits the nucleus Nicholson supposes that it oscillates in a direction at right angles to the direction
or circulation.

from differences between the energies or frequencies of two different levels in

the atom. Bohr’s spectral frequencies do not correspond directly to any actual or-
bital frequency that an electron possesses. And it was this new understanding of
the relationship between spectra and energy levels that provided Bohr with one
of the main ingredients of his own theory. On the face of things, Nicholson was
therefore simply wrong, since he based his whole theory on what we now know
to be incorrect physics.
But such a view is a typical example of Whiggism and remains at the level of
“right” and “wrong” that I will be aiming to move beyond. Parts of Nicholson’s
theory seem to have succeeded very well, given that many scientists were
impressed by his explanation of the nebular spectrum and his successful predic-
tion of new lines before they had been observed. In addition, Nicholson also pro-
posed the notion of quantization of angular momentum, which Niels Bohr very
soon embraced, and to much effect.
It is not easy to dismiss Nicholson’s accommodation of so many spectral
lines and especially his predictions of some unknown lines. It would not
be the first time that progress had been gained on the basis of what later
seemed like an insecure foundation. Perhaps just enough of Nicholson’s
overall view was sufficiently correct to allow him to do some useful science.
After all, it would be unreasonable to expect there to be uniform progress
in every single aspect of a theory. Typically some parts may be regarded as
being progressive while others may be degenerating, to use the Lakatosian
terminology.
And if we take an even wider perspective and consider the longue durée in
the history of science, surely all scientific progress has been gained on the basis
of what later turned out to be incorrect foundations when seen in the light of
later scientific views. What really matters is that science, in the form of the sci-
entific community, should progress as a whole. Attributing credit to a particular
 Nicholson’s Proto-Element Theory 111

scientist may be important in deciding who prizes should be awarded to, but
does not matter in the broader question of how the scientific community gains a
better understanding of the world.

6.4 Accommodating the Spectra of Four Nebula, Including

Orion Nebula

In this section the manner in which Nicholson was able to assign many unknown
lines in the spectrum of the Orion nebula will be examined. First I present a figure
containing the spectral lines that had been accounted for in terms of terrestrial hy-
drogen and helium (Figure 6.7). The dotted lines signify the spectral lines that
had not yet been assigned, or identified, in any way. This situation therefore pro-
vided Nicholson with another opportunity to test his theory of proto-​atoms and
proto-​elements.
As in many other features of Nicholson’s work his approach was rather simple.
He began by assuming that ratios of spectral frequencies correspond to ratios of
mechanical frequencies among the postulated electron motion.24 In mathemat-
ical terms he assumed

Figure 6.7 Spectrum of Orion nebula showing many unassigned lines.

24 The detailed calculations can be found in Nicholson’s articles, J.W. Nicholson, “The Spectrum

of Nebulium,” Monthly Notices of the Royal Astronomical Society, (London), 72, 49–​64, (1911); “A
Structural Theory of the Chemical Elements,” Philosophical Magazine, 6, (22), 864–​89, (1911); “The
Constitution of the Solar Corona I, Protofluorine,” Monthly Notices of the Royal Astronomical Society
(London), 72, 139–​50, (1911); “The Constitution of the Ring Nebula in Lyra,” Monthly Notices of the
112 Historical Cases for the Debate

Figure 6.8 Nicholson’s accommodation of 9 of the 11 unidentified lines in Figure 6.7.

νspectral line 1 frotation A

=
νspectral line 2 frotation B

where the f values emerged from his calculations, while one of the ν values
was obtained empirically from the spectra in question and the other one was
‘predicted’.

In his 1911 article “The Spectrum of Nebulium,” Nicholson also predicted the
existence of a new spectral line for the nebulae in question:

Now the case of k = −2 for the neutral atom has been seen to lead to another
line which will probably be very weak. Its wavelength should be 5006.9 ×
.86939 = 4352.9. It does not appear in Wright’s table.25

Remarkably enough this prediction was also soon confirmed. In a short note
in the same journal in the following year, 1912, Nicholson was able to report
that the spectral line had been found at a wavelength of 4353.3 Ångstroms

Royal Astronomical Society (London), 72, 176–​77, (1912); “The Constitution of the Solar Corona II,
Protofluorine,” Monthly Notices of the Royal Astronomical Society, (London), 72, 1677–​692, (1912);
“The Constitution of the Solar Corona III,” Monthly Notices of the Royal Astronomical Society
(London), 72, 729–​39, (1912).
25 J.W. Nicholson, “The Spectrum of Nebulium,” Monthly Notices of the Royal Astronomical Society,

72 (1), 49–​64 (1911). https://​doi.org/​10.1093/​mnras/​72.1.49.

 Nicholson’s Proto-Element Theory 113

with an error of just 0.009% or roughly 1 in 11,110 by comparison with his

prediction.

At the meeting of the Society of 1912 March the writer announced the dis-
covery of the new nebular line at λ4353 which had been predicted in his paper
on “The Spectrum of Nebulium.” A plate of the spectrum of the Orion nebula,
on which the line was found, and which had been taken with a long exposure
at the Lick Observatory in 1908 by Dr. W.H. Wright, was also exhibited. In the
meantime the line has been recorded again by Dr. Max Wolf, of Heidelberg,
who has, in a letter, given an account of its discovery, and this brief note gives a
record of some of the details of the observation.
The plate on which the line is shown was exposed at Heidelberg between
1912 January 20 and February 28, with an exposure of 40h 48m. The most
northern star of the Trapezium is in the center of the photographed region, and
the new line is visible fairly strongly, especially in the spectrum of the star and
on both sides.
The wave-​length in the Orion nebula, obtained by plotting from an iron
curve, is 4353.9, which is of course, too large, as all the lines in this nebula are
shifted to greater wave-​lengths, on account of the motion of the nebula. But the
correction is not so large as a tenth-​metre.
The wave-​length of the line on the Lick plate, as measured at the Cambridge
Observatory by Mr. Stratton, is 4353.3, the value calculated in the paper being
4352.9.26

Nicholson experienced a similar triumph with the prediction of a new spec-

tral line, which he believed was due to proto-​fluorine and which he estimated to
have a wavelength of 6374.8 Ångstroms. A new spectral line was soon discovered
in the solar corona with a wavelength of 6374.6.
Considered together, these successes by Nicholson are indeed rather remark-
able. Just to recap, he accounted for 9 of 11 previously unidentified lines in the
spectrum of the Orion Nebulae; 14 of the unidentified spectral lines in the solar
corona. In addition, and perhaps more impressively, Nicholson predicted two
completely unknown lines, one in each of these spectra, both of which were
promptly discovered and found to have almost exactly the wavelengths he
predicted:

Nebulium prediction:   4352.9A, observation: 4353.3A, error: 1 in 11,111

Solar corona prediction: 6374.8A, observation: 6374.6A, error: 1 in 31,745

26 J.W., Nicholson, “On the New Nebular Line at λ4353,” Monthly Notices of the Royal Astronomical

Society (London), 72, 693, (1912).

114 Historical Cases for the Debate

6.5 Nicholson’s Calculations on the Spectrum of

the Solar Corona

Nicholson next turned his attention to the spectrum of the solar corona
that had been much studied and that showed numerous lines that had
not yet been accounted for (Figure 6.9). In this study Nicholson was even
more successful than he had been with the spectrum of the Orion nebula,

Figure 6.9 Observed lines in the solar corona spectrum.

 Nicholson’s Proto-Element Theory 115

Figure 6.10 Nicholson’s accommodation of 14 of the lines from Figure 6.9 using
proto-​fluorine and ionized forms of this proto-​atom.

because he succeeded in accounting quantitatively for as many as 16

unexplained lines.
Figure 6.10 shows the observed frequencies of the lines, along with Nicholson’s
assignments in terms of the atom of proto-​fluorine and various ionized forms of
the same atom.

6.6 Nicholson and Planck’s Constant

The manner in which Nicholson arrived at the all-​important Planck constant

was by calculating the ratio of the energy of a particle to its frequency and finding
that this ratio was equal to a multiple of Planck’s constant. Nicholson concluded
that this constant therefore had an atomic significance and indicated that an-
gular momentum could only change in discrete amounts when electrons leave or
return from an atom. The relevance of this finding lies in the fact that up to this
point the quantum had only been associated with energy and not with angular
momentum. Nicholson was the first person to make this association, in what
would soon become an integral aspect of Bohr’s theory of the hydrogen atom. In
116 Historical Cases for the Debate

the case of the proto-​fluorine atom, Nicholson calculated the ratio of potential
energy to frequency to be approximately

Potential energy/frequency = 154.94 ×10 −27 erg seconds =25 h

In arriving at his result Nicholson had used the measured values of e and m,
the charge and mass of the electron. However, his method still did not provide
a means of estimating the radius of the electron, and he was forced to eliminate
this quantity from his equations, a problem that he would overcome a little later.
Nicholson proceeded to calculate the ratio of potential energy to frequency
in proto-​fluorine ions with one or two fewer electrons and found 22 h and 18 h,
respectively. He noted that the three values for Pf, Pf+1, and Pf+2 were members of
a harmonic sequence,

25, 22, 18, 13, 7, 0.

By dividing each value by the number of electrons in the atom he found the
Planck units of angular momentum per electron to be,

5, 5.5, 6, 6.5, and 7.

Nicholson thus arrived at the general formula for the angular momentum of a
ring of n electrons as

1
2 (15 − n) n

This formula then allowed him to fix the values of the atomic radius in each case,
and since angular momentum did not change gradually, he took this to mean
that atomic radius would also be quantized.
Several authors have traced the manner in which Bohr picked up this hint
of quantizing angular momentum.27 This feature was not present in Bohr’s
initial atomic model, and he only incorporated it over a series of steps fol-
lowing a close study of Nicholson’s papers. Bohr also spent a good deal of
time trying to establish the connection between his own and Nicholson’s
atomic theory.

27 J. Heilbron, T.S. Kuhn, “The Genesis of Bohr’s Atom,” Historical Studies in the Physical Sciences,

1, 211–​90, (1969).
 Nicholson’s Proto-Element Theory 117

6.7 Reactions to the Work of Nicholson

As the historian of physics John Heilbron stated in a recent plenary lecture to

the American Physical Society, the success of Nicholson’s work on nebulium had
been “spectacular.” Heilbron also commented on how it had served as a motiva-
tion for Bohr’s work. But looking at the literature in physics and the history of
physics one finds a remarkable range of views expressed concerning Nicholson’s
work. The following is a brief survey of these varied reactions.
Initially the commentators tended to praise Nicholson. For example, after a
meeting held in Australia in 1914, W. M. Hicks remarked,

Nicholson’s calculated frequencies and the observed lines were “so close
and so numerous as to leave little doubt of the general correctness of the
theory . . . Nicholson’s theory stands alone as a first satisfactory theory of one
type of spectra.”28

In a paper published at the end of 1913, William Wilson observed that

Nicholson had

used the quantum hypothesis with extraordinary success in his valuable

investigations of the sun’s corona.29

Physics historian Abraham Pais saw the relationship between Bohr and
Nicholson sometime later in the following terms,

Bohr was not impressed by Nicholson when he met him in Cambridge in 1911
and much later said that most of Nicholson’s work was not very good. Be that as
it may, Bohr had taken note of his ideas on angular momentum, at a crucial mo-
ment for him . . . He also quoted him in his own paper on hydrogen. It is quite
probable that Nicholson’s work influenced him at that time.30

Returning to Heilbron:

The success of Nicholson’s atom bothered Bohr. Both models assumed a nu-
cleus, and both obeyed the quantum; yet Nicholson’s radiated—​and with un-
precedented accuracy—​while Bohr’s was, so to speak, spectroscopically mute.

28 McCormmach, “The Atomic Theory of John William Nicholson,” Archives for History of Exact

Sciences, 3, 160–​184, (1966), p. 183.

29 Ibid, p. 184.
30 A. Pais, Niels Bohr’s Times, In Physics, Philosophy, and Polity, Oxford, Oxford University Press,

1991, p. 145.
118 Historical Cases for the Debate

By Christmas 1912, Bohr had worked out a compromise: his atoms related to the
ground state, when all the allowed energy had been radiated away; Nicholson’s
dealt with earlier stages in the binding. . . . Just how a Nicholson atom reached
its ground state Bohr never bothered to specify. He aimed merely to establish
the compatibility of the two models. The compromise with Nicholson was to
leave an important legacy to the definitive form of the theory.31
Later in the same paper Bohr proposed other formulations of his quantum
rule, including, with full acknowledgement of Nicholson’s priority, the quanti-
zation of the angular momentum.32

Another historian-​philosopher of physics, Max Jammer, writes

It should also be pointed out that Nicholson’s anticipations of some of Bohr’s

conclusions were based, as Rosenfeld has pointed out, on the most questionable
and often even fallacious reasoning.33 (emphasis added)

Now for one last commentator, Leon Rosenfeld, who reveals some further
aspects of how Nicholson has been regarded. In his introduction to a book by
Niels Bohr to celebrate the 50th anniversary of the 1913 theory of the hydrogen
atom, Rosenfeld writes:

The ratio of the frequencies of the two first modes happens to coincide with that
of two lines of the nebular spectra: this is enough for Nicholson to see in this
system a model of the neutral “nebulium” atom; and as luck would have it, the
frequency of the third mode, which he could then compute, also coincided with
that of another nebular line, which—​to make things more dramatic—​was not
known when he made the prediction in his first paper, but was actually found
somewhat later. . . .
From the mathematical point of view Nicholson’s discussion of the stability
conditions for the ring configurations and of their modes of oscillation is an
able and painstaking piece of work; but the way in which he tries to apply the
model . . . must strike one as unfortunate accidents.34
In the third paper, however, published in 1912, occurs the first mention of
Planck’s constant in connection with the angular momentum of the rotating

31 J. Heilbron, “Lectures in the History of Atomic Physics, 1900–​1922.” In History of Twentieth

Century Physics, C. Weiner, ed., 40–​108, Academic Press, New York, 1977, p. 69.
32 Ibid., p. 70.
33 M. Jammer, The Conceptual Development of Quantum Mechanics, New York, McGraw-​ Hill,
1966. p., 73.
34 L. Rosenfeld, in preface to N. Bohr, On the Constitution of Atoms and Molecules, New York, W.E.

Benjamin, 1963, p. xii.

 Nicholson’s Proto-Element Theory 119

electrons: again here there is no question of any physical argument, but just a
further display of numerology. . . .
Bohr did not learn of Nicholson’s investigations, as we shall see, before the end
of 1912, when he had already given his own ideas of atomic structure their fully
developed form.35
By contrast [with Nicholson] the thoroughness of Bohr’s single-​handed attack
on the problem and the depth of his conception will appear still more impressive.36

There is clearly no “fence sitting” here to give Nicholson any benefit of the
doubt. Rosenfeld does not even believe that Nicholson’s apparent early success
was due to a cancellation of errors. But perhaps some aspects of Rosenfeld’s
life serve to explain some of his reaction. Rosenfeld was without doubt one of
Bohr’s leading supporters and also acted as a leading spokesperson for Bohr’s
Copenhagen interpretation of quantum mechanics for the last 30 or so years of
Bohr’s life. Rosenfeld is also known to have been an especially vitriolic and harsh
critic in spite of his apparently shy and retiring personality. His fellow Belgian
and one-​time collaborator, the physical chemist Ilya Prigogine, described him
as a “paper tiger.”37 It is hardly surprising therefore that Rosenfeld championed
Bohr against any claims from anyone, such as Nicholson, who he regarded as
imposters, or anyone who might try to steal some of the thunder from Bohr.
The views of Rosenfeld can be contrasted this with those of Kragh, the con-
temporary historian and a Dane like Bohr,

No wonder Bohr, when he came across Nicholson’s atomic theory found it to

be interesting as well as disturbingly similar to his own ideas. Nicholson’s atom
was a rival to Bohr’s and Nicholson was the chief critic of Bohr’s ideas of the
quantum atom.38

But let us say, for the sake of argument, that Rosenfeld is right and that Nicholson’s
work was completely worthless. Even if this were true, I contend that Nicholson’s
publications contributed to Bohr’s developing his own atomic theory for the
simple reason that Nicholson served as his foil. In some places Bohr is quite dis-
missive of Nicholson’s work, such as when writing to his Swedish colleague Carl
Oseen, where he describes Nicholson’s work as “pretty crazy” while adding,39

35 Ibid., p. xiii.
36 Ibid., p. xiii–​xiv.
37 Prigogine was born in Russia but emigrated to Belgium.
38 H. Kragh, Niels Bohr and the Quantum Atom: The Bohr Model of Atomic Structure 1913–​1925,

Oxford, Oxford University Press, 2012, p. 27.

39 This comment by Bohr refers to Nicholson’s earlier theory of electrons in metals, although one

gathers the impression that Bohr continued to hold this crucial view about Nicholson. I am grateful
for a reviewer for drawing my attention to this qualification.
120 Historical Cases for the Debate

I have also had discussion with Nicholson: He was extremely kind but with him
I shall hardly be able to agree about very much.40

In other places Bohr shows Nicholson considerably more respect, such as

when writing to Rutherford while he was on the point of submitting his famous
trilogy paper that was published in 1913.

It seems therefore to me to be a reasonable hypothesis, to assume that the

state of the systems considered in my calculations is to be identified with that
of atoms in their permanent (natural) state . . . According to the hypothesis
in question the states of the system considered by Nicholson are, on the con-
trary, of a less stable character; they are states passed during the formation of
the atoms, and are states in which the energy corresponding to the lines in the
spectrum characteristic for the element in question is radiated out. From this
point of view systems of a state as that considered by Nicholson are only pre-
sent in sensible amount in places in which atoms are continually broken up and
formed again; i.e. in places such as excited vacuum tubes or stellar nebulae.41

In another passage from a letter to his brother Harald, Niels Bohr writes

Nicholson’s theory is not incompatible with my own. In fact my calculations

would be valid for the final chemical state of the atoms, whereas Nicholson
would deal with the atoms sending out radiation, when the electrons are in
the process of losing energy before they have occupied their final positions.
The radiation would thus proceed by pulses (which much speaks well for) and
Nicholson would be considering the atoms while their energy content is still
too large that they emit light in the visible spectrum. Later light is emitted in the
ultraviolet, until at last the energy which can be radiated away is lost.42

After Bohr had published his three-​part article, Nicholson continued to press
him in a number of further publications. If we must speak in terms of winners
and losers, Bohr would be regarded as a winner and Nicholson as a loser. But as
in all walks of life, it is not just about winning, but more about partaking. There
would be no athletic races for spectators to watch if the losers were not even to

40 Bohr to Oseen, December 1, 1911, cited in N. Bohr, L. Rosenfeld, E. Rüdinger, & F. Aaserud,

“Collected Works.” In Early Work (1905–​1911), Vol. 1, J.R. Nielsen, ed., North-​Holland Publishing,
Amsterdam, 1972, pp. 426–​431.
41 N. Bohr to Rutherford, 1913, cited in L. Rosenfeld, preface to N. Bohr, On the Constitution of

Atoms and Molecules, New York, W.E. Benjamin, 1963, p. xxxvii.

42 Bohr to Harald, in N. Bohr, L. Rosenfeld, E. Rüdinger, & F. Aaserud, Early Work (1905–​1911).

In both instances where I have written “my,” Bohr had actually written “his”—​which must surely
be typos.
 Nicholson’s Proto-Element Theory 121

participate in the race. The very terms winner and loser are necessarily code-
pendent in the context of any scientific debate, as were the roles of Bohr and
Nicholson.
Now this picture that I have painted would seem to raise at least one obvious
objection. If all competing theories are allowed to bloom because there is no such
thing as a right or wrong theory, how would scientists ever know which theo-
ries to utilize and which ones to ignore? Indeed this is the kind of criticism that
was levelled against Feyerabend43 when he claimed that “anything goes” and was
promptly criticized by numerous authors.44 I think the answer to this question
can be found in evolutionary biology. Nature has the means of finding the best
way forward. Just as any physical trait with an evolutionary advantage eventu-
ally takes precedence, so the most productive theory will eventually be adopted
by more and more scientists in a gradual, or trial and error fashion. The theo-
ries that lead to the most progress will be those that garner the largest amount
of experimental support and which provide the most satisfactory explanations
of the facts. This entire process will not be rendered any the weaker even if one
acknowledges an anti-​personality and anti–​“right or wrong view” of the growth
of science.
More generally, I believe that the two aspects can coexist quite happily.
Scientists can, and regularly do, argue issues out to establish the superiority of
their own views, as well as their claims to priority. But the march of progress, to
use an old-​fashioned term, does not care one iota about these human squabbles.
And it is the overall arc of progress that really matters, not whose egos are bruised
or who obtains the greater number of prizes and accolades.

6.8 How Was Any of the Success Possible Given

the Limitations of Nicholson’s Theory?

Having examined the apparent successes of Nicholson’s theory we must still ask
how any of this was even possible given what we now know of his ideas. Here is a
brief list of what seems to be problematical in Nicholson’s scheme: First of all, the
proto-​elements like nebulium that he postulated do not exist. Second, he identi-
fied mechanical frequencies of electrons with spectral frequencies and assumed
that these oscillations took place at right angles to the direction of electron cir-
culation. Third, Nicholson’s electrons were all in one single ring, unlike the sub-
sequent Bohr model in which they are distributed across different rings or shells.

43 P. Feyerabend, Against Method, London, Verso, 1975.

44 J. Preston, Paul Feyerabend entry in Edward N. Zalta, ed., The Stanford Encyclopedia of
Philosophy, Winter 2016 Edition. https://​plato.stanford.edu/​archives/​win2016/​entries/​feyerabend/​.
122 Historical Cases for the Debate

So in the light of modern knowledge, Nicholson seemed to be making several

false assumptions. According to the traditional way that a realist might regard
such matters, Nicholson’s ideas were not even approximately correct since they
would be regarded as downright false.45 And yet he achieved remarkable success,
at least according to most of the commentators whose views were quoted here.
Are cases such as Nicholson exceptional or can other examples be found in the
history of science? If one accepts that all, or most, theories are eventually refuted,
one has to concede that the progress of science implies that “wrong theories” reg-
ularly lead to progress!
Having gone into some details about Nicholson’s work in early 20th century
atomic physics as well as astrophysics, I would now like to return to the general
philosophical question and try to make some sense of the apparent success his
view seemed to enjoy, at least initially.
I am proposing an evolutionary theory of the development of science in a lit-
eral biological sense.46 While appealing to some of the ideas that have emerged
from evolutionary epistemology as supported by Donald T. Campbell and many
others, my account also involves a new departure as I will be explaining later. The
general idea that is common to most approaches in evolutionary epistemology
is that since evolution drives all biological development, it also drives the way
in which we human agents think and how we develop theories and experimen-
tation. It is a central tenet of evolutionary epistemology that all the knowledge
of the natural world that we possess is essentially determined by evolutionary
biology. Consequently, it seems plausible to further suggest that scientific know-
ledge is neither right nor wrong at any epoch in history but instead better or
worse suited to the environment that science finds itself working in, namely
the way in which scientific theories are suited to or correspond with the natural
world as revealed by experimentation.
Needless to say, I do not dispute such scientific facts as the view that the earth
is round rather than flat. Of course I accept that there are truths with a small “t”
such as it is true that the earth is round or that grass is green. My concern is more
with the notion that there may be a scientific truth with a capital “T.” It is also im-
portant to distinguish theories from facts or observational statements. Whereas
it is true that grass is simply green, this is a far cry from the question of whether
Bohr’s theory or the theory of evolution is true or correct. Theories, especially
those in the physical sciences, consist of mathematical relationships. They do
not just assert whether the earth circles the sun or vice versa. Similarly, Newton’s
theory is based on his famous three mathematical laws, rather than making a

45 As I have already indicated I reject such an over-​simplification, especially when it comes to

assumptions and theories.

46 E.R. Scerri, A Tale of Seven Scientists and A New Philosophy of Science, New York, Oxford

University Press, 2016.

 Nicholson’s Proto-Element Theory 123

simple assertion on whether or not the moon is made of blue cheese or anything
quite so specific.
I take it that one would never wish to claim that biological evolution is either
right or wrong, but rather that any biological developments are either suited to
their environment or not. Those developments that are suited to their biological
niches result in their being perpetuated in future generations while those that are
not simply wither away. So, I claim, it is with the development of science.
In passing, I should note some kinship with the views of the influential anti-​
realist philosopher Van Fraassen, who supports an evolutionary view of the
progress of science when he writes,

For any scientific theory is born into a life of fierce competition, a jungle red in
tooth and claw. Only the successful theories survive—​the ones which in fact
latched onto the actual regularities in nature.47

This view that has been criticized in particular by Kitcher,48 and Stanford,49 al-
though it would take me too far afield to comment on this exchange here.
My own brand of evolutionary epistemology also holds that the study of sci-
entific development should not be approached by concentrating on individual
discoverers nor through individual theories. I claim that science essentially
develops as one unified organism, while fully anticipating that this aspect will
meet the greatest resistance from critics. Needless to say, many people have real-
ized the societal/​collective nature of science, and I cannot claim any originality
on that score. For example, there have been many programs such as the Strong
Program, Science Studies, and the Sociology of Science, which all take a more
holistic approach to studying how science develops. However, these programs
generally hold that social factors determine scientific discoveries. My interest lies
primarily in the actual science, while at the same time maintaining a radical form
of sociology that considers scientific society to be a single organism which is es-
sentially developing in a unified fashion.
I propose that scientific research is conducted by a tacit network of scholars,
and researchers, who frequently appear to be at odds with themselves, and fre-
quently are involved in bitter priority disputes, while unknowingly partaking in
the same overall process. For example, in the case of John Nicholson, and several
others that I have discussed in a recent book, the protagonists were seldom in
direct contact with the Bohrs, Paulis and other luminaries in the world of early

47 B. Van Fraassen, The Scientific Image, Oxford, Clarendon Press, 1980, p. 40.
48 P. Kitcher, The Advancement of Science: Science without Legend, Objectivity without Illusions,
New York, Oxford University Press, 1993.
49 K. Stanford, An Antirealist Explanation of the Success of Science, Philosophy of Science, 67, 266–​

84, (2000).
124 Historical Cases for the Debate

20th century atomic physics. Nevertheless, one sees an entangled, organic devel-
opment in which ideas compete and collide with each other, even if one famed
individual is eventually associated with any particular scientific discovery or
episode.
I regard science as very much proceeding via modification by trial and error
rather than via “cold rationality.” My view is aligned with biological evolution
rather than with an enlightenment view of the supremacy of rationality and the
powers of pure deduction. I regard scientific development as being guided by
evolutionary forces, blind chance, and random mutation. When seen from a dis-
tance science appears to develop as one unified organism. Under a magnifying
glass there may be little sign of unity, so much so that some philosophers and
modern scholars have been driven to declare the dis-​unity of science, a view that
I believe to be deeply mistaken.50
From the perspective that I propose, one can better appreciate how what a
realist regards as wrong theories, as in the case of most of John Nicholson’s scien-
tific output, can lead to progress. I do not need to explain wrong theories away in
my account. Nor do I believe that simultaneous or multiple discoveries deserve
to be explained away as being aberrations. Similarly, priority disputes, which are
so prevalent in science, can be seen as resulting from the denial of the unity of
science and an underlying body-​scientific that I allude to.
Developments that might normally be regarded as wrong ideas frequently
produce progress, especially when picked up and modified by other scientists.
Nicholson’s notion of angular momentum quantization, which he developed in
the course of what has generally been regarded as an incorrect theory, was picked
up by Niels Bohr who used it to transform the understanding of atomic physics.
The image of science that I envisage resembles a tapestry of gradually evolving
ideas that undergo mutations and collectively yield progress. In some cases the
source of the mutation is random, just as mutations are random in the biolog-
ical world. This view stands in sharp contrast with the traditional notion of pur-
posive and well thought out ideas on the part of scientists.51
Consequently, the kinds of inconsistent theories that Peter Vickers and others
have examined no longer seem to be in such urgent need of explanation and
should not be regarded as being so puzzling. Indeed, such developments should

50 P. Galison, D. Stump (eds.), The Disunity of Science, Palo Alto, Stanford University Press, 1996.
51 I regard mutations in scientific concepts to occur on the level of individual scientists and
not primarily in entire social groups. My evolutionary epistemology thus also differs from Kuhn’s
in this respect since his unit of evolutionary change is the social group of scientists. See some in-
teresting discussions regarding Kuhn’s evolutionary account of scientific progress in B.A. Renzi,
“Kuhn’s Evolutionary Epistemology and Its Being Undermined by Inadequate Biological Concepts,”
Philosophy of Science, 76, 143–​59, (2009); T.A.C. Raydon, P. Hoyningen-​Huene, “Discussion: Kuhn’s
Evolutionary Analogy in The Structure of Scientific Revolutions and ‘The Road since Structure,’ ”
Philosophy of Science, 77, 468–​76, (2010).
 Nicholson’s Proto-Element Theory 125

be regarded as the rule rather than exceptions. I agree completely with Kuhn that
ideas and theories that survive should be regarded as facilitating progress and
not as tracking the truth.
As so many evolutionary epistemologists have urged, I regard knowledge as
being presumptive, partial, hypothetical, and fallible. If scientific progress is far
more organic than usually supposed, we can make better sense of why Nicholson
was able to contribute to the growth of science. None of the scientists I have
examined really knew what they were doing, in a sense.52 Their crude ideas de-
veloped in an evolutionary trial and error manner instead of in a strictly rational
manner, or so I propose.
Nicholson and many people like him contributed very significantly to the de-
velopment of science. Nicholson was not simply wrong, since he inadvertently
helped Bohr to begin to quantize angular momentum. Nicholson is as much part
of the history as Bohr is, and minor contributors matter as much as the well-​
known ones. In anticipation of the following section, I believe that the early
Kuhn’s focus on revolutions may have served to diminish the importance of such
marginal figures.

6.9 Is the Proposed View Compatible

with Kuhnian Revolutions?

Is an evolutionary view of the kind I propose compatible with the revolutions for
which Kuhn is so well known? First of all, as many authors have claimed, Kuhn’s
own work on historical episodes sometimes points to continuity rather than rev-
olution, such as in the case of quantum theory and the Copernican Revolution.53
Of course, Kuhn also acknowledges the role of normal science that in very
broad terms can be seen as upholding the role of minor historical contributors
to the growth of science. Nevertheless, as I see it, there was no sharp revolution
in the development of quantum theory only an evolution. Paradoxically, such
an evolution is easier to appreciate from the wider perspective of science as one
unified whole than from the perspective of individual contributors or individual
theories. Viewing theory-​change as revolutionary, on the other hand, may mask
the essentially biological-​like growth of science that I am defending.
Nevertheless, I am in complete agreement with Kuhn in supposing that sci-
ence does not drive toward some external truth. In this respect I am on the side
of anti-​realism. I prefer to regard scientific process as driven from within, by

52 Scerri 2016.
53 T.S. Kuhn, Black-​Body Radiation and the Quantum Discontinuity, 1894–​1912, Chicago, Chicago
University Press, 1987; J. Heilbron, T.S. Kuhn, “The Genesis of the Bohr Atom,” Historical Studies in
the Physical Sciences, 1, 211–​290 (1969).
126 Historical Cases for the Debate

evolutionary forces which look back to past science.54 This is not to say that the
world does not constrain our theorizing. Kuhn just wants us to see that the scope
of our theories is not determined by nature in advance of our inquiring about it.
Indeed, I find that I agree with Kuhn on a great number of issues, the main excep-
tion being the occurrence of scientific revolutions.
As scholars generally concur, Kuhn’s later work was aimed toward developing
an evolutionary epistemology.55 As Kuhn developed his epistemology of science,
he saw increasingly more similarities between biological evolution and scientific
change. Consequently as he developed his epistemology of science it became a
more thoroughly evolutionary epistemology of science. Wray makes the very
perceptive remark that while Kuhn was one of the key philosophers of science
who initiated the historical turn in the philosophy of science in the early 1960s,
he later came to adopt what he termed a historical perspective. According to Wray
this historical perspective, or developmental view, is nothing less than an evo-
lutionary perspective on science. In accord with Wray and others, Kuukkanen
(2013, 134) points out that the later Kuhn felt that his evolutionary image of sci-
ence did not get the amount of attention that it deserved. In the course of his last
interview, Kuhn deplored this situation while saying, “I would now argue very
strongly that the Darwinian metaphor at the end of the book [SSR] is right and
should have been taken more seriously than it was.”56
Just as evolution lacks a telos and is not driven towards a set goal in advance,
science is not aiming at a goal set by nature in advance. Kuhn continued to re-
gard his evolutionary view as important to the end of his life, or as Wray writes,
“Whatever else he changed he did not change this aspect.” Kuhn also claims
that in the history of astronomy, the earth-​centered models held the field back
for many years. Similarly, he claimed that, truth-​centered models of scientific
change were holding back philosophy of science. Somewhat grandiosely, Kuhn
notes a similarity between the reception of Darwin’s theory, and the reception of
his own theory his own view on the evolution of science, in that both meet the
greatest resistance on the point of elimination of teleology.

54 Kuhn prefers the phrase “pushed from behind.” I believe that “driven from within” better

conveys the way that scientific knowledge is being generated in an outward fashion rather than being
“pulled” toward the truth.
55 B. Wray, Kuhn’s Evolutionary Social Epistemology, Cambridge, UK: Cambridge University Press,

2011, 84.
56 K. Jouni-​Matti, “Revolution as Evolution: The Concept of Evolution in Kuhn’s Philosophy.”
In Kuhn’s The Structure of Scientific Revolutions Revisited, Theodore Arabatzis, ed., 134–​
152, Routledge, New York, 2013, p. 134. See T.S. Kuhn, J. Conant, & J. Haugeland, The Road
Since Structure: Philosophical Essays, 1970-​ 1993 (with an autobiographical interview), 2000,
Chicago: University of Chicago Press. Kuhn’s first thoughts on epistemology based on evolutionary
lines appear at the end of The Structure of Scientific Revolutions, 1962, Chicago: University of Chicago
Press, 169–​172.
 Nicholson’s Proto-Element Theory 127

Whereas Kuhn did not initially have an explanation for the success of sci-
ence, he later proposed that scientific specialization was the missing factor. Kuhn
argued that just as biological evolution leads to an increasing variety of species, so
the evolution of science leads to an increasing variety of scientific sub-​disciplines
and specializations.57 For Kuhn science is a complex social activity, and the unit
of explanation is the group, not the individual scientists. This resonates with my
own view that I alluded to earlier that the growth of science is not successfully
tracked by considering individual scientists or individual theories. Kuhn asks
us to judge changes in theory from the perspective of the research community
rather than the individual scientists involved. Both early converts to a theory and
holdouts aid the community in making the rational choice between competing
theories. This, I suggest, is how the work of Nicholson should be viewed, namely
as a step toward the new theory in the context of Bohr and others.58
However, there is another respect in which I would want to qualify this view to
also reaffirm the importance of change on the level of individual scientists rather
than change predominantly within social groups. Kuhn’s evolutionary account
of the growth of science appears to be somewhat half-​hearted. He does not say
very much on the mechanism of evolutionary change in scientific theories except
for his talk of specialization. Kuhn believes that the development of new sub-​
disciplines such as the emergence of biochemistry from chemistry, for example,
is analogous to the evolution of new species which become incapable of mating
with members of the species from which they have evolved. Similarly Kuhn
holds that the members of the new scientific sub-​discipline, which has branched
off from an older one, are unable to communicate with members of the mother
discipline. Kuhn even makes a virtue of this reconceptualized incommensura-
bility in suggesting that such isolation allows for a greater development within a
particular sub-​discipline.
While authors like Renzi have criticized Kuhn’s evolutionary analogy, Raydon
and Hoyningen-​Huene defend him, in suggesting that Kuhn did not intend
his analogy to be taken literally. Be that as it may, I do intend the evolutionary
analogy to be literal. Since the evolution of scientific theories is part of an under-
lying biological evolution of the human species I do not regard this suggestion to
be too far-​fetched, even though these evolutionary processes may be occurring
at levels that are far removed from each other.

57 Out of physics and chemistry there emerges physical chemistry. Biology and chemistry give rise

to biochemistry. Biochemistry in turn gives rise to physical-​biochemistry.

58 This project has shown me that contrary to what I believed for about 25 years, there is much

merit in Kuhn’s work. I had been too stuck on the cartoon Kuhn (the best-​seller Kuhn) who is sup-
posed to deny progress and who is often taken to be at the root of all evil, such as science wars and the
sociological turn—​but only because I arrived at the idea of an evolutionary epistemology through my
own work in asking how a “wrong” theory can be so successful in so many cases.
128 Historical Cases for the Debate

In the case of the evolution of organisms, modern biology has taught us that
the underlying mechanism is one of random mutation on the level of errors that
occur in the copying of DNA sequences. The more accurate biological analogy
for the development of scientific theories would seem to be for the “mutation”
of ideas to occur in the minds of individual scientists such as Bohr or Nicholson.
The unit involved in the evolution of scientific theories would therefore be indi-
vidual scientists and not the social group that scientists belong to. Here then is
where I depart from Kuhn’s evolutionary epistemology. In my account the evolu-
tion of scientific theories takes place in a literally biological sense and is mainly
situated at the level of individuals rather than social groups.

6.10 Can Kuhn Have it Both Ways?

Can revolution coexist with evolution in science as Kuhn seems to believe? In

the later Kuhn, revolutions are no longer paradigm changes. They are now taxo-
nomic or lexical changes, thus raising the question of whether the revolutions he
gave as examples no longer count as revolutions.59

The early Kuhn’s wholesale and psychologically drastic revolution becomes a

gradual and piecemeal communitarian evolution in the later Kuhn, something
that may show simultaneous continuity and discontinuity between prerevolu-
tionary and revolutionary stages.60

Brad Wray has claimed that revolutions are essential to Kuhn because they
are incompatible with the view that scientific knowledge is cumulative and that
scientists are constantly marching ever closer to the truth. I must say that I dis-
agree with this view. Scientists may not be moving toward a fixed truth, but the
development may still be gradual and not revolutionary. After all, biological ev-
olution is not teleological but is generally thought to be gradual (pace Stephen
Jay Gould et al.). To me the main insight from Kuhn is his evolutionary episte-
mology not his notion of discontinuous theory-​change.
For example, why consider the quantum revolution to have ended in 1912 as
Kuhn does? Surely an equally important revolution due to Bohr began around
this time of 1912. Or was the true revolution the coming of quantum mechanics à
la Heisenberg and Schrödinger in 1925–​26 or maybe QED sometime later in the

59 A quick reply to this question is that Kuhn gives very few examples of what he considers to be

scientific revolutions in the sense of lexically driven revolutions.

60 J.M. Kuukkanen, “Revolution as Evolution.” In, Kuhn’s The Structure of Scientific Revolutions

Revisited, eds., V. Kindi and T. Abratzis, Routledge, London 2012.

 Nicholson’s Proto-Element Theory 129

late 1940s or QCD in the 1970s? Or could it be that there are so many revolutions
that the very concept ceases to be helpful?

6.11 Conclusions

Nicholson is somewhat neglected in the history of science, but I don’t believe it

is because he was simply wrong in many of his basic assumptions.61 In spite of
holding some assumptions concerning the structure of the atom that were sub-
sequently abandoned, Nicholson was still able to make a number of highly suc-
cessful accommodations of known data and predictions of completely unknown
data. Furthermore, Nicholson’s idea of the quantization of angular momentum
was key to Bohr’s subsequent progress in the development of atomic physics.
Nicholson was part of the organic manner in which science evolves or, in this
case, the way that atomic physics evolved. He was an important “missing link”
between the old classical physics and the new quantum theory and the way that it
was applied to the atom.
Needless to say, if Nicholson had not been the first to propose the quantiza-
tion of angular momentum somebody else would probably have done so. I am
not trying to rehabilitate Nicholson’s role, but merely wishing to highlight the
crucial and catalytic role that is often played by the “little people” in science.
Moreover, it is quite conceivable that the history of atomic physics might have
taken a different path, perhaps one not involving the quantization of angular
momentum. The fact remains that it did, and that Nicholson played an unde-
niable role in what actually took place. My main point is to try to highlight the
organic and evolutionary way in which science develops and that it is only in
retrospect that priority is attributed to certain contributors. Given our limita-
tions in attempting to reconstruct such an organic and interconnected growth
process, it is hardly surprising that we tend to simplify the story by latching
on to the leading players in any particular scientific episode. As the reader will
have noticed, I am not unduly puzzled by what may appear to be inconsistent
theories and, in the case I have examined, even some bizarre theories can lead
to scientific progress. If one accepts the notion that scientific work is at root a
collective exercise even if not literally so, then inconsistencies can be thought
of as temporary road-​blocks which delay progress but which get ironed out in
the longue durée.

61 I am speaking mainly of popular science and science textbooks. Professional historians of sci-

ence know better of course.

 7
Selective Scientific Realism
and Truth-​Transfer in Theories
of Molecular Structure
Amanda J. Nichols1 and Myron A. Penner2

1Division of Natural and Health Sciences

Oklahoma Christian University

[email protected]
2Department of Philosophy

Trinity Western University

[email protected]

7.1 Introduction

According to scientific realists, the predictive success of mature theories

provides a strong epistemic basis for thinking that such theories are approx-
imately true. However, we know that many theories once regarded as well
confirmed and predictively successful were eventually replaced with suc-
cessor theories, and some claim this undermines the epistemic confidence we
should have in the approximate truth of current science. Selective scientific
realists in turn argue that if one can show that the predictive success of some
rejected theory T is a function of theoretical claims consistent with current
science, then T’s failure doesn’t undermine the claim that current successful
theories are approximately true. As such, Selective Scientific Realism (SSR)
can be tested through historical examples. Showing that the predictive suc-
cess of a failed theory is the result of theoretical features later rejected provides
a counterexample to SSR. Conversely, SSR is supported if its explanation of
the predictive success of failed theories—​namely that the factors which lead to
predictive success in the failed theory survive in the successor theory—​is able
to handle a wide array of historical cases.
In what follows, we look at one such historical case not currently discussed
in the literature on SSR: theoretical advances in understanding molecular

Amanda J. Nichols and Myron A. Penner, Selective Scientific Realism and Truth-​Transfer in Theories of Molecular
Structure In: Contemporary Scientific Realism. Edited by: Timothy D. Lyons and Peter Vickers, Oxford University Press.
© Oxford University Press 2021. DOI: 10.1093/​oso/​9780190946814.003.0007
 Selective Scientific Realism and Truth-Transfer 131

structures at the turn of the 20th century which resulted from the Blomstrand-​
Jørgensen/​Werner Debate about the structure of cobalt complexes. Both the
Blomstrand-​Jørgensen chain theory and Werner’s coordination theory, which
ultimately replaced it, make predictions about the number of ions that will be
dissociated when the molecule undergoes a precipitation reaction. While chain
theory makes correct predictions in many cases, Werner’s coordination theory
correctly predicted the number of ions in cases where chain theory failed. If the
predictive success of chain theory depended on features that were later rejected
in coordination theory, we’d have a case which undermines SSR’s explanation of
truth transfer between failed theories and their successors. Conversely, the rise
of coordination theory supports SSR if the features of chain theory that resulted
in its predictive successes were also a part of coordination theory. It turns out the
latter is the case. We conclude by applying the lessons from the debate about the
structure of cobalt complexes to philosophical debate about scientific realism.
In brief, the chain theory utilized by Jørgensen involved three rules for mod-
eling the structure of inorganic compounds. Jørgensen applied this framework
with some experimental success: his models predicted certain outcomes of ion
dissociation which were confirmed in a series of precipitation experiments.
However, there were some experiments in which Jørgensen’s model failed to
accurately predict ion dissociation. During the same time period, Werner
was developing a geometric model for cobalt complexes much different from
Jørgensen’s. It turns out that Werner’s model was able to succeed experimentally
where Jørgensen’s had failed. This led, ultimately, to Blomstrand-​Jørgensen chain
theory being replaced by Werner’s coordination theory.
Those who are skeptical about whether experimental success indicates the
truth of scientific theories might look at this historical episode as one which
justifies such skepticism. After all, in retrospect we can see that Jørgensen
achieved a degree of experimental success even though he was mistaken
about the structure of the cobalt complexes in question. Why should we
think that the experimental success of contemporary scientific theories
is in virtue of their truth, when experimental success obviously does not
guarantee truth?
However, we think that for this particular historical episode, SSR provides the
correct historical analysis. It turns out that the factors which led to Jørgensen’s
predictive success are also features of Werner’s coordination theory that ulti-
mately replaced Jørgensen’s chain theory. Jørgensen’s predictive success was
in virtue of him latching on to certain truths about how cobalt complexes are
structured. We further argue that the lessons from this historical case can illumi-
nate how two contemporary objections to realism—​P. Kyle Stanford’s Problem
of Unconceived Alternatives and Timothy D. Lyons’ pessimistic modus tollens
argument—​fall short as arguments against realism.
132 Historical Cases for the Debate

7.2 Molecular Structures and 19th Century Chemistry

By the middle of the 19th century, chemists were routinely investigating both
physical and chemical properties of different compounds as well as identifying
ratios between elements in different compounds. Masses for different elements
were well established by this time, which meant that chemists could carefully
weigh samples before and after experiments involving a reaction and isolation
of a compound in order to determine the loss or gain of mass. An active area
of inquiry was molecular structure and trying to determine how atoms are ar-
ranged and connected to each other. Since J. J. Thomson didn’t discover electrons
until 1897, the modern theory of molecular bonds consisting of electrons had
not yet emerged.1 While chemists at this time did not have a theory explaining
how atoms connect, there was consensus that atoms connected in some way to
form molecules. Moreover, chemists were drawing connections from the phys-
ical and chemical properties to structural features of the molecule, primarily in
organic chemistry because of the work being done with carbon.2 It was discov-
ered that the way atoms were connected to one another seemed important, and
the arrangement that atoms have within space—​their molecular geometry—​
also seemed to affect chemical and physical properties. For example, Jacobus
Henricus van’t Hoff ’s work demonstrated that there are compounds that have
the same chemical formula with the same connectivity between atoms, but their
three-​dimensional orientations differ, making the compounds behave differently
in polarized light. These optically active compounds introduced the idea of the
geometry of a molecule.3 Molecules were discovered to be not mere planar enti-
ties, but rather three-​dimensional structures where each constituent atom’s ori-
entation in space can change the structure of the whole molecule.
Chemists had an early idea of valence, which can be understood as the af-
finity an atom has to combine with other atoms. Valence is expressed as a quan-
tity reflecting the number of bonds an atom may have. For instance, carbon is
said to have a valence of 4 (tetravalence) because it has four connecting points.
Hydrogen is univalent because it has one connecting point. Therefore, one
carbon atom will combine with four hydrogen atoms forming the compound
methane (CH4) (Figure 7.1). How elements bond to form compounds is a func-
tion of each constituent element’s valence, because not anything can bond to just
anything. Valence helps to predict molecular structure. Drawing on both these
valence patterns and the more developed branch of organic chemistry, Friedrich
August Kekulé predicted that the arrangement of atoms in molecules could be

1 See Douglas and McDaniel (1965: 338).

2 See Day and Selbin (1962: 183).
3 See Moore (1939: 294–​295).
 Selective Scientific Realism and Truth-Transfer 133

C H
H
H

Figure 7.1 Methane (CH4).

H H H H H H H H

H C C C C C C C C H

H H H H H H H H

Figure 7.2 Octane (C8H18).

known. In 1858, he claimed that some carbon compounds consist of hydro-

carbon chains: carbon atoms are connected together with a certain number of
hydrogen atoms on each carbon because each carbon atom is tetravalent.4 Basic
organic molecules like octane follow this chain-​like structure (Figure 7.2).

7.3 Cobalt Complexes: Chain-​Like Structures?

Christian Wilhelm Blomstrand was an inorganic chemist who worked with

metal ammines, specifically compounds that had a metal combined with am-
monia molecules and halogen atoms, such as chlorides. In 1871, Blomstrand
proposed that ammonia molecules are structured in chains like hydrocarbons
where ammines are composed of a metal atom, a chain of ammonia molecules,
and a certain number of chlorides. Drawing on the bonding principles of or-
ganic chemistry, Blomstrand assumed that the whole metal ammine would
have a chain-​like structure. But how is the metal ammine “chain” put to-
gether? Depending on the metal’s valence, Blomstrand reasoned that a certain
number of ammonia molecules could form a chain off of the metal. Chlorides
would then be connected to either the ammonia chain or directly to the metal
itself. Blomstrand performed precipitation experiments in order to figure out
whether the chlorides were bonded either to the metal or to the ammonia

4 See Farber (1952: 164–​165).

134 Historical Cases for the Debate

chain.5 He thought that if a chloride came off the compound easily to react
with silver nitrate (and precipitate silver chloride as a product) it was because
the chloride was directly connected to the end of the ammonia chain, not the
metal.6
Sophus Mads Jørgensen continued the work with metal ammines in the
1890s, specifically with cobalt ammine complexes. A cobalt ammine complex
is a molecule made up of the metal cobalt, ammonia groups, and chlorides.
Jørgensen followed three bonding rules for structures when theorizing the
structure of these cobalt complexes. First, according to the valence rule,
bonding is constrained by an element’s valence. Thus, because the metal co-
balt is trivalent, a cobalt atom will have three connecting points. Second, ac-
cording to the dissociation rule (established from lab experiments involving
precipitation reactions with silver nitrate), chloride dissociation, that is, the
separation of chloride atoms from the metal ammine, will not occur if chloride
is directly bonded to metal. And third, according to the chain rule, the en-
tire structure of the compound would be chain-​like, and thus, ammine chains
will form off a cobalt atom. While we are not claiming that Jørgensen explic-
itly formulated these rules in his model for cobalt complexes, it is clear from
his dependence on Blomstrand’s work and his own published research that
Jørgensen followed what we are calling the valence, dissociation, and chain
rules. Much of Jørgensen’s work centered on the compounds that are known
today as coordination complexes, and he was one of the first chemists to
provide experimental evidence that the cobalt ammine complexes were not
dimers, contrary to what Blomstrand hypothesized. Jørgensen’s 1890 paper
outlines this evidence and treats cobalt as trivalent, demonstrating that
Jørgensen is following the valence rule.7 Kauffman points out that in 1887
Jørgensen published his work using the precipitation experiment (discussed
in the next section), reasoning that if the chloride is directly bonded to a
metal atom, the chloride will not dissociate. This is what we have dubbed the
dissociation rule.8 Continuing the ideas originally proposed by Blomstrand,
Jørgensen thought of the ammines to be in a chain structure (the chain
rule), and Werner responds to these ideas directly in his paper proposing
his new coordination theory.9,10 This is relevant to what we later will argue

5 See Kauffman (1965).

6 See Meissler and Tarr (2011: 322).
7 See Jørgensen (1890: 429) cited in Kauffman (1965).
8 See Jørgensen (1887: 417) cited in Kauffman (1965).
9 See Kauffmann (1965: 186).
10 See Werner (1893: 267–​330) translated by Kauffman (1968: 16–​20).
 Selective Scientific Realism and Truth-Transfer 135

H Cl
H
N H H H H H H H H
H
Co N N N Cl
N
H
N H H H H
H
Cl
H

Figure 7.3 Pre-​experiment of hexammine complex: Jørgensen’s proposed structure.

Note: Here, Jørgensen’s work involved cobalt(III) hexammine chloride [Co(NH3)6]Cl3 complexes.

in establishing both continuity and discontinuity between Blomstrand-​

Jørgensen chain theory and Werner’s coordination theory.

7.4 The Precipitation Experiment

Applying these three rules results in the Blomstrand-​Jørgensen chain theory of cobalt
ammine complexes, which includes models for structures of different compounds;
the accuracy of these proposed structures can be tested in precipitation experiments.
During a precipitation experiment, cobalt complex molecules split into ions (separated
charged particles). The experiment begins by mixing a cobalt complex solution with a
silver nitrate solution. A solid (the precipitate) forms by chloride separating from the
cobalt complex and bonding with the silver; the resulting precipitate is silver chloride.
Jørgensen’s three bonding rules make predictions with respect to the number of ions
that will result from a precipitate experiment.
An early example from Jørgensen is his postulated chain structure for co-
balt (III) hexammine chloride, further referred to as the hexammine complex.
Jørgensen’s structure of this cobalt complex, following the valence rule, consists of
a cobalt center with three groups connected to it (Figure 7.3). The three chloride
atoms are connected to ammine groups (NH3Cl), but none of the chloride atoms
are attached directly to the cobalt atom. Following the chain rule, the third group
is a chain of ammines with a chloride on the end. The dissociation rule predicted
that there should be four ions produced in the precipitation experiment: all three
chlorides will dissociate because they are each connected to an ammine group, not
the cobalt atom. Three chlorides plus the leftover part of the molecule would result
in four ions. That prediction was supported by measurement when four ions were
confirmed after the hexammine complex underwent a precipitation experiment
(Figure 7.4).
136 Historical Cases for the Debate

H Cl−

H
N H H H H H H H
H
H
Co+3
N N N N Cl−
H
N H H H H
H
H Cl−

Figure 7.4 Post-​experiment of hexammine complex. The precipitation experiment

confirmed Jørgensen’s structure: 4 ions result.

Alfred Werner, a younger chemist only 26 years old, responded to Jørgensen’s

work with a different possible structure for the hexammine complex: a geomet-
rically symmetrical coordination compound.11 Werner proposed two unique
features for his model: (1) an octahedral arrangement around the cobalt center
and (2) an additional type of valence called secondary valence where the metal
center is directly attached to six ammine groups instead of three chains—​making
an inner sphere. He considered the inner sphere of the compound as “a discrete
unit, a complex cation.”12 A cation is a positively charged ion, and a charged ion
can be any charged molecular or atomic entity. These six attachments between
the cobalt atom and an ammine group are examples of the additional valence that
Werner suggests. This additional valence is referred to as secondary valence or co-
ordination number. Werner, like Jørgensen, followed the valence rule in his model
as well, according to which bonding is constrained by an element’s valence. Because
cobalt is trivalent, three chlorides are needed to fulfill the valence rule.13 However,
in Werner’s model, cobalt’s trivalence is structured where the three chlorides are at-
tached to the inner sphere, but not attached in the same way as the ammine groups;
he described these chlorides as being “at large.” Following the dissociation rule,
Werner’s proposed structure predicted four ions: chlorides would only dissociate if
they were not directly connected to the cobalt atom (Figure 7.5). The experimental
work supported this structure as well because all three chlorides can be precipi-
tated off, also resulting in a total of four ions (Figure 7.6).
Note that at this time in the history of chemistry, there are two mutually exclu-
sive models for the structure of this particular hexammine complex—​and thus at

11 See Werner (1893: 267–​330) translated by Kauffman (1968: 48).

12 See (Kauffman 1965: 188).
13 This primary type of valence is later understood to be an ionic bond where the positive part, the

cobalt ion, is attracting the negative parts, the chlorides. Ammines are neutral in charge, and their
attachment to the cobalt is not like the primary valence.
 Selective Scientific Realism and Truth-Transfer 137

Cl–
H 3+
H H
H H
H N H
H N N
H
H Cl–
Co H
H N
N H
H N
H
H H
H

Cl–

Figure 7.5 Pre-​experiment of hexammine complex: Werner’s proposed structure.

3+
H
H H
H H
H N Cl–
H
H N N
H Cl–
H Co H
H N Cl–
N H
H N
H
H H
H

Figure 7.6 Post-​experiment of hexammine complex. The precipitation experiment

confirmed Werner’s structure: 4 ions result.

least one model must be mistaken. But note also, that both models at this stage
are making successful predictions—​entailing results that are subsequently con-
firmed by experiment. And it’s this phenomenon—​the phenomenon of incorrect
models making correct predictions—​that seems to provide some prima facie jus-
tification for skepticism about a realist interpretation of current science. “Why,”
so the thinking goes, “should we be confident that the successful predictions of
current science are a marker of truth, if we have all these examples from history
of the predictive success of failed models?” In fact, one might extend this worry
about the shortcomings of current science to include the potential superiority of
“unconceived alternatives” to current theories.14 We’ll look at specific arguments

14 See Stanford (2006).

138 Historical Cases for the Debate

Cl H H H H H H

Co N N N Cl

Cl H H H

Figure 7.7 Pre-​experiment of triammine complex: Jørgensen’s proposed structure.

Note: Here Jørgensen’s work involved cobalt(III) triammine chloride [Co(NH3)3Cl3] complexes.

Cl H H H H H
H

Co1+ N N N Cl–

Cl H H H

Figure 7.8 Prediction of triammine complex after the precipitation experiment.

Jørgensen’s proposed structure predicted 2 ions, but no ions resulted from the
precipitation experiment.
Note: Jørgensen still thought of the cobalt as trivalent, but in terms of charge balance, after one
chloride dissociates, that leaves a positive charge on the leftover part of the complex.

against scientific realism in a later section of the paper and address these lines
of thinking more directly. However, it’s worth raising at this juncture in order to
remind ourselves what’s at stake, philosophically, by examining the 19th century
debate over the structure of cobalt complexes.
During this same time through 1899, Jørgensen and Werner worked
separately in synthesizing and characterizing other metal ammines. Like
experiments with hexammines, work with penta-​and tetra-​ammines yielded
experimental data that supported both Jørgensen’s chain theory and Werner’s
coordination theory. However, it was the triammines that forced the triumph
of one theory over the other.15 The metal triammine has a cobalt atom, three
ammine groups, and three chlorides. Jørgensen predicted a chain structure
for the triammine complex with a chain of three ammines: one chloride on
the end of the ammine group, and two chlorides directly attached to the co-
balt atom (Figure 7.7). Again following the dissociation rule, his structure
predicted two ions after precipitation: the chloride off the ammine chain
and then the remaining complex (Figure 7.8). Experimental work did not

15 See Kauffman (1965).

 Selective Scientific Realism and Truth-Transfer 139

H
H
Cl
Cl
H N
H Co H
H N
N H
H Cl
H

Figure 7.9 Pre-​experiment and post-​experiment of triammine complex

support Jørgensen’s chain structure. It turns out that zero ions precipitated
during the experiment.
However, Werner’s proposed structure for the triammine complex success-
fully predicted the number of ions for the precipitation experiment. Keeping
with the same two features of the hexammine model, Werner implemented an
octahedral, symmetric shape and secondary valence in his triammine model.
The cobalt atom is attached to three separate ammine groups and three separate
chlorides. The primary valence of cobalt is satisfied with the cobalt atom being
connected to three chlorides (Figure 7.9). Because the chlorides are directly at-
tached to the cobalt, they should not separate during precipitation. Therefore, no
ions are predicted, and precipitation experiments confirmed this prediction: no
ions were measured.
While we don’t have space for an in-​depth discussion of the epistemic status of
models with respect to the physical systems being represented, a few explanatory
comments are in order. Margaret Morrison notes the “problem of inconsistent
models” arises when multiple inconsistent models provide useful information
about a physical system, and this problem undermines the degree to which one
should give a realist interpretation of a model—​particularly if one lacks the
ability to tweak models in ways that would allow for empirical confirmation/​
disconfirmation. Says Morrison, “In these contexts we usually have no way to
determine which of the many contradictory models is the more faithful repre-
sentation, especially if each is able to generate accurate predictions for a certain
class of phenomena.”16 This is exactly the situation during the early experimenta-
tion and modeling of cobalt complexes, when both Jørgensen’s chain theory and
Werner’s coordination complex model made predictions that were supported by
precipitation experiments.
However, note the conditions that Stathis Psillos presents concerning when a
model M can be interpreted as a realist depiction of a physical system X:

16 See Morrison (2011: 343).

140 Historical Cases for the Debate

I think one can, in principle, take a realist stance towards particular models. For
although scientists do not start off with the assumption that a particular model
gives a literal description of a physical system X, there may be circumstances
in which a model M of X offers an adequate representation of X. These
circumstances relate to the accuracy with which a given model represents the
target system . . . . Amassing more evidence, such that novel correct predictions
for X derived from M, may be enough to show that M represents X correctly. In
sum, taking a realist attitude towards a particular model is a matter of having
evidence warranting the belief that this model gives us an accurate representa-
tion of an otherwise unknown physical system in all, or in most, causally rele-
vant respects.17

The conditions specified by Psillos for supporting a realist interpretation of a

model M were met by Werner’s model, specifically with respect to Werner cor-
rectly predicting ion precipitation in the triammine complex experiments.
Our survey of work on cobalt ammine complexes in 19th century chemistry
reviewed a series of experiments that led to the replacement of the Blomstrand-​
Jørgensen chain theory by Werner’s coordination theory. While both theories were
supported by precipitation experiments involving hexa-​, penta-​, and tetraammine
complexes, only one theory was supported after the precipitation experiment done
on the triammine complex. Historians of chemistry point out that Jørgensen resisted
coordination theory even after the triammine complex experiments. According to
Kauffman, Jørgensen questioned Werner’s experimental methods viewing coordi-
nation theory “as an ad hoc explanation insufficiently supported by experimental
evidence.”18 Kauffman further states, “Jørgensen always preferred facts to bold
hypotheses, and his controversy with Werner clearly reflects his attitude.”19 While
Jørgensen ultimately admitted defeat when Werner was able to prepare two isomers
of a different complex that could not be explained by chain theory,20,21 Werner’s co-
ordination theory was not accepted widely until after Thomson’s discovery of the
electron and a more robust electronic theory of valency was able to further explain
the different types of valency proposed by Werner.22

17 See Psillos (1999: 144).

18 See Kauffman (1965: 186).
19 See Kauffman (1965: 185). There are larger issues at stake here—​beyond the scope of this paper

to discuss—​concerning the practice of scientific research and the nature of theoretical justification.
For example, Day and Selbin (1962: 264) suggest “Werner had no theoretical justification for his two
types of valency.” Werner’s first paper (1893) proposing his theory outlines his justifications.
20 See Meissler and Tarr (2011: 323).
21 The widely-​accepted idea that Jørgensen admitted defeat to Werner is debatable. Jørgensen

never formally stated defeat or acceptance of coordination theory, rather, it seems he ignored the
controversy after 1899. S.P.L. Sørensen reported in his eulogy that Jørgensen admitted defeat in a
conversation in 1907. See Kragh (1997).
22 See Day and Selbin (1962: 264).
 Selective Scientific Realism and Truth-Transfer 141

7.5 Implications for Scientific Realism

In this section, we apply lessons from the scientific debate on the structure of
cobalt complexes to the philosophical debate about whether some version of sci-
entific realism is justified. Scientific realism reflects a certain cluster of intuitions
and beliefs about the nature and purpose of science. At the heart of the realist
sensibility is the notion that the success of scientific research and theory de-
velopment consists of a theory explaining the data and making accurate novel
predictions. Realists view the fit between data and theory as justifying an in-
ference to the truth of theoretical statements. As such, science is normed and
constrained by the mind-​independent reality it seeks to map.
Although versions of scientific realism abound, we adopt as a starting point
Stathis Psillos’s elegant and comprehensive description of scientific realism as
a conjunction of three stances. According to Psillos, the realist adopts a meta-
physical stance that the world has a definite, mind-​independent, natural kind-​
like structure, a semantic stance that theories are truth-​governed descriptions
in which putative referring terms do, in fact, refer, even to unobservable entities,
and an epistemological stance according to which mature and predictively suc-
cessful theories are held to be approximately true.23 Each stance has its philo-
sophical detractors.
Post-​structuralists of various sorts will deny that the world has any accessible
mind-​independent structure. Non-​realist philosophers of science will deny that
successful science requires accepting that theoretical terms in scientific theories
are successful referring terms, and will further deny that predictive success is a
mark of, or best explained by, truth.

7.5.1 The Problem of Unconceived Alternatives

One of the most prominent arguments against the claim that predictive success
is a marker of truth comes from P. Kyle Stanford’s argument based on the possi-
bility of unconceived alternatives. According to Stanford,

[T]‌he most troubling historical pattern for scientific realism is the repeated
and demonstrable failure of scientists and scientific communities throughout
the historical record to even conceive of scientifically serious alternatives to the
theories they embraced on the strength of a given body of evidence, that were

23 See Psillos (1999: xix). More specifically our version of realism is best described as a type of di-

vide et impera realism; for the purpose of this paper we remain open to various options concerning
what, exactly, constitutes the truth-​conducive constituents of scientific theories.
142 Historical Cases for the Debate

nonetheless also well-​confirmed by that same body of evidence. This “Problem

of Unconceived Alternatives” arises because the inferential engine of much
fundamental theoretical science is essentially eliminative in character, pro-
posing and testing candidate hypotheses and then selecting from among them
the one best supported by the evidence as that in which we should invest our
credence . . . But the reliability of such inferences requires that particular ep-
istemic conditions be satisfied. One such condition is that we must have all of
the likely or plausible alternative possibilities in view before proceeding to em-
brace the winner of such an eliminative competition as the truth of the matter.24

Stanford’s idea can be explicated as follows. Suppose that for some body of evi-
dence E, there are n number of theories being considered by the relevant scien-
tific community to account for E. Suppose further that through a combination of
testing and theoretical work, consensus emerges within the scientific commu-
nity about which of the candidate theories can be eliminated and about which
theory among the remaining candidates best explains E. The problem, says
Stanford, is that we have good reason based on the history of science to think
that the epistemic control condition for determining when eliminative inference
is reliable—​namely, the condition that we must have all of the likely or plausible
alternative possibilities in view before proceeding to embrace the winner—​is never
satisfied. Stanford continues:

[T]‌he historical record of scientific inquiry itself gives us compelling empir-

ical grounds for doubting whether [the epistemic control condition] is typi-
cally satisfied when we formulate and test what we might call “foundational”
scientific theories concerning the constitution of entities, the underlying causal
mechanisms, and the dynamical principles at work in otherwise inaccessible
domains of nature . . . . What the historical record reveals instead . . . is a ro-
bust pattern of theoretical succession in which such foundational theories are
accepted on the strength of a given body of evidence, only to be ultimately
superseded by alternatives that were also well-​confirmed by that evidence, but
nonetheless simply remained unconceived at the time of the earlier theory’s
acceptance.25

Stanford’s Problem of Unconceived Alternatives has a direct bearing on our anal-

ysis of the Jørgensen/​Werner debate about the structure of cobalt complexes.
Recall that we’re defending a selective scientific realist interpretation of the tri-
umph of Werner’s coordination theory that modeled molecular structures with

24 See Stanford (2018: 213).

25 Ibid.
 Selective Scientific Realism and Truth-Transfer 143

symmetrical geometric shapes over Jørgensen’s theory that modeled molecular

structures in chain-​like shapes. Jørgensen’s theory, though ultimately rejected,
did enjoy a certain amount of predictive success—​a phenomenon that skeptics
about whether predictively successful science is a marker of truth can point to
as justification for such skepticism. However, we’ve argued that a better way to
understand the predictive success of Jørgensen’s failed theory is to recognize that
the factors that lead to his theory making successful predictions (what we earlier
called the valence rule and the dissociation rule) are also a part of Werner’s co-
ordination theory. As a result, on a selective realist interpretation, the transition
from Jørgensen’s chain theory to Werner’s coordination theory is an example of
the onward and progressive march of science toward truth. Jørgensen’s theory,
while mistaken in some respects, wasn’t completely mistaken (hence the limited
experimental success he did achieve). Now recall Stanford’s description of elimi-
native inference: “proposing and testing candidate hypotheses and then selecting
from among them the one best supported by the evidence as that in which we
should invest our credence.”26 This type of eliminative inference is an accurate
description of the process by which chemists came, eventually, to reject chain
theory in favor of coordination theory—​the two main theories on the table—​as
experimental evidence continued to mount in favor of Werner’s theory.
But notice that if we take into account Stanford’s Problem of Unconceived
Alternatives, then the eliminative inference undertaken by chemists in the late
19th century is incomplete and necessarily constrained by the conceptual limits
of their time. Nineteenth-​century chemists working in this area were debating
whether cobalt complexes were chain-​like structures or symmetrical geometric
shapes, and the differing proposed structures reflected theoretical views about
valence and the nature of chemical bonds. But according to the Problem of
Unconceived Alternatives, a lesson from the history of science is that we need
to acknowledge the possibility that there exists some alternative theory distinct
from both chain theory and coordination theory—​an alternative such that it
(a) gives a better explanatory framework for understanding the structure of co-
balt complexes, and (b) is conceptually inaccessible to chemists both in the 19th
century and the present day. Thus, even though in this case, selective scientific
realists can tell a plausible story about truth-​transfer from Jørgensen’s theory
to Werner’s, the Problem of Unconceived Alternatives shows that the realist’s
story is incomplete and doesn’t support a realist picture for current theories con-
cerning the structure of cobalt complexes.
Stanford’s significant objection to scientific realism has generated much dis-
cussion in the past decade and raises interesting questions in epistemology in
general, as well as for the epistemology and metaphysics of science. While it’s

26 Ibid.
144 Historical Cases for the Debate

beyond the scope of this paper to address all the nuances of Stanford’s project,
we feel that there are strong defenses that the realist can present in response to
Stanford. We further note that whether or not Stanford’s Problem of Unconceived
Alternatives is a successful objection to scientific realism will depend in large
measure on more fundamental epistemological assumptions, including whether
the logical possibility of an unconceived alternative does in fact constitute a
defeater for evidence one has in hand.
A charitable interpretation of Stanford’s philosophical objection to scientific
realism has an empirical consequence that seems implausible, a consequence
that is part of what we’re calling the No Final Theory Objection to the Problem
of Unconceived Alternatives. Suppose it’s always true, as Stanford maintains,
that the eliminative inference scientists perform at any given time is necessarily
constrained by the inability to consider unconceived alternative theories which
provide a better explanation for the data in question. This can be stated with the
following:

(1) For any current theory T that purports to explain some evidence E, there
is another theory T* such that:
(i) T* is unconceivable at present.
(ii) T and T* are inconsistent—​they cannot both be true. And,
(iii) T* is a better explanation E than T.

But notice a consequence of (1) is that there can be no “final theory,” a term
that can be used to denote different things. Some, following Nobel Laureate
Steven Weinberg, use “final theory” to denote a final physical theory that
provides a unified explanation for gravity, quantum mechanics, and the strong
and weak nuclear forces.27 However, suppose, as suggested by Weinberg him-
self, possessing a final theory that provides a unified explanation of funda-
mental forces might still leave problems about consciousness and mind/​body
interaction unsolved.28 In that case, having a final theory wouldn’t constitute
having a “theory of everything,” if “everything” included consciousness. For
ease of reference, we’ll use final theory to denote the type of unifying theory
that Weinberg has in view—​nothing of consequence in our objection turns
on whether one has a wider scope in view for final theory. Thus, considering
candidates for a final theory as a substitution instance for the schema presented
in (1) it follows that:

27 See Weinberg (1994).

28 Weinberg interviewed by John Horgan, https://​blogs.scientificamerican.com/​cross-​check/​
nobel-​laureate-​steven-​weinberg-​still-​dreams-​of-​final-​theory/​.
 Selective Scientific Realism and Truth-Transfer 145

(2) For any candidate for a final theory F, there will be a theory F* such that:
(i) F* is unconceivable at present.
(ii) F and F* are inconsistent—​they cannot both be true. And,
(iii) F* is a better explanation than F.

Thus, if a final theory is one that provides a complete and accurate unification of
whatever fundamental forces there be, from (1) and (2) it follows that:

(3) There is no final theory.

However, why should we think (3) is true? Why should epistemological consid-
erations based on the history of science tell us whether there is an accurate and
unified way of explaining how fundamental forces interact? Instead, it seems
that (1)–​(3) serve as premises in a reductio against the Problem of Unconceived
Alternatives, for the negation of (3) seems much more plausible—​asserting
(3) commits one to accepting a fundamental inconsistency between physical
reality and the logical possibility of, ultimately, representing physical reality
through comprehensive theory. If Stanford is right, not even an omniscient being
could have a final theory—​which indicates that something is amiss in the initial
premises.
To be sure, the truth of a theory isn’t the only available explanation for a
theory’s predictive success. Suppose a theory T generates a testable observation
statement O and we do in fact observe that O. Observing O certainly doesn’t en-
tail that T is true. But would observing that O provide an epistemologically re-
spectable reason for thinking that T is true, or approximately true? A defeasible
reason, but a reason nonetheless?
Timothy D. Lyons doesn’t think so. In his 2017 “Epistemic selectivity, histor-
ical threats, and the non-​epistemic tenets of scientific realism,” Lyons presents
two arguments against what he terms the realist meta-​hypothesis:

Realist Meta-​Hypothesis: Those constituents that are genuinely deployed in the

derivation of successful novel predictions are approximately true.29

Lyons’s first argument concludes that the meta-​hypothesis is false; we’ve dubbed
it the False Basis Argument, because it employs the observation that successful
predictions are sometimes derived from constituents of theories that are not true.
In those cases, successful predictions are derived from theoretical bases that turn
out to be false, and the successful predictions made by the Blomstrand-​Jørgensen

29 See Lyons (2017).

146 Historical Cases for the Debate

chain theory are examples of this.30 Here’s a slightly modified version of Lyons’s
argument:

7.5.2 The False Basis Argument

1. If the realist meta-​hypothesis were true, then all the constituents gen-
uinely deployed in the derivation of successful novel predictions are
approximately true.
2. However, we do find constituents genuinely deployed toward success that
cannot be approximately true.
3. Therefore, the realist meta-​hypothesis is false.

In the False Basis Argument, much is going to turn on what, precisely, is

meant by “genuine deployment.” Notice that we can distinguish between those
constituents one takes to be relevant in generating an inference or a novel predic-
tion and those constituents that are in fact relevant in generating an inference or
a novel prediction.
Suppose S believes that the conjunction of p, q, and r entails some observa-
tion O. Suppose also that, unknown to S, it is in fact the conjunction p and q that
entails O, and r is an explanatory free-​rider; r is in fact predictively idle. It’s a
good thing, too, because it turns out that r is false and later comes to be rejected.
Now, in S’s mind, she is genuinely deploying each constituent in the conjunction
when deriving O. Deploying r does some psychological work for S, and factors
into the story S tells as to why one should expect O, given p, q, and r. So there is a
psychological sense of “genuine deployment” according to which S is genuinely
deploying r.
But there’s also a more causally robust sense according to which r isn’t being
genuinely deployed in deriving O, for r doesn’t genuinely factor into the actual
causal explanation for O. And as far as the realist meta-​hypothesis is concerned,
from a realist’s perspective, “genuine deployment” should be understood in the
causally robust sense of deployment.

30 One might think that the phenomenon of false theories making successful predictions parallels

what’s going on in Gettier cases in epistemology. In Gettier cases, we have examples of justified
true beliefs, but such that the truth of the belief is a “happy accident.” For example, the successful
predictions made by Jørgensen might be Gettier-​like in that regard because it’s merely a “happy acci-
dent” that the false theory correctly predicted the number of ions that were dissociated in precipita-
tion experiments. However, the selective scientific realist wants to say that that is not so—​rather, what
led to the predictive successes of Jørgensen was the fact that he followed accurate rules that survived
into the successor theory. Thus, while there’s a superficial similarity to Gettier cases (in the appear-
ance of truth being a function of “mere coincidence”), the similarity disappears on closer inspection.
 Selective Scientific Realism and Truth-Transfer 147

What we mean by “causally robust” is similar to Peter Vickers’s distinction be-

tween “working posits” and “idle posits” that are used within a theory to derive
a novel prediction.31 Vickers’s clear and comprehensive paper looks to flesh out
the general divide et impera position of SSR while taking seriously anti-​realist
criticisms.32 In analyzing historical cases, the selective scientific realist attempts
to determine whether the predictive success of failed theories is a function of
portions of the failed theory which both approximated truth and survived into
successor theories. Vickers notes that when analyzing the role of theory in
making predictions, we can identify Derivation External Posits (DEPs) as those
posits that might “play some role in guiding the thoughts of scientists but do not
deserve realist commitment because they cannot be considered part of the der-
ivation of the prediction in question.”33 Vickers contrasts DEPs with Derivation
Internal Posits (DIPs), where DIPs are the posits the scientist uses to derive the
prediction. However, DIPs can be further subdivided into working posits and
idle posits.34 Idle posits in Vickers’s sense are what we mean by r in our earlier
example, where r is an explanatory free-​rider.
With these distinctions in hand, let’s review the Blomstrand-​Jørgensen/​
Werner debate about the structure of cobalt complexes. In order to do so, it is
helpful to point out that Werner’s theory is still accepted today within the larger
theory of chemical bonding. Though Werner’s notion of secondary valence, the
inner sphere around the cobalt, was required for the derivation of his octahedral
model, there was not a clear understanding of different types of bonding. While
both Jørgensen and Werner observed the dissociation rule, that is, the separa-
tion of chloride atoms from the metal ammine, will not occur if chloride is directly
bonded to metal, they both recognized that this did not fit with observations
about all metal salts. Many, but not all, metal salts dissociate in water. Chemists
during Jørgensen and Werner’s time would probably consider the metal salt, so-
dium chloride (NaCl), as a molecule with chloride directly bonded to the sodium
metal atom. Observations for these types of metal salts showed that the metal
and chloride would readily dissociate. Metal ammines, like the cobalt ammine
complexes Werner and Jørgensen studied, have cobalt, ammines, and chlorides,
and follow the dissociation rule. While this dissociation rule for metal ammines
is still part of the larger theory of chemical bonding today, as Werner points out,
neither he nor Jørgensen had an explanation as to why cobalt ammine complexes
behaved differently than the other type of metal salts.35 Modern theory of

31 See Vickers (2013).

32 The classic presentation of divide et impera is Psillos (1999:108–​114); for an alternative view, see
Lyons (2006).
33 See Vickers (2013).
34 Ibid.
35 See Werner (1893: 267–​330) translated by Kauffman (1968: 16–​17).
148 Historical Cases for the Debate

chemical bonding explains the difference.36 Metal salts that would dissociate
in water are classified as ionic, where opposite charges (e.g. the positive sodium
and negative chloride) attract. Another type of bonding is classified as covalent,
where the electrons in the bond are shared (either equally or unequally), usu-
ally seen in bonding between two nonmetals. Transition metal complexes, like
the cobalt ammine complexes, have both types of bonding. Werner’s secondary
valence, the inner sphere around the cobalt, are covalent bonds whereas the pri-
mary valence (e.g. the chlorides in the outer sphere) are ionic bonds. Therefore,
the chlorides that will dissociate are the same type of bonds seen in other metal
salts: ionic. Ligand Field Theory further expands this theory,37 and one can say
that Werner’s coordination theory is approximately true, exhibiting novel pre-
dictive success, and has been preserved in present-​day chemistry.
Jørgensen was deploying the valence rule, dissociation rule, and the chain rule
in order to, ultimately, make predictions about the number of ions that would
separate during precipitation experiments. In the psychological sense of deploy-
ment, he was genuinely deploying all three rules. But what turns out to be the
case is that the chain rule was an explanatory free-​rider—​it didn’t factor into the
actual causal story concerning the number of ions that separated during the pre-
cipitation experiments. Thus, this particular historical example doesn’t serve as
an example referenced in premise (2) of the False Basis Argument, for the rel-
evant sense of genuine deployment doesn’t apply. Moreover, as surveyed in the
previous paragraph, the successful working posits within Blomstrand-​Jørgensen
chain theory survived into Werner’s model and through to the present day.
We have only looked at one case where it’s plausible to interpret the historical
sequence of events in a way that is friendly to scientific realism. In order to seri-
ously undermine (2) of the False Basis Argument, more work would need to be
done in order to look at the historical cases cited by anti-​realists in support of the
pessimistic induction.38 However, we think that the transition from Blomstrand-​
Jørgensen chain theory to Werner’s geometrical structures and the coordina-
tion theory that support it can be seen as the ever-​increasing glory of science
at work. Blomstrand and Jørgensen were wrong, importantly wrong, about
some things but not everything. And Jørgensen did deploy a false constituent
in generating predictions which were then observed in experiments. However,
when faced with cases where false theoretical constituents lead to successful
predictions, it seems like one is faced with an interpretive choice: either there’s
a truth-​conducive explanation lurking nearby that we don’t have clearly in view,
or there isn’t, in which case it turns out that the successful prediction happens to

36See Meissler and Tarr (2011: ­chapters 3, 9, 10).

37Ligand Field Theory is based upon crystal field theory and molecular orbital theory. Griffith and
Orgel (1957), as cited in Meissler and Tarr (2011), give a complete description of their theory.
38 For a recent list of such historical cases, see Lyons (2017: 3215).
 Selective Scientific Realism and Truth-Transfer 149

be a mere coincidence.39 And if one isn’t inclined to attribute theoretical success

to mere coincidence, it seems perfectly reasonable to look for the more robust,
realistic, truth-​conducive explanation lurking nearby in order to discover those
causally robust features that are actually being deployed in generating the novel
prediction. This is what we have done in the debate over the structure of cobalt
complexes, and it is a strategy that holds much promise in adjudicating other
historical examples of theory change in science and the alleged problem for sci-
entific realists of false constituents generating novel predictions.
Lyons’s second argument concludes that the realist meta-​hypothesis lacks epi-
stemic justification—​it can’t be justifiably believed. Here follows a slightly modi-
fied version of the argument.

7.5.3 The Weak Evidence Argument

4. We are justified in believing the realist meta-​hypothesis (given the evi-

dence) only if we do not have greater evidence for those that oppose it.
5. However, we do have greater evidence for those that oppose it.
6. Therefore, given the evidence, it is not the case that we are justified in
believing the realist’s meta-​hypothesis.

(4) is a plausible epistemological principle which ties epistemic justification to

evidence; while a deeper analysis would want to flesh out notions of evidence,
we’ll grant (4) for the sake of argument. What about (5)?
Lyons considers the following alternatives to the realist meta-​hypothesis,
dubbed by Lyons as ContraSRs, and which we subdivide accordingly:

ContraSR1: Those constituents that are genuinely deployed in the deriva-

tion of successful novel predictions are statistically unlikely to be approx-
imately true.
ContraSR2: Those constituents that are genuinely deployed in the derivation of
successful novel predictions are not-​even-​approximately-​true.

Compare again the skeptical meta-​hypotheses with the realist meta-​hypothesis:

Realist Meta-​Hypothesis: Those constituents that are genuinely deployed in the

derivation of successful novel predictions are approximately true.

39 We prefer “mere coincidence” as opposed to “miracle,” for presumably miracles, if any there be,

wouldn’t be random, accidental occurrences.

150 Historical Cases for the Debate

Why does Lyons think that we have more evidence for the ContraSRs than for
the realist meta-​ hypothesis? Avoiding language of confirmation/​ disconfir-
mation, and opting instead for a hypothesis having “correlatively precise posi-
tive instances,” or “correlatively precise negative instances,” it seems like Lyons
supports (5) with something like the following No Instances Argument.

7.5.4 The No Instances Argument

7. A predictively successful theory that is not approximately true would be a

correlatively precise positive instance of the ContraSRs.
8. We have a list of predictively successful theories that are not
approximately true.
9. Thus, we have correlatively precise positive instances of the ContraSRs.
10. We have no correlatively precise negative instances of the ContraSRs.
11. Moreover, we have no correlatively precise positive instances of the realist
meta-​hypothesis.
5. Thus, we have greater evidence for the ContraSRs.40

The controversial premises in the No Instances Argument are (8) and (11).
According to (8), we have a list of predictively successful theories that are not
approximately true. That is, we have a list of theories where false constituents
were genuinely deployed in deriving novel predictions. However, what our pre-
ceding discussion about genuine deployment in the Blomstrand-​Jørgensen/​
Werner case has shown, is that there’s a difference between a psychological sense
of genuine deployment and a causally robust sense of genuine deployment that
involves the working posits within a theory. Recall that in cases where it appears
that false constituents were genuinely deployed to generate predictive success,
one is faced with an interpretive choice between explaining the predictive suc-
cess in terms of approximate truth or mere coincidence. Either the predictive
success is accounted for because there’s a truth-​conducive explanation lurking
nearby, or there isn’t (in which case the predictive success is a mere coincidence).
Certainly more investigative work would need to be done, along the lines of what
we’ve done with the story of cobalt complexes. However, if a long string of mere
coincidences is itself unlikely, than we have reason to think that the evidence for
(8) is not as strong as it’s taken by Lyons to be.41

40This is just a slightly differently worded version of (5) listed earlier.

41Thanks to Peter Vickers for reminding us that realists can still acknowledge the existence of
minor coincidences that occur on occasion. Moreover, acknowledging minor coincidences does not
undermine the overall force of the claim that empirical success is a marker of theoretical truth.
 Selective Scientific Realism and Truth-Transfer 151

What about (11)? In setting up the Weak Evidence Argument, Lyons states the
following:

Moreover, without granting victory in advance to the realist’s meta-​hypothesis,

we have no data that stand as correlatively precise negative instances of a
ContraSR. By contrast, the realist’s meta-​hypothesis has no correlatively pre-
cise positive instances without already establishing it . . . . I emphasize that
these restricted tasks of, first, tallying such positive/​negative instances for the
ContraSRs and the realist meta-​hypothesis and, second, recognizing the evi-
dential superiority of the former over the latter involve no inference to, accept-
ance of, let alone belief in, any such empirically ampliative meta-​hypothesis
about successful theories.42

He further states:

First, it is admittedly logical considerations that reveal that, on the one hand,
instances of the correlates of the ContraSRs can be secured without presup-
posing the truth of the ContraSRs, and, on the other, that such a feature is not
shared by the realist meta-​hypothesis.43

Lyons’s worry here seems to be that there is no non-​circular way of assessing pu-
tative positive correlates of the realist hypothesis which wouldn’t already presup-
pose the truth of the realist hypothesis. As such, those cases can’t really count as
evidence in favor of the hypothesis.
But can the realist hypothesis really be defeated so easily by a priori consid-
erations in epistemology? Here the realist meta-​hypothesis is in a position sim-
ilar to analogous positions about the reliability of one’s cognitive faculties—​sense
perception, say. Consider the following realist hypothesis about sense perception:

Sense Perception Meta-​Hypothesis: In the typical case, sense perception is

generally reliable.

What possible evidence could one provide in support of the Sense Perception
Meta-​ Hypothesis that doesn’t assume the reliability of sense perception?
Not much. But here we’ll just point out that a number of philosophers, recog-
nizing the challenge of finding non-​circular justification for our belief-​forming
practices, have concluded that if there are any justified beliefs at all, then cer-
tain types of circularity are unavoidable and do not undermine justification. And

42 See Lyons (2017).

43 Ibid.
152 Historical Cases for the Debate

because it is extremely difficult to deny that there are any justified beliefs, scien-
tific or otherwise, the inevitable conclusion is that some types of circularity are
unavoidable but not such that they undermine justification.
For example, William Alston argues that our inability to give a non-​circular
justification for the reliability of sense perception does not undermine the justi-
fied deliverances of sense perception.44 Michael Bergmann argues if one accepts
the foundationalist principle that some beliefs are justified non-​inferentially,
one is thereby committed to approving of epistemically circular arguments;
Bergmann goes on to argue that because foundationalism is vastly superior to
competing accounts of the structure of knowledge, epistemic circularity can’t be
all bad.45 Psillos argues that “rule-​circularity”—​that is, arguing for some con-
clusion p via an inference rule R when p entails that R is reliable—​is not an epis-
temologically vicious type of circularity.46 Thus, if there are benign instances of
circularity, like Alston, Bergman, and Psillos agree is the case, then it’s not the
case as Lyons asserts in (11), that there can be no correlatively precise positive
instances of the realist meta-​hypothesis. True, realism will require interpreting
evidentially significant cases in ways that adopt some version of circularity, but
perhaps in a way that is not vicious circularity, particularly if the alternative to
realist interpretations of novel predictions requires appealing to a string of mere
coincidences.
In essence, what we are claiming is that there is lots of evidence for the realist
meta-​hypothesis and lots of evidence against the ContraSRs if one understands
genuine deployment in a causally robust sense of working posits that lends itself
naturally to a realist interpretation of scientific practice, including practice that
involves theory change over time. Granted, there is a kind of circularity here.
But suppose, as Psillos, Alston, Bergmann, and others have argued, that some
instances of circularity are benign and are simply a feature of the epistemological
landscape, a landscape that includes the epistemological territories of scientific
practice and philosophy. In that case, more work would need to be done to es-
tablish that the type of circularity we are acknowledging is vicious, not benign.47
Moreover, interpreting cases of genuine deployment in ways that favor anti-​
realism may also involve an appeal to circularity of the sort that requires some
question-​begging assumptions about base-​rates and the probability space of

44 See Alston (1986).

45 See Bergmann (2004).
46 See Psillos (1999: 83–​85).
47 Some of the issues relevant to this section but beyond the scope of our paper to explore in any

detail involve internalism vs. externalism in epistemology. As Psillos (1999: 85) notes, those with ex-
ternalist intuitions will not be bothered by a type of rule-​circularity that uses an inference rule R to
justify R. Moreover, externalists will be open to criteria for genuine deployment that reflect working
posits which actually carve nature’s joints, regardless of whether one is able to demonstrate successful
deployment according to internalist criteria.
 Selective Scientific Realism and Truth-Transfer 153

theories from which the relevant base-​rates are drawn. Peter Dicken (2013)
looks at ways in which the realism/​anti-​realism debate is subject to exempli-
fying instances of the base-​rate fallacy.48 Dicken uses the common example of
circumstances involving medical diagnosis in order to illustrate the base-​rate fal-
lacy and then proceeds to apply the statistical lessons learned to the realism/​anti-​
realism debate. In order to make the analogy explicit, let us define the ALPHA
condition as one where the facts on the ground make it very likely that the rele-
vant indicator conditions are present. Moreover, let the BETA condition be one
where the relevant indicator conditions are taken to indicate certain facts on the
ground.49 Here then, is the analogy:

Case 1: Testing for Disease (95% Accuracy)

ALPHA: If you have the disease, then testing positive for the disease is a strong
indicator that you have the disease.
BETA: If you (a random person drawn from a population) test positive for the
disease, then whether this is a good indicator of whether you have the di-
sease depends on the relevant base-​rate—​that is, the percentage of people
within a population that have the disease.

Dicken illustrates the base-​rate fallacy by noting that if, say, we’re looking at a
population of 100,000 and a base-​rate of 0.1%, then the BETA condition is not a
strong indicator of the presence of disease. In that scenario, a positive test result
on a random subject yields only a 2% chance that the subject has the disease. This
is because even though the test has a 95% success rate, given the very low number
of actual disease carriers relative to the large population, the 5% rate of misdi-
agnosis means that the number of uninfected people misdiagnosed as having
the disease (i.e. 4995) is much, much higher than the number of infected people
accurately diagnosed as having the disease (i.e. 95). So, a positive test result on a
randomly drawn subject from this population has a 95/​4995 (i.e. 2%) chance of
being accurate.
Applying the lesson to the realism/​anti-​realism debate, we get the following
two analogous conditions:

Case 2: Testing for the Truth of Theories

ALPHA: If a theory T is true, then the chance that T will generate accurate
novel predictions is high.
BETA: If T is successful in generating accurate novel predictions, then
whether this is a good indicator that T is true will depend on the relevant

48 See Dicken (2013).

49 Using ALPHA and BETA to specify these conditions is our convention, not Dicken’s.
154 Historical Cases for the Debate

base rate—​that is, the percentage of theories in the relevant “population”

being true. But how should one determine the base-​rate in a non-​circular
way? According to Dicken, the prospects are not good:

So the initial threat of circularity remains . . . the scientific realist can no

longer justify his initial philosophical argument on the basis of the relia-
bility of our first-​order inferences to the best explanation, since this is in
fact precisely what his intended argument attempts to establish . . . . To put
the point even more succinctly, any evidence the scientific realist can offer
for the truth of our current scientific theories will be swamped by the back-
ground probability of an arbitrary scientific theory being true; therefore in
order for the no-​miracles argument to make any positive justificatory con-
tribution to scientific realism, one must first assume that the base-​rate likeli-
hood of a predictively successful theory being true is actually quite high; but
to assume that is just to assume what the scientific realist was attempting to
establish all along.50

However, Dicken goes on to observe that a similar type of base-​rate challenge

effects certain forms of anti-​realism, the cogency of which requires that the base-​
rate be determinately low. In his treatment of Stanford’s problem of unconceived
alternatives, Dicken states:

Yet while the threat of unconceived alternatives may well be more robust
than the threat of the standard pessimistic meta-​induction, and while it may
indeed avoid the need for knowing the underlying base-​rate of any arbitrary
scientific theory being true it seems that for the inference to be compelling
we must simply presuppose a different base-​rate concerning the reliability of
scientists: maybe our historically unconceived alternatives are all heavily biased
towards researchers working in a very specific domain of inquiry; or maybe
while the history of science furnishes us with a great number of instances of
scientists failing to exhaust the relevant possibilities, this is due to the relatively
large number of scientific practitioners, rather than the likelihood of any arbi-
trarily successful researcher failing to consider another alternative theoretical
formulations.51

Are we at a stalemate with respect to worries about circularity more broadly, and
base-​rate fallacies in particular? We think not, at least so far as defending SSR
through our particular historical example is concerned.

50 See Dicken (2013, 565).

51 Ibid., 569.
 Selective Scientific Realism and Truth-Transfer 155

First, notice that one key determination in establishing the relevant base-​rates
is determining the relevant “population” from which a candidate theory is being
assessed. There are many different ways to determine the relevant population
base for which one is attempting to determine base-​rates. Given that there do not
seem to be clear and neutral criteria for determining base-​rate thresholds in ways
that clearly privilege realism or anti-​realism, it is unlikely for there to be such
clear and neutral criteria for determining the relevant population base.
Second, note that while the analogy between the medical diagnosis case and
the scientific realism case is helpful for illustrating the relevance of base-​rates
to the realism/​anti-​realism debate, notice also that the two sorts of cases are
not completely analogous. With respect to medical diagnosis, the presence or
absence of disease is a binary, “on/​off ” affair: one either has the disease or one
doesn’t. But with respect to the truth or approximate truth of any given theory
T, things are not cast in such binary terms—​at least not in binary terms that the
scientific realist is obligated to accept. This is because, on the selective scientific
realist framework we are adopting, theories that are, strictly speaking, false when
taken in their entirety, can have true working posits. This further complicates the
task of determining the relevant population for determining base-​rate threshold,
as determining base-​rate isn’t as simple as determining which members of the
population “have or do not have the disease.”
Third, recall where we are in the dialectic with Lyons’s arguments against the
realist meta-​hypothesis. The main focus of this section of our paper has been to
determine whether Lyons’s criticisms of realism undermine our interpretation
of the transition from the Blomstrand-​Jørgensen model of cobalt complexes to
Werner’s coordination theory. We have argued that this episode of theory change
in 19th century chemistry can be plausibly interpreted using the tools of SSR,
noting that the features that lead to predictive success of the failed theory—​its
working posits that were genuinely deployed in a causally robust sense—​survived
and were part of the successful theory that replaced it. Lyons has identified a re-
alist meta-​hypothesis that is required by realists—​including the divide et impera
realism we are advocating in this paper:

Realist Meta-​Hypothesis: Those constituents that are genuinely deployed in the

derivation of successful novel predictions are approximately true.

We then examined two arguments by Lyons to the conclusion that the realist
meta-​hypothesis is false.
According to the False Basis Argument, some theoretical constituents genu-
inely deployed in deriving successful novel predictions turn out to not be true—​
a consequence that is inconsistent with the realist meta-​hypothesis. However,
we argue that if we distinguish between psychological deployment and a more
156 Historical Cases for the Debate

causally robust sense of genuine deployment, something akin to the actual

working posits in a theory, then, minimally, it’s plausible to read Jørgensen’s suc-
cess in generating novel predictions as consistent with, and not a counterexample
to, the realist meta-​hypothesis. While not conclusive refutation of the False Basis
Argument, we suggest that this strategy is likely to yield similar fruit in applying
it to other alleged counterexamples to the realist meta-​hypothesis.
According to the Weak Evidence Argument, there is greater evidence for the
meta-​hypotheses that contradict the realist meta-​hypothesis (the ContraSRs)
than there is for the realist meta-​hypothesis itself. However, if one restricts gen-
uine deployment to map onto the causally robust working posits of a theory, then
pace Lyons, both the evidence for the ContraSRs is undermined and evidence for
the realist meta-​hypothesis accumulates. Moreover, it is hard to avoid some type
of circularity for both realists and anti-​realists in determining whether particular
instances of deployment count as evidence for the ContraSRs or for the realist
meta-​hypothesis.
There are interesting and important epistemological issues that require fur-
ther attention here related to the epistemic authority of science as a guide for
what one should think is true about the world. Lyons (2016) suggests that we’d
be better off and more reflective of scientific practices if we looked at science as
a “mode of inquiry” and not as being “fixated on belief.” He suggests that realists
give up the epistemic facet of scientific realism and be “axiological realists” where
science, as a mode of inquiry is aimed at truth, but stop short of being epistemic
realists who claim that we should look to science for epistemic justification.
We think that conceiving of science as a mode of inquiry is exactly right, how-
ever, even conceiving of science along these lines doesn’t preclude the relevance
of science as an epistemic guide for belief. If, say, science as a mode of inquiry
is focused on explanation, and explanations can have epistemic good-​making
properties, it’s not a stretch to also say that the reach of scientific explanations
can serve as an epistemic guide to what we should think is true about the world—​
that is, as an epistemic guide to belief. Retaining the epistemic feature of realism
explains why this is so. Well-​confirmed and mature scientific theories give us a
good reason to think that the world is being accurately, approximately described
by those theories. As such, we can look to well-​confirmed and mature science as
a fallible, but reliable, source of epistemic justification.

7.6 Conclusion

Evolving views of the structure of cobalt complexes in 19th century chemistry

provide an interesting case study through which we can test SSR. It turns out
that in this case the factors which lead to predictive success of the failed model
 Selective Scientific Realism and Truth-Transfer 157

survived into the successor theory, providing a positive instance in which

the predictive success of a failed theory can be accounted for along plausible,
realist lines.

Acknowledgments

This work was substantially improved by comments received when this paper
was presented at the Canadian Society for History and Philosophy of Science
2017 annual meeting, the 2017 Quo Vadis Selective Scientific Realism confer-
ence, the University of Notre Dame Centre for Philosophy of Religion, and the
Purdue University philosophy department colloquium. Thanks in particular for
detailed written comments by Tim Lyons and Peter Vickers, as well as by anon-
ymous reviewers. Also, work on this project was supported by a grant from the
Templeton Religion Trust.

References
Alston, William P., “Epistemic Circularity.” Philosophy and Phenomenological Research 47
[1] (Sep., 1986), 1–​30.
Bergmann, Michael, “Epistemic Circularity: Malignant and Benign.” Philosophy and
Phenomenological Research 69 [3] (Nov., 2004), 709–​727.
Day, M. Clyde, Jr., and Joel Selbin, Theoretical Inorganic Chemistry. Reinhold Publishing
Corporation (1962), 183.
Dicken, Peter, “Normativity, the Base-​ Rate Fallacy, and Some Problems for Retail
Realism.” Studies in History and Philosophy of Science 44 (2013), 563–​570.
Douglas, Bodie E., and Darl H. McDaniel, Concepts and Models of Inorganic Chemistry.
Blaisdell Publishing Company (1965), 338.
Farber, Eduard, The Evolution of Chemistry. New York: The Ronald Press Company
(1952), 164–​165.
Griffith, J.S., and L.E. Orgel. “Ligand-​Field Theory.” Quarterly Reviews, Chemical Society
11 [4] (1957): 381–​393.
Jørgensen, Sophus Mads, “Beiträge zur Chemie der Kobaltammoniakverbindungen: VII
I. Ueber die Luteokobaltsalze.” Journal für praktische Chemie 35 [2] (1887): 417, cited
in Kauffman.
Jørgensen, Sophus Mads, “Zur Constitution der Kobaltbasen.” Journal für praktische
Chemie 41 [2] (1890): 429, cited in Kauffman.
Kauffman, George B., Sophus Mads Jorgensen (1837–​1914) A Chapter in Coordination
Chemistry History in Selected Readings in the History of Chemistry. Compiled by Aaron
J. Ihde and William F. Kieffer, editor, reprinted from the 1933–​1963 volumes of Journal
of Chemical Education, Easton, Pennsylvania: Division of Chemical Education of the
American Chemical Society (1965).
Kragh, Helge, “S.M. Jørgensen and His Controversy with A. Werner: A Reconsideration.”
The British Journal for the History of Science 30 [2] (1997), 203–​219.
158 Historical Cases for the Debate

Lyons, Timothy D. “Scientific Realism and the Strategema de Divide et Impera.” British
Journal for the Philosophy of Science 57 (2006), 537–​560.
Lyons, Timothy D., “Structural Realism versus Deployment Realism: A Comparative
Evaluation.” Studies in History and Philosophy of Science 59 (2016), 95–​105.
Lyons, Timothy D., “Epistemic Selectivity, Historical Threats, and the Non-​Epistemic
Tenets of Scientific Realism.” Synthese 194 [9] (2017): 3203–​3219.
Meissler, Gary L., and Donald A. Tarr, Inorganic Chemistry, 4th ed. Prentice Hall
(2011), 322.
Moore, F. J., A History of Chemistry, 3rd ed. McGraw-​Hill Book Company, Inc. (1939),
294–​295.
Morrison, Margaret. “One Phenomenon, Many Models: Inconsistency and
Complementarity,” Studies in History and Philosophy of Science 42 (2011): 342–​351.
Psillos, Stathis. Scientific Realism: How Science Tracks Truth. Oxford: Oxford University
Press (1999).
Stanford, P. Kyle, Exceeding Our Grasp: Science, History, and the Problem of Unconceived
Alternatives. Oxford (2006).
Stanford, P. Kyle, “Unconceived Alternatives and the Strategy of Historical Ostension,”
in Juha Saatsi (ed), The Routledge Handbook of Scientific Realism. Routledge
(2018): 212–​224.
Vickers Peter, “A Confrontation of Convergent Realism.” Philosophy of Science 80 [2]
(April 2013), 189–​211.
Weinberg, Steven, Dreams of a Final Theory. Vintage Books (1994).
Weinberg, Steven, interviewed by John Horgan, https://​blogs.scientificamerican.com/​
cross-​check/​nobel-​laureate-​steven-​weinberg-​still-​dreams-​of-​final-​theory/​.
Werner, Alfred, Zeitschrift für anorganische Chemie 3 (1893): 267–​330, translated in
George B. Kauffman, Classics in Coordination Chemistry Part I: The Selected Papers of
Alfred Werner. New York: Dover Publications, Inc. (1968), 17.
 8
Realism, Physical Meaningfulness,
and Molecular Spectroscopy
Teru Miyake1 and George E. Smith2

1Department of Philosophy

Nanyang Technological University

[email protected]
2Department of Philosophy

Tufts University
[email protected]

8.1 Introduction

The question of the reality of microphysical entities, such as atoms, molecules,

and their constituents, has always been central to the scientific realism de-
bate. In particular, the experiments of Jean Perrin on Brownian motion (Perrin
1910) have been the focus of work by realists,1 who present them as having
successfully shown that molecules exist. On the other hand, Bas van Fraassen
(2009) has recently offered an anti-​realist reading of Perrin’s work, and indeed
the entire tradition in which it is situated. According to van Fraassen, the aim of
the nineteenth-​and early twentieth-​century tradition from Dalton through to
Perrin was not, as the realists would have it, to establish the reality of atoms and
molecules at all, but the development and enrichment of theory so as to make
what he calls “empirical grounding” possible for the theoretical quantities of ki-
netic theory.
Empirical grounding of a theory, according to van Fraassen, consists of three
main features (van Fraassen 2009, p. 11): first, any significant parameter of the
theory must be measurable under certain conditions; second, measurements may
be theory-​relative, that is, the connections between the theoretical parameters
and observable parameters may be theoretically posited; and third, there must

1 See, for example, Wesley Salmon (1984), Deborah Mayo (1996), Peter Achinstein (2001), Stathis

Psillos (2011a, 2011b, 2014), and Alan Chalmers (2009, 2011).

Teru Miyake and George E. Smith, Realism, Physical Meaningfulness, and Molecular Spectroscopy In: Contemporary
Scientific Realism. Edited by: Timothy D. Lyons and Peter Vickers, Oxford University Press. © Oxford University Press
2021. DOI: 10.1093/​oso/​9780190946814.003.0008
160 Historical Cases for the Debate

be concordance in the values of such parameters determined by different means.

There are some further constraints that van Fraassen takes from Glymour (1980)
to avoid ways of trivially satisfying these requirements. The empirical grounding
of kinetic theory, as van Fraassen envisions it, was a great achievement, but it did
not require any commitment to the claim that atoms or molecules exist. Stable
concordant measurement of theoretical parameters does not, by itself, show that
the parameters being measured correspond to definite properties of entities that
actually exist.
Whether, in fact, van Fraassen is correct about the aims of this tradition is
a rather difficult interpretational issue, which we shall not attempt to answer
here. Instead, in setting the stage for the rest of this paper, we shall simply point
out that even if successful, Perrin’s experiments did not, in fact, give us much
access at all to the structure and properties of atoms and molecules. If we take
Perrin’s claims at face value, he established a new method for the determination
of Avogadro’s number, which he claimed was the most accurate method yet. In
retrospect, however, this was just one of several different methods for the meas-
urement of this quantity that were developed in the early twentieth century. The
values measured by Perrin, moreover, turned out to be 10% or more higher than
other values measured, for example, by Planck (1900), Rutherford and Geiger
(1908), Millikan (1911, 1913), Boltwood and Rutherford (1911), and others,
which are closer to our modern values. By 1913, Bohr and the Braggs were al-
ready adopting values for molecular magnitudes that were in agreement with
significantly lower values for Avogadro’s number than Perrin’s. In particular, in
Bohr’s landmark 1913 paper, the sole direct empirical test of his model compared
his theoretically calculated value of Rydberg’s constant, 3.1 × 1015 Hz, with the
measured value, 3.29 × 1015 (Bohr 1913, p. 9). Had Bohr adopted the announced
value for the fundamental charge that Perrin had inferred from his value of
Avogadro’s number, 4.3 × 10–​10 esu (Perrin 1912, 248), instead of the value he
chose, 4.7 × 10–10, his theoretically calculated value would have been 2.0 × 1015,
too far removed from the measured value to provide much in the way of a pos-
itive test of his model. So, exactly what Perrin achieved in his determination of
molecular magnitudes is more open to question than is generally realized.
Realists would, of course, be quick to point out that Perrin’s experiments did
not just give us another measurement of Avogadro’s number. For example, a
further accomplishment of Perrin’s experiments, as Deborah Mayo (1996) has
pointed out, is that one of them (Chaudesaigues 1908) established the stochastic
nature of Brownian motion, thus indicating that Brownian motion, whatever its
cause, must arise from innumerable uncoordinated momentum exchanges with
the medium surrounding Brownian particles, which certainly seems to indi-
cate that the medium is composed of uncoordinated microscopic parts. Even so,
these results are consistent with such momentum-​exchanging microscopic parts
 Realism and Molecular Spectroscopy 161

having only a fleeting existence. Absent any evidence that such microscopic parts
retain their integrity over a significant length of time, this result falls short of
establishing the existence of atoms or molecules.
An undeniably large epistemic gap exists, moreover, between having good
reason to think that the medium in which Brownian particles are suspended is
composed of microscopic parts in vigorous stochastic motion, and having know-
ledge about specific details of those parts. How much access to molecules was, in
fact, gained through Perrin’s experiments? According to physicists and chemists,
we now know that molecules have a great deal of structure, as we shall discuss
later. How much of this structure could have been inferred from measurements
of Avogadro’s number? The sizes of molecules could be calculated from a de-
termination of Avogadro’s number, but such calculations were based on the
assumption that molecules are spherical or at least have a distinct sphere of in-
fluence. As van Fraassen points out (2009, p. 8), dependence on such unrealistic
assumptions throws considerable doubt onto whether such calculations are truly
yielding information about molecules. Moreover, this shortcoming was expressly
recognized at the time. James Jeans adopts values of the molecular magnitudes
corresponding to those of Rutherford and Geiger in the 1916 edition of his The
Dynamical Theory of Gases and then concludes, “If, however, the molecules are
assumed as a first approximation to be elastic spheres, experiment leads to dis-
cordant results for the diameters of these spheres, shewing that the original as-
sumption is unjustifiable” (Jeans 1916, p. 9). Simply put, accordingly, whatever
grounds the various determinations of Avogadro’s number and the evidence for
fine-​grained stochastic motion in the substrate supporting Brownian motion
had provided for molecular-​kinetic theory, it had provided no information on
the sizes and shapes of molecules. In the 1921 edition of Theoretical Chemistry
from the Standpoint of Avogadro’s Rule and Thermodynamics, Walther Nernst was
still making a claim he made in the first edition of 1893: “At present scarcely an-
ything definite is known regarding either the nature of the forces which bind the
atoms together in the molecule and which hinder them from flying apart in con-
sequence of the heat of motion, or regarding their laws of action” (Nernst 1895,
p. 237; 1923, p. 327).
Some information can nevertheless be obtained about the structure of
molecules through kinetic theory. If equipartition of energy is assumed, then
kinetic energy should be distributed evenly through all the degrees of freedom
of a molecule. A spherical molecule has three degrees of freedom, while a mol-
ecule in the shape of a dumbbell can rotate along two axes, so it has two addi-
tional degrees of freedom. If the parts of the molecule are free to move relative to
each other, there will be additional degrees of freedom. The ratio of the specific
heat of a gas at constant pressure and the specific heat at constant volume, ac-
cording to kinetic theory, is a function of the number of degrees of freedom of
162 Historical Cases for the Debate

the molecular constituents of the gas. Measurements of the ratio of specific heats
might, then, provide some insight into the shapes of molecules. Unfortunately,
at the beginning of the twentieth century, measurements of the ratio of specific
heats of various gases could not be reconciled with what was then known about
molecules. In a lecture delivered to the Royal Institution in 1900, Lord Kelvin
named the specific heat anomaly one of two “Nineteenth Century Clouds over
the Dynamical Theory of Heat and Light” (Kelvin 1904, pp. 486–​527).
How, then, did scientists first come to gain access to the structure and proper-
ties of atoms and molecules? Our view is that this was first made possible by the
development, in the late 1920s and after, of theories and techniques that enabled
theory-​mediated access to the structure of atoms and molecules through spec-
troscopy. The analysis of spectra has a history that predates quantum mechanics
by more than a century, but it was the development of quantum mechanics
that first allowed physicists to make stable theory-​mediated measurements
of parameters that they took to represent the real properties and structures of
molecules. In this paper, we thus aim to answer the following question:

(1) How did spectroscopists first come to have the ability to make stable
theory-​mediated measurements of parameters that are taken to represent
the real properties and structure of molecules?

Anti-​realists such as van Fraassen will grant that such concordant measurement
has been achieved but will deny that spectroscopists have thereby gained access
to the real structure and properties of molecules. For that reason, we shall ask the
further question:

(2) What evidence is there that when spectroscopists measure such parameters
they are gaining access to the real properties and structures of molecules—​
in other words, that the parameters are physically meaningful and not just
artifacts of the particular way spectroscopists have chosen to represent
whatever lies beyond our senses?

Let us illustrate what we mean by an artifact by way of an example. A familiar

molecular parameter from kinetic theory that is an artifact of a particular way
of representing the molecular realm is mean-​free-​path. Values for this param-
eter were derived from Maxwell’s theory of viscosity in the latter part of the
nineteenth century under two assumptions that together entail that molecules
have a definite spherical “collision cross-​section”: (1) molecules are spherical,
and (2) each molecule moves entirely without constraint until it collides with
the “sphere of influence” of another molecule, a “spherical surface” which it
cannot penetrate at all. Both of these are now regarded as fictions (Hecht 1990,
 Realism and Molecular Spectroscopy 163

pp. 129–​174). Moreover, the values for the mean-​free-​paths obtained from vis-
cosity when extended to other phenomena—​especially, to deviations from the
ideal gas law and refraction of light through gases—​yielded values for the sizes
of the spheres that were not only discordant with one another, but also in vio-
lation of Avogadro’s rule (Nernst 1895, pp. 347–​352; Meyer 1899, pp. 149–​221,
299–​348). Van Fraassen would call this a failure to ground the theory; we see it as
a representation of the molecular realm that never had clear claim to being phys-
ically meaningful. As this example suggests, the answer to the second question
is likely to be rather complicated, and so in this paper we do our best to give an
initial attempt to answer it. In particular, we believe that one of the keys to an-
swering this question consists in understanding a distinction that we will make
between representations that are physically meaningful and those that are not.
Molecular spectroscopy is a little-​known topic among philosophers of science,
so we will take some care in explaining how it works. Section 8.2 of this paper
will be devoted to providing enough background on molecular spectroscopy to
be able to answer question (1)—​that is, to explain how stable theory-​mediated
measurement of parameters that are taken to represent properties of molecules
has been achieved. In order to keep things manageable, we will limit our discus-
sion to the molecular spectroscopy of diatomic molecules and will focus on the
early period of molecular spectroscopy, from the 1920s through around 1950. In
Section 8.3, we will consider ways in which one might attempt to answer ques-
tion (2). We will conclude that a proper answer to question (2) will require an
understanding of physical meaningfulness, which we will explain through an
examination of an example from celestial mechanics. A full answer to question
(2) will go beyond the scope of this paper, for it would require an examination of
the history of molecular spectroscopy in the period after 1950, but we will pro-
vide, in Section 8.4, a brief look at further work in molecular spectroscopy that
we believe supports our contention that spectroscopists indeed are gaining ac-
cess to physically meaningful parameters of molecules.

8.2 Molecular Spectroscopy and Stable Measurement

of Molecular Constants

Molecular spectroscopy is a massive field. It grew rapidly from the late 1920s,
when quantum mechanics was first successfully applied to the analysis of spec-
troscopic data. This early period culminated with the publication of Gerhard
Herzberg’s landmark Molecular Spectra and Molecular Structure, published in
four volumes between 1939 and 1979. In this paper, we will mostly focus on re-
search in the field up through 1950, the year of publication of the second edi-
tion of Volume 1, on the spectra of diatomic molecules (Herzberg 1950). As an
164 Historical Cases for the Debate

indication of the amount of research that was being done at the time on dia-
tomic molecules, we note that the second edition of Volume 1 contains a table
of constants for diatomic molecules that goes on for 80 pages (pp. 501–​581)
and lists 1574 references in its bibliography. Molecular spectroscopy has been
growing steadily ever since. Volume 4 of Herzberg, published in 1979, consists
solely of tables of constants for diatomic molecules. It is 716 pages long and
lists its references by molecule. The ordinary hydrogen molecule, 1H2, has 169
references, while 1H2H, a molecule consisting of one ordinary hydrogen atom
and one deuterium atom, has 51 references; 2H2, or molecular deuterium, has
60 references. There is now an online database of papers on diatomic molecules
called DiRef (Bernath and McLeod 2001), which covers the period after Volume
4; it contains 30,000 references for the period 1974–​2000.
The molecular constants are parameters that can be determined from spec-
troscopic data, and they are taken to represent certain properties of molecules.
When a molecule undergoes a transition between two energy states, it absorbs or
releases a photon, and the frequency of the photon is taken to correspond to the
difference in energy levels between these two states. Thus, an examination of the
spectrum of light absorbed or released by a gas can provide information about
the energy states of the molecules in the gas. The interpretation of this spectrum
requires some knowledge about the possible energy states of a diatomic mole-
cule. A diatomic molecule at room temperature is taken to be a body that is not
merely in translational motion, but also undergoing vibrational and rotational
motion. The two nuclei in the molecule vibrate anharmonically in accord with
an asymmetric potential function that depends in large part upon the distance
between the nuclei. The shape of the potential function depends upon the ar-
rangement of the positive charges of the nuclei and the negative charges of the
electrons in the molecule. The rotation has the effect of centrifugally stretching
out the distance between the nuclei, but the effect is only significant at high levels
of rotation, so we shall omit discussion of it here. Energy can be emitted and
absorbed by the molecule through transitions between rotational, vibrational,
and electronic states. Energy is quantized, with the consequence that the spectral
lines correspond to a transition from one energy state to another, whether it be a
change in the rotational, vibrational, or electronic state of the molecule.
The energies of these states can be calculated for idealized systems by solving
the Schrödinger equation for such systems. For example, the simplest idealiza-
tion of a diatomic molecule is the rigid rotor, a system consisting of two point-​
masses separated by a fixed distance, rotating about an axis perpendicular to the
line connecting the masses (Herzberg 1950, pp. 66–​73). The Schrödinger equa-
tion for such a system has solutions for only certain energy values, customarily
given in units of wavenumber (cm–​1) by
 Realism and Molecular Spectroscopy 165

F ( J ) = BJ ( J + 1) , where B = h / (8π 2 cI ).

Here, J is the rotational quantum number, and can take the integral values 0, 1,
2, . . . , while h is Planck’s constant, c is the speed of light, and I is the moment of
inertia of the rigid rotor. Thus B, called the rotation constant, essentially gives
the inverse of the moment of inertia of the system, normalized and converted to
units of wavenumber. A further analysis of the selection rules for the quantum
number J shows that the spectrum of a rigid rotor should consist of a series of
evenly spaced lines, the first of which lies at wavenumber 2B, with the distance
between successive lines also being 2B. Thus, the value of B, and hence that of the
moment of inertia, of a rigid rotor, can be read directly off of its spectrum, pro-
vided that the spectrum has been interpreted correctly. Given the masses of the
rigid rotor, the distance between the two point-​masses can be determined from
the moment of inertia.
A diatomic molecule is not, however, a rigid rotor—​the distance between its
nuclei is constantly changing. If a diatomic molecule were, instead, a rotating
harmonic oscillator, it would produce a spectrum with a single peak that, at
higher resolution, contains peaks spaced uniformly corresponding to dif-
ferent rotational transitions. In fact, actual diatomic molecules are anharmonic
oscillators, and this gives rise to non-​uniformities in the spectral lines. These
non-​uniformities provide ways of determining constants characterizing both
the anharmonicity of the vibration and the effect of vibration on rotation, as
well as other effects. The values of these constants depend upon the potential
functions between the two nuclei, which in turn depend upon whether the
electrons in the molecule are in the ground state or excited states. In order to
keep things simple, we will only be considering the ground state—​that is, a
state from which spontaneous transitions never occur—​but no small part of
the evidence that the constants are characterizing a structure of the sort speci-
fied comes from comparing the values of the constants for different electronic
states.
With all of this in mind, the vibrational (G) and rotational (Fν) energy levels
for a diatomic molecule may be calculated theoretically, and are customarily
given by the following expressions (Bernath 2016, p. 225), again in units of
wavenumber (cm-​1):

G(ν) = ω e (ν + 1 / 2) − ω e xe (ν + 1 / 2)2 + ω e ye (ν + 1 / 2)3 + ω e z e (ν + 1 / 2)4 + …

( ) (
Fν ( J ) = Bν J ( J + 1) − Dν J ( J + 1) + H J ( J + 1) + … )
2 3
166 Historical Cases for the Debate

The vibrational energy levels G(ν) are thus expressed as a series expansion in
ν + 1/​2, where ν is the vibrational quantum number, while the rotational en-
ergy levels Fν(J) are expressed as a series expansion in J(J + 1), where J is the
rotational quantum number. The rotation constant Bν is the first coefficient in
the expansion for the rotational energy levels, and it depends on the vibrational
energy state as follows:

Bν = Be − α e (ν + 1 / 2) + γ e (ν + 1 / 2)2 + …

The constant Be, the first term in the expansion of Bν, is the value that the rota-
tion constant would have if the nuclei were not vibrating at all, in the manner
of a rigid rotor, with the distance between the nuclei being equal to the equi-
librium distance re, that is, the distance between the nuclei at the bottom of the
anharmonic potential. We will omit discussion of coefficients of higher-​order
terms here. Spectroscopists have developed techniques for calculating the values
of these constants from spectra. In published tables of constants, the values given
are typically for the equilibrium values Be, αe, re, ωe, ωexe, ωeye. Note that these
constants are usually given in units of wavenumber, except for re.
Table 8.1 shows the values for some of these constants for a few diatomic
molecules. The three columns on the left show constants for different isotopic
forms of the hydrogen molecule. The first column shows a molecule consisting
of two ordinary hydrogen nuclei, the second column shows a diatomic mole-
cule consisting of an ordinary hydrogen and a deuterium nucleus, and the third
column shows a diatomic molecule consisting of two deuterium nuclei. The two
columns on the right show constants for two isotopic variants of the hydrogen

Table 8.1 Values of molecular constants for several diatomic molecules.

Molecular constant 1H 1H 2H 2H 1H35Cl 2H35Cl

2 2

Be (cm–​1) 60.809 45.655 30.429 10.5909 5.445

αe (cm–1) 2.993 1.9928 1.0492 0.3019 0.1118
re (cm × 10–8) 0.7416 0.7413 0.7416 1.2764 1.274
ωe (cm–​1) 4395.2 3817.09 3118.4 2989.74 2145.163*
ωexe (cm–1) 117.99 94.958 64.09 52.05 27.1825*
ωeye (cm–1) [0.29] 1.4569 1.254 0.2243* 0.08649*

Source: From Herzberg (1950), except * = from Huber and Herzberg (1979), based on results from
1953 and after.
 Realism and Molecular Spectroscopy 167

chloride molecule. There are complications that we are omitting for the moment,
but by 1950, spectroscopists had achieved stable theory-​mediated measurement
of parameters that are taken to represent the structure and properties of a large
number of diatomic molecules. As already noted, Herzberg (1950) contains a
table that stretches over 80 pages and lists constants for a huge number of dia-
tomic molecules. These constants vary systematically from one molecule to an-
other, in strict correspondence to differences in their spectra.

8.3 Gaining Access to Molecular Structure

As we have mentioned, however, even if we grant that stable theory-​mediated

measurement of the molecular constants has been achieved, we might yet
ask: What reason is there to think that those constants are giving us access to
definite physical details of actual molecules, and that they are not just artifacts of
the way spectroscopists have chosen to represent molecules? The answer to this
question is complicated, but we will start by describing several ways in which one
might try to answer it.
First, note that spectroscopists have measured the molecular constants for a
large number of different molecules. In particular, as indicated in Table 8.1, the
constants for isotopic variants of various molecules such as the hydrogen mole-
cule and the HCl molecule have been measured. This gives spectroscopists one
way of checking whether the constants have been interpreted correctly. The iso-
topic variants effectively give them a way of varying the masses in the diatomic
molecule while holding other things, such as the electromagnetic field of the
molecule, constant. Different isotopic variants for the hydrogen molecule should
give rise to virtually the same potential function, because only the masses of the
nuclei change, while no changes are made to the electromagnetic properties of
the nuclei. Sure enough, if we look at the values in Table 8.1, we see that re, the
constant that is taken to represent the equilibrium distance between the nuclei,
is very nearly the same for all the isotopes of the hydrogen molecule, given in the
first three columns. What does change are the constants that are taken to rep-
resent the frequency of vibration—​ωe, ωexe, ωeye—​which is to be expected, be-
cause the frequency of vibration of the anharmonic oscillator ought to vary if the
masses change. Also, Be, which can be interpreted as the inverse of the moment
of inertia of the molecule at equilibrium, changes as well, as it should, since the
masses of the nuclei differ for each of the isotopic variants of the hydrogen mol-
ecule. We note further that Be for 2H2 is very nearly half that for 1H2, while the
value for 1H2H is very nearly 3/​4 the value for 1H2, which is to be expected given
the interpretation of Be as the inverse of the moment of inertia. We see a similar
pattern for the isotopic variants of HCl. Of course, these are just two molecules.
168 Historical Cases for the Debate

We claim that if we examined all the molecules in Herzberg’s table of molecular

constants, we would see a similar pattern for other molecules as well. It’s worth
pointing out that there is a lot of work on polyatomic molecules as well. In this
case, spectroscopists must deal with further molecular constants, because now
there are more degrees of freedom in the molecule. Success in the interpretation
of spectra for these molecules in terms of rotation and vibration ought to lead to
further evidence that the given interpretation of spectra for diatomic molecules
is correct. In short, one way of confirming whether the molecular constants are
indeed measuring what spectroscopists say they are measuring is by comparing
naturally occurring variants in which virtually the only difference is the mass of
one or more of the nuclei, and confirming that the spectra change in the way they
are expected to.
Another reason to think that the molecular constants are giving us access to
real physical details of molecules is that the values of the molecular constants
agree with facts about molecules that are known chemically or thermodynam-
ically. For example, an approximation to the anharmonic potential function
can be calculated from the equilibrium distance re and the vibration constants
ωe, ωexe, ωeye. From this anharmonic potential, the dissociation energy, or the
amount of energy required to take a diatomic molecule in its ground state and
pull the two nuclei apart, can be calculated. The energy of dissociation can also
be determined empirically through chemical means, and the agreement between
the energy of dissociation measured spectroscopically and chemically is evidence
that the molecular constants are indeed yielding information about the actual
potential of the molecule. So, in short, another check on whether spectroscopists
are indeed measuring what they claim to be measuring is to compare spectro-
scopic measurements with chemical or thermodynamic measurements.
Finally, spectroscopic measurements have a high degree of redundancy.
As mentioned earlier, the internal energy states of diatomic molecules can be
separated into three types: rotational, vibrational, and electronic (Figure 8.1).
The pure rotational spectra can be interpreted in terms of a rigid rotor, from
which the rotational constant B and internuclear distance r may be determined
from the spectra. Vibrational-​rotational spectra, which are typically in the near
infrared, arise due to transitions between vibrational energy levels, without
any change in the electronic energy levels, and exhibit fine structure that can
be interpreted in terms of a vibrating anharmonic oscillator, from which rota-
tional and vibrational constants can be determined. There is thus redundancy in
the determination of the rotation constants—​they can be determined from both
the pure rotation spectra and from the vibration-​rotation spectra. Electronic
spectra, which are typically in the visible range, arise due to transitions between
electronic energy levels and have much more complicated structure. These
spectra represent a great number of transitions and may appear more like bands
 Realism and Molecular Spectroscopy 169

Figure 8.1 Rotational, vibrational, and electronic energy states of diatomic

molecules. The three double arrows indicate a pure rotational transition (shortest
double arrow, seen toward the lower left corner), a vibrational-​rotational transition
(medium-​length double arrow), and a transition between electronic energy levels
(longest double arrow).
Source: Herzberg (1971, p. 17).

than lines. High resolution is needed to resolve the structure. Because the struc-
ture in these bands corresponds to transitions between rotational and vibra-
tional energy states within different electronic states, they contain information
about the rotational and vibrational constants. The electronic spectra thus give
spectroscopists another independent way of determining both the rotational and
vibrational constants.
Further, Raman spectroscopy provides yet another independent way of cal-
culating the rotational and vibrational constants. Raman spectroscopy is done
by shining light of a particular wavelength, usually from a laser, at a sample and
examining the scattered light. When a photon with energy hν′ is incident on a
diatomic molecule, the molecule can go from a lower energy state E″ to a higher
one E′ by taking from the photon an amount of energy ΔE = E′ − E″, and then
170 Historical Cases for the Debate

scattering a photon with an energy hν′− ΔE. On the other hand, the molecule can
go from a higher to a lower state by imparting the energy ΔE to the photon, scat-
tering a photon with an energy hν′+ ΔE. The spectrum of the scattered light will
accordingly consist of a peak at the frequency of the incident light, with lines on
either side of the peak at frequencies given by ν′− (ΔE/​h) and ν′ + (ΔE/​h). Thus,
Raman spectroscopy provides another way of obtaining the differences between
the various energy levels of the diatomic molecule, from which the rotational
and vibrational constants can be determined. We note here that the spectra we
have previously mentioned involve the absorption and emission of light, whereas
Raman spectra involve scattering, an entirely different phenomenon. Raman
spectra appear in a different frequency range, and the theoretical link between
the observed lines in spectra and the molecular constants is different. In addi-
tion, there are other types of spectroscopy that were developed after the 1950s,
but we omit discussion of them here. Our point is that there is a massive amount
of redundancy in the determination of the molecular constants.
An anti-​realist such as van Fraassen would nevertheless claim that redun-
dancy, or concordance in the values of the molecular constants that are measured
using various different kinds of spectroscopy, does not by itself show that these
molecular constants are actually giving us information about the real structure
of molecules. But evidence comes not merely from concordance in these values.
For each different method of determination of molecular constants, the molec-
ular structure is being accessed through a different causal pathway.
The different ways in which the molecular structure can be accessed ap-
pear as details in the molecular spectra that allow spectroscopists to decide
whether they are, indeed, accessing the molecular structure that they claim to be
accessing. Here is one example. The selection rules for rotational spectra imply
that there should be no pure rotation spectrum for homonuclear molecules,
but heteronuclear molecules should produce such a spectrum. This is be-
cause in order for there to be absorption or emission of a photon, there needs
to be a changing dipole moment. Homonuclear molecules always have a di-
pole moment of zero, so they cannot absorb or emit photons through rotation.
Spectroscopists have found that this is indeed the case—​no pure rotational
spectra can be found for molecules such as H2, but such spectra can be found
for molecules such as HCl. In Raman spectroscopy, however, there is a different
causal pathway by which spectroscopists claim to be gaining access to the struc-
ture of molecules. The theory of Raman spectroscopy says that there ought to be
rotation lines even for molecules without a dipole moment, and this is indeed
what is found by spectroscopists—​the Raman spectra of homonuclear molecules
yields rotational lines.
Spectra are extremely complicated, and they have many more features that
we have not mentioned. Spectra exhibit line splitting, shifts in the positions
 Realism and Molecular Spectroscopy 171

of the lines, broadening of the lines, variations in intensity of the lines, and so
on. Without getting into specifics here, we claim that these details can often
be used to check whether the molecular constants really are measuring what
spectroscopists claim they are measuring, and, in fact, examination of those
details can yield further information about either the molecules themselves or
the causal pathways used to probe them.
With regard to the question of what reason there is to think that molec-
ular constants are giving us access to definite physical details of molecules,
and they are not just artifacts, we believe we can offer an impressive number
of different ways of checking whether the interpretation assigned to spectra by
spectroscopists is indeed correct. After having given all of these reasons, how-
ever, we nevertheless hesitate to answer this question affirmatively, because we
think there is still a problem along a different dimension. This is in trying to an-
swer the question of exactly what those molecular constants represent. We have
so far glossed over some details in our description of the molecular constants.
Take the rotation constant B, for example. If a molecule were a classical rigid
rotor, the distance between the nuclei would not vary, and the rotation constant
B could be interpreted straightforwardly as corresponding to the inverse of the
moment of inertia. But when we go beyond the rigid rotor to a vibrating rotator
with an anharmonic potential, the rotation constant is now dependent on the
vibrational quantum number, and its interpretation becomes more complicated.
Usually what is given in tables of constants is Be, which is what the value of the
rotation constant would be if the distance between the two nuclei at all times
was re, the distance between the two nuclei at the bottom of the potential. But
the molecule is never actually in a state with no vibration and hence never in a
state in which the distance re persists for any significant length of time.2 A way of
thinking about what is happening here is that the distance between the nuclei of
the molecule always fluctuates around the re value, and so the rotation constant,
which is dependent on this distance, must be fluctuating as well. So there is some
ground for saying that there is no feature of the actual molecule to which Be cor-
responds, even according to the theory that enables the measurement to be made
in the first place.
To take the point further, we said that the rotational, vibrational, and elec-
tronic energy states of a molecule can be separated. The separability is justified by
the Born-​Oppenheimer approximation and the Born adiabatic approximation,
in which the assumption is made that the timescales involved in the motions
of the nuclei are much slower than the timescales involved in the motions of
electrons. The timescale at which electrons interact with one another and the

2 This is the case even in the lowest vibrational energy state, for which the energy is not zero but a

positive value, called the zero-​point energy.

172 Historical Cases for the Debate

nuclei is of the order of 10–​16 seconds. The period for vibrational motion is of the
order of 10–​14 seconds, while the period for rotational motion is of the order of
10–​11 seconds. Now, consider what this means. Each time the molecule rotates,
it has undergone of the order of a thousand vibrations, and since the rotation
constant depends on the distance between the nuclei, this must mean that the
rotation constant that is being measured can only be a time-​averaged value—​
it does not correspond to any static or instantaneous state of the molecule. In
other words, the molecular constants are a way of capturing a situation that is
enormously more complicated, with time-​averaging smoothing or glossing over
fluctuations of much shorter periods.
Here, we claim that, nevertheless, it may be possible to argue that the mo-
lecular constants such as Be are physically meaningful, even though they do not
represent some actual state of a molecule. We believe that an analogy with ce-
lestial mechanics is helpful here. In celestial mechanics the orbits are character-
ized in terms of Keplerian parameters defining an ellipse and area swept out per
unit of time. These parameters define a counterfactual situation—​they describe
the elliptical trajectory that would occur if there were only two bodies. To quote
Newton, “if the sun were at rest and the remaining planets did not act on one an-
other, their orbits would be elliptical, having the sun at a common focus, and they
would describe areas proportional to the times” (Newton 1999, p. 817f, emphasis
added). We contend that the Keplerian parameters, notwithstanding their coun-
terfactual status, are, even from a modern perspective, physically meaningful.
First, an examination of the 250-​year history of celestial mechanics be-
tween the times of Newton and Einstein shows that deviations between
predictions made based on calculations from these orbital parameters, and ac-
tual observations of planetary motions, have been used to make inferences about
the existence and location of other bodies that might be influencing the motions
(Smith 2014). In other words, the Keplerian representation served as a research
instrument to expose physical sources of deviations from this representation.
The most famous example of this was the discovery of Neptune, but this one ex-
ample grossly understates how much was discovered about the motions in our
planetary system from this approach.
Second, at the time Newton began work on what became his Principia, he was
aware of four alternatives to the Keplerian representation that fit the observations
just as well as it did. Indeed, Newton was more familiar with both the planetary
theories, and the tables of their motions produced by Thomas Streete (1661),
Vincent Wing (1669), and Nicholaus Mercator (1676), than he was with any of
Kepler’s writings. These three employed three different alternatives to Kepler’s
rule that orbiting bodies sweep out equal areas in equal time, a rule which
Newton likely had learned of from reading Mercator’s systematic defense of his
approach over Kepler’s and Streete’s (Mercator 1676, pp. 144–​164). A few pages
 Realism and Molecular Spectroscopy 173

in the Principia after the counterfactual claim we just quoted, Newton went on to
show that the then most prominent violation of Kepler’s rule, namely the motion
of our moon, results from the action of solar gravity alternately accelerating and
decelerating the moon twice over the course of its monthly orbit, in the process
markedly distorting that orbit in a way theretofore unrecognized (Newton 1999,
pp. 840–​848). In thus identifying a robust physical source for the deviation of the
moon from the area rule, Newton gave evidence supporting the counterfactual
claim that the moon would conform to the rule at least more closely than it does,
and the orbit would be closer to a Keplerian ellipse, were it not for the perturbing
effect of the sun.
This is but one of hundreds of examples from celestial mechanics in which the
Keplerian representation of orbital motion, counterfactual though it is, revealed
deviations that were then shown to have physically identifiable sources. The pre-​
Newton alternatives to Kepler’s representation remind us that there can always
be many other comparably accurate representations, but not all of them would
yield deviations meeting the demand of having identifiable robust physical
sources. We thus distinguish between physically meaningful representations and
parameterizations and those that are not through the following criterion: physi-
cally meaningful representations and parameterizations yield deviations that have
physically identifiable sources.

8.4 Realism, Physical Meaningfulness,

and Molecular Spectroscopy

Are the molecular constants such as Be physically meaningful in the same

way that the Keplerian orbital elements are? That is, have deviations between
predictions about spectra based on calculations from the molecular constants,
and actual spectra, been informative? Have they led to the identification of fur-
ther physical sources that had previously been unaccounted for? Fully answering
these questions will require a detailed examination of developments in molecular
spectroscopy in the period after 1950, especially microwave spectroscopy, which
would exceed the scope of this paper, but we offer the following considerations.
Consider first what is giving rise to these questions. We have emphasized
that the standard constants characterizing the rotation and vibration of dia-
tomic molecules presuppose the Born-​Oppenheimer and the Born adiabatic
assumptions. As a consequence, these constants represent—​in van Fraassen’s
sense (van Fraassen 2008)—​diatomic molecules as rotating dumbbell-​shaped
bodies in which the two nuclei vibrate in an anharmonic fashion along a line
between their point-​mass centers, as if there were no interaction at all between
the electrons and the two nuclei. The only respect in which the electrons are
174 Historical Cases for the Debate

relevant is in the distinction between different electronic states, with different

values of the molecular constants and different potential functions for each elec-
tronic state. (We remind readers that, for simplicity, we have restricted the dis-
cussion to the ground state, but the Herzberg tables of constants for diatomic
molecules are not so restricted.) Ignoring electron-​nuclei interaction allows a
non-​fluctuating potential function between the two nuclei to be defined for each
electronic state. But, of course, there is electron-​nuclei interaction, and hence
this non-​fluctuating potential function represents, at best, a time-​averaged rep-
resentation of a potential function that in fact is fluctuating at frequencies several
orders of magnitude greater than those in which the molecule can be rotating
and its nuclei vibrating with respect to one another. Put this way, our questions
can be viewed as prompted by the question: How physically meaningful are the
time-​averaged potential functions?
Further complicating the situation are the spectral data themselves, from
which the values of the time-​averaged molecular constants are derived. The
spectral lines, of course, are themselves aggregates of multiple photon-​molecule
interactions over time. They are, moreover, aggregates as well over the emission
and absorption of photons by large numbers of molecules. The spectral data are
thus, in fact, data models, as Patrick Suppes (1962) would put it, and so they, too,
as employed in the determination of the values of the molecular constants, are
representations in van Fraassen’s sense. In saying this, we are not denying that the
spectral data provide experimental access to the structure of diatomic molecules,
their rotational and vibrational motions, and the relation between these motions
and radiation. We are only pointing out that the experimental access they pro-
vide is not to the dynamical behavior of an individual molecule at a time scale
that involves no time-​averaging over still smaller time scales. To recognize this is
to recognize what is giving force to our questions about the physical meaningful-
ness of the constants for diatomic molecules.
Some may think that the time-​averaging of any physical process cannot help
but be physically meaningful. One need merely turn to the history of less than
successful efforts to represent turbulent fluid flow in terms of time-​averaged
parameters (like Reynolds stresses) to appreciate that this is not true (see Tritton
1988, p. 299ff, for example). That there is a time-​averaged set of parameters
that can serve to represent, for a higher-​order time scale, a temporally evolving
physical process involving intense unsteadiness at smaller time scales is some-
thing that has to be established. The seminal Born-​Oppenheimer paper of 1927
was thus important in justifying this under certain assumptions in the case
of molecules; what is missing in the case of turbulent flow is a counterpart to
this paper. Born-​Oppenheimer by itself, however, does not establish that the
constants for diatomic molecules are physically meaningful, for this is not a the-
oretical question—​that is, a question about the consistency of such averaging
 Realism and Molecular Spectroscopy 175

with quantum theory—​but rather an empirical question about deviations from

the averaging.
For discrepancies between calculation and observation to give support to the
physical meaningfulness of those constants, accordingly, some sort of empir-
ical access is needed to time scales eliminated from consideration by the Born-​
Oppenheimer and Born adiabatic approximations. In the decades following
Herzberg, the primary approach to gaining such access has been to pursue subtle
anomalies—​in particular, minor violations of integral spacing of lines—​in some
of the rotational and rotational-​vibrational spectral bands. Such anomalies are
termed “perturbations” in the literature. The goal has been to link them to specific
terms defining the potential function that are ignored in the Born-​Oppenheimer
approximation because they involve electron-​nuclei interactions. This research
has been summarized in two books published during the first decade of the pre-
sent century, the thousand-​page Rotational Spectroscopy of Diatomic Molecules
(Brown and Carrington 2003) and the seven-​hundred-​page The Spectra and
Dynamics of Diatomic Molecules (Lefebvre-​Brion and Field 2004). We are not
yet in a position to assess how much support the results presented in these
books provide for the physical meaningfulness of the standard diatomic molec-
ular constants. Clearly, nevertheless, these provide a source for addressing our
questions about any parallel between those constants and those of Keplerian
orbits.
Another approach to gaining experimental access to the time scales of
electron-​nuclei interactions that emerged late in the twentieth century is so-​
called “femtochemistry.” The perturbation of molecules in this approach results
from external excitation by tunable laser pulses of duration within a femto-​
second (10-​15 second) time scale. The primary focus of this research is, in the
words of the leading figure of the field Ahmed Zewail, to probe “the ultrafast
dynamics of the chemical bond.” This process nevertheless can yield results on
fluctuations in the potential function between the nuclei of diatomic molecules.
For example, in one of the early results in the field, the response of NaI molecules
to short-​term laser pulses involved an interaction between covalent-​bond and
ionic-​bond potential functions, yielding an oscillatory response in the 300-​
femtosecond range to the excitation (Zewail 1994, pp. 219–​245).3 Again, we are
not yet in a position to assess how much support femtochemical results might
provide for the physical meaningfulness of the standard diatomic molecular
constants. That they yield short-​time information about the actual potential

3 The two specific papers to which we are referring are Todd S. Rose, Mark J. Rosker, and Ahmed

H. Zewail, “Femtosecond Real-​Time Probing of Reactions. IV. The Reactions of Alkali Halides,”
Journal of Chemical Physics, vol. 91, 1989, pp. 7415–​7436; and P. Cong, A. Mokhtari, and A. H.
Zewail, “Femtosecond Probing of Persistent Wave Packet Motion in Dissociative Reactions: Up to 40
ps,” Chemical Physics Letters 172, 1990, pp. 109–​113.
176 Historical Cases for the Debate

function does make them a possible source for such support that needs to be
examined.
We have been emphasizing research after 1950 as the most instructive way
for comparing the physical meaningfulness of the diatomic molecular constants
with that of the Keplerian orbital parameters. Evidence supporting such a par-
allel between the two had emerged even before 1930. Table 8.1 shows the con-
sistency of the values of the molecular constants obtained from spectra with the
change of the isotopic mass of the hydrogen nucleus. The spectroscopic search
that resulted in the 1931 discovery of deuterium was prompted by the discovery
two years earlier of first one and then a second isotope of oxygen besides the
16O that had previously been used, under the assumption of its being unique, as

the basis for the system of atomic weights. Specifically, long-​exposure absorp-
tion spectra of air had yielded an anomalous band spectrum of faint lines shifted
slightly in frequency from a long-​established band spectrum. Analysis showed
that the shift corresponds to a 16O18O molecule under the assumption of a single
time-​averaged potential function (Giaque and Johnston 1929a, 1929b). Still
more extended time exposure then revealed as well an even more rare 17O iso-
tope (Giaque and Johnston 1929c and 1929d). Band spectra of other molecules
involving oxygen, such as NO, subsequently independently confirmed both
(Naudé 1930), as did mass spectrometry (Aston 1932).
The parallel between the discovery of Neptune from anomalies in the motion
of Uranus and the discovery of 18O from an anomalous band in the spectrum
of oxygen extends beyond this. Neptune was widely heralded at the time as ev-
idence in support of Newtonian gravity. So too was the discovery of 18O for the
selection rules of the new quantum theory:

this observation must be regarded as a particularly impressive confirmation of

quantum mechanical predictions; for example, in the bands of the O16O18, all
the lines appear, whereas every second line is missing for O16O16. (Herzberg
1939, p. 150)

The evidence here involved isotopic shifts only in the time-​averaged values of
the respective molecular constants, but some subtleties in these shifts were at
least attributed back then to short-​time interactions between electrons and nu-
clei ignored in the time-​averaging (Van Vleck 1936).
Let us return to the beginning of this paper. Realists have tried to respond to
van Fraassen by showing how, in the case of Perrin, concordant measurements of
parameters such as Avogadro’s number can be achieved and examining what this
concordance of measurements is telling us about the reality of molecules. The
measurements in question dated from 1908 to 1913. Whatever support for the re-
ality of molecules resulted from them at the time, if research over the subsequent
 Realism and Molecular Spectroscopy 177

decades had failed to determine anything more about the structure and shape of
molecules beyond the spherical hypothesis, this would have rendered the con-
clusion that molecules are real to be of, at best, limited empirical significance.
In fact, however, the rotational and vibrational spectra of diatomic molecules,
and in some cases of polyatomic molecules as well, have yielded seemingly im-
pressive conclusions about detailed specifics of their structure and shape. We
say “seemingly” because the values of the constants characterizing their struc-
ture and shape involve a time-​averaging that glosses over extremely complicated
short-​term fluctuations in their structure. From the point of view of the realist-​
instrumentalist debate, the question therefore arises whether these constants are
physically meaningful (in the sense that their values have causal implications)
or whether to the contrary they amount to nothing more than a heuristically
useful representation of the spectral data with no physical significance beyond
these data. We claim that the question of the physical meaningfulness of these
constants is now a far more appropriate focus for the realist-​instrumentalist de-
bate than are the concordant measurements of the 1908–​1913 period.
We can put this point in a slightly different way that might be more helpful.
We believe van Fraassen (2009) is right to emphasize that the primary aim of
researchers in microphysics in the early twentieth century was making stable
concordant measurement, or empirical grounding, to use van Fraassen’s term,
possible for microphysical parameters. Where we differ from van Fraassen is in
how we view the development of evidence after stable concordant measurement
has been achieved. In all the cases we know of from the history of the physical
sciences, scientists have continued to pursue discrepancies from concordant
measurements, no matter how small they may be, for these discrepancies have
always led to discoveries of new physical sources that had, in some cases, not
even been considered before. This methodology requires a distinction to be
made between representations that yield deviations that have physically identi-
fiable sources, and those that do not yield such deviations, which we term phys-
ical meaningfulness. We believe that a potentially important step forward in the
realist-​instrumentalist debate can be achieved by focusing on the questions of
how exactly to characterize physical meaningfulness and how it might be pos-
sible to distinguish parameters that are physically meaningful from those that are
not, even as one must acknowledge that they do not correspond to any situation
that actually occurs.
Focusing the realism-​instrumentalism debate on the question of the physical
meaningfulness of the parameters in the standard representation of diatomic
molecules has some virtues beyond those we have stressed earlier. As we have
provisionally drawn the distinction, whether a parameterization is physically
meaningful turns on how strong the evidence is in support of a specific phys-
ical source for each systematic discrepancy between theoretical calculation and
178 Historical Cases for the Debate

observation. Correspondingly, one virtue of focusing on diatomic molecules

is the amount of evidence that has been developed over the course of the last
90 years of research on them—​at least comparable to, if not substantially ex-
ceeding, the scope of the evidence developed over the last three centuries in ce-
lestial mechanics. Another virtue is that the assessment of whether the evidence
has established specific physical sources is something on which instrumentalists
and realists should agree, even though their interpretations of what this phys-
ical source amounts to will likely differ. We do not see how van Fraassen can
deny that the evidence has established the isotope effects we noted earlier, even
though he will continue to reject the reality of atomic nuclei. Equally, we do not
see how realists can deny that the diatomic molecular constants provide a repre-
sentation, in van Fraassen’s sense, that glosses over details and is in some respects
counterfactual. If both sides agree that the evidence confirms that the representa-
tion is nevertheless physically meaningful, the debate comes down to whether an
agreed on physical source for a discrepancy shows, for example, that the source is
perturbing an actual rotation or vibration.
Thirty years ago, in an article more widely cited than heeded, Howard Stein
proposed “that between a cogent and enlightened ‘realism’ and a sophisticated
‘instrumentalism’ there is no significant difference—​no difference that makes a
difference” (Stein 1989, p. 61). Focusing on the last 90 years of research on the
structure of diatomic molecules might well prove to be the ideal testing ground
for Stein’s proposal.4

References
Achinstein, Peter. 2001. The Book of Evidence. Oxford: Oxford University Press.
Aston, F. W. 1932. “Mass-​Spectra of Helium and Oxygen,” Nature, vol. 130, p. 21f.
Bernath, Peter F. 2016. Spectra of Atoms and Molecules, Third Edition. Oxford: Oxford
University Press.
Bernath, Peter F., and Sean McLeod. 2001. “DiRef, A Database of References Associated
with the Spectra of Diatomic Molecules,” Journal of Molecular Spectroscopy, vol. 207,
p. 287.
Bohr, Niels. 1913. “On the Constitution of Atoms and Molecules,” Philosophical Magazine
Series 6, vol. 26, pp. 1–​25.
Boltwood, Bertram B., and Ernest Rutherford. 1911. “Production of Helium by Radium,”
Philosophical Magazine Series 6, vol. 22, pp. 586–​604.
Born, Max, and Robert Oppenheimer. 1927. “On the Quantum Theory of Molecules,”
tr. H. Hettema, in H. Hettema, Quantum Chemistry: Classic Scientific Papers.
Singapore: World Scientific, 2000, pp. 1–​24.

4 Part of the research for this paper was supported by the Ministry of Education, Singapore,

under its Academic Research Fund Tier 1, No. RG156/​18, as well as the Radcliffe Institute for
Advanced Study.
 Realism and Molecular Spectroscopy 179

Brown, John, and Alan Carrington. 2003. Rotational Spectroscopy of Diatomic Molecules.
Cambridge: Cambridge University Press.
Chalmers, Alan. 2009. The Scientist’s Atom and the Philosopher’s Stone: How Science
Succeeded and Philosophy Failed to Gain Knowledge of Atoms. Dordrecht: Springer.
Chalmers, Alan. 2011. “Drawing Philosophical Lessons from Perrin’s Experiments on
Brownian Motion: A Response to van Fraassen,” British Journal for the Philosophy of
Science, vol. 62, pp. 711–​732.
Chaudesaigues, M. 1908. “Le mouvement brownien et le formule d’Einstein,” Comptes
Rendus, vol. 147, pp. 1044–​1046.
Giauque, W. F., and H. L Johnston.1929a. “An Isotope of Oxygen, Mass 18,” Nature, vol.
123, p. 318.
Giauque, W. F., and H. L Johnston.1929b. “An Isotope of Oxygen, Mass 18. Interpretation
of the Atmospheric Absorption Bands,” Journal of the American Chemical Society, vol.
51, pp. 1436–​1441.
Giauque, W. F., and H. L Johnston.1929c. “An Isotope of Oxygen of Mass 17 in the Earth’s
Atmosphere,” Nature, vol. 123, p. 831.
Giauque, W. F., and H. L Johnston.1929d. “An Isotope of Oxygen, Mass 17, in the Earth’s
Atmosphere,” Journal of the American Chemical Society, vol. 51, pp. 3528–​3534.
Glymour, Clark. 1980. Theory and Evidence. Princeton: Princeton University Press.
Hecht, Charles E. 1990. Statistical Thermodynamics and Kinetic Theory. Mineola,
NY: Dover.
Herzberg, Gerhard. 1939. Molecular Spectra and Molecular Structure, I. Diatomic
Molecules. New York: Prentice-​Hall.
Herzberg, Gerhard. 1950. Molecular Spectra and Molecular Structure: I. Spectra of
Diatomic Molecules, Second Edition. Princeton: D. Van Nostrand.
Herzberg, Gerhard. 1971. The Spectra and Structures of Free Simple Radicals: An
Introduction to Molecular Spectroscopy. Mineola, NY: Dover.
Huber, Klaus Peter and Gerhard Herzberg. 1979. Molecular Spectra and Molecular
Structure: IV Constants of Diatomic Molecules. New York: Van Nostrand Reinhold.
Jeans, James. 1916. The Dynamical Theory of Gases 2nd edition. Cambridge,
UK: Cambridge University Press.
Kelvin, Lord. 1904. Baltimore Lectures on Molecular Dynamics and the Wave Theory of
Light. Cambridge: Cambridge University Press.
Lefebvre-​Brion, Hélène, and Robert W. Field. 2004. The Spectra and Dynamics of Diatomic
Molecules. Amsterdam: Elsevier Academic Press.
Mayo, Deborah G. 1996. Error and the Growth of Experimental Knowledge.
Chicago: University of Chicago Press.
Mercator, Nicholaus. 1676. Institutionum Astronomicarum Libri Duo, De Motu Astrorum
Communi & Proprio. London: Samuel Simpson.
Meyer, Oskar Emil. 1899. The Kinetic Theory of Gases. tr. Robert E. Baynes.
London: Longmans, Green; Die Kinetische Theorie der Gase. Breslau: Marschke &
Berendt, 1899.
Millikan, Robert A. 1911. “The Isolation of an Ion, A Precision Measurement of Its
Charge, and the Correction of Stokes’s Law,” The Physical Review, vol. 32, pp. 349–​397.
Millikan, Robert A. 1913. “On the Elementary Electrical Charge and the Avogadro
Constant,” The Physical Review, Second Series, vol. II, pp. 109–​143.
Naudé, S. Meiring. 1930. “The Isotopes of Nitrogen, Mass 15, and Oxygen, Mass 18 and
17, and their Abundances,” Physical Review, vol. 36, pp. 333–​346.
180 Historical Cases for the Debate

Nernst, Walther. 1895. Theoretical Chemistry from the Standpoint of Avogadro’s

Rule and Thermodynamics, tr. Charles Skeele Palmer of first German edition.
London: Macmillan; 1923. tr. L. W. Cobb of the eighth-​ tenth German edition.
London: Macmillan.
Newton, Isaac. 1999. The Principia: Mathematical Principles of Natural Philosophy, tr. I.
Bernard Cohen and Anne Whitman. Berkeley: University of California Press.
Perrin, Jean. 1910. Brownian Movement and Molecular Reality, tr. F Soddy. London: Taylor
and Francis.
Perrin, Jean. 1912. “Les Preuves de la Réalité Moléculaire (Études spécial des émulsions),”
in Théorie du Rayonnement et Les Quanta: Rapports et Discussions de la Réunion Tenue
à Bruxelles,du 30 Octobre au 3 Novembre 1911, Sous Les Auspices de M. E. Solvay, ed. P.
Langevin and M. de Broglie, Paris: Gauthier-​Villars, pp. 150–​250.
Planck, Max. 1900. “On the Theory of the Energy Distribution Law of the Normal
Spectrum,” tr. D. ter Haar, in D. ter Haar, The Old Quantum Theory. Oxford: Pergamon
Press, 1967, pp. 82–​90.
Psillos, Stathis. 2011a. “Making Contact with Molecules: On Perrin and Achinstein,”
in Philosophy of Science Matters: The Philosophy of Peter Achinstein, ed. Gregory J.
Morgan. Oxford: Oxford University Press, pp. 177–​190.
Psillos, Stathis. 2011b. “Moving Molecules Above the Scientific Horizon: On Perrin’s Case
for Realism,” Journal for General Philosophy of Science, vol. 42, pp. 339–​363.
Psillos, Stathis. 2014. “The View from Within and the View from Above: Looking at van
Fraassen’s Perrin,” in Bas van Fraassen’s Approach to Representation and Models in
Science, ed. Wenceslao J. Gonzalez. Dordrecht: Springer, pp. 143–​166.
Rutherford, Ernest. and Hans Geiger. 1908. “The Charge and Nature of the α-​Particle,”
Proceedings of the Royal Society of London, Series A, vol. 81, No. 546, pp. 162–​173.
Salmon, Wesley. 1984. Scientific Explanation and the Causal Structure of the World.
Princeton: Princeton University Press.
Smith, George E. 2014. “Closing the Loop: Testing Newtonian Gravity, Then and Now,” in
Newton and Empiricism, ed. Zvi Biener and Eric Schliesser. Oxford: Oxford University
Press, pp. 262–​351.
Stein, Howard. 1989. “Yes, but . . . Some Skeptical Remarks on Realism and Anti-​Realism,”
Dialectica, vol. 43, pp. 47–​65.
Streete, Thomas. 1661. Astronomia Carolina, A New Theory of the Celestial Motions.
London: Lodowick Lloyd.
Suppes, Patrick. 1962. “Models of Data,” in Logic, Methodology, and the Philosophy of
Science: Proceedings of the 1960 International Conference, ed. Ernest Nagel, Patrick
Suppes, and Alfred Tarski. Stanford: Stanford University Press, pp. 252–​261.
Tritton, D. J. 1988. Physical Fluid Dynamics, 2nd ed. Oxford: Clarendon Press.
Van Fraassen, Bas C. 2008. Scientific Representation. Oxford: Oxford University Press.
Van Fraassen, Bas C. 2009. “The Perils of Perrin, in the Hands of Philosophers,”
Philosophical Studies, vol. 143, No. 1, pp. 5–​24.
Van Vleck, J. H. 1936. “On the Isotope Corrections in Molecular Spectra,” Journal of
Chemical Physics, vol. 4, pp. 327–​338.
Wing, Vincent. 1669. Astronomia Britannica. London: Georgii Sawbridge.
Zewail, Ahmed H. 1994. Femtochemistry: Ultrafast Dynamics of the Chemical Bond,
Volume I. Singapore: World Scientific.
 PART II
C ON T E MP OR A RY S C I E NT I F IC
R E A L ISM
 9
The Historical Challenge to Realism
and Essential Deployment
Mario Alai

Department of Pure and Applied Sciences

University of Urbino Carlo Bo
[email protected]

9.1 Deployment Realism

Scientific realists use the “No Miracle Argument” (NMA): it would be a miracle
if theories were false, yet got right so many novel and risky predictions. Hence,
predictively successful theories are true. Of course, one could easily make up a
theory with completely false theoretical assumptions which predicted a phenom-
enon P (call it a FP-​theory) if she knew P in advance and used it in framing the
theory. But how could she think of a FP-​theory, without knowing P? Or knowing
P but without using it in building the theory? In fact, it is puzzling how one could
have built a FP-​theory even if she used P inessentially: suppose Jill built a FP-​
theory by knowing and using P, but she could have done without it, because, quite
independently of her, John built the same theory without using P. This I call Jill
using P inessentially, and it is something hard to explain, because it is under-
standable how the theory was built by Jill, but not by John (Lipton 1991, 166; Alai
2014c, 301).
So, how could one find a FP-​theory without using P essentially? Well, one
could do this if P were very poor in content, i.e., very probable: any false theory
could predict the existence of some new planet somewhere in the universe. But
it took Newton’s theory to predict precisely the existence of a planet of such and
such mass with such and such orbit as Neptune. So, the riskier a prediction is (i.e.,
the more precise, rich in content, hence improbable it is), the less likely it is that it
has been made by a completely false theory.
It might be objected that since a falsity may entail a truth, even false theo-
ries might make novel predictions. Granted, in the Hyperuranion of possible
theories there are some with completely false theoretical assumptions which

Mario Alai, The Historical Challenge to Realism and Essential Deployment In: Contemporary Scientific Realism.
Edited by: Timothy D. Lyons and Peter Vickers, Oxford University Press. © Oxford University Press 2021.
DOI: 10.1093/​oso/​9780190946814.003.0009
184 Contemporary Scientific Realism

(together with appropriate auxiliary assumptions) make true novel and improb-
able predictions (call them FNIP-​theories). However, their rate is very small, for
it is a logical fact that the less probable a consequence is, the fewer the premises
entailing it. Hence, it is extremely improbable that FNIP-​theories are found by
chance. But since scientists look for true theories, only by chance can they pick a
FNIP-​theory. Therefore, it is extremely unlikely that scientists find a FNIP-​theory.1
For instance, quantum electrodynamics predicted the magnetic moment of the
electron to be 1159652359 × 10–​12, while experiments found 1159652410 × 10–​12:
hence John Wright (2002, 143–​144) figured that the probability to get such an
accuracy by chance, i.e., through a false theory, is as low as 5 × 10–​8.
Here (following Alai 2014c) I will speak of “novel predictions” (NP) when the
predicted phenomenon is either unknown (historically novel) or not used (use-​
novel) or not used essentially (functionally novel) by the theorist. So, it is prac-
tically impossible that a novel prediction which in addition is risky, i.e., a highly
improbable, was derived from a completely false theory.
Against the claim that success warrants truth, Laudan (1981) had objected that
many false theories in the history of science were successful, so realists replied
that however those theories didn’t have novel predictive success.2 Unfortunately,
it turns out that also many false theories had novel success (see the list in Lyons
2002, 70–​72). Selective realists acknowledged the problem and argued that the
NMA does not warrant the complete truth of theories, but only their partial truth.
In particular, for Kitcher’s (1993) and Psillos’s (1999) “Deployment Realism”
(DR), we should be committed only to those particular hypotheses which were
deployed in deriving novel predictions: this is Psillos’s “divide et impera” move
against Laudan’s “reductio” (Psillos 1999, 102, 108, passim).
A rejoinder came from Timothy Lyons: first, he explained that Laudan’s orig-
inal objection was not “an articulation of the pessimistic meta-​induction toward
the conclusion that our present theories are probably false,” but a modus tollens
refutation of the NMA (a “meta-​modus tollens”) (Lyons 2002, 64–​65; 2006, 557).
Secondly, he criticized DR by listing a number of false hypotheses which were ac-
tually deployed in the derivation of various novel risky predictions—​including
Neptune’s discovery (2002; 79–​83; 2006). Deployment realists, however, can an-
swer this new challenge by arguing that the NMA does not warrant the truth
of all hypotheses actually used in deriving novel predictions, but only of those
which were essential, i.e., strictly necessary to those predictions. False hypoth-
eses are no counterexample to the NMA if they were used de facto, but actually
superfluous, i.e., non-​essential.

1 Alai 2014c, §1; 2014d, §5; 2016, §4. More on this objection at §9.7.1.
2 Musgrave (1985, 1988); Lipton (1993, 1994); Psillos (1999); Sankey (2001).
Historical Challenge to Realism and Essential Deployment 185

Thus, essentiality is crucial to two key claims of DR: (1) a predicted phenom-
enon is novel only if it was not used essentially in building the theory which
predicts it; and (2) a hypothesis is most probably true only if it was essential in
deriving a novel prediction. Here we shall focus on the latter claim.
According to Psillos (1999, 110) a hypothesis H is deployed essentially in
deriving a novel prediction NP if

(1) NP follows from H, together with some other hypotheses OHs and auxil-
iary assumptions AA, but not from OHs & AA alone;
(2) no other hypothesis H* is available which can do the same job as H, viz. is
(a) compatible with OHs and AA,
(b) non-​ad hoc,
(c) potentially explanatory, and
(d) together with OHs and AA predicts NP.

Conditions (1) and (2) capture the idea that H is essential, i.e., strictly necessary,
when (1) OHs without H could not predict NP, and (2) no alternative hypothesis
is available which could substitute H in the derivation of NP.

9.2 Lyons’s Criticism of Psillos’s Criterion of Essentiality

However Lyons (2006; 2009) criticized this criterion of essentiality, in particular

condition (2). First, he argued that it is too vague to be applicable to any histor-
ical case (2006, 542). For instance, it is unclear when H* should not be avail-
able: when the theory is put forward? when the prediction is derived? when the
prediction is confirmed? at some point in the future? Further unclear points are
what “potentially explanatory” means; whether each of the elements of OHs and
AA also need to be “essential” by this criterion; whether the replacement hypo-
thesis needs to be consistent with those components of OHs and AA which are
“essential” for other predictions but unnecessary for the prediction of concern;
whether H* can result in the loss of other confirmed predictions.
Secondly, according to condition (2) a competitor H* makes H inessential
only if H* is compatible with OHs and AA, and it is non-​ad hoc (i.e., predicts
phenomena not used in constructing H*: Psillos 1999, 106). But Lyons notices
that in actual history “most (if not all) well-​respected theories” were either in-
compatible with OHs and AA, or ad hoc in Psillos’s sense, or both (Lyons 2009,
149). So the requirement placed by (2) on admissible competitors is “too strict
to permit even some of the most exemplary scientific theories . . . to qualify as
competitors” (150), and this means that (2) is too weak (or practically empty), for
most (if not all) possible hypotheses H will lack admissible competitors. Psillos
186 Contemporary Scientific Realism

didn’t understand his conditions for essentiality as a definition, but just as a crite-
rion: he writes “H indispensably contributes to the generation of P if . . . [(1) and
(2) are met],” not “if and only if . . . [(1) and (2) are met]” (1999, 110). Yet, Lyons’s
criticisms are at least prima facie well-​founded, and must be taken seriously.
But Lyons also claims that his criticisms force deployment realists to
abandon the essentiality requirement altogether. Besides, he claims that the
“crucial” or “fundamental” insight of DR is that we should credit “those and
only those constituents that were genuinely responsible for, that actually led
scientists to, specific predictions” (2006, 543), no matter whether they were
indispensable or not. In fact, he argues that Psillos, after introducing his cri-
terion of essentiality (Psillos 1999, 109–​110), “never mentions that rigid crite-
rion again”3 and fails to show that his case studies comply with it (Lyons 2009,
143). On the contrary, according to Lyons, Psillos often talks of commitment
to constituents which simply “generated,” or “brought about,” or were “respon-
sible for,” or “genuinely contributed to,” or “really fueled” (Psillos 1999, 108–​
110) the success of the theory (Lyons 2006, 540). Moreover, Lyons says that
the same “fundamental insight” is also embraced, in some form or another, by
Kitcher, Niiniluoto, Leplin, and Sankey (538). Thus, he holds, DR should get
rid of the essentiality requirement and “return to the crucial insight” of actual
deployment (543).
At this point, however, Lyons argues that many assumptions which in his-
tory were actually employed and led to novel predictions are false. In fact, in
various papers (2002, 2006, 2016, 2017) he discusses a number of striking novel
predictions, toward each of which many false assumptions were employed (we
shall briefly review some of them later). Hence, he concludes the NMA does not
work even when restricted to particular hypotheses, and DR is false: “If we take
seriously the no miracles argument [i.e., if novel predictions derived from false
assumptions are a miracle] . . . we have witnessed numerous miracles” in history
(2006, 557).
However, in reply, a number of points should be noticed: first, by accepting
such purported “miracles,” one is left with the problem of explaining how they
are possible. If one wonders how scientist S made the novel risky prediction NP
which was then confirmed, a possible answer is that NP followed from true
assumptions in S’s theory, for all consequences of true assumptions are true. But
one cannot answer that NP followed from some false hypothesis in S’s theory,
because that would entail that S had luckily assumed a false hypothesis entailing
NP, and as mentioned, this is extremely improbable.4 Lyons (2002; 2003) and

3 Personal communication, but see 2006, 542.

4 More on this at section 9.7.1. One might undercut the NMA and DR also by rejecting the need or
possibility of explaining unlikely events, or of explanations in general. But I cannot discuss that rad-
ical objection here.
Historical Challenge to Realism and Essential Deployment 187

others proposed alternative explanations, not involving the truth of the deployed
hypotheses, but I have argued elsewhere that they all fail (Alai 2012, 88–​89;
2014a).
Secondly, as concerns which one is the “fundamental insight” of DR that
cannot be abandoned vs. what is merely accessory and should be dropped,
two questions must be distinguished: (I) the exegetic question whether Psillos
understands “deployment” as merely actual deployment, or as essential deploy-
ment; and (II) the theoretical question of how deployment should be understood
if we are to give DR its best currency in order to correctly assess its tenability.
As to (I), when Psillos claims that we should credit constituents that are “re-
sponsible” for success (1999, 108–​109), Lyons understands “responsibility” as ac-
tual deployment, suggesting that this was the original core intuition of DR (2006,
540, 543). But this is incorrect. As Lyons himself acknowledges (2006, 539) on
the next page Psillos specifies that constituents to be credited are those “that
made essential contributions” to success and that “have an indispensable role” in
it, and right after this he spells out his conditions (1)–​(2) (109–​110). So, by re-
sponsibility Psillos clearly means essential responsibility.
Even apart from this, when an author says two different things, in the exegesis we
are allowed to discount one of them only if they are contradictory. But if they are not
contradictory we must keep both, and understand how they can be reconciled. In
this case, sometimes Psillos talks of responsibility or contribution, or deployment,
etc., without further qualifications, sometimes he introduces the essentiality or in-
dispensability qualification. But these two talks are not contradictory, and can be
easily reconciled, since the latter is just a specification of the former. In fact, when
talking plainly of responsibility or contribution, Psillos never adds “no matter
whether essential or not” or the like. Therefore, even his plain talk of responsibility
or contribution must be understood as implying “essential” or “indispensable.”
Even if in his case studies (1999, ch. 6) Psillos fails to explicitly refer back to
his criterion of essentiality and to explicitly show how it applies, he does use the
essentiality requirement. For instance, he argues that we need not be committed
to the caloric hypothesis, although it was actually deployed in Laplace’s predic-
tion of the speed of sound in air, because that prediction “did not depend on
this hypothesis” (121). Hence, he obviously would not give up his essentiality
requirement. Moreover, in the absence of any indication to the contrary, and
as this quotation suggests, we must assume that he understands the essentiality
requirement precisely in terms of his essentiality criterion.
Summing up, from the exegetic point of view, ignoring the essentiality re-
quirement would be misinterpreting Psillos’s position. But apart from this, since
here we must assess the tenability of DR, what matters more is not the exeget-
ical question, but (II) the theoretical question of how DR should be understood
in order to give it the most favorable interpretation. Now again, the answer is
188 Contemporary Scientific Realism

the essential deployment interpretation, for it immunizes DR from Lyons’ pur-

ported historical counterexamples.
In fact, to begin with, by Occam’s razor essentiality is required to the cogency
of the NMA: in abductions we can assume only what is essential, i.e., the weakest
hypothesis sufficient to explain a given effect; but if a hypotheses, although
deployed, was not essential in deriving NP, it is not essential in explaining its
derivation either; therefore deployment realists need not (and must not) be com-
mitted to its truth.
Moreover, Psillos’s conditions (1)–​(2) are just one way to spell out essentiality.
Therefore, even if Lyons succeeds in criticizing them, he has not shown that no
other characterization of essentiality is possible, hence he cannot conclude that
this condition must be abandoned. In fact, he only suggests that it be abandoned,
so proposing a version of DR deprived of the essentiality requirement which he
(wrongly, I argued) believes to be Psillos’s “fundamental” version. Therefore, al-
though his refutation of this latter version by meta-​modus tollens from histor-
ical cases is effective, it does not affect actual DR, i.e., the essential deployment
version.
Granted, Lyons’s analysis of Psillos’s conditions (1)–​(2) is correct, and his
criticisms of (2) are convincing. But I will propose a different condition, which
performs the task Psillos had in mind, yet is simpler, less troublesome, and
escapes Lyons’s criticisms. This condition is not altogether new, having been used
or presupposed at various places, I suggest, by Psillos himself (e.g., 1999, 119–​
121) and explicitly proposed by Vickers, who writes:

H does not merit realist commitment whenever H is doing work in the deri-
vation solely in virtue of the fact that it entails some other proposition which
itself is sufficient, when combined with the other assumptions in play, for the
relevant derivational step (2016, §3).

9.3 An Improved Criterion of Essentiality Escaping

Lyons’s Criticisms

I propose to substitute the crucial condition (2) in Psillos’s criterion with:

(2’) Th
 ere is no other hypothesis H* which is proper part of H (hence weaker
than H) which together with OHs and AA entails NP.

Here in section 9.3 I explain (2’), spelling out the intuitive notion of indispen-
sability more precisely through Yablo’s (2014) concept of proper parthood, and
arguing that the troublesome requirements of Psillos’s criterion can be dropped
Historical Challenge to Realism and Essential Deployment 189

without loss, thus escaping Lyons’s criticisms; in section 9.4 I show how my crite-
rion immunizes DR from Lyons’s historical counterexamples; in section 9.5 I hold
that essentiality in this sense cannot be detected prospectively, and Vickers’s op-
timism on the identifiability of non-​essentiality should be complemented by
some less optimistic considerations; but far from causing problems for realists,
this frees them from unreasonable obligations; in section 9.6 I explain in more
details why my criterion is enough, although apparently weaker than Psillos’s; in
section 9.7 I answer some actual and potential objections.
Intuitively, a hypothesis H* is a proper part of H iff its content is part of
the content of H but does not exhaust it. At first approximation and for most
purposes, a part of H is therefore a hypothesis entailed by H but not entailing it.
But this is not always enough: for instance consider

(i) It’s Spring

and

(ii) It’s April.

As (i) is entailed by (ii) without entailing it, (i) is a proper part of (ii). But

(iii) It’s April or the streets are wet

is implied by (ii) without entailing it, however it is not a (proper) part of it, for
it brings in a different subject matter. Therefore, just characterizing parthood
in terms of entailment will not always do for our purposes. Instead, we must
understand parthood as Yablo (2014):5 H* is a part of H iff it is entailed by H
and preserves its subject matter. The subject matter of a proposition consists
of two classes: that of its truthmakers (or reasons why it is true) and that of its
falsemakers (or reasons why it is false). A proposition H* entailed by H preserves
H’s subject matter, hence, is part of it, iff every truthmaker/​falsemaker of H* is
entailed by a truthmaker/​falsemaker of H. Moreover, H* is a proper part of H iff
every truthmaker/​falsemaker of H* is entailed by a truthmaker/​falsemaker of H,
but not vice versa. This explains the intuitive idea that (iii), although entailed by
(ii), is not part of it: because some truthmakers of (iii) (e.g., “it’s raining”) are not
entailed by any truthmaker of (ii).6

5 I thank Matteo Plebani for pointing this out to me. See Plebani (2017).
6 Although Yablo’s approach helps us to properly explain essentiality, it was not designed to this
purpose, but to answer the general question of “aboutness,” i.e., the subject-​matter of sentences: why
and how sentences with the same truth-​conditions say different things, like “here is a sofa” and “here
is the front of a currently existing sofa, and behind it is the back.” Thus, it offers a natural solution
to a wide variety of problems concerning agreement, orders, possibilities, priority, explanation,
190 Contemporary Scientific Realism

Condition (2’) expresses precisely the above-mentioned Occam’s requirement,

namely that H is not redundant, i.e., that one could not explain the derivation of
NP by assuming the truth of something less than H. For instance, suppose that in-
stead of his actual gravitation theory Newton had advanced the following theory:

(H) inside each massive body there resides a demi-​god, which attracts the
demi-​gods dwelling in each other body by a force F = Gm1m2/​r2.

H would predict the same novel phenomena as Newton’s actual theory, but we
wouldn’t need to believe it, because it is redundant, i.e., not essential to those
predictions. In fact, H consists in the following assumptions:

(H*) each body attracts all other bodies by a force F = Gm1m2/​r2

(Hd-​g) F is exerted by a demi-​god residing inside each body.

H would violate condition (2’), because there would be another hypothesis H*

(i.e., Newton’s actual theory) which is a proper part of H and sufficient to derive
its novel predictions.
This alternative condition (2’) avoids Lyons’s vagueness and emptiness
objections to Psillos’s condition (2). First, (2’) is not too vague because of five
reasons: (i) there is no question about when the alternative H* is available,
because (2’) does not exclude just that H* is available, but even that it is logi-
cally possible (for if it were possible, H would be redundant in the prediction
of NP, hence possibly false); (ii) there is no need to specify what “explanatory”
means, since (2’) has no such requirement; (iii) it is not required that the elem-
ents of OHs and AA are also essential;7 (iv) since (2’) excludes only hypotheses
H* which are part of H, ipso facto such hypotheses are compatible with all the
components of OHs and AA with which H is compatible; (v) alternative hypoth-
eses H* are excluded no matter whether substituting H by H* results in the loss of
other confirmed predictions or not: if H allows us to derive two predictions NP1
and NP2, and the weaker H* allows us to derive NP1 but not NP2, then H is not es-
sential with respect to NP1, but it may be essential with respect to NP2, in which
case it is most probably true.
Lyons’s second criticism was that (2) is practically empty, because it excludes
only alternative hypotheses which are compatible with OHs and AA, while in

knowledge, partial truth, and confirmation. For instance, a white marble does not confirm (1) “All
crows are black” because (1) does not say the same as (2) “All non-​black things are non-​crows,” and
Yablo explains why (2014, pp. xiv, 7–​8).

7 Whether they are or not can be decided in the same way as for H. In section 9.6 I shall notice that

different but compatible assumptions may all be essential for a given novel prediction.
Historical Challenge to Realism and Essential Deployment 191

most actual cases alternative hypotheses are not compatible with OHs and AA.
My condition (2’) escapes this problem, since it excludes, without exceptions, all
the alternative hypotheses which are part of H and can derive NP. In fact, in sec-
tion 9.4 I shall mention some historical examples of hypotheses which violated
condition (2’), confirming that it is not empty.
Yet, it might seem that here lurks a further difficult problem: as just said, all
the alternative hypotheses excluded by (2’) are compatible with OHs and AA,
since they are part of H. But how about those alternative hypotheses which are
not compatible with OHs and AA, because they are not part of H? Lyons stresses
that they exist in many historical cases, and if so, don’t they make the original
hypothesis H superfluous, i.e., non-​essential? Why doesn’t my criterion exclude
such hypotheses?
In other words, Psillos’s condition (2) for essentiality is of medium strength: it
excludes all the hypotheses which can derive NP when joined with OHs and AA.
Lyons’s criticism is that it is too weak, and his implicit suggestion is that it should
be strengthened to exclude even the hypotheses which can derive NP in con-
junction with different collateral assumptions. My condition (2’), instead, is even
weaker than Psillos’s (for it excludes only a subset of the alternative hypotheses
excluded by (2), i.e., only those which are part of H), and this is why it escapes
Lyons’s first criticism: how can it be enough? In particular, shouldn’t we require
that H is unique, i.e., that there is no alternative hypothesis H’ incompatible
with H, which (no matter whether compatible and joined with OHs and AA,
or incompatible with them and joined with different hypotheses and auxiliary
assumptions) can derive NP? For if it existed, we couldn’t know which of them
is true.
I answer that we don’t need to explicitly exclude any other H’ except those
which are part of H (so, we don’t need Psillos’s condition (2) with its drawbacks,
nor its strengthening implicitly suggested by Lyons), because the required
uniqueness of H is already ensured by condition (2’) together with the risky char-
acter of the prediction NP. In fact, just as it happens for theories, it is practically
impossible to find a completely false hypothesis making risky novel predictions.
A false hypothesis H may yield a risky novel prediction NP, only if it is partly
true: i.e., if it has a true part H* from which NP can be derived. But in this case,
the essential role in the derivation of NP is played by H*, H being inessential by
my condition (2’).
This much we know a priori from the NMA (as spelled out in the five ini-
tial paragraphs of section 9.1), and anti-​realists have not been able to spot any
inferential fallacy in that argument. However, they challenged it by apparent
counterexamples, i.e., by citing historical instances of false hypotheses which
were supposedly deployed essentially in deriving novel predictions. But in each
of those cases it can be shown that either (a) those hypotheses were not actually
192 Contemporary Scientific Realism

false or the predictions not actually true, or (b) the predictions were not actually
novel and risky, or (c) they were not deployed essentially (Alai 2014b).
Condition (2’) is designed precisely to exclude cases of type (c), and I shall
offer six examples in section 9.4 and three in section 9.5. Therefore, if H fulfills
(2’), it must be completely true. But if so, we don’t need in addition to exclude
that there is any incompatible assumption H’ from which NP could be essentially
derived, because any H’ incompatible with H would be (at least partly) false,
hence it could not (except by a miracle) be essential in deriving NP: H’ could
play a role in deriving NP only if it had a completely true part H’* sufficient to
derive NP, and obviously H’* would be compatible with H, since both would be
completely true.
This is a nutshell account of why my condition (2’) is enough to ensure the
required uniqueness of H, but I shall explain it in more detail in section 9.6. This
account, anyway, entails an empirical claim: in the history of science there have
never been a couple of incompatible hypotheses both essential in deriving the
same novel risky prediction. So, my proposal is empirically testable: it will be fal-
sified if any such couple is found, but confirmed if it is not.

9.4. How the Essentiality Condition Rules out Lyons’s

Purported Counterexamples

Here is how my condition (2’) rules out as inessential five false assumptions
which according to Lyons were deployed to derive novel predictions (the first
three are more extensively discussed in Alai 2014b, §7). Three further examples
will be discussed in section 9.5. Yet another striking case is Arnold Sommerfeld’s
1916 prediction of the fine structure energy levels of hydrogen: for decades it was
thought to be derived from utterly false assumptions, and so considered “the ul-
timate historical challenge” to the NMA, for even physicists called Sommerfeld’s
success a “miracle.” But Vickers (2018) shows that the wrong assumptions were
not essential in deriving it, since the true part of Sommerfeld’s premises was
enough.

9.4.1 Dalton

Lyons argues that Dalton derived his true Law of Multiple Proportions (LMP)
from his false Principle of Simplicity:

PS: “Where two elements A and B form only one compound, its compound
atom contains 1 atom of A and 1 of B. If a second compound exists, its atoms
Historical Challenge to Realism and Essential Deployment 193

will contain 2 atoms of A and 1 of B, and a third will be composed of 1 of

A and 2 of B, etc.” (Lyons 2002, 81; Hudson 1992, 81).

But this false PS has a true part: a weaker principle we might call “of Multiple
Quantities”:

PMQ: The quantity of atoms of B combining with a given number of atoms of

A is always a multiple of a given number.

Besides, PMQ is enough to derive the Law of Multiple Proportions:

LMP: The weights of one element that combine with a fixed weight of the other
are in a ratio of small whole numbers.

Therefore, although actually employed in deriving LMP, the false PS was not es-
sential, since its true part PMQ was sufficient to derive LMP.

9.4.2 Mendeleev

Mendeleev derived his predictions of new elements from his false Periodic Law:

PL: atomic weights determine the chemical properties of elements.

PL is false, since chemical properties are rather determined by atomic numbers

(Lyons 2002, 80, 84). However, PL entails that

Q: an atomic quantity approximately proportional to atomic weight determines

chemical properties (where Mendeleev thought this quantity was atomic
weight itself, while we know it is the atomic number).

Now, Q is a part of PL, true, and sufficient to derive Mendeleev’ predictions. So,
although actually employed, PL was not essential, while the essential assump-
tion, Q, was true.

9.4.3 Caloric

Lyons (2002, 80) points out that the caloric theory truly predicted that

ER: The rate of expansion is the same for all gases.

194 Contemporary Scientific Realism

But ER was derived from a number of false claims, viz.:

(1) heat is a weightless fluid called caloric;

(2) the greater the amount of caloric in a body, the greater its temperature;
(3) gases have a high degree of caloric;
(4) caloric, being a material itself, is composed of particles;
(5) caloric particles have repulsive properties which, when added to a sub-
stance, separate the particles of that substance;
(6) the elasticity of gases is caused by this repulsive property of caloric heat
particles (Carrier 1991, 31).

However, these false claims were not essential in predicting ER. In fact, each of
them includes a true part not referring to caloric, from which ER could still be
derived, respectively

(1*) heat is weightless, and it expands like fluids;

(2*) the greater the amount of heat (whatever it consists in) in a body, the
greater its temperature;
(3*) gases have a high amount of what heat consists in;
(4*) particles play a crucial role in the constitution of heat;
(5*) heat involves repulsive forces, which separate the particles of substances
to which it is added;
(6*) the elasticity of gases is caused by the repulsive forces involved by heat.8

So, while the false (1)–​(6) were used, they were not essential, since their parts
(1*)–​(6*) are sufficient to derive ER (and true, as far as we know).

9.4.4 Kepler

In (2006, 545–​553) Lyons discusses a number of false premises from which

Kepler derived (P1) that the Sun spins, (P2) that a planet’s speed is highest at
its perihelion and lowest at its aphelion, and (P3) his three laws. Further, since
Kepler’s laws were used by Newton, those false premises indirectly yielded also
the successes of Newton’s gravitation theory, including Neptune’s discovery.
An exhaustive analysis of Lyons’s discussion would require a different paper.
Here I can only sketch how my essentiality condition rescues the NMA even in
this case.
Kepler’s false assumptions were:

8 See Psillos (1999, 115–​130) for an extensive presentation of this strategy.

Historical Challenge to Realism and Essential Deployment 195

(#1) the planets tend to rest, and they move only when forced to move;
(#2) the Sun has an Anima Motrix which spins it around its axis;
(#3) this spinning is transmitted to the planets through the Sun’s rays, which
push the planets in their orbits (thus the Sun exerts on the planets a di-
rective rotational force, not an attractive one);
(#4) the force of the Sun’s rays is inversely proportional to the distance from
the Sun, like the intensity of their light.

The model of the universe described by (#1)–​(#4) is crazy in the light of our
mechanics (#1), metaphysics (#2), and physics (#3), but it was not in Kepler’s
time. Moreover, it was a natural abductive conclusion from two facts known at
that time:

(a) the planets move around the Sun approximately on the same plane and in
the same wise;
(b) the order of their velocities is inverse to that of their distances from
the Sun.9

These facts plausibly suggested a model analogous to mechanisms well known

to Kepler, like a sling, where a central hub rotates, and through a strap transmits
its motion to a peripheral body, which is then set into circular motion as well. In
Kepler’s model the hub is the Sun, the peripheral body is a planet, and the strap
represents the Sun’s rays. The sling has just one projectile, while the planets are
many; but there are similar earthly mechanisms which preserve this analogy, like
wool winders, wheels to lift water, or clock gears (although the sling model is
better from the point of view of fact (b), for (initially) the projectile rotates more
slowly than the hand, and the more slowly the longer the strap is).
The predictions (P1) that the Sun spins and (P2) that the velocity of the
planets is highest at the perihelion and lowest at the aphelion, are immediate
consequences of Kepler’s model (#1)–​(#4). Of course the model is wrong, for his
abduction from the known facts (a) and (b) included some wrong guesses. But it
is also unnecessarily strong as an explanation of those facts. Moreover, the wrong
part was not essential to Kepler’s novel predictions or to the discovery of his three
laws, since the model had a weaker core which was both true and sufficient to
derive them. Kepler might have restricted his assumptions to that weaker core if

9 This was already clear to Copernicus, who knew the times of the orbits and the relative distances

from the Sun, hence the relative lengths of the orbits. Kepler also knew that the times grow more than
the lengths, hence the velocity diminishes as the distance from the Sun increases. This suggests that
the planets be driven by a force emanating from the Sun, so varying inversely with the distance from
it (Koyré 1961, II, §1). Kepler thought that the light varies inversely with the distance from its source
(rather than with its square), hence he thought the Sun’s motrix force did the same.
196 Contemporary Scientific Realism

he had been more cautious in extrapolating from (a) and (b); but unfortunately
there were no technological analogies, like the sling, the wool winder, etc., to
suggest to him such a model. This true core is the following:

(#5) the solar system moves around the centre of the Sun as a coherent but
non-​rigid disk;
(#6) its coherence is mainly due to a force exerted by the Sun on the planets
(Kepler thought it was a directive rotational force, and the only one in
play; while we know it is the attractive gravitational force, supplemented
by the planets’ own attractive forces and the effects of inertia);
(#7) the velocities of the planets are also proportional to the same force
(Kepler thought this happens because that force is responsible for their
motion, by overcoming their tendency to rest; instead we know this
happens because their motion is due to their tangential inertia, which
must be equal to that (gravitational) force, otherwise the planet would
either fly off on its tangent, or collapse on the Sun);
(#8) that force is in an inverse relation with the distance from the Sun (for
Kepler that relation was F=1/​d, we know it is F=1/​d2).

I am not saying that (#5)–​(#8) by themselves constitute a viable model of the

universe: we get one only by supplementing them either by the rest of Kepler’s
assumptions or by the rest of our assumptions10 (in the first case, of course, we
get a viable, but false model, in the second case a true model). For instance, we
wouldn’t get a viable model just by adding to (#6) that the force is an attractive
one, without adding the planets’ inertia, for then the planets would collapse on
the Sun. This is probably why Kepler himself, after considering the idea of an at-
tractive force, rejected it (Lyons 2006, 547). I am only claiming that (#5)–​(#8) are
a core shared by two otherwise very different models; that as far as we know they
are true; and that they are enough to derive Kepler’s novel predictions (P1)–​(P3).
In fact, (P1) the spinning of the Sun is already implicit in (#5); (P2) that a
planet’s speed is highest at its perihelion and lowest at its aphelion is implied by
(#7) and (#8); finally, (P3) Kepler’s three laws are just kinematic laws, and once
given (#5) they were found mainly by making hypotheses on which curves and
functions would fit the data and checking them back with the data; but their
discovery was helped by Kepler’s revolutionary intuition that the orbits must
not be found by geometrical models alone, but reasoning on the forces which
govern them, as suggested by (#7) and (#8) (Hoskin 1999, ch. V). So, there were
no miracles: the true model (#5)–​(#8) is part of the false model (1#)–​(#4) and

10 Some of which are mentioned within parentheses after (#6), (#7), and (#8) respectively.
Historical Challenge to Realism and Essential Deployment 197

enough to generate (P1)–​(P3); therefore the model (#1)–​(#4) was not essential
by condition (2’), hence the NMA does not commit to it.
For the sake of simplicity, in (#6), (#7), and (#8) I spoke of a force, as if Newton’s
gravitational theory were true; but, one could object, General Relativity shows
that there are no gravitational forces, and the planets’ motion is due to the curva-
ture of spacetime caused by the Sun. The objection is fair, but it only shows that
we must circumscribe more accurately the true core of Kepler’s model, which we
can do by generalizing (#6), (#7), and (#8) as follows:

(#5) the solar system moves around the centre of the Sun as a coherent but
not rigid disk;
(#6*) its coherence is due to an effect (i.e., for us, the curvature of spacetime)
which is mainly brought about by the Sun;
(#7*) the velocities of the planets are also proportional to the same effect;
(#8*) that effect is in an inverse relation with the distance from the Sun (hence
so are the planetary velocities).

This restricted core is shared by Kepler, Newton, and Einstein and is still
sufficient to derive Kepler’s novel predictions. But what if someday General
Relativity is replaced by a better theory and this by another one, etc.?11 If we
constantly weaken our hypotheses to keep them compatible with succes-
sive theories, won’t we reduce them to mere descriptions of observed phe-
nomena, and scientific realism to empiricism?12 I don’t think this pessimism is
warranted, because history shows that each successive theory T’, while giving
up part of the theoretical content of the earlier theory T, adds some new theo-
retical content. Even if in turn T’ is superseded by T” and some of its content
is given up, it is quite possible that (a) some of the original content of T is pre-
served even in T”, and (b) the theoretical content of T’ preserved in T” is even
larger than that of T preserved in T’.
For instance, in the case discussed here, the directive force has been replaced
by the attractive force, and this by a curvature of spacetime, but all of them are
unobservable entities or properties. Each theorist while discarding some theo-
retical assumptions adds some new ones (Newton adds inertia and gravitation
force, Einstein spacetime curvature). Tomorrow it might no longer be spacetime
curvature, but something else; but whatever it is, the Sun will have a major role in
it, and it will have an inverse relation with the distance from the Sun and a direct
relation with the velocities of the planets. Besides, each new model is probably
closer to the truth, because it has a better fit to the data and greater predictive

11 I deny that our current theories are completely true and will never be rejected: Alai (2017).
12 I owe this objection to an anonymous reviewer.
198 Contemporary Scientific Realism

power. Thus our theoretical knowledge (our theoretical true justified beliefs)
increases over time. While acknowledging that our beliefs are not yet “the whole
truth and nothing but the truth,” we can trust to be on the right track. I shall fur-
ther discuss similar worries in §§9.7.3–​9.7.5.

9.4.5 Neptune

Finally, Lyons claims that the following false claims were used in the prediction
of Neptune (2006, 554):

(1) the sun is a divine being and/​or the center of the universe (Kepler);
(2) the natural state of the planets is rest;
(3) a non-​attractive emanation coming from the sun pushes the planets in
their paths;
(4) the planets have an inclination to rest, which resists the solar push, and
contributes to their slowing speed when more distant from the sun;
(5) the planets are pushed by a “directive” magnetic force;
(6) there exists only a single planet and a sun in the universe (Newton);
(7) each body possesses an innate force, which, without impediment,
propels it in a straight line infinitely;
(8) between any two bodies there exists an instantaneous action-​at-​a-​
distance attractive force;
(9) the planet just beyond Uranus has a mass of 35.7 earth masses
(Leverrier)/​50 earth masses (Adams);
(10) that planet has an eccentricity 0.10761 (Leverrier)/​0.120615 (Adams);
(11) the longitude of that planet’s perihelion is 284°, 45’ (Leverrier)/​299°, 11’
(Adams), etc.

However, (1)–​(5) were no longer held in the 18th century. They were involved
in Neptune’s discovery only to the extent that they were involved in the dis-
covery of Kepler’s laws, which in turn were used by Newton. But I argued that
(1)–​(5) were not involved essentially in Kepler’s discoveries. (9)–​(11) concern
Neptune, so they could not be premises from which Neptune’s existence or lo-
cation was derived: they are only (partially wrong) parts of a global hypothesis
on the planet’s existence and behavior. (6) was practically true in this context,
given the negligible influence of the other planets. To my knowledge (7) was
never held by anybody; anyway, it entails the true claim that inertial motion con-
tinues indefinitely in a straight line. Finally, I just argued that (8), gravitational
force, was not essential in Newton’s predictions. So, even Neptune ceases to be a
counterexample to DR.
Historical Challenge to Realism and Essential Deployment 199

9.5 Essentiality Cannot be Detected Prospectively

For deployment realists the components of discarded theories which were es-
sentially involved in novel predictions are most probably true; but Stanford
(2006), Votsis (2011), and Peters (2014) argue that it is not enough to identify
these components just retrospectively, i.e., as those preserved in current theories.
Rather, they must be identified prospectively, from the viewpoint of their con-
temporaries. In fact, deployment realists claim that the hypotheses of past the-
ories still preserved today are true because they were essential. Thus they resist
the Pessimistic Meta-​Induction (PM-​I), maintaining that both past and current
theories are at least partly true. But explaining that those hypotheses were es-
sential because they are preserved today, would be explaining that they are true
because they are preserved today, so begging the question of the truth of current
theories. In fact, since a hypothesis H is preserved today because we believe it is
true, saying that it is true because it is essential, and it is essential because it is pre-
served today, would be saying it is true because we believe that it is true.
Equally, deployment realists claim that the now rejected hypotheses deployed
in novel predictions were not essential, hence the NMA did not commit to their
truth. In this way they block Laudan and Lyons’s meta-​modus tollens (M-​MT).
But explaining that they were not essential because they are not preserved today
would be stipulating that the NMA warrants all and only the hypotheses ac-
cepted today. Thus the realist’s defense against both the PM-​I and the M-​MT
would be circular.
This is why it would seem desirable that the components of past theories
which are essential for their novel predictions are identified prospectively; more-
over, Votsis (2011), Peters (2014), Cordero (2017a, 2017b), and others claim that
such prospective identification is possible. On the opposite side, Stanford (2006,
167–​180; 2009, 385–​387) claims this is impossible, therefore the arguments for
DR are circular.13
I take an intermediate position: any time we believe a hypothesis is essential,
it may in fact be, but neither at that time nor later can we ever be certain that
it is. Yet, I argue that this does not make the arguments for DR circular. That
we cannot ever be certain is shown by many hypotheses which were firmly
believed by past scientists because they appeared to be essential to certain novel
predictions, but subsequently turned out to be inessential, in fact false.
For instance, Fresnel and Maxwell derived various novel predictions from the
hypothesis that

AV: aether vibrates.

13 See also Vickers (2013, 207).

200 Contemporary Scientific Realism

Today we know that AV was inessential in these derivations, because it can be

substituted by its weaker consequence

VM: there is a vibrating medium.14

Fresnel and Maxwell could easily have seen that VM was a proper part of AV and
enough to derive those predictions. However, they didn’t consider VM (probably
they didn’t even think of it), no doubt because they presupposed that

P1: all mediums are material,

and/​or that

P2: all waves are produced by the oscillations of particles.

Hence, given their presuppositions, any vibrating medium was a material me-
dium composed of particles (i.e., either water, or air, or aether). Therefore from
their viewpoint, AV was required by their predictions as much as VM.
Again, Laplace predicted the speed of sound in air starting from the false
hypothesis that

H: the propagation of sound is an adiabatic process, in which some quantity of

caloric contained by air is released by compression.

Psillos (1999, 119–​121) argues that H was not essential to Laplace’s prediction,
which could also be derived from a part of H, viz.

H*: the propagation of sound is an adiabatic process, in which some quantity

of latent heat contained by air (whatever be the nature of heat) is released by
compression.

However, at that time adiabatic heating could only be explained as the disen-
gagement of caloric from ordinary matter, caused by mechanical compression
(Chang 2003, 904), for it was presupposed that

P1: gases can be heated without exchanges with the environment only if they
contain heat in a latent form,

and

14 Today we call it electromagnetic field.

Historical Challenge to Realism and Essential Deployment 201

P2: only material substances can be contained by material substances in a

latent form.

But the material substance of heat was just caloric: therefore, given P1 and P2, H*
entailed H, hence H too had to be considered essential.
Generalizing, we might say that at any time in the history of science, when
a novel prediction NP is derived from an assumption H, there may be an as-
sumption H* which is proper part of H and sufficient—​from a purely logical
viewpoint—​to derive NP; however, given certain current explicit or implicit
presuppositions PRS, H may “appear to be conceptually or metaphysically
entailed by” H* (Vickers 2016, §4), hence scientists may falsely believe that H is
essential to NP. At a later time, however, the advancement of science may show
that the PRS are false, so scientists can (retrospectively) recognize that H was
inessential, after all. This recognition is facilitated if meanwhile experimental
or theoretical doubts have arisen about H, so that scientists started wondering
whether H was really necessary. Yet, this recognition is logically independent of
those doubts, as it follows already from the falsity of the PRS; hence, although
retrospective, it is not circular.
Vickers, however, suggests that although prospectively we cannot ever be
certain that H is essential to a prediction, in some cases we could (still prospec-
tively) recognize it is not essential, although “metaphysically or conceptually” re-
quired by other considerations. This may (i) suggest the realist to “restrict her
commitments to what is directly confirmed by the predictive successes” (i.e., H*),
and (ii) supply “a worthwhile heuristic,” i.e., show which commitments should be
abandoned first in case of empirical or theoretical refutations (i.e., H).
For instance, Bohr predicted (NP) the spectral lines of ionized helium by
assuming that

H: The electron orbits the nucleus only on certain specific orbital trajectories,
each characterized by a given quantized energy.

H turns out to be false, but NP could have been derived by the weaker
hypothesis that

H*: The electron can only have certain, specific, quantized energy states.

Moreover, says Vickers, Bohr could have seen (prospectively) that H was ines-
sential, because it entailed H*, which was enough to derive NP. But he still held
H because “it may have been inconceivable at the time to think that electrons
could have quantized electron energies without having associated quantized
orbital trajectories (cf. Stanford 2006, 171)”. Nonetheless, Bohr might have
202 Contemporary Scientific Realism

distinguished between his reason for believing H* (i.e., the essentiality of H* in

predicting NP), and his reason for believing the content of H exceeding H* (i.e.,
his presupposition that electrons could have quantized energies only by having
orbital trajectories). Thus he might have realized that “the latter were not as se-
cure as the former” (Vickers 2016, §4).
I agree, but with the following qualifications:

(α) in many cases prospective recognition of inessentiality will be impossible;

(β) essential hypotheses cannot be identified with certainty, neither prospec-
tively (as argued by Stanford) nor retrospectively;
(γ) however, this is not a problem, because the PM-​I can be blocked even
without identifying essential hypotheses, and the M-​MT can be blocked
even by identifying inessential hypotheses retrospectively.
(δ) at any rate, recognizing inessentiality (especially prospectively) is not a
task for (realist) philosophers, but for scientists;
(ε) even when inessentiality can be recognized prospectively, this does not
help the realist defense against the PM-​I and M-​MT.

Here are my arguments:

(α) Perhaps in Bohr’s case the extra-​content of H could be distinguished
from H*; but in other cases it may consist in principles so obvious or deeply en-
trenched to pass unnoticed, such as the principles of conservation of energy and
mass, isotropy and homogeneity of space, physical causal closure, etc. We usu-
ally presuppose so many principles of this kind, that even by paying close at-
tention one cannot be certain to have ruled out all of them. Moreover, it may be
extremely difficult to imagine an assumption H* that still entails NP once all of
them have been discarded.
Besides, Vickers grants that H may be “conceptually” entailed by H*, i.e., that
the contemporaries may lack the conceptual resources needed to distinguish
H* from H. For instance, those who used Newtonian forces to predict Neptune
couldn’t even conceive any “effect” (such as the curvature of spacetime) which
could cause the planetary motions except a force. Therefore, when H seems to be
essential in deriving NP, it may be impossible (either prospectively or retrospec-
tively) to see that it is not.
(β) A fortiori, we cannot ever be certain that there is no weaker component H*
from which NP could have been derived. Therefore, we cannot ever be certain
that H is essential, neither prospectively (as Stanford claims) nor retrospectively.
This is just natural, for if we could identify essential components we would know
that they are completely true. But if so, we could anticipate scientific progress
much more than we actually can: while trusting that current theories are largely
Historical Challenge to Realism and Essential Deployment 203

true, scientists grant that some of their assumptions are probably wrong, but only
future research will tell which ones.
Besides, the PM-​I is probably right that no theory or component older
than 100 years or so was completely true (but not that none was at least partly
true). Cordero, Peters, and Votsis are also right that we know when a hypo-
thesis H is “true,” if this means “at least partly true”: but even if deployed in
novel predictions, H may have been inessential, and even if otherwise well
supported it can be partly false, and sooner or later replaced by another more
completely true.
The progress of research allows us to drop more and more false
presuppositions. Thus we may discover, retrospectively, that H was inessential,
for it had some proper part H* which (i) can be true even if H is false and (ii)
is sufficient to derive NP. That weaker H* may also be actively sought for, if H
encounters empirical or theoretical refutations, which suggest that it cannot pos-
sibly be essential (since essential components are completely true). In this case
(still retrospectively), one can recognize that H was inessential even before iden-
tifying its weaker substitute H*.
(γ) Neither the impossibility of recognizing essentiality, nor the difficulties in
prospectively recognizing non-​essentiality are a problem for deployment realists,
because these recognitions are not necessary to resist the PM-​I and the M-​MT.
In fact, as soon as a risky novel prediction NP derived from H is confirmed, we
know (prospectively, and a fortiori retrospectively) that, short of miracles, there
is some truth in H. This is enough to refute the PM-​I’s claims that both past and
present theories are completely false.
The M-​MT, in turn, is always used retrospectively, therefore only a retro-
spective recognition of inessentiality is needed to resist it: if at time t hypo-
thesis H was firmly believed because considered essential to derive NP, but at
t’ it is shown to be false, the M-​MT uses it as a counterexample to the NMA.
But realists can reply by two moves: (i) arguing that probably H was inessen-
tial, for only by a miracle could NP have been derived from a completely false
assumption; (ii) showing H was inessential, by identifying a proper part H*
sufficient to derive NP, and the false presuppositions PRS that at t prevented
us from distinguishing H from H*. Both moves are retrospective (at t’ or at a
later t”), yet sufficient to block the M-​MT. Besides, move (ii) is independent
of the refutation of H, therefore the realist’s defense is not circular, as Stanford
claims.
(δ) Even when it is possible, the recognition of non-​essentiality (especially
prospectively) is a task for scientists: philosophers just don’t have the necessary
expertise. The burden of scientific realists is arguing that certain criteria (e.g., es-
sential deployment) can justify our beliefs in the at least partial truth of theories
204 Contemporary Scientific Realism

H
H*

Figure 9.1

or hypotheses. But Fahrbach (2017) asks too much when he requires that they
also apply those criteria to actual research, so teaching practitioners which are
the working hypotheses and which are the idle parts in their theories, urging
changes, suggesting directions of research, etc.
Therefore, qua philosophers, realists need not be committed to any partic-
ular theory or hypothesis, not even to the best current ones: as argued by Smart
(1963, 36), this is a task for scientists. At most, they may argue that science is con-
vergent, hence, in general, current theories are probably more largely true than
past ones. Therefore, pace Stanford (2017), they need not be more conservative
than anti-​realists. Actually, since they set for hypotheses a higher standard than
anti-​realists—​truth, rather than empirical adequacy or the like—​for them it is
even more likely that any particular hypothesis fails to reach that standard, hence
must be substituted by a better one.
(ε) The discussion at (γ) shows that even when inessentiality can be recog-
nized prospectively, that is superfluous in resisting the PM-​I and the M-​MT. In
fact, the at least partial truth of H can be recognized prospectively, even if H is
not essential, and this is enough to block the PM-​I. Besides, prospective recog-
nition of inessentiality is superfluous against the M-​MT, because the latter is al-
ways used retrospectively.

9.6 A More Complete Account of the Dispensability

of Psillos’s Condition (2)

Here I explain in more detail why my condition (2’) is enough and we can
do without the greater complexity of Psillos’s condition (2) or its possible
strengthening implicitly suggested by Lyons. When Psillos excludes any “other
hypothesis H*,” H* may be compatible or incompatible with H, and in different
mereological relations to it. The following cases are included:

(a) H* is proper part of H. This is precisely the kind of alternative hypothesis H*
excluded by my condition (2’) (Figure 9.1)
(b) H is a proper part of H* (Figure 9.2)
Historical Challenge to Realism and Essential Deployment 205

H* H

Figure 9.2

H = H*

Figure 9.3

H H*

Figure 9.4

(c) H* and H coincide (Figure 9.3)

(d) Neither hypothesis is part of the other, and they have no common content
(Figure 9.4)
(e) Neither hypothesis is part of the other, since they have some common con-
tent but two alternative extensions (Figure 9.5)

Now, Psillos’s condition (2) excludes alternative hypotheses H* of all these five
kinds, and here the troubles arise, while my condition (2’) excludes only alter-
native hypotheses H* of kind (a), and I claim that this is enough for deploy-
ment realists. In cases (b), (d), and (e) H* has some content outside H; so, in
these cases H* might be incompatible with OHs and AA, and Lyons implicitly
suggests they should be excluded along with the cases in which H* is compat-
ible with OHs and AA. But we don’t need to worry about any of these cases,
and here is why:
206 Contemporary Scientific Realism

H-H* H+H* H*-H

Figure 9.5

In case (b) H is weaker than and part of H*. Therefore the existence of H*
does not make H inessential, so there is no need to exclude it. In fact, there have
always been such unnecessarily rich alternatives to the true hypothesis (like e.g.
the ether theory with respect to the simple assumption of Maxwell’s equations, or
the aforementioned demi-​gods theory with respect to Newton’s actual theory).
From a purely logical viewpoint, for any H there are infinite such H*, but they do
not make H inessential.
In case (c) H* coincides with H, so again it does not make H inessential and we
don’t need to exclude it.
In case (d) H and H* have no common content, and this may happen in
two ways:

(d1) H and H* are compatible, hence, they may be both true. In general it
is quite possible that a prediction is reached by independent inferential
routes starting from different hypotheses. This is just natural if we con-
ceive the world as a tightly connected whole, in which each phenom-
enon, at each scale, is causally linked to many other phenomena even
at different scales. For instance, Perrin famously derived the same value
for Avogadro’s number reasoning from tenets of chemistry, thermody-
namics, electrical theory, the theory of Brownian motions, and others.
In such cases, each hypothesis Hi deployed in a different inferential
route to NP is not “essential” in the sense that without Hi one could not
have predicted NP (for NP could have been predicted through some
different inferential routes involving some alternative hypothesis Hj).
However, Hi can be essential in the sense that NP could not have been
predicted along that particular inferential route without Hi, i.e., in the
sense that Hi cannot be substituted in the same inferential route by one
Historical Challenge to Realism and Essential Deployment 207

of its proper parts. This is precisely what my condition (2’) requires, and
it is all we need from a criterion of essentiality, for it is enough to war-
rant the complete truth of Hi (for we saw that, short of miracles, a hy-
pothesis deployed in predicting NP cannot be completely false, and it
can be partly false only if its false part is dispensable in the derivation).
Therefore we need not exclude case (d1), for in this case H is essential in
the required sense if it fulfills (2’). By the way, since H entails NP only in
conjunction with AA and OHs, also AA and OHs are essential if they
satisfy (2’).15
(d2) Otherwise, H and H* may have no common part and be incompatible.
One might think that we need to exclude this case: for if we had two in-
compatible hypotheses both predicting NP, we couldn’t know which of
them is true.16 However, as explained in my nutshell account at section
9.3, this case is already excluded by the risky character of NP. In fact, as
argued earlier, it is practically impossible to find a completely false hypo-
thesis entailing a risky novel prediction. But if H is essential in the sense
of (2’), it must be completely true. Therefore, if H* is incompatible with
it, we know that (i) H* is at least partly false and (ii) it cannot possibly
predict NP, for it contradicts an assumption (i.e., H) which is essential to
that prediction.17 Therefore, my condition (2’) (together with the risky
character of NP) already excludes that any hypothesis incompatible with
H also predicts NP.

Finally, case (e): H and H* partially overlap. Like in case (d), here too we must
distinguish:

(e1) If H and H* are compatible, it is possible, though unlikely, that they

are both are essential, hence certainly true, as explained in case (d1).
Otherwise, the essential assumption is neither H nor H*, but some part
of their common content.
(e2) If H and H* are incompatible (like e.g. Rutherford’s and Bohr’s theo-
ries of the atom), neither of their non-​overlapping parts is required
by the derivation: in fact, NP is predicted by H without using H*-​
H, and it is predicted by H* without using H-​H*. Therefore both

15 Then we know they are true (unless NP was used in building them).
16 Besides, we would have a counterexample to the NMA, for there would be at least one false hy-
pothesis allowing us to derive a novel prediction.
17 Instead, if H is not essential in my sense, it consists of an essential part H and an inessential part
e
Hi. If H* also has a role in predicting NP, it may consist in an essential part H*e and an inessential
part H*i. Of course H*e cannot contradict He, but H*i may contradict Hi, in which case H and H* are
incompatible, yet have both a (non-​essential) role in predicting NP, and are both partly true. Instead
He and H*e are two essential but compatible hypotheses as considered in case (d1).
208 Contemporary Scientific Realism

H and H* have a superfluous part, hence neither is essential. If H

were essential in my sense, it would be practically impossible that H*
predicted NP, for it would contradict an assumption which is needed
to predict it. So, even in case (e2) condition (2’) is enough to warrant
the truth of H.

Summing up, my condition (2’) is enough to guarantee that a hypothesis H is

essential in the derivation of a novel prediction NP, in a sense which warrants its
complete truth, and we don’t need Psillos’s more complex and troublesome con-
dition (2) or its strengthening implicitly suggested by Lyons.

9.7 Objections and Answers

Here I reply to some potential objections.

9.7.1 The “False May Entail True” Objection

Since a falsity may entail a truth, even false theories might make novel
predictions.
Answer:
As argued at the beginning of section 9.1, in the Hyperuranion of possible
theories there are some with wholly false theoretical assumptions which make
true novel and improbable predictions (FNIP-​theories). But their rate is exceed-
ingly small, hence it is extremely unlikely that scientists pick one of them (be-
cause scientists look for true theories, hence they could find FNIP-​theories only
by chance). On the opposite, all true theories have true consequences, and those
sufficiently fecund have true novel consequences. Granted, true (and fecund)
theories are much fewer than false ones; however, they are not found by chance,
but on purpose and through reliable methods (White 2003; Alai 2016, 552–​554;
2018, §III).

9.7.2 The Disjunctive Objection

Your condition (2’) cannot be fulfilled, hence no hypothesis is essential by

your criterion, because any hypothesis H always entails some weaker claim H’,
like e.g.,
Disjct: H or the Moon is made of cheese.
Answer:
Historical Challenge to Realism and Essential Deployment 209

First, condition (2’) excludes that NP is entailed by any hypothesis which is

part of H (in Yablo’s sense), not just by any hypothesis entailed by H. In fact,
Disjct is not part of H. Second, disjunctions are weaker than their disjuncts.
Therefore, if H entails NP, in most cases H*: «H or H’» does not. Hence, in such
cases H fulfills (2’). For instance, Newton’s Gravitation Law entailed the existence
of Neptune; but the disjunction of this law with ‘The Moon is made of cheese’
does not. Therefore in certain cases NP cannot be entailed by any hypothesis
which is part of H, hence H is essential.
In particular instances the disjunction of H with another hypothesis H’ might
still be strong enough to entail NP (Vickers 2016, §3). Obviously, however, this
can happen only if even H’ alone entails NP. But in this case H and H’ must be in
one of the five mereological relations examined in section 9.6, and I argued that
in each of them condition (2’) is enough to decide whether H (and H’) is essen-
tial or not.

9.7.3 The “Theoretician’s Dilemma” Objection 1

The author’s condition (2’) seems to make realism redundant for purely logical
reasons. A statement is derivable from a set of statements iff the content of that
statement is already included in that set of statements. The only proper part of
that set of statements that’s required to derive the statement is in fact the state-
ment itself. Thus, a novel prediction NP needs only NP to be entailed. But surely
we want to be realists about more things than the content corresponding to NP.18
Answer:
Most frequently and typically NP is not part of H, because, as explained by
Vickers19 and others, it is not entailed by H alone, but only in conjunction with
certain other hypotheses OHs of T, and various auxiliary assumptions AA.
Therefore (2’) does not allow us to dispense with H in favor of NP.
But the objection might become that since H&OHs&AA entails NP, we can
dispense with H&OHs&AA in favor of NP. This is reminiscent of Hempel’s
(1958) “theoretician’s dilemma”: theoretical hypotheses H are required to con-
nect the observable initial conditions IC to the observable final conditions FC,
and they are justified only if they succeed in this, i.e., if IC→H→FC. But then
why not drop H and just keep the empirical laws IC→FC? Hempel answered that
this would work from a purely logical point of view, but not from an epistemolog-
ical point of view: H does not predict only IC→FC, but many other phenomena
as well, most of which we would never have discovered without assuming H.

18 Objection raised by an anonymous referee.

19 Vickers (2013, p. 202 footnote 9; 2016, footnote 8).
210 Contemporary Scientific Realism

The answer here is similar: we would never have found NP, and many other
predictions, unless H&OHs&AA were true. In fact, NP is novel and risky, hence
it cannot have been found just by chance. Besides, the auxiliary assumptions
AA are the consequences of many independent theories T’, T”. . . Tn, typically
dealing with matters different from T. Therefore, it would be a miraculous co-
incidence if even one of them were completely false, yet the conjunction of their
consequences entailed NP. The only plausible explanation is that T, T’, T”. . . Tn
are all at least partially true (Alai 2014c, §4). Besides, each of T, T’, T”. . . Tn,
in conjunction with still different assumptions, issues many other predictions,
which are only explainable in the same way.
One might reply that guessing n true theories is even less likely than guessing
n false theories with one true joint consequence. This is right, but guessing n false
theories with one true, novel, and risky joint consequence is still too improb-
able to be a minimally plausible explanation of our frequent predictive successes.
So, how was it possible to derive NP and so many other novel predictions? The
only plausible explanation is that theories are not guessed, but found by a method
which is reliable in tracking truth. Scientific realists have a reasonable account
of why the scientific method is reliable, while anti-​realists have never been suc-
cessful in providing an account of how predictive success might be achieved by
pure chance or “miracles” (Alai 2014a). Notice, here the truth-​conduciveness of
the scientific method is not a petitio principii, but the conclusion of my inference
to the only plausible explanation in this paragraph (the second in this section).20

9.7.4 The “Theoretician’s Dilemma” Objection 2

By your criterion no hypothesis will ever be essential, because whenever H to-

gether with OHs and AA entails prediction P, there is always a weaker hypo-
thesis doing the same, viz.

H*: If OHs&AA, then P.21

Answer:
Again, here H* is not part of H, hence its existence does not make H inessen-
tial. For instance, suppose the following:

H: The atoll is loaded with deadly radioactive material;

20 White (2003); Alai (2016, 552–​554; 2018 §III).

21 Objection raised in discussion by John Worrall, who cited a topos from the debates on
confirmation.
Historical Challenge to Realism and Essential Deployment 211

OHs&AA: Carl lands on the atoll;

P: Carl will die;
H*: If Carl lands on the atoll he will die.

Of course H* together with OHs&AA entails P. However H* is not part of H,

because there are truthmakers of H* (e.g., «There will be a disastrous tsunami on
the atoll») not implied by any truthmaker of H. One relevant consequence is that
while H explains P, H* does not. Even more importantly, in cases like this H* is a
purely “empirical” conditional, connecting the initial conditions with an empir-
ical consequence. Therefore, unless it were inferred from the “theoretical” hypo-
thesis H, H* could be discovered only a posteriori, by witnessing P. Therefore P
would not be novel, and H* would not have been deployed in a novel prediction.

9.7.5 The “Ramsey Sentence” Objection

Given that whatever follows from H also follows its Ramsey sentence (RSH), and
that H implies RSH, wouldn’t it follow that every hypothesis is dispensable in
favor of its Ramsey sentence?22
Answer:
No: in general, the essential part H* of H does not coincide with RSH, because
it contains both something more and something less than RSH. It must contain
something more because RSH would not allow to derive NP. For instance, as-
suming in Newton’s gravitation law (N) the terms “force” and “mass” are theoret-
ical and “distance” is observational, its Ramsey sentence is

( )
RS N : ∃x , ∃y x=Gy 1 y 2 /D2 .

But RSN is not enough to predict Neptune. We also need to say what kind of
properties or relations are x and y, how they behave, etc.: we need an expanded
Ramsey sentence like

( )
RS NEx : ∃x, ∃y ....x.... & ....y.... & x=Gy 1 y 2 /D2 ,

where ‘....x...’ and ‘...y...’ are various laws concerning x and y (which, in practice,
characterize them respectively as force and mass).

22 Objection raised by an anonymous referee.

212 Contemporary Scientific Realism

But typically the essential part H* contains also something less than RSH or
RSHEx. For instance, Newton’s Gravitation Law is partly false (since there exists
no gravitation force). Consequently, also RSN and RSNEx are partly false, hence,
inessential. Therefore, in general, the essential H* is different from RSH. It might
still be representable by an expanded Ramsey sentence RSH*Ex (a proper part of
RSHEx), but this wouldn’t be a problem for realism, because:

(i) Like any Ramsey sentence, RSH*Ex is not a merely empirical claim, it
states the existence of unobservable entities, even without naming them.
(ii) In general, RSHEx will have a very complex structure, hence it might be
practically impossible to write it out completely, so to dispense with H.
Besides, as argued in section 9.5, identifying the essential part of hypoth-
eses is very difficult, so even if we could spell out RSHEx, often we would
be unable to dispense with it in favor of its essential part RSH*Ex.
(iii) At any rate, my claim is that the NMA commits only to essential hypoth-
eses, not that only essential hypotheses can be true. So, if the essential
hypothesis turns out to be something like RSH*E, since Ramsey sentences
characterize only partially the entities to which they refer, realists may
hold that something stronger than RSH*E is true.

9.7.6 The “Modus Ponens” Objection

By your criterion no hypothesis will ever be essential, because whenever H to-

gether with OHs and AA entails NP, numberless but intuitively irrelevant
hypotheses do the same, for instance,

H**: (If God exists, then H) & God exists.23

Answer:
As explained above, H** is neither part of H nor implied by it. Therefore H**
does not make H inessential.

Acknowledgments

I am grateful to the organizers of the “Quo Vadis Selective Scientific Realism?”

conference in Durham (2017) for generously helping my participation with

23 Objection raised in discussion by John Worrall, referring to another topos from the debates on

confirmation.
Historical Challenge to Realism and Essential Deployment 213

a travel bursary. I received helpful comments and suggestions from Gustavo

Cevolani, José Díez, Vincenzo Fano, Ludwig Fahrbach, Greg Frost-​Arnold,
Pablo Lorenzano, Giovanni Macchia, Flavia Marcacci, Matteo Plebani, Andrea
Sereni, John Worrall, and especially Peter Vickers and Timothy Lyons. Thanks
to Maria Grazia Severi and Peter Vickers for checking my English grammar.
This research was supported by the University of Urbino Carlo Bo through the
Contributo d’Ateneo progetti PRIN2015 and the DiSPeA Research Projects 2017 and
by the Italian Ministry of Education, University and Research through the PRIN
2017 project “The Manifest Image and the Scientific Image” prot. 2017ZNWW7F_​
004. For this work, I profited from a leave of absence granted by the University of
Urbino (Department of Pure and Applied Sciences) in 2020, during which I was
affiliated with the International Academy for Philosophy of Science.

References
Alai, M. (2012) “Levin and Ghins on the ‘No Miracle’ Argument and Naturalism,”
European Journal for Philosophy of Science vol. 2, n. 1, 85–​110. http://​rdcu.be/​mTcs.
Alai, M. (2014a) “Why Antirealists Can’t Explain Success,” in F. Bacchini, S. Caputo
and M. Dell’Utri (eds.) Metaphysics and Ontology Without Myths, Newcastle upon
Tyne: Cambridge Scholars Publishing, pp. 48–​66.
Alai, M. (2014b) “Defending Deployment Realism against Alleged Counterexamples,”
in G. Bonino, G. Jesson, J. Cumpa (eds.) Defending Realism. Ontological and
Epistemological Investigations, Boston-​Berlin-​Munich: De Gruyter, pp. 265–​290.
Alai, M. (2014c) “Novel Predictions and the No Miracle Argument,” Erkenntnis vol. 79,
n. 2, 297–​326. http://​rdcu.be/​mSra.
Alai, M. (2014d) “Explanatory Realism,” in E. Agazzi (ed.) Science, Metaphysics,
Religion: Proceedings of the Conference of the International Academy of Philosophy of
Science, Siroki Brijeg 24–​24 July 2013 Milano, Franco Angeli, pp. 99–​116.
Alai, M. (2016) “The No Miracle Argument and Strong Predictivism vs. Barnes,” in L.
Magnani and C. Casadio (eds.) Model-​Based Reasoning in Science and Technology.
Logical, Epistemological, and Cognitive Issues, Cham: Springer, pp. 541–​556.
Alai, M. (2017) “Resisting the Historical Objections to Realism: Is Doppelt’s a Viable
Solution?” Synthese vol. 194, n. 9, 3267–​3290. http://​rdcu.be/​mSrG.
Alai, M. (2018) “How Deployment Realism Withstands Doppelt’s Criticisms,”
Spontaneous Generations vol. 9, n. 1, 122–​135.
Carrier, M. (1991) “What is Wrong with the Miracle Argument?” Studies in History and
Philosophy of Science vol. 22, 23–​36.
Chang, H. (2003) “Preservative Realism and Its Discontents: Revisiting Caloric,”
Philosophy of Science vol. 70, 902–​912.
Clarke, Steve, and Timothy D. Lyons (eds.) (2002) Recent Themes in the Philosophy of
Science. Scientific Realism and Commonsense. Dordrecht: Kluwer.
Cordero, A. (2017a). Retention, truth-​content and selective realism, in E. Agazzi (ed.)
Varieties of Scientific Realism, Cham: Springer, pp. 245–​256.
Cordero, A. (2017b) “Making Content Selective Realism the Best Realist Game in Town,”
Quo Vadis Selective Scientific Realism? Conference, Durham, UK 5–​7 August 2017.
214 Contemporary Scientific Realism

Fahrbach, L. (2017) “How Philosophy Could Save Science,” Quo Vadis Selective Scientific
Realism? Conference, Durham, UK 5–​7 August 2017.
Hempel, Carl G. (1958) “The Theoretician’s Dilemma: A Study in the Logic of Theory
Construction,” Minnesota Studies in the Philosophy of Science vol. 2, 173–​226.
Hoskin, M. (ed.) (1999) The Cambridge Concise History of Astronomy,
Cambridge: University Press.
Hudson, J. (1992) The History of Chemistry. London: Macmillan Press.
Kitcher, P. (1993) The Advancement of Science. New York: Oxford University Press.
Koyré, A. (1961) La révolution astronomique. Paris: Hermann.
Laudan, L. (1981) “A Confutation of Convergent Realism,” Philosophy of Science vol.
48, 19–​49.
Lipton, P. (1991) Inference to the Best Explanation, London: Routledge.
Lipton, P. (1993) “VI—​Is the Best Good Enough?,” Proceedings of the Aristotelian Society
vol. 93, 89–​104.
Lipton, P. (1994) “Truth, Existence, and the Best Explanation,” in A. A. Derksen (ed.), The
Scientific Realism of Rom Harré. Tilburg: Tilburg University Press, pp. 89–​11.
Lyons, T. D. (2002) “The Pessimistic Meta-​Modus Tollens,” in S. Clarke and T. D. Lyons
(eds.) Recent Themes in the Philosophy Science. Australasian Studies in History and
Philosophy of Science, vol 17, Dordrecht: Springer, pp. 63–​90. doi:10.1007/​978-​94-​017-​
2862-​1_​4.
Lyons, T. D. (2003) “Explaining the Success of a Scientific Theory,” Philosophy of Science
vol. 70, 891–​901.
Lyons, T.D. (2006) “Scientific Realism and the Stratagema de Divide et Impera,” The
British Journal for the Philosophy of Science, vol. 57, 537–​560.
Lyons, T.D. (2009) “Criteria for Attributing Predictive Responsibility in the Scientific
Realism debate: Deployment, Essentiality, Belief, Retention,” Human Affairs vol. 19,
138–​152.
Lyons, T.D. (2016) “Structural Realism versus Deployment Realism: A Comparative
Evaluation,” Studies in History and Philosophy of Science Part A vol. 59, 95–​105.
Lyons, T.D. (2017) “Epistemic Selectivity, Historical Threats, and the Non-​Epistemic
Tenets of Scientific Realism,” Synthese vol. 194 n. 9, 3203–​3219.
Musgrave, A. (1985) “Realism versus Constructive Empiricism,” in P. Churchland and C.
Hooker (eds.) Images of Science, Chicago, Chicago University Press, pp. 197–​221.
Musgrave, A. (1988) “The Ultimate Argument for Scientific Realism,” in Robert Nola (ed.)
Relativism and Realism in Science, Dordrecht: Springer, pp. 229–​252.
Peters, D. (2014) “What Elements of Successful Scientific Theories are the Correct Targets
for “Selective” Scientific Realism?” Philosophy of Science vol. 81, 377–​397.
Plebani, M. (2017) “Aboutness for Impatients,” Academia.edu, https://​www.academia.
edu/​34295884/​Aboutness_​for_​the_​impatient.
Psillos, S. (1999) Scientific Realism: How Science Tracks Truth.
London-​New York: Routledge.
Sankey, H. (2001) “Scientific Realism: An Elaboration And A Defense,” Theoria, vol.
98, 35–​54.
Smart, J.J.C. (1963) Philosophy and Scientific Realism. London: Routledge.
Stanford, P. K. (2006). Exceeding Our Grasp: Science, History, and the Problem of
Unconceived Alternatives. Oxford: Oxford University Press.
Stanford, K. P. (2009) “Author’s Response,” in “Grasping at Realist Straws,” a Review
Symposium of Stanford (2006), Metascience vol. 18, 379–​390.
Historical Challenge to Realism and Essential Deployment 215

Stanford, P. K. (2017) “A Difference That Makes a Difference: Stein on Realism,

Instrumentalism, and Intellectually Nourishing Snacks,” Quo Vadis Selective Scientific
Realism? Conference, Durham, UK 5–​7 August 2017.
Vickers, P. (2013) “A Confrontation of Convergent Realism,” Philosophy of Science vol. 80,
n. 2, 189–​211.
Vickers, P. (2016) “Understanding the Selective Realist Defence against the PMI,” Synthese
vol. 194, 3221–​3232. https://​doi:10.1007/​s11229-​016-​1082-​4.
Vickers, P. (2018) “Disarming the Ultimate Historical Challenge to Scientific Realism,”
The British Journal for the Philosophy of Science axy035, https://​doi.org/​10.1093/​bjps/​
axy035
Votsis, I. (2011) “The Prospective Stance in Realism,” Philosophy of Science vol. 78,
1223–​1234.
White, R. (2003) “The Epistemic Advantage of Prediction over Accommodation,” Mind
vol. 112 n. 448, 653–​683.
Wright, J. (2002) “Some Surprising Phenomena and Some Unsatisfactory Explanations
of Them,” in Clarke and Lyons (eds.), Recent Themes in the Philosophy of Science.
Australasian Studies in History and Philosophy of Science, vol. 17. Dordrecht: Springer,
pp. 139–​153. https://​doi:10.1007/​978-​94-​017-​2862-​1_​4.
Yablo, S. (2014) Aboutness. Princeton: Princeton University Press.
 10
Realism, Instrumentalism, Particularism
A Middle Path Forward in the Scientific Realism Debate
P. Kyle Stanford

Department of Logic and Philosophy of Science

University of California, Irvine
[email protected]

10.1 Introduction

Here I propose a particular conception of both the current state of play and what
remains at stake in the ongoing debate concerning scientific realism. Recent
decades have witnessed considerable evolution in this debate, including a wel-
come moderation of both realist and instrumentalist positions in response to ev-
idence gathered from the historical record of scientific inquiry itself. In fact, I will
suggest that a set of commitments has gradually emerged that are now embraced
by many (though by no means all) who call themselves scientific realists and also
by many (though by no means all) who would instead characterize themselves as
instrumentalists or (perhaps less helpfully) “antirealists.” I will go on to suggest
that these commitments collectively constitute a shared “Middle Path” on which
many historically sophisticated realists and instrumentalists have already made
substantial progress together, perhaps without realizing how much closer their
respective views have thereby become to one another than either is to the clas-
sical forms of realism and instrumentalism whose labels and slogans they none-
theless conspicuously retain. I will then go on to suggest, however, that at least
one crucial disagreement still remains even between realists and instrumentalists
who walk this Middle Path together, and that this remaining difference actually
makes a difference to how we should go about conducting scientific inquiry itself.
Recognizing this difference poses a further challenge for instrumentalists both
on and off the Middle Path, however, and responding to it will illuminate a fur-
ther dimension of the realism debate itself: namely, the level(s) of abstraction or
generality at which we should seek and expect to find useful epistemic guidance
concerning scientific theories or beliefs. Along this dimension of the debate,

P. Kyle Stanford, Realism, Instrumentalism, Particularism In: Contemporary Scientific Realism.

Edited by: Timothy D. Lyons and Peter Vickers, Oxford University Press. © Oxford University Press 2021.
DOI: 10.1093/​oso/​9780190946814.003.0010
 Realism, Instrumentalism, Particularism 217

I suggest, Middle Path realists and instrumentalists are once again united, but
this time against both the shared presuppositions of their classical predecessors
on the one hand and radical or extreme forms of particularism on the other.

10.2 Finding a Middle Path

The first of the shared commitments that I suggest constitute a Middle Path be-
tween classical forms of both realism and instrumentalism is to what I’ve else-
where (Stanford 2015) called “Uniformitarianism,” in parallel with the great battle
in 19th century geology between Catastrophists and Uniformitarians concerning
the causes and pattern of changes to the Earth. Famously, Uniformitarians (like
Charles Lyell) argued that the broad topographic and geographic features of the
Earth were the product of natural causes like floods, volcanoes, and earthquakes
operating over immense stretches of time at roughly the same frequencies and
magnitudes at which we find them acting today. Their Catastrophist opponents
(like Georges Cuvier) held instead that such natural causes had operated in
the past with considerably greater frequency and/​or magnitude than those we
now observe, on the order of the difference between a contemporary flood and
the great Noachian deluge reported in the Christian Bible. Catastrophists thus
held that the Earth had steadily quieted down over the course of its history, that
truly fundamental and wide-​ranging changes to its topography and geography
are now confined to the distant past, and that contemporary natural causes will
further modify that topography and geography only in comparatively marginal
and limited ways. By contrast, Uniformitarians held that if given enough time
to operate present-​day natural causes would continue to transform the existing
topography and geography of the Earth just as profoundly as it was transformed
in the past.
Likewise in the case of the realism dispute, Uniformitarians take the view that
the future of the scientific enterprise will continue to be characterized by the-
oretical revolutions and conceptual transformations just as profound and fun-
damental as those we have witnessed throughout the history of that enterprise.
By contrast, a textbook description of what I will call classical or Catastrophist
realism holds that “Our mature scientific theories, the ones used to underwrite
our scientific projects and experiments, are mostly correct” and “[w]‌hat errors
our mature theories contain are minor errors of detail” (Klee 1999 313–​4). Such
Catastrophists take the view that truly profound and fundamental revisions to
our scientific understanding of the world are either largely or completely con-
fined to the past and that the future of scientific inquiry will not be characterized
by the sorts of fundamental revolutions or transformations by which theo-
ries like Newton’s mechanics, Dalton’s atomism, and Weismann’s theory of the
218 Contemporary Scientific Realism

germ-​plasm ultimately came to be profoundly modified, qualified, amended, or

simply replaced. That is, Catastrophist realists hold that the theoretical ortho-
doxy embraced by future scientific communities will include what seem both to
them and us to be simply expanded, amended, and more sophisticated versions
of the most successful theories we ourselves have already adopted. Thus, while
Uniformitarians see us as being in the midst of an ongoing historical process
of fundamental revolution or transformation in our scientific beliefs, rather
than as having the enviable good fortune of living at or near the end of that pro-
cess, Catastrophist scientific realists instead seem forced to adopt some form of
exceptionalism concerning at least the most successful scientific theories of the
present day.
Such classical or Catastrophist realism has not only been influentially
championed by philosophical luminaries like Smart (1963) and Putnam
(1975) but is also often claimed to be the view of the matter favored by
common sense itself. It is therefore striking that in recent years many prom-
inent historically sophisticated scientific realists have abandoned such
Catastrophism in favor of the Uniformitarian alternative while insisting
that this concession does not undermine the more modest realist epistemic
entitlements that they themselves defend. So-​called “selective” scientific
realists have argued, for example, that although we should indeed anticipate
further radical and fundamental changes in our theoretical conception of
the natural world, we can nonetheless identify particular elements, aspects,
or features of our best scientific theories that we can justifiably expect to
find preserved throughout the course of such further changes, whether those
privileged elements are held to be their claims about the “structure” of na-
ture (Worrall 1989), their “working posits” (Kitcher 1993), their “core causal
descriptions” (Psillos 1999), their “detection properties” (Chakravartty 2007,
Egg 2016), the posits that “unify the accurate empirical claims” of a theory
(Peters 2014), the verae causae they identify (Novick and Scholl 2020), or
something else altogether. This commitment to Uniformitarianism would
seem to unite such selective realists with historically motivated defenders
of instrumentalism or anti-​realism and against classical, commonsense, or
Catastrophist varieties of scientific realism itself.
But realists and instrumentalists on the Middle Path are similarly united in
rejecting central commitments of many classical forms of instrumentalism or
anti-​realism as well, including the idea, perhaps most familiar from Thomas
Kuhn in his most exuberant moods, that the changes still to come in our scien-
tific conception of the world will periodically be so radical and profound that
they will render it impossible to impartially, neutrally, or fairly compare the-
ories on either side of such a revolutionary divide either to one another or to
the available body of empirical evidence. Instead, Middle Path instrumentalists
 Realism, Instrumentalism, Particularism 219

allow not only that we can perfectly well articulate the competing claims of two
scientific theories about a single shared world using language that privileges
neither theory, but also that we can characterize the available evidence in sup-
port of each theoretical alternative in ways that are at least neutral between those
two competing theories (if not independent of any and all “theorizing” whatso-
ever) in reaching an impartial judgment that one is better or worse confirmed by
the existing evidence, or even that one or both theories are definitively refuted
(within the bounds of fallibilism) by that evidence. They similarly reject the fur-
ther consequence drawn by various interest-​driven theorists of science that the
outcomes of such theoretical competitions are therefore typically determined
less by the available evidence in support of each theoretical alternative than by
the comparative power, standing, resourcefulness, and determination of the so-
cial groups who advance and defend them. Where Uniformitarianism assures
us that further profound and fundamental changes are still to come for our sci-
entific picture of the world, a further commitment to what we might naturally
call commensurability assures us that such changes will neither prevent us from
fully understanding the competing conceptions of nature thereby proposed nor
from impartially adjudicating the character and strength of the evidence in sup-
port of each of those competing conceptions. This is certainly not to suggest that
resolving such competitions is simple or straightforward, or that an individual
scientist’s own theoretical (and other) sympathies do not influence her evalua-
tion of the evidence, or even that such comparisons can always be convincingly
resolved by the particular body of evidence available at a given time. But it is
to claim that their ultimate resolution typically depends on the accumulation of
evidence that the relevant scientific community rightly sees as objectively and
impartially favoring one proposal over the other. Such a commitment to com-
mensurability would seem to similarly unite instrumentalists and realists on the
Middle Path together with one another and against such radically Kuhnian var-
ieties of instrumentalism or anti-​realism.
Realists and instrumentalists who travel this Middle Path together are also
united against other classical varieties of instrumentalism or anti-​realism by
their commitment to what I will call the Maddy/​Wilson Principle, which I first
encountered when my colleague Penelope Maddy skeptically responded to my
own instrumentalist sympathies with this especially pithy formulation: “well,
nothing works by accident or for no reason.” That is, she seemed to regard instru-
mentalism as committed not simply to the view that our best scientific theories
are powerful cognitive instruments that guide our predictions, interventions,
and other forms of practical engagement with nature successfully, but also that it
is misguided or somehow illegitimate even to ask, much less try to discover or ex-
plain, how they manage to achieve that success when and where they do. Having
recognized the ability of our best theories to guide our practical engagement
220 Contemporary Scientific Realism

with nature successfully, such instrumentalism rejects even the demand for any
explanation of how or why they are able to do so.
When I reminded her of this conversation some years later, she was quick to
credit the influence of Mark Wilson’s Wandering Significance as having inspired
this particular expression of her disquiet. And indeed, Wilson does articulate
something very like this same concern in his own inimitable way:

However, I regard this [instrumentalist] terminology as misleading because

successful instrumentalities, whether they be of a mechanical or a symbolic na-
ture, always work for reasons, even if we often cannot correctly diagnose the na-
ture of these operations until long after we have learned to work profitably with
the instruments themselves. (2006 220, original emphasis)

Both Maddy and Wilson, then, endorse the broad principle that when a sci-
entific theory enjoys a track record of fine-​grained and wide-​ranging success
in guiding our practical engagement with the world, it typically does so in
virtue of some systematic connection or relationship between the description
of the world offered by that theory and how things actually stand in the world
itself.
It is no accident that Maddy and Wilson both regard instrumentalism as com-
mitted to the idea that there is simply nothing more to say or know about how
and why our instrumentally powerful scientific theories work as well as they do
when and where they do. Scientific instrumentalism has had a wide variety of
incarnations over the last several centuries, and some of the most influential
are indeed characterized by explicit or implicit versions of this commitment.
Perhaps most famously, some logical positivist and logical empiricist thinkers
held that the claims of our best scientific theories are not even assertions about
the world in the first place, nor, therefore, even candidates for truth or falsity,
but instead simply “inference tickets” allowing us to predict or infer some ob-
servable states of affairs from others. More recently, van Fraassen’s Constructive
Empiricism argues that it is illegitimate to demand anything more than the em-
pirical adequacy of a theory as an explanation of its success, insisting instead
“that the observable phenomena exhibit these regularities, because of which they
fit the theory, is merely a brute fact, and may or may not have an explanation
in terms of unobservable facts ‘behind the phenomena’ ” (1980 24). Thus, there
is indeed a long and distinguished philosophical tradition advocating forms of
instrumentalism that either reject or remain agnostic concerning the Maddy/​
Wilson Principle, insisting instead that once we have recognized the instru-
mental utility of our scientific theories it is somehow misguided or illegitimate
even to ask how or why those theories manage to be so instrumentally useful
when and where they do.
 Realism, Instrumentalism, Particularism 221

The influential legacy of such classical instrumentalism makes it critical to

recognize not just the plausibility but also the foundational importance of the
Maddy/​Wilson Principle. In science as in ordinary life, it is generally true that
when things work, they work for reasons, and when scientific theories are able
to achieve robust empirical and practical success, it is surely at least reasonable
to think that the reasons for that success will consist in some systematic rela-
tionship or connection between how the theory represents (some part of) the
world as being and how things actually stand there—​otherwise that empirical
success really would be miraculous! Indeed, the Middle Path not only embraces
the Maddy/​Wilson Principle’s insistence that there must be some reason why
a cognitive instrument that works well does so, but also goes so far as to insist
that the historical record itself provides us with abundant exemplars of at least
the broad sorts of such systematic relationships that presumably also constitute
the reasons for the successes of various contemporary scientific theories as well,
whether or not we are ever in a position to specify those reasons more precisely
in particular cases.
To see how, suppose for a moment that Einstein’s relativistic mechanics
represents a true and complete account of how things stand in the otherwise in-
accessible domain of nature that it seeks to describe. On that assumption, we
can explain how and why Newton’s gravitational mechanics is so spectacularly
successful when and where it is by describing in detail how aspects or elements
of Newtonian mechanics are systematically related to features of the actual world
(as described by General Relativity) and how these systematic relationships
permit the theoretical apparatus of Newtonian mechanics to make accurate
predictions and guide interventions successfully across a wide though not un-
limited range of circumstances. If contemporary theoretical orthodoxy in me-
chanics simply describes how things really stand in nature, the details of those
systematic relationships themselves constitute the “reasons” why Newtonian me-
chanics works as well as it does. Assuming the truth of contemporary theoretical
orthodoxy more broadly would similarly allow us to identify the actual reasons
for the systematic instrumental utility of many other successful but ultimately
rejected scientific theories of the past such as the caloric theory of heat, phlo-
gistic chemistry, the wave theory of light, Weismann’s theory of the germ-​plasm,
and many other familiar examples. And even if we think all contemporary theories
are fundamentally mistaken, we should fully expect that highly successful con-
temporary scientific theories stand in one or more systematic relationships to the
truth of the matter regarding various otherwise inaccessible domains of nature
that are like the relationship between Newtonian and relativistic mechanics, or
like that between caloric theory and contemporary thermodynamics, or like that
between Weismann’s theory of the germ-​plasm and contemporary molecular
genetics, and so on. Of course, even this small collection of familiar examples
222 Contemporary Scientific Realism

makes clear that there is no single such relationship holding between all earlier
successful theories and their contemporary successors: realists would need to
appeal to a wide and heterogeneous array of different systematic relationships or
“reasons” of this sort in explaining the diverse particular empirical successes of
the various instrumentally powerful, past scientific theories that have since been
abandoned or replaced.
I suggest that many realists and instrumentalists alike, especially those who
rely on evidence drawn from the historical record, have gradually come to em-
brace these shared commitments to Uniformitarianism, commensurability,
and the Maddy/​Wilson Principle, perhaps without fully realizing how much
more they have thereby come to agree with one another than either of them
does with the more classical doctrines whose labels and slogans they nonethe-
less retain. Their broadly shared conception of the past, present, and future of
scientific inquiry anticipates many substantial continuities between future sci-
entific orthodoxy and that of the present day (just as there are between present
theoretical orthodoxy and that of the past), but also many substantial discon-
tinuities just as profound and significant as those we now find in past historical
episodes of fundamental revolution or upheaval. Moreover, I suggest that realists
and instrumentalists who travel this Middle Path have already made consider-
able progress in using further historical evidence to refine and elaborate that
broadly shared conception. Important recent work conducted by self-​described
realists and antirealists alike, for example, has revealed that the mistaken (i.e.,
ultimately rejected) claims or components of successful past theories have often
played crucial and ineliminable roles in generating those theories’ empirical
successes (e.g., Lyons (2002, 2006, 2016, 2017), Saatsi and Vickers (2011); for
a fairly comprehensive listing, see Vickers (2013) and the references contained
therein). Likewise, Peter Vickers (2013, 2017) has recently argued that we can
at least identify posits (including some that do genuine work in generating a
theory’s successful implications) that are nonetheless inessential or eliminable
from those theories in a sense that renders them poor candidates for realist com-
mitment. Such increases in the nuance and sophistication of their broadly shared
vision of the past, present, and future of scientific inquiry should be recognizable
as important forms of progress by both realists and instrumentalists alike on the
Middle Path.

10.3 Trouble in Paradise: A Remaining Difference that

Makes a Difference

I do not mean to suggest, however, that these shared commitments of realists and
instrumentalists on the Middle Path leave no room for significant disagreement
 Realism, Instrumentalism, Particularism 223

between them. As noted earlier, “selective” scientific realists are largely moti-
vated by the idea that the historical record itself (perhaps together with other
considerations) puts us in a position to reliably distinguish the particular elem-
ents, aspects, or features of our own best scientific theories (e.g., working posits,
core causal descriptions, structural claims, etc.) genuinely responsible for their
empirical successes from those that are instead “idle” or otherwise not required
for those successes (e.g., the electromagnetic ether). Even if we expect further
profound and dramatic changes in theoretical orthodoxy as scientific inquiry
proceeds into the future, Middle Path realists insist that we can nonetheless jus-
tifiably expect these privileged elements, aspects, or features to be retained and
ratified in some recognizable form throughout the further course of scientific in-
quiry itself. Note that without a commitment to at least this minimal form of sta-
bility or persistence, the selective realist loses not only her grounds for claiming
to have identified secure epistemic possessions on which we can safely rely as in-
quiry proceeds, but also her grounds for claiming that these particular elements,
aspects, or features of our best scientific theories were those genuinely respon-
sible for their various empirical successes (for which they have turned out not to
be essential or required after all).
By contrast, those on the Middle Path who embrace the “instrumentalist”
label certainly accept the Maddy/​Wilson Principle’s insistence that there must be
such reasons for the successes of our best scientific theories, but they nonetheless
deny that we ourselves are generally in a position to know what those reasons are
in particular cases, or to reliably identify which particular elements or features
of our best scientific theories we should therefore expect to find persisting
throughout the course of further inquiry. Such instrumentalists will see the wide
variety of such reasons that (as we noted earlier) contemporary scientific ortho-
doxy would need to invoke in explaining the various successes of various past
scientific theories as piling up counterexamples rather than confirmation for the
idea that any single aspect, element, or feature of a successful scientific theory is
invariably or even just reliably preserved in its successors: sometimes equations
or claims about the “structure” of nature are preserved from one successful scien-
tific theory to its successors, but other times it is the fundamental entities posited
by that theory, or the “core causal descriptions” of those entities, or something
else altogether—​no one of these forms of continuity (or corresponding versions
of selective realism itself) seems to capture even an especially wide range of cen-
tral historical examples.
Rather than seeking to adjudicate this remaining central point of disagree-
ment dividing even realists and instrumentalists traveling together on the Middle
Path, I will instead argue for its importance. It truly matters, I suggest, because
it actually makes a difference to how we should go about conducting further sci-
entific inquiry itself. To see why, recall first the case of classical, commonsense,
224 Contemporary Scientific Realism

or Catastrophist realists, who are confident that the theoretical orthodoxy

embraced by future scientific communities will include what seem both to us and
to the members of those communities to be simply updated, expanded, and more
sophisticated versions of at least the most successful theories that we ourselves
have already embraced. At least with respect to those successful theories, then,
the classical or Catastrophist realist simply does not see any real need for what
the National Science Foundation and other granting agencies call “transform-
ative science” and characterize explicitly as “revolutionizing entire disciplines;
creating entirely new fields; or disrupting accepted theories and perspectives”
(Bement 2007). The realist should be perfectly happy (or at least systematically
happier than her instrumentalist counterpart, as will become clear below) for
review boards to reject lines of research or theoretical proposals that fundamen-
tally contradict or violate existing successful scientific theories, as she thinks it
quite unlikely that any such alternative will ever become part of the theoretical
orthodoxy we come to embrace in the future. Indeed, the more conflict there is
between a given theoretical proposal and the central claims of existing successful
theoretical orthodoxy, the more confident she will be that it is misguided in some
fundamental and fatal way.
Of course, Middle Path realists instead hold only that we can justifiably pre-
dict which specific elements, aspects, or features of our best scientific theories
are responsible for their successes, but they similarly expect those same elem-
ents, aspects, or features of our own theories to persist and be preserved in some
recognizable form in any future scientific theory we ultimately adopt. (Again, if
that expectation is defeated, the realist loses her claim to have picked out either
the trustworthy elements of contemporary scientific theories or those actually
responsible for their empirical successes.) Accordingly, it seems that they too
should be systematically skeptical of theoretical proposals or avenues of research
that contradict or fail to preserve whatever elements, aspects, and/​or features of
our best scientific theories engender those realist commitments. In both cases,
realists seem to have systematic grounds for prioritizing investments in finding,
exploring, and testing theoretical alternatives that preserve whatever it is that
they are realists about over the pursuit of theoretical alternatives that fail to do so.
The realist may, of course, have instrumental or strategic reasons for taking se-
riously or exploring particular theoretical possibilities that conflict with what she
takes herself to already know: as Ronald Fisher famously suggested, for example,
“No practical biologist interested in sexual reproduction would be led to work
out the detailed consequences experienced by organisms having three or more
sexes, yet what else should he do if he wishes to understand why the sexes are, in
fact, always two?” (1930 ix). She might even be convinced (for whatever reason)
that the best way to make incremental progress in improving existing theories is
by trying to articulate radically different alternative theoretical proposals with
 Realism, Instrumentalism, Particularism 225

which to compare them. But the instrumentalist shares any such strategic or in-
strumental reasons the realist may have for exploring or developing a theoretical
alternative conflicting with (some privileged part of) existing theoretical ortho-
doxy. The instrumentalist also has another that is ultimately far more impor-
tant: she believes that even more instrumentally powerful alternatives radically
distinct from contemporary scientific theories are actually out there still waiting
to be discovered.
Evaluating the promise, interest, or appeal of any particular theoretical pro-
posal is, of course, a complex and multi-​dimensional affair, so the suggestion
here is certainly not that realists must always favor investing in the pursuit of
any theoretical proposal (about anything) consistent with existing orthodoxy
over the pursuit of any theoretical proposal (about anything) contradicting
that orthodoxy. Nor is there some threshold degree of theoretical conservatism
that realists must meet or exceed, either in general or in any particular case. The
point instead is that those who are realists regarding some particular scientific
theory (or privileged part thereof) should be prepared to treat the inconsistency
of a given alternative theoretical proposal with the central claims (or otherwise
privileged elements) of that theory as a reason to doubt that the alternative in
question is even a viable candidate for representing the truth about the do-
main of nature it seeks to describe. All else being equal, this in turn constitutes
a reason (for the realist) to discount or disfavor that alternative in competition
for funding or support against otherwise equally attractive alternatives that do
not similarly contradict (the relevant parts of) existing orthodoxy and therefore
remain more plausible avenues for successfully extending, expanding, sophisti-
cating, or supplementing our existing scientific conception of ourselves and the
world around us. But this same reason for increased skepticism remains in force
for the realist even when the alternatives being compared are anything but oth-
erwise equally attractive, that is, even when it is simply one among a wide range
of considerations bearing on the comparative attractions of investing in the pur-
suit of one theoretical proposal rather than another. Thus, holding all such fur-
ther considerations fixed in any particular case reveals that realists have reasons
instrumentalists lack for skepticism about just those theoretical proposals that
violate (the relevant parts of) existing theoretical orthodoxy.
Of course, the instrumentalist no less than the realist will need to make dif-
ficult choices about how to invest and distribute the scarce resources available
to support scientific inquiry itself. Many of the considerations she will weigh in
making such judgments or decisions will operate for her in precisely the same
way that they do for the realist: both, for example, will see alternative theoret-
ical proposals as more attractive or promising targets for investment the better
able they are to recover and/​or explain whatever empirical consequences or
implications of existing theories have already been independently confirmed by
226 Contemporary Scientific Realism

experiment and observation. Realists and instrumentalists will likewise appeal

in much the same way to a wide array of further considerations to which granting
agencies already direct the attention of their referees and review boards, such as
the extent to which a proposal is well-​reasoned, well-​organized, and based on
a sound rationale; the extent to which it promises to benefit society or advance
desired societal outcomes; and the extent to which the proposers are well quali-
fied and have access to the resources needed to carry out the proposed activities.
And both realists and instrumentalists are entitled to make nuanced, discrim-
inating judgments driven by the details of the theory and evidence in question
concerning the extent to which the exploration and development of any partic-
ular theoretical alternative or line of investigation is comparatively more or less
likely to help us make progress either in refining and extending existing theoret-
ical orthodoxy or in finding and developing even more empirically successful
theories that will ultimately supplant those we now embrace. But for those who
are realists about a given scientific theory (or particular elements, aspects, and
features of that theory) that calculation should reflect what she already takes her-
self to know about the claims, elements, aspects, or features of contemporary the-
oretical orthodoxy that will be retained and ratified in some recognizable form
as part of any theoretical orthodoxy we come to embrace in the future. This same
consideration simply does not arise for those who are instead instrumentalists
about that theory.

10.3.1 Theoretical Conservatism: A Double-​Edged Sword

Although realists may bristle at the suggestion that they defend a systematically
more theoretically conservative form of scientific inquiry than instrumentalists
do, this difference seems to generate an even more direct and immediate chal-
lenge to instrumentalism instead. After all, if such commitments to various parts
of contemporary theoretical orthodoxy are what enable realists to dismiss alter-
native theoretical possibilities out of hand when they conflict with too much of
what we think we already know about the world, it seems natural to worry that
the instrumentalist will, by contrast, wind up forced into an absurd permissive-
ness with respect to the alternative theoretical possibilities that she is prepared
to take seriously and/​or consider potentially deserving investments of the time,
energy, money, and other scarce resources available for pursuing scientific in-
quiry itself. Surely even the instrumentalist thinks we should take a dim view
of investing scarce resources in finding and/​or developing alternative theoret-
ical proposals contradicting claims like “fossils are the remains of once-​living
organisms,” “many diseases are caused by bacterial or viral infections,” or “the
world around us is filled with microscopic organisms” as well as many others
 Realism, Instrumentalism, Particularism 227

that seem put beyond serious question or reasonable doubt by the evidence we
already have. Although the instrumentalist remains open to the possibility that
future scientific communities may ultimately express these firmly established
facts using a very different theoretical vocabulary than our own,1 she should
nonetheless take a skeptical view of alternative theoretical proposals asserting
or implying the falsity of such claims (even as we ourselves express them). (Here
and throughout I assume that the theoretical alternatives in question simply con-
tradict, explicitly or implicitly, the relevant claims concerning fossils, infections,
microscopic organisms, etc., and do not offer convincing alternative explanations
of the evidence we presently take to support those claims.) But if the instru-
mentalist restricts her beliefs to only the empirical consequences or observable
implications of her best scientific theories, she might seem to lose any ground for
thinking that theoretical alternatives contradicting such claims should be taken
any less seriously or regarded with any more suspicion than those which instead
simply contradict, say, the far more speculative claims of our best scientific theo-
ries concerning the nature of dark matter or dark energy.
Recall, however, that the instrumentalist’s view is not that all claims of con-
temporary theoretical orthodoxy in science will ultimately be abandoned, but
instead simply that many central and fundamental claims will be and that we
are not generally in a position to predict in advance just which claims these will
be. What she rejects is the realist’s claim to have identified general or categorical
features of scientific theories and/​or their supporting evidence from which their
approximate truth (or some analogue) can be reliably inferred. That is, the re-
alism debate itself has been most fundamentally concerned with whether there
is any general or categorical variety of empirical success or evidential support
that serves as a reliable indicator that a theory (or its privileged parts) will be
retained and ratified throughout the course of further inquiry. But realists and
instrumentalists alike can recognize exceptions to their general or generic ex-
pectations in particular cases based on evidence or other considerations specific
to the case in question. Many realists, for example, are prepared to make such an
exception in the case of quantum mechanics, whose empirical success is extraor-
dinary but whose very intelligibility to us as a description of how things actually
stand in some otherwise inaccessible domain of nature remains controversial.
Moreover, many realists are inclined to see the theory of evolution by natural se-
lection as extremely well confirmed despite the fact that the evidence supporting
that theory includes little in the way of the “novel predictive success” that they
argue is generally required to justify the claim that a theory (or its privileged

1 For example, when Joseph Priestly reported that after breathing “dephlogisticated air . . . my

breast felt peculiarly light and easy for some time afterwards” we judge that he made a true claim
about the effects of breathing oxygen using flawed or dated theoretical vocabulary, not a false or
empty claim about a substance that does not exist (see Kitcher 1993 100).
228 Contemporary Scientific Realism

parts) will persist throughout the course of further inquiry. While the instru-
mentalist holds instead that no generic characteristic or category of theories or
the evidence available in support of them entitles us to any more than the expec-
tation that the theory in question is a useful conceptual tool or instrument for
guiding our practical interaction with nature, she nonetheless remains just as
free as the realist to recognize particular cases as exceptions to this generic ex-
pectation on the basis of considerations specific to those cases. What she cannot
do is respond to the challenge by specifying generic epistemic characteristics or
categories that distinguish trustworthy from untrustworthy scientific beliefs, for
such generic characteristics and categories are just what she thinks we have yet to
identify successfully.
Accordingly, her judgment that a particular belief or claim is established be-
yond a reasonable doubt will have to be a function of her evaluation of the details
of the specific evidence she has in support of that particular belief. I have argued
at length elsewhere (Stanford 2010), for example, that the details of the evidence
we now have supporting the (once highly contentious) hypothesis that fossils
are the remains of previously living organisms should lead us to conclude that
this hypothesis is not merely a useful cognitive instrument but is in addition an
accurate description of how things stand in nature itself. Of particular impor-
tance in that case, I suggested, was an abundance of what I called “projective”
evidence in support of this hypothesis (especially from the field of experimental
taphonomy) in addition to merely eliminative and abductive forms of evidence.
In a similar fashion, it will be the details of the evidence available in particular
cases which convince the instrumentalist that any particular belief is established
beyond a reasonable doubt. (Indeed, this same response might also appeal to se-
lective or Middle Path realists, who face their own version of the problem insofar
as proposed theoretical alternatives might well preserve the “working posits,”
“structural claims,” or other privileged elements of contemporary theoretical or-
thodoxy while nonetheless disqualifying themselves from serious consideration
by contradicting claims like “many diseases are caused by bacterial infections” or
“fossils are the remains of once-​living organisms.”)
Here an important strategic difference emerges between the realist’s and the
instrumentalist’s respective engagements with the historical record. The realist
begins from the attempt to explain the success of science and seeks to defend the
credibility of one particular explanation against potentially undermining histor-
ical counterevidence, in the process refining whatever criterion of epistemic se-
curity she proposes for identifying just which categories of scientific claims and
commitments she thinks can be trusted to persist throughout further scientific
inquiry. By contrast, the instrumentalist begins from the demonstrably serious
threat to the persistence of our scientific beliefs posed by the historical record
and tries to find ways to restrict our beliefs so as to obviate or sufficiently mitigate
 Realism, Instrumentalism, Particularism 229

that threat. Because she knows of no general way to distinguish contemporary

scientific claims or beliefs that will be abandoned or overturned from those that
will instead be ratified and retained in future theoretical orthodoxy, she will pro-
ceed cautiously, presuming that a generically successful scientific theory is simply
a useful conceptual tool or instrument (like Newtonian mechanics or caloric
thermodynamics) unless and until presented with compelling specific reasons
to regard a particular theory, belief, or commitment as an accurate description
of some otherwise inaccessible part of nature (like the organic origins of fossils).
That is, she starts with strict constraints regarding the beliefs to which a generi-
cally successful theory entitles us (such as the theory’s “empirical implications”
or its claims about “observable” matters of fact, though cf. Stanford 2006 Ch.
8) and then looks for reasons to relax these restrictions in particular cases. In the
meantime, to oversimplify, given a motley collection of suspects with checkered
pasts and conflicting evidence, realists are generously presuming our successful
theories (or privileged parts thereof) innocent unless and until proven guilty,
while instrumentalists are cynically presuming guilt unless innocence can be
convincingly established.
This freedom of both realists and instrumentalists on the Middle Path to coun-
tenance exceptions to their respective general expectations or inferential entitle-
ments concerning the fates of generically successful theories does, however,
represent a further departure from their classical predecessors, who instead (as
we noted earlier) typically saw themselves as articulating competing proposals
concerning the appropriate epistemic attitude to take toward “successful scien-
tific theories” as such. In fact, recognizing this further divergence of Middle Path
realism and instrumentalism from their classical counterparts invites renewed
attention to a much less widely recognized further dimension of disagreement
in the modern realism debate concerning the level of abstraction or generality at
which we should seek or expect to find useful epistemic guidance concerning
our scientific beliefs. Along this dimension, I will now suggest, Middle Path
realists and instrumentalists are once again largely united with one another, not
only in contrast to their classical predecessors’ fully general ambitions and ex-
pectations, but also to the diametrically opposed expectations of those who de-
fend radical forms of what we might call “particularism” regarding the scientific
realism debate.

10.3.2 Particularism and Generality in the Realism Debate

Since its inception in the writings of Smart (1963), Putnam (1975), van Fraassen
(1980), Boyd (1984), Laudan (1981), and others, the modern realism debate has
been predicated on the assumption that there is some point to ascending to the
230 Contemporary Scientific Realism

levels of abstraction at which we generalize about our “mature scientific theories”

and their “empirical successes” or “approximate truth.” For realists, the point of
that ascent was to explain the successes of such theories in a way that revealed
general or abstract epistemic categories we might use to reliably pick out those
particular theories or otherwise privileged elements of contemporary scientific
orthodoxy that represent secure epistemic possessions we can justifiably expect
to persist in some recognizable form throughout the remaining course of sci-
entific inquiry itself. The instrumentalist thinks we learn quite different lessons
from considering matters at this level of abstraction and generality: she is con-
vinced by the historical evidence, for example, that we should expect many of
even the most fundamental commitments of our most successful contemporary
theoretical orthodoxy to be eventually overturned and that there are no such ge-
neral epistemic features or categories we might use to form reliable expectations
regarding which of those commitments will or will not be preserved in some
recognizable form throughout the course of our further scientific investigation
of the world. Of course, if the instrumentalist goes on to claim that we should
believe only a successful theory’s empirical implications, or what it says about
observable matters of fact, or some such (cf. Stanford 2006 Ch. 8), she too is of-
fering a (competing) abstract and general criterion of epistemic security for our
scientific beliefs.
In contrast to both realism and instrumentalism, particularism holds that
there is not now nor was there ever any point in ascending to these heights of
abstraction and generality in the first place and suggests that no useful guidance
for evaluating the epistemic status of scientific claims can be gleaned from doing
so. The particularist thinks the very best we can do in deciding whether some
particular scientific claim or commitment is true and/​or will be retained and rat-
ified throughout the course of further inquiry is to carefully evaluate the details
of the specific evidence we have for and against that particular claim or com-
mitment. That is, she thinks that the delicate and painstaking scientific work of
evaluating particular claims and commitments already represents our most so-
phisticated efforts to determine the appropriate level of confidence we should
have in particular claims about what things exist and how they behave, and she
has more confidence in the outcome of those efforts than in any far less specific
guidance we might hope to find by seeking broad patterns in the historical re-
cord or from any generic conception of how science works. That is, she thinks
that the evidential import of any purported general relationship between em-
pirical success and truth, or facts about how often past successful theories have
turned out to be not even approximately true, or how reliably we’ve failed to con-
ceive of well-​confirmed theoretical alternatives when they existed, or how fre-
quently past scientists themselves have held spectacularly mistaken beliefs about
which parts of their own theories were conclusively confirmed by the available
 Realism, Instrumentalism, Particularism 231

evidence, simply pales into comparative insignificance when confronted with

the ordinary first-​order evidence we have in favor of or against any particular
scientific claim. In this way the particularist sees the existing debate between sci-
entific realists and instrumentalists as simply superfluous to our real efforts to
find out anything about the world: broad reflections on the scientific enterprise
as a whole or patterns in the historical record simply add nothing of substance to
the outcomes of those investigations in particular cases. Moreover, such general-
ities and abstractions are simply insensitive to precisely the sorts of variation in
the details of the evidence we have in different cases that the particularist thinks
really should generate varying degrees of confidence concerning various claims
about the existence and character of fossils, or dark matter, or electrical charge,
or bacterial infections.
Particularist sentiments of this sort, I suggest, constitute an important
part of Arthur Fine’s motivation for embracing what he calls the Natural
Ontological Attitude (NOA). Fine insists that realists and antirealists alike
make a profound mistake when they seek to provide universalizing interpret-
ations that purport to characterize the general aim of science, what our sci-
entific claims really mean or say, and/​or the epistemic accomplishments of
science as a whole. “What binds realism and antirealism together,” he says, is
that “[t]‌hey see science as a set of practices in need of an interpretation, and
they see themselves as providing just the right interpretation. But science is
not needy in this way” (1986a 147–​8). Instead, he suggests, the first-​order sci-
entific claims and counterclaims of ordinary scientific practice already repre-
sent our most careful efforts to decide what entities exist and what claims are
true (in the humble, quotidian, philosophically unanalyzed senses of those
terms), and we must treat such claims and beliefs as standing on their own
bottom rather than standing in need of any further, distinctively philosophical
analysis of what they really mean, or which ones we should actually believe, or
even the point of making such claims in the first place. In fact, Fine rejects the
idea that the scientific enterprise has any general aim or goal. Although sci-
entific activity is replete with important goals and purposes, these are specific
to particular contexts and practices, such as “For what purpose is this partic-
ular instrument being used, or why use a tungsten filament here rather than
a copper one?” But he notes that it is simply a gross fallacy in quantifier logic
to move from “They all have aims” to “There is an aim they all have” (1986b
173). When we go on to try to identify the general goal, aim, or purpose of sci-
ence itself, he says, “we find ourselves in a quandary, just as we do when asked
‘What is the purpose of life?’ or indeed the corresponding sort of question for
any sufficiently rich and varied practice or institution” (1984 148): “the quest
for a general aim [for science], like the quest for the meaning of life, is just
hermeneuticism run amok” (1986b 174).
232 Contemporary Scientific Realism

Fine insists that it is similarly misguided to appeal to any global philosophical

analysis or interpretation of science in order to decide which scientific claims to
believe (e.g., those concerning “observable” states of affairs) or how much confi-
dence to invest in them. Instead,

NOA’s attitude makes it wonder whether any theory of evidence is called for.
The result is to open up the question of whether in particular contexts the evi-
dence can reasonably be held to support belief (regardless of the character of the
objects of belief). Thus NOA, as such, has no specific ontological commitments.
It has only an attitude to recommend: namely, to look and see as openly as one
can what it is reasonable to believe in and then to go with the belief and com-
mitment that emerges. Different NOAers could, therefore, disagree about what
exists, just as different, knowledgeable scientists disagree. (1986b 176–​7)

What binds these NOAers together, it seems, is simply their skepticism about
whether ascending to some more general or abstract philosophical level of in-
vestigation, analysis, or reflection on science itself will help them make any
progress in deciding what to believe about the world. Scraping away or refusing
to indulge in the universalizing interpretations offered by both realist and anti-​
realist philosophers of science simply leaves us with the first order practices of
advancing, challenging, and defending particular scientific claims in particular
scientific contexts. “The general lesson,” Fine suggests, “is that, in the context of
science, adopting an attitude of belief has as warrant precisely that which science
itself grants, nothing more but certainly nothing less” (Fine 1986a 147).
In a similar fashion, Penelope Maddy sees the trouble at the root of both re-
alism and instrumentalism as their shared inclination to try to ascend to some
higher philosophical court of epistemic evaluation in which we leave behind the
ordinary sorts of evidence and evaluative standards that are characteristic of sci-
ence itself. She sees van Fraassen, for example, as conceding that the evidence in
favor of the existence of atoms is perfectly adequate and convincing for scientific
purposes but insisting that this does not settle whether or not we have sufficient
epistemological or philosophical grounds for holding such beliefs:

As far as methodology goes, the actual practice of science, it is perfectly rea-

sonable for our scientist to take the Einstein/​Perrin evidence as establishing the
real existence of atoms. But for the proper “interpretation” of atomic theory, we
must adopt a point of view other than that of the practicing scientist: “stepping
back for a moment,” we adopt an “epistemic attitude” towards the theory ([van
Fraassen 1980] 82). Only then, answering the question as epistemologists, do
we determine that the Einstein/​Perrin evidence is not enough, and indeed, that
no evidence can be enough to establish the existence of entities that cannot
 Realism, Instrumentalism, Particularism 233

be perceived by unaided human senses. Here we have yet another two-​level

theory: at the ordinary scientific level, we have good evidence that atoms are
real; at the interpretive, epistemic level, we do not. (Maddy 2001 43–​4)

She suggests that realists like Boyd mistakenly take the bait here, trying to rise
to the challenge of defending our scientific beliefs in this higher, philosophical
court of inquiry where ordinary scientific evidence is disallowed, and thereby
being pushed “away from the details of the local debate over atoms and towards
global debates over such questions as whether or not the theoretical terms of ma-
ture scientific theories typically refer” (2001 46).
By contrast, Maddy’s “Second Philosopher” (2007) simply declines the invi-
tation to leave ordinary scientific evidence and methods of evaluation behind
and ascend to any such extrascientific or philosophical court of evaluation for
scientific beliefs. Indeed, she feels no temptation to follow realists and antirealists
down this shared rabbit hole unless and until someone can convincingly explain
what this further distinctively epistemological or philosophical inquiry is sup-
posed to achieve or accomplish and why the ordinary scientific evidence that
actually convinces her of the reality of atoms should be treated as inadequate or
irrelevant to that inquiry. She insists instead that scientific inquiry itself already
represents our best efforts to decide (for any purposes we do or should actually
care about) whether or not particular scientific claims are true and/​or how much
confidence in any particular claim is warranted, and that such questions can only
be convincingly answered in the case-​by-​case or piecemeal manner that science
itself employs:

where the constructive empiricist issues a blanket rejection of all unobserv-

able posits, the Realist issues an equally blanket endorsement; the Second
Philosopher faults both for passing over the details of the evidence for each par-
ticular posit, for shirking the responsibility to evaluate each case individually.
(2007 310n)

Although both Fine and Maddy deny that there are any useful answers to
questions about science at the level of generality at which philosophers have
traditionally sought them, they are both careful to leave room for at least
the bare possibility that we might secure claims of somewhat greater gen-
erality (regarding confirmation, explanation, and the like) “bottom up,” as
it were, by generalizing over and abstracting away from the details of par-
ticular investigations (Fine 1986a, 179–​80; Maddy 2007 403n). But at least
some particularists are explicitly skeptical regarding this possibility. Magnus
and Callender, for example, have argued influentially that “profitable” re-
alism debates must be conducted at the “retail” level of individual, particular
234 Contemporary Scientific Realism

scientific inquiries rather than “wholesale.” And although they concede that
it might be “logically possible” to abstract away from or generalize over the
results of such retail investigations, they doubt that the resulting guidance
will be “either interesting or useful,” suggesting instead that “[w]‌e should pay
attention to particular cases for their own sake and not as proxies for some-
thing else” and that “the great hope for realism and anti-​realism lies in re-
tail arguments that attend to the details of particular cases” (2004 336). And
Maddy herself has argued that when any such general epistemic guidance we
embrace conflicts with the more detailed, informed, and contextual judgments
of confirmation we make regarding particular cases, we typically prioritize
the normative force of the particular judgments over that of the more general
guidance. This is in part because such general guidance must itself remain suf-
ficiently vague, indeterminate, and open-​ended to permit further specification
or adjustment in response to new developments: even if we generalized and
abstracted our way bottom-​up from historical or other evidence to the con-
clusion that we should only believe in entities we can “detect,” for instance, she
suggests that we will subsequently adjust the boundaries of what we count as
“detection” (thereby including or excluding some new method) so as to pre-
serve the normative force of our reflective confirmational judgments about in-
dividual scientific cases (including those making use of the new method). In
this way, as Maddy says it, “the ordinary science, not the [general] criterion, is
doing the work” (2007 402).
Moreover, even many of those who do still seek “interesting and useful” ge-
neral epistemic guidance concerning our scientific beliefs have in recent decades
done so in ways that reflect growing particularist sympathies, suggesting that
such guidance may well consist in the accumulation and synthesis of principles
each of which applies to only a limited range of cases rather than the sort of fully
general criterion of epistemic security sought by their classical predecessors.
Richard Miller (1987), for example, argues that convincing realist commitments
must be grounded in “topic-​specific truisms” whose force is restricted to partic-
ular contexts or domains of inquiry, while Jamin Asay (2019) suggests that re-
alism debates should be conducted at the level of particular sciences rather than
science as a whole. More generally, Larry Sklar argues that “by far the most inter-
esting issues will be found in the detail of how the global, skeptical claims . . . be-
come particularized and concrete when they appear as specific problems about
theories within the context of ongoing physical theorizing” (2000 78). And more
recently, Juha Saatsi has advocated a similarly modest particularism, arguing
first that it is “a manifestation of philosophical arrogance to think that as a re-
alist philosopher one commits oneself to providing a global recipe—​largely inde-
pendently of the science steeped in relevant details—​for revealing what aspects
 Realism, Instrumentalism, Particularism 235

of theory-​world correspondence makes any given theory in mature science tick,”

but explicitly contending nonetheless that we should seek to generalize from the
way particular exemplars latch onto the world to other cases “relevantly similar
to those exemplars” (2017 3240–​1; see also Saatsi and Vickers 2011).
Those who harbor modest particularist sympathies of this sort will presumably
welcome my earlier suggestion that Middle Path realists and instrumentalists
alike remain free to recognize exceptions to the general expectations they de-
fend concerning the fates of generically successful scientific theories. Like such
modest particularists, Middle Path realists and instrumentalists are prepared to
abandon their classical predecessors’ aspirations to fully general criteria of epi-
stemic security for our scientific beliefs (what Saatsi calls a “global recipe”), but
they regard the radical particularist’s contrary conviction that there is little or
no useful epistemic guidance to be found at any level of generality or abstrac-
tion higher than that of the individual case (what Asay calls “hyperlocalism”)
as a counsel of despair. After all, “to look and see as openly as one can what it
is reasonable to believe in and then to go with the belief and commitment that
emerges” (as Fine and NOA recommend) is presumably what scientists them-
selves have been doing all along, and many of their resulting sincere and care-
fully considered judgments have been among those ultimately overturned and
abandoned in the course of further inquiry. Moreover, scientists’ own explicit
judgments of conclusive confirmation for particular scientific theories, par-
ticular parts or aspects of those theories, and particular scientific claims have
repeatedly turned out to be spectacularly mistaken (Stanford 2006 Ch. 7). Our
hope is to make more reliable judgments of this sort, and Middle Path realists
and instrumentalists alike think that we can and do find considerable useful ep-
istemic guidance informing those judgments at levels of abstraction and gener-
ality higher than that of the individual scientific investigation.
My own earlier examination of the case of organic fossil origins (2010), for
example, suggested a tentative general moral: the greater the extent to which
the supporting evidence for a given scientific theory or belief is eliminative or
abductive in character, the greater its vulnerability to the problem of unconceived
alternatives and the more cautious we should be about endorsing its truth. The
historical record might also seem to support the view that highly successful sci-
entific theories are more likely to be subsequently overturned in favor of pre-
viously unconceived alternatives when they concern questions of fundamental
ontology in domains of nature far removed from ordinary human experience
(like particle physics and cosmology as opposed to ecology or geology). And ear-
lier we noted the detailed case made by Lyons and by Vickers for the claim that
the subsequently abandoned components of successful past theories have regu-
larly played crucial and ineliminable roles in generating the empirical successes
236 Contemporary Scientific Realism

(including the novel predictive successes) of those theories. None of these claims
amounts to anything like the fully general criteria of epistemic security sought
by classical realists and antirealists, but each is nonetheless a clear example of
epistemic guidance intermediate in generality between those classical ambitions
and the radical particularist’s competing conviction that useful epistemic guid-
ance for science can only be found in the details of the evidence we have in sup-
port of particular scientific claims. Middle Path realists and instrumentalists
are thus once again united, this time charting a course between the fully general
commitments and expectations of their classical predecessors on the one hand
and the sharply opposed expectations of radical or extreme forms of particu-
larism on the other.

10.4 Conclusion

I’ve suggested here that the modern scientific realism debate is most cen-
trally concerned with whether or not we have yet discovered some particular
form(s) of empirical success or evidential support allowing us to reliably pre-
dict whether particular theories (or privileged parts thereof) will be retained
and ratified throughout the course of further scientific inquiry. As I’ve tried
to emphasize, realists and instrumentalists fundamentally disagree on the an-
swer to this question (and thus on the predictability of further changes in our
theoretical conception of the world around us) in ways that actually make
a difference to how we should pursue our further scientific investigation of
the world. But I have also argued that in recent decades many contempo-
rary realists and instrumentalists have come to share a set of fundamental
commitments that unite them more closely with one another than either is to
their classical predecessors, including Uniformitarianism, commensurability,
and the Maddy/​Wilson principle, as well as to seeking useful guidance con-
cerning the epistemic security of our scientific theories and beliefs at a level
of abstraction and generality intermediate between that anticipated by radical
or extreme forms of particularism and the perfectly general criteria of epi-
stemic security envisioned by classical forms of realism and instrumentalism
alike. For such Middle Path realists and instrumentalists, arguably more im-
portant than any remaining point of debate or disagreement between them
is the shared epistemic project in which they are jointly engaged: realists and
instrumentalists alike on the Middle Path are already actively seeking to iden-
tify, evaluate, and refine candidate indicators of epistemic security for our sci-
entific beliefs, and both see the historical record of scientific inquiry itself as
the most important source of evidence we have available to us for pursuing
that joint project together.
 Realism, Instrumentalism, Particularism 237

References
Asay, J. (2019) Going local: A defense of methodological localism about scientific realism.
Synthese 196: 587–​609.
Bement, A. L., Jr. (2007) Important notice 130: Transformative research, National Science
Foundation, Office of the Director. https://​www.nsf.gov/​pubs/​issuances/​in130.pdf.
Boyd, R. (1984) The current status of scientific realism, in Jarrett Leplin (ed.) Scientific
Realism. Berkeley: University of California Press, 41–​82.
Chakravartty, A. (2007) A Metaphysics for Scientific Realism: Knowing the Unobservable.
Cambridge: Cambridge University Press.
Egg, M. (2016) Expanding our grasp: Causal knowledge and the problem of unconceived
alternatives. British Journal for the Philosophy of Science 67: 115–​141.
Fine, A. (1986a) The Shaky Game: Einstein, Realism, and the Quantum Theory.
Chicago: University of Chicago Press.
Fine, A. (1986b) Unnatural attitudes: Realist and instrumentalist attachments to science.
Mind 95: 149–​179.
Fischer, R. (1930) The Genetical Theory of Natural Selection. Oxford: Clarendon Press.
Kitcher, P.S. (1993) The Advancement of Science. New York: Oxford University Press.
Klee, R. (1999) Scientific Inquiry: Readings in the Philosophy of Science. New York: Oxford
University Press.
Laudan, L. (1981) A confutation of convergent realism. Philosophy of Science 48: 19–​48.
Lyons, T. (2002) Scientific realism and pessimistic meta-​modus tollens, in S. Clarke and
T.D. Lyons (eds.) Recent Themes in the Philosophy of Science: Scientific Realism and
Commonsense. Dordrecht: Springer, 63–​90.
Lyons, T. (2006) Scientific realism and the stratagema de divide et impera. British Journal
for the Philosophy of Science 57: 537–​560.
Lyons, T. (2016) Structural realism versus deployment realism: A comparative evaluation.
Studies in History and Philosophy of Science Part A 59: 95–​105.
Lyons, T. (2017) Epistemic selectivity, historical threats, and the non-​epistemic tenets of
scientific realism. Synthese 194: 3203–​3219.
Maddy, P. (2001) Naturalism: Friends and foes. Philosophical Perspectives 15: 37–​67.
Maddy, P. (2007) Second Philosophy: A Naturalistic Method. Oxford: Oxford UP.
Magnus, P.D., and Callender, C. (2004) Realist ennui and the base rate fallacy. Philosophy
of Science 71: 320–​338.
Miller, R. (1987) Fact and Method: Explanation, Confirmation, and Reality in the Natural
and the Social Sciences. Princeton: Princeton University Press.
Novick, A., and Scholl, R. (2020) Presume it not: True causes in the search for the basis of
heredity. British Journal for the Philosophy of Science 71: 59–​86.
Peters, D. (2014) What elements of successful scientific theories are the correct targets for
“selective” scientific realism? Philosophy of Science 81: 377–​397.
Putnam, H. (1975) Mathematics, Matter, and Method (Philosophical Papers Vol. 1).
London: Cambridge University Press.
Psillos, S. (1999) Scientific Realism: How Science Tracks Truth. London: Routledge.
Saatsi, J.T. (2017) Replacing recipe realism. Synthese 194: 3233–​3244.
Saatsi, J.T., and Vickers, P. (2011) Miraculous success? Inconsistency and untruth in
Kirchhoff ’s diffraction theory. British Journal for the Philosophy of Science 62: 29–​46.
Sklar, L. (2002) Theory and Truth: Philosophical Critique within Foundational Science.
Oxford: Clarendon Press.
238 Contemporary Scientific Realism

Smart, J.J.C. (1968) Between Science and Philosophy. New York: Random House.
Stanford, P.K. (2006) Exceeding Our Grasp: Science, History, and the Problem of
Unconceived Alternatives. New York: Oxford University Press.
Stanford, P.K. (2010) Getting real: The hypothesis of organic fossil origins. Modern
Schoolman 87: 219–​243.
Stanford, P.K. (2015) Catastrophism, Uniformitarianism, and a Scientific Realism Debate
That Makes a Difference. Philosophy of Science 82: 867–​878. doi:10.1086/​683325
Van Fraassen, B.C. (1980) The Scientific Image. Oxford: Clarendon Press.
Vickers, P. (2013) A confrontation of convergent realism. Philosophy of Science
80: 189–​211.
Vickers, P. (2017) Understanding the selective realist defence against the PMI. Synthese
194: 3221–​3232.
Wilson, M. (2006) Wandering Significance: An Essay on Conceptual Behaviour.
Oxford: Clarendon Press.
Worrall, J. (1989) Structural realism: The best of both worlds? Dialectica 43: 99–​124.
 11
Structure not Selection
James Ladyman

Department of Philosophy
University of Bristol
[email protected]

11.1 Introduction

Structural realism is the best of both worlds, according to John Worrall (1989),
because it takes account of the most powerful argument against scientific re-
alism, as well as the motivations for it. The problem for scientific realism that
he takes most seriously is that posed by the actual historical record of scientific
theories in physics and chemistry, because it shows, all philosophical argument
aside, that not everything that is supposed by the highly empirically successful
theories of the past is real by the lights of current science. He was not troubled
by other arguments against scientific realism (such as the argument from the un-
derdetermination of theory by evidence), and the problem of theory change is
the only problem for standard scientific realism that Worrall sought to solve by
adopting structural realism. It is the problem that others, notably Stathis Psillos
(1999), seek to solve by what David Papineau dubbed “selective realism” (1996),
which involves analyzing case studies from the history of science to find a for-
mula that restricts the epistemic commitment of scientific realists to parts of
theories that will be retained. The idea is that close examination of the history
of science will reveal what it is about abandoned theoretical constituents that
distinguishes them from those that are retained, so that a selective commitment
to current theories can be then be applied in the confidence that the relevant on-
tology will not subsequently be abandoned. (Psillos’s formula is roughly “only
believe in the reference of the central theoretical terms of theories that play an
essential role in generating novel predictive success.”) While selective realism is
motivated and tested by studying the history of science, it is a form of realism
that involves criteria that can be applied to our best current theories to select in
advance what of their ontology will be preserved. If selective realism is under-
stood as selecting part of science which we can be confident will be retained on

James Ladyman, Structure not Selection In: Contemporary Scientific Realism. Edited by: Timothy D. Lyons and Peter
Vickers, Oxford University Press. © Oxford University Press 2021. DOI: 10.1093/​oso/​9780190946814.003.0011
240 Contemporary Scientific Realism

theory change, it is simple to think of structural realism as the kind of selective

realism that selects structure. Hence, naturally enough structural realism is often
interpreted as a form of selective realism (as indeed it was by Papineau).
It might therefore be surprising that ontic structural realism (OSR) as
I proposed and developed it (including with Don Ross in Ladyman and Ross
2007) is not intended as a form of selective realism in the sense just outlined.
Furthermore, Steven French (whose recent work (2014) defends a somewhat dif-
ferent version of OSR) also makes it clear that he does not regard OSR as a form
of selective realism, and nor does John Worrall understand his original struc-
tural realism that way. For Worrall, structural realism is a general kind of epi-
stemic humility about what we know on the basis of our best scientific theories,
not a criterion of epistemic commitment that can be applied within theories past
or present. While OSR is supposed to deal with the problem of theory change, it
is also motivated by the need to solve problems with the ontology of standard sci-
entific realism as applied to physics. For French, Ladyman, and Ross, OSR is not
an epistemological modification of standard scientific realism wholly, primarily
or even partly (see also, for example, Esfeld and Lam (2008), where a moderate
version of OSR is proposed solely for the metaphysical interpretation of space-
time). OSR is an epistemic thesis to the extent that it incorporates the epistemic
commitment to our best science that all forms of scientific realism involve, but it
is distinctive in proposing a metaphysics along with it.
Of course, authors have no ultimate authority over the readings of their texts,
but, in any case, the assimilation of structural realism to selective realism creates
pseudo-​problems and is otherwise unhelpful. The next section (which forms
the bulk of the paper) explains that structural realism does not respond to the
problem of theory change by offering a way of demarcating in advance the parts
of theories that are likely to be retained in future as a response. Rather struc-
tural realism in both its epistemic and ontic guises solves the problem of theory
change with general departures from standard scientific realism.1 (Epistemic
structural realism is also associated with Russell’s structuralism and epistemo-
logical and logical issues not addressed here.)
There are several ontological problems for standard scientific realism that do
not figure in the canonical debate between realists and antirealists, which was
the context for the development of both Worrall’s structural realism and the cur-
rent literature on selective realism, and which are not discussed in early work
on OSR. The form of OSR developed by Ladyman and Ross (2007) addresses
them (as does French 2014). Hence, there are two reasons why their OSR is not a

1 A referee claims that epistemic structural realism involves looking at our current best theories

and trying to identify the structure to which we then make a realist commitment. However, nobody
has ever actually done this with our best science in the attempt to tell what will be retained on theory
change.
 Structure not Selection 241

kind of selective realism. The first is that it does not offer a criterion for selective
realist commitment to current theories (as argued in the next section), and the
second is that it addresses ontological problems other than the problem of theory
change.
Section 11.3 argues that the construction of a positive realist metaphysics that
can deal with the problem of theory change, must also take account of ontolog-
ical issues, most importantly the lack of a fundamental level and scale-​relativity
(and explains how these problems are related to the problem of theory change).
Section 11.4 reviews the form of OSR involving the theory of real patterns de-
veloped by Ladyman and Ross, and shows how it addresses the ontological
problems for scientific realism, emphasizing the centrality of the idea of objec-
tive modal structure (where this is also retained on theory change even when
ontology changes radically). It is the latter that makes OSR a form of realism, and
differentiates it from van Fraassen’s structural empiricism, as well as allowing
OSR to avoid collapsing the distinction between abstract and concrete structure
(van Fraassen’s problem of “pure” structuralism, 2006), and making an account
of causation and other modal aspects of scientific knowledge possible. The paper
concludes with some remarks about the relationship between structural realism
and the historiography of science.

11.2 Structural Realism is not Selective Realism

The problem of theory change is an empirical challenge to scientific realism

pressed in Larry Laudan’s work (1977 126, 1984), following Henri Poincaré
(1905/​1952 160), Ernst Mach (1911 17), and Hilary Putnam (1978 25). In its
most basic form the problem is that the history of science is littered with laws,
propositions, and theories that are now regarded as only approximately true and/​
or true in a restricted domain, or outright false theories. This provides a reason
for skepticism about the first-​order methods of science that recommend belief
in entities such as black holes and electrons. The problem of theory change is
not addressed much in the debate about scientific realism centered on the work
of Bas van Fraassen. In that context the big issue is the underdetermination of
theory by evidence. Realists argue that beliefs about observables that have not
been observed and inductive generalizations in everyday life are in general just
as underdetermined as beliefs about unobservables. The fundamental argument
of many scientific realists is that the observable/​unobservable distinction is of no
philosophical significance (see Churchland and Hooker 1985). In particular, it is
of no epistemic significance and does not demarcate the knowable from the un-
knowable, and it is of no ontological significance in the sense that it has nothing
to do with what things, or kinds of things, exist. Hence, realists argue that the
242 Contemporary Scientific Realism

fallible methods, in particular inference to the best explanation, used in everyday

life to arrive at beliefs about the unobserved are just as legitimate when extended
and refined to arrive at beliefs involving unobservables in science. On this way
of thinking, skepticism about unobservables based on underdetermination is
analogous to skepticism about the external world or other minds (and may be
disregarded by the scientifically minded philosopher as “merely philosophical”
(Worrall 1989)). On the other hand, the problem of theory change is based on
what we know about the actual history of science.2
The simplest form of argument from theory change against scientific realism
is the Pessimistic Meta-​Induction that can be rendered as follows:

(i) There have been many empirically successful theories in the history of
science that have subsequently been rejected, and whose theoretical
terms do not refer according to our best current theories.
(ii) Our best current theories are no different in kind from those discarded
theories, and so we have no reason to think they will not ultimately be
replaced as well.

So, by induction we have positive reason to expect that our best current theories
will be replaced by new theories according to which some of the central theoret-
ical terms of our best current theories do not refer, and hence, we should not be-
lieve in the approximate truth or the successful reference of the theoretical terms
of our best current theories.
Both (i) and (ii) are dubious. The number of theories that can be listed in sup-
port of (i) is drastically reduced if the notion of empirical success is restricted
to making successful novel predictions (Psillos 1999). (ii) is undermined by the
unprecedented degree of quantitative accuracy and practical and predictive
success of current science. Furthermore, it may be argued that only since the
twentieth century have the physical sciences been fully integrated because of two
developments, namely the atomic valence theory of the periodic table and elec-
tronic theory of the chemical bond. Science is now very mature in the sense that
testing hypotheses at the cutting edge in one field relies upon background the-
ories from the basics of other sciences that are fully integrated with each other.
The common system of quantities and units, and many of the core background
theories used across all the sciences have been stable for a long time. There is no
precedent in history for the current extent of overlap and reinforcement among

2 Of course, they can be related by asking whether knowledge of the observable phenomena is

the limit of what is retained on theory change. Obviously observable phenomena are lost on theory
change, in so far as new theories are more empirically adequate than their predecessors.
 Structure not Selection 243

theories in the natural sciences, nor for the commonality of statistical and other
methods.
However, there is another form of the problem of theory change that does not
have the form of an induction based on Laudan’s much-​discussed list. Rather it
is an argument against the main argument for scientific realism namely the no
miracles argument. Laudan’s paper was also intended to show that the successful
reference of its theoretical terms is not a necessary condition for the novel pre-
dictive success of a theory (1981 45), and so there are counterexamples to the
no miracles argument. No attempt at producing a large inductive base need be
made; rather, one or two cases are argued to be counter-​arguments to the re-
alist thesis that novel predictive success can only be explained by successful ref-
erence of key theoretical terms. As Psillos (1999 108) concedes, even if there are
only a couple of examples of false and non-​referring, but mature and strongly
successful theories, then the “explanatory connection between empirical success
and truth-​likeness is still undermined.” Hence, Laudan’s ultimate argument from
theory change against scientific realism is not really an induction of any kind, but
a reductio.

The Argument from Theory Change:

(I) Successful reference of its central theoretical terms is a necessary condi-
tion for the approximate truth of a theory.
(II) There are examples of theories that were mature and had novel predictive
success but whose central theoretical terms do not refer.

So there are examples of theories that were mature and had novel predictive suc-
cess but which are not approximately true.
Approximate truth and successful reference of central theoretical terms is not
a necessary condition for the novel-​predictive success of scientific theories. So,
the no miracles argument is undermined since, if approximate truth and suc-
cessful reference are not available to be part of the explanation of some theories’
novel predictive success, there is no reason to think that the novel predictive suc-
cess of other theories has to be explained by realism (Ladyman and Ross 2007
84–​85). Recall that the realist claims that novel predictive success is only intelli-
gible on the realist view so once it is established that it is possible where realism is
untenable then the realist’s argument is much less if at all compelling (assuming
it was to some extent compelling in the first place).
The ether theory of light and the caloric theory of heat seem to have enjoyed
novel predictive success by anyone’s standards.

If their central theoretical terms do not refer, the realist’s claim that approx-
imate truth explains empirical success will no longer be enough to establish
244 Contemporary Scientific Realism

realism, because we will need some other explanation for success of the caloric
and ether theories. If this will do for these theories then it ought to do for others
where we happened to have retained the central theoretical terms, and then
we do not need the realist’s preferred explanation that such theories are true
and successfully refer to unobservable entities. (Ladyman and Ross 2007 84, see
also Ladyman 2002)

Stathis Psillos (1999) first argues that these are the only two truly problematic
cases for the scientific realist. He then adopts his “divide and conquer strategy,”
which is to adopt a different solution for each though both are ways of denying
(II). In the case of the ether, he argues that according to his causal-​descriptivist
theory of reference, the term “ether” refers, while in the case of caloric he argues
that there was never any reason why a realist of the time had to commit to it
being a material substance and that it was not after all a “central” theoretical term
in the sense that demands successful reference. Hence, he argues that we can be
confident that all the central theoretical terms of our current science do in fact
refer to things in the world.
Psillos argues that Fresnel’s use of the term ether referred to the electromag-
netic field and that continuity of reference being assured the problem posed
by that case is solved. This is contentious but anyway other cases are also prob-
lematic. For example, consider the relationship between classical and quantum
physics. Classical physics describes phenomena in diverse domains to very high
accuracy and made predictions of qualitatively new phenomena. It had a vast
amount of positive evidence in its favor. The rationalists and Kant had in var-
ious ways attempted to derive its principles a priori. However, it turned out to
be completely wrong for very fast relative velocities and for very small particles,
as was shown by relativistic and quantum physics respectively. This is surely the
problem of theory change in its most stark form, yet the problem is nothing to
do with the abandonment of theoretical terms, or difficulty in securing refer-
ence.3 Clearly the reference of the term “energy” at least overlaps in classical and
quantum theories because quantities in each coincide in the limit as the number
of particles increases. This does not solve the problem that the radical ontological
and metaphysical differences between the theories pose for the realist. Different
theories often use the same terms, to say very different things about what the
world is like. The question of which terms refer does not settle whether it is

3 A referee objects that this may be thought to be a less serious form of theory change because of

the precise mathematical relationships of approximation between the laws of the respective theories
but these can only be interpreted in terms of approximate truth if the ontological discontinuities in
question are not taken to be significant, while the laws are taken to have more than the purely empir-
ical content van Fraassen and other antirealists take them to have. This is just what ontic structural
realism recommends.
 Structure not Selection 245

plausible to say that the metaphysics and ontology of the theory is even approxi-
mately true, because in many cases the same theoretical term is still used despite
radical changes.
For this reason, Worrall (1989) argues that the project of defending realism
by securing reference for abandoned theoretical terms is missing the point of the
problem as well as the key to its solution.4 Arguably “field” is just the new word
for what the ancients and early moderns termed “aether” or “ether.” Fields like
the ether permeate all of space and are posited to explain certain phenomena.
Anaxagoras is said to have introduced the idea of the ether to propel the planets,
and subsequently there were many ethers to account for various things, notably
the propagation of electricity and light through empty space. Fresnel’s optical
ether was supposed to be a solid and Maxwell’s field behaves very differently and
is said to be immaterial. However, from our current perspective the ether has
much more in common with the field than atoms do with the indivisible particles
of antiquity (Stein 1989).
Structural realism is not motivated by the need to secure reference across
theory change but by Heinz Post’s (1971) General Correspondence Principle, ac-
cording to which the well-​confirmed laws of old theories are retained by their
successors as approximations within certain domains. Even comparing the well-​
confirmed laws of theories that differ wildly in their metaphysical dimensions
such as quantum mechanics and classical mechanics, or general relativity and
Newtonian gravitation, the past theories are limiting cases of the successor the-
ories. This confirms Poincaré’s idea of physics as ruins built on ruins in the sense
that the old theories form the foundations of the new ones rather than the ground
being cleared before work begins. The key point is that more than the empirical
content and phenomenological laws of past theories is retained. For example, it
is not just Kepler’s laws and Galileo’s kinematical laws that are preserved as lim-
iting cases of general relativity, but Newton’s inverse square force law that unifies
and corrects those laws is retained in the form of the Poisson equation as a low-​
energy limit of Einstein’s field equation. Similarly, the classical limit of quantum
physics gives rise to the approximate truth of the laws of Newtonian mechanics
for macroscopic bodies.5
The laws take mathematical form, and there are special cases, such as that of
Fresnel’s equations, where the very same equations are reinterpreted in terms
of different entities. However, of course this is not the norm, and in many
cases mathematical structure is lost and radically modified on theory change.

4 Whether or not Worrall is right there is less much emphasis on reference in the subsequent liter-

ature on scientific realism.

5 The structure of the world described by science and preserved on theory change can be taken as

just the occurrent regularities between actual events. This is structural empiricism as defended by
van Fraassen.
246 Contemporary Scientific Realism

Structural realism does not require that all mathematical (or any other kind of)
structure is preserved on theory change. If it did it would be refuted by the fact
that the mathematical form of the physics of Maxwell is different from that of
the physics of Einstein. As French and Ladyman (2011) put it, “[t]‌he advocate of
OSR is not claiming that the structure of our current theories will be preserved
simpliciter but rather that the well-​confirmed relations between the phenomena
will be preserved in at least approximate form and that the modal structure of the
theories that underlies them, and plays the appropriate explanatory role, will also
be preserved in approximate form” (32–​33). However, this does not enable any
kind of selective realism because it says nothing at all about how the laws of the
theory are only approximate, and so does not enable the selection of what will be
retained in advance. French and Ladyman (2003) emphasized that they under-
stood OSR as the view that the mathematical structure of theories represents,
to some extent, the modal structure of the world. Hence, when the mathemat-
ical structure changes this often reflects the fact that some of how the world’s
modal structure was represented to be was wrong. For example, the modal struc-
ture of the world is different according to Special Relativity than it is according
to Newtonian mechanics because, for example, the velocity addition law of the
latter is not even approximately true in case of frames moving relative to each
other at velocities close to that of light, and this is reflected in the mathematical
change to the Lorentz transformations from the Galilean transformations.
The point of OSR in the context of theory change is that it inflates the ontolog-
ical importance of relational structure to take account of the fact that the ontolog-
ical status of the relevant entities may be very different in the different theories,
yet the relationship between the way they represent the modal structure of the
world can nonetheless be studied in depth by investigating the relevant mathe-
matical structures (even in the case of geocentrism and heliocentrism, as shown
by Saunders 1993). Ladyman (1997) argues that mathematical representation is
ineliminable in much of science and takes this to be key to OSR (and work by
French, Ross, and Wallace on OSR is also predicated on it). However, this does
not mean that this kind of structural realism only applies to mathematicised the-
ories, as shown by the following case.
Kuhn and Feyerabend popularized the view that theory change in the history
of science disrupts narratives of continuous progress and realism. The response
of many philosophers of science to this challenge, particularly in Popper’s school,
was to examine episodes in the history of science in detail to see if the histori-
cally inspired critiques were really supported by the evidence. Noretta Koertege
(1968) considered phlogiston theory as an example supporting the general cor-
respondence principle because the well-​confirmed empirical regularities stated
in terms of phlogiston theory (such as that air saturated with phlogiston by com-
bustion does not support respiration) are true when translated into the language
 Structure not Selection 247

of oxygen (de-​oxygenated air does not support respiration). George Gale (1968)
argued that phlogiston theory was also a good explanatory theory that explained
the loss of weight of wood, coal, and ordinary substances when burnt, among
many other things. Furthermore, and crucially for structural realism, the fact
that combustion, respiration, and calcination of metals are all the same kind of
reaction and there is an inverse kind of reaction too is an example of a theoretical
relation that is retained from phlogiston theory. Even though there is no such
thing as phlogiston, the tables of affinity and antipathy of phlogistic chemistry
express real patterns (of which more in the next section) that we now express in
terms of reducing and oxidizing power.
Indeed as also discussed by Gerhard Schurz (2009), the terms “phlogistication”
and “dephlogistication,” meaning the assimilation and release of phlogiston re-
spectively when they were used in chemistry, can now be regarded as referring
to the processes of oxidation and reduction (where oxidation of X = the forma-
tion of an ionic bond with an electronegative substance and reduction of X is the
regaining of electrons). If the oxidizing agent is oxygen, and the oxidized com-
pound is a source of carbon then the product is carbon dioxide (“fixed air”), and
if the oxidizing agent is an acid, then hydrogen (“inflammable air”) is emitted, in
keeping with phlogiston theory. One could go further and allow that “phlogiston
rich” and “phlogiston deficient” refer too, namely to strongly electro-​negative
and electro-​positive molecules respectively. One could even argue that “phlo-
giston” refers to electrons in the outer orbital of an atom (as suggested by Andrew
Pyle, see Ladyman 2011). However, none of this changes the fact that phlogiston
theory is wrong and that there is nothing contained in all flammable materials
that leaves them on combustion, and nothing that metals lose when they become
calces. No form of selective realism could have told us in advance that phlogiston
would be abandoned, because it was essential to much empirical success.
The examples from mathematical physics cited earlier and the case of phlo-
giston theory show that even though the ontology of science may change quite
radically at the level of objects and properties (for example, there is no elastic
solid ether, no principle of combustion), there can be continuity of the modal
structure that is attributed to the world. OSR is not selective realism; rather, it is
the generic modification of standard scientific realism as exemplified by Psillos
(1999). The point is that “[t]‌heories, like Newtonian mechanics, can be literally
false as fundamental physics, but still capture important modal structure and re-
lations” (Ladyman and Ross 2007 118). Likewise important features of the modal
structure of the world can be represented in terms of phlogiston, as explained
earlier. When Worrall made his dialectical move in the realism-​anti-​realism
debate, philosophers of science had given a lot of attention to the problem of
the reference of theoretical terms, because it had been realized that they could
be discarded from even very successful science. In the light of this discussion
248 Contemporary Scientific Realism

it should be clear that the real problem is not the reference of theoretical terms
but that the notion of approximate truth is hard to apply to the ontology and
metaphysics of scientific theories. Atoms are simply not indivisible in meta-
physical terms, though they are of course in practical terms without a great deal
of very sophisticated engineering. OSR does not give an account of metaphys-
ical approximation, but it begins from the insight that more than the empirical
content of theories is preserved. The claim that modal structure is preserved is
compatible with radical ontological discontinuity, even in the extreme case of
phlogiston, which Psillos did not seek to redeem.
As mentioned in section 11.1, OSR was also introduced to address ontological
problems other than the problem of theory change. The next section outlines
several that are not discussed in the literature about selective realism. Among the
most important are the issues of fundamentality and scale-​relativity.

11.3 Ontological Problems for Scientific Realism

The problems explained next are only specifically ontological problems for scien-
tific realism, because scientific realism involves ontological commitment on the
basis of science, but they are problems more generally for any philosophy of sci-
ence. There are a number of ontological problems for scientific realism that are
not discussed in the context of selective realism or epistemic structural realism.
Like the problem of theory change, these problems with standard scientific re-
alism arise from how science actually is.6

(a) Scientific realism often incorporates metaphysical ideas of fundamen-

tality and reduction and an epistemologically and methodologically
impoverished view of scientific practice.7 Nancy Cartwright (1983)
criticizes ideas of fundamentality and reduction by showing how scien-
tific practice involves approximation and idealization in ways that do not
fit with the standard scientific realist picture of how theoretical laws are
related to empirical content.
(b) Much, most, or even all of science is not about what is ontologically
fundamental. There has always been a tension within scientific realism

6 As mentioned earlier OSR is also supposed to take account of the problems with the standard

scientific realist interpretation of physics, in particular, to take account of the irreducibility of rela-
tional structure and the problems of identity and individuality in quantum mechanics and space-
time physics. These issues are not discussed in the present paper but are central for Ladyman (1998),
French and Ladyman (2003), Ladyman and Ross (2007), and French (2014). For a recent discussion
see Ladyman (2015, 2016).
7 Of course, many scientific realists are not fundamentalists and reject the idea that ontology

should be based on fundamental physics, including Cartwright herself.

 Structure not Selection 249

because: on the one hand it is defended as the extension to science of

common sense forms of reasoning that are used to arrive at ontolog-
ical commitments (such as inference to the best explanation); and on
the other hand, many take scientific realism to motivate or even require
eliminativism about common sense ontology. More generally, there is the
problem of how the ontologies of the sciences relate, and the question
of whether there is a fundamental level, and if so whether anything else
exists. This is equivalent, for most philosophers, to the question as to how
the special sciences relate to fundamental physics.
(c) The scale-​relativity of ontology is the thesis defended by Ladyman and
Ross (2007) according to which entities may exist at some energy, length,
or time scales and not at others. The scientific realist who does not em-
brace the scale relativity of ontology must otherwise accommodate the
fact that within science in general there are entities that are recognized at
one scale of description but not at others. In physics this is formalized by
the mathematical relationships among the “effective field” theories that
form the Standard Model of particle physics. Entities such as mesons and
hadrons are not part of the high-​energy description of quantum chromo-
dynamics, but they can be described by a lower-​energy description, and
the way that energy cut-​offs can be used to relate such theories at different
scales is relatively well-​understood (see Ladyman 2015; Wallace 2011;
Franklin forthcoming). Beyond physics, entities are similarly bound to
scale. For example, in economics prices only exist at long time scales rel-
ative to individual exchanges of goods and services, and in geography a
torrent of water is not a river if it does not persist.8

These problems are related. (a) suggests that a straightforward realist interpre-
tation of scientific theories and models is naïve. For example, the equator, the
orbitals of atoms, and temperature of a gas are all abstractions and idealizations
to a degree. (a) relates to (b) because something discrete may be represented
at a less fundamental level as something continuous (as with ordinary fluids),
and something heterogeneous at one scale may be idealized as something ho-
mogeneous at another, as with any polyatomic molecular substance. (b) relates
to (c) because the special sciences are often associated with restricted scales of
energy, length, and time. These problems must be faced by any scientific re-
alist. These problems are also related to the problem of theory change, because
past theories are also often those that we now regard as working at particular
scales and as describing the emergent entities that are now described by a part of

8 Entities like rivers and waves are arguably necessarily extended in time but nothing here depends

on that claim.
250 Contemporary Scientific Realism

physics that has the status of a special science compared to fundamental physics.
For example, atoms are now part of atomic physics, which is not fundamental
and describes reality at a restricted scale. The ontology of such theories remains
“effective,” in the sense that the entities it posits are part of empirically successful
descriptions and models.
Defenders of scientific realism such as Psillos (1999) and Anjan Chakravartty
(2017) are primarily concerned with epistemological issues rather than the
problems just outlined. However, in other contexts scientific realism is argued
to lead to the elimination of all but the most fundamental physical entities (see,
for example, Rosenberg 2011). In this way, scientific realism becomes anti-​
naturalistic and cannot be claimed to take the ontological commitments of
science at face value. On the other hand, those who reject eliminativism or re-
ductionism need a version of realism that takes account of (a)–​(c). The next sec-
tion reviews how the real patterns account of ontology incorporated into OSR by
Ladyman and Ross makes for a unified solution to the ontological problems for
scientific realism.

11.4 A Realistic Metaphysics

For French and Ladyman (2003) and in their later publications jointly and sep-
arately, OSR is to be understood as the claim that science represents the objec-
tive modal structure of the world. Ladyman and Ross (2007) and French (2014)
argue that incorporation of the idea that the world has a modal structure that is
described by science into OSR makes it a realist philosophy of science, and dis-
tinct from the structural empiricism of van Fraassen (2006). They also agree that
OSR so construed can accommodate the modal aspects of scientific knowledge
including causation, law, and symmetry.9 Modal structure is to be understood
roughly as an abstraction of causal and nomological structure. For example,
it is part of the modal structure of the world that metals expand when heated,
that mass is conserved in chemical reactions, and that heat cannot be con-
verted into work with perfect efficiency. These are all modal claims that support
counterfactuals and explanations, and they have all survived very different ideas
about the ontology in terms of which they are stated.
However, the forms of OSR that they defend also differ significantly espe-
cially in respect of how they relate to problems (a)–​(c) of the last section. French
(2014) is an eliminativist about all ontology except the structures of physics, and

9 Berenstain and Ladyman (2012) also argue that the arguments for scientific realism are undercut

without it. Note that this makes OSR an immodest and to some extent speculative philosophical po-
sition, unlike epistemic structural realism which is quietist about metaphysics and very modest about
epistemology.
 Structure not Selection 251

in this way avoids problems (b) and (c) by denying that anything that is not fun-
damental exists. This is a big price to pay for a naturalist because it departs so
radically from taking science at face value.10
Ladyman and Ross (2007) develop a version of OSR involving the theory of
real patterns in order to address (a)–​(c) and develop a position that takes the
special sciences to give us irreducible knowledge of the modal structure of the
world.11 The commitment to objective modal structure is essential to making
the resulting naturalized metaphysics a form of realism, since the theory of real
patterns can also be developed as a form of instrumentalism. We take it that
all our scientific knowledge is about effective ontology in the sense of the pre-
vious section, and accordingly seek a unified way of thinking about the entities
of physics and the special sciences. We propose the idea of real patterns to this
end. Real patterns are the features that are used in everyday life and in science
to describe the world in terms of projectible regularities. As such they are not
confined to any ontological category and may be events, objects, processes, or
properties. For example, Mount Everest is a real pattern that features in a host
of projectible regularities about the world. To be genuine entities they need to
be non-​redundant, so that, for example, the disjunction of real patterns is not a
new real pattern. Redundancy can be captured in different ways, the most ob-
vious being information theory. Real patterns compress information (though
at the cost of loss of precision). Much can be said about Everest that would be
much more costly to state otherwise. Describing the world without talking about
Mount Everest would be possible, but it would be harder. Real patterns simplify
the description of the world relative to some background already established lan-
guage (of more basic real patterns). Alternatively, redundancy can be understood
dynamically. Some coarse-​grained variables feature in a much simpler dynamics
than that of the underlying variables; think of a rolling ball describable by the
position and momentum of its center of mass and its angular velocity, instead of
all the positions and momenta of its parts. However, some such coarse-​grained
variables do not feature in any simpler dynamics and they are redundant and an-
ything defined in terms of them is not a real pattern.12

10 See Ladyman (2019) for discussion of French (2014).

11 David Wallace (2015) also argues for the real patterns account of effective emergent entities
such as quasi-​particles. Ladyman (2017) provides a summary of the main theses of Ladyman and
Ross (2007), some of which concern topics that are not addressed at all here, including the unity of
science and the primacy of physics. Ladyman (2017) argues that the further ontological problems of
vagueness of composition and identity over time, and the problems of generation and corruption,
apply to both special science objects and everyday ones and that the real patterns account of ontology
offers a unified solution for them.
12 Of course, this is a grey area. A cloud is a real pattern to a very limited extent because it is so

ephemeral. A rainbow is a real pattern but only with respect to vision perception. Real patterns can
also include events such as parties or revolutions which are of course notoriously difficult to individ-
uate. The theory of real patterns does not solve those problems of individuation but takes them to be
practical problems in some cases and pseudo-​problems otherwise.
252 Contemporary Scientific Realism

The real patterns account of ontology is compatible with the scale relativity
of ontology, because it requires only that there be some regime in which the real
pattern in question is projectible (and non-​redundant) for it to exist. Different
emergent structures are found at different scales, and accordingly there are real
patterns that exist at some scales and not at others. For example, Mount Everest
does not exist at atomic scale because there are no projectible regularities about
entities at that scale that involve Mount Everest. The real patterns account is com-
patible with taking the ontologies of the special sciences at face value because
insofar as the science in question captures the modal structure of the world, its
ontology involves real patterns just as much as that of physics. Hence, problems
(b) and (c) can be solved without adopting eliminativism about everything ex-
cept fundamental physics.
As Dennett (1991) makes clear in his original discussion, being a real pat-
tern is not an all or nothing matter. Mountains don’t have exact boundaries.
As many philosophers find mystifying, almost all events, objects, and proper-
ties commonly recognized as real involve some if not many kinds of vagueness.
The real patterns account accommodates (a) because it is not based on unreal-
istic ideas of a pristine scientific ontology to accompany exact and fundamental
laws. Representations can be more or less good at capturing the modal struc-
ture of the phenomena, and real patterns are indispensable to such successful
representations, even though approximation and idealization are essential to the
way the representation works.
The real patterns account fits well with the history of theory change in sci-
ence. The well-​confirmed laws of past theories describe aspects of the modal
structure of the world, however, the ontologies in terms of which they are
stated are not apt for describing other aspects of the modal structure of the
world. In so far as past theories are empirically adequate, the entities they in-
volve are real patterns and it may be appropriate to continue to refer to them
in some regimes. For example, the force of gravity, the light waves of Fresnel,
the electric and magnetic fields of classical electrodynamics, and the heat
flows of Carnot are all real patterns and part of an effective ontology in the
respective regime. However, sometimes the real patterns in question are best
not described in terms of the old ontology, as with the case of phlogiston,
because it does not get enough of the modal structure right. However, even
in such cases we can say that there were some real patterns to the old theory,
in particular, the commonality of different forms of dephlogistication is a
real part of the modal structure of the world, as explained earlier. The real
patterns account allows that continuity of reference across theory change can
always be secured to some extent whenever there are real patterns that are
carried over as approximations.
 Structure not Selection 253

11.5 Realism and the Historiography of Science

Despite Kuhn’s subsequent clarifications of his own view of science (1977), his
Structure of Scientific Revolutions (1962) inspired schools of history and sociology
of science that demur explanations of theory change in terms of evidence or exper-
imental results (internalism). Rather, it became standard for explanations of theory
choice in science, in so far as they are offered, to appeal to economic, psychological,
and social (external) factors rather than the tribunal of experiment (a classic of the
genre being Shapin and Shaffer 1985). In the history of science, celebrated recent
studies have emphasized the rationality of the losers and the psychological and so-
cial influences on the victors, sometimes going as far as claiming that abandoned
theories should have been retained contrary to the theory-​choice made by the scien-
tific community (for example, Chang (2012) argues that phlogiston theory should
have been retained despite its many failings; see Blumenthal and Ladyman (2017,
2018). In this and other cases, notably that of quantum mechanics (see Cushing
1995), it is claimed on the basis of the general considerations of underdetermination
that an alternative history could have delivered the same or more scientific success.
The orthodoxy in current historiography is that actors and their social
networks in the history of science should always be represented sympathetically,
and that their perspective be adopted in describing the relevant evidence and
theories. The opening lines of Harvard’s STS website describe the field as an “ap-
proach to historical and social studies of science, in which scientific facts were
seen as products of scientists’ socially conditioned investigations rather than as
objective representations of nature”.13 Realist historiography is often said to be
Whiggish and triumphalist. However, there is no reason why realists must be
Whiggish in the relevant sense and no reason why celebrating the success of sci-
ence is not compatible with recognizing the bias, error, and missteps that are also
part of its history, and the role of external factors where they are relevant.
While Whiggism is much derided it is less often clearly defined (see
Alvargonzález (2013) for a judicious and informative discussion). In political
history it is associated with an implausible teleology. In the history of science
it is similarly associated with the idea that the development of theories is an in-
evitable progression toward the truth. However, the idea that scientific meth-
odology should track the truth is compatible with a degree of contingency.
Furthermore, it is not unreasonable to think that there is a degree of inevitability
to some scientific discoveries, because so many people have come close to the
same ideas around the same time. For example, the inverse square law of gravity
may not have been known to Hooke, but he at least came near to it (Westfall

13 http://​sts.hks.harvard.edu/​about/​whatissts.html
254 Contemporary Scientific Realism

1967). In any case, acknowledging the actuality of scientific progress does not
require believing it is inexorable, so the teleological element of Whiggism is not
required for a realist historiography of science.
Triumphalism is associated with judging past theories by the lights of pre-
sent ones and with dismissing the successes of abandoned theories and ridi-
culing the reasoning of the historical actors who defended them. However,
celebrating the success of theory-​choice and development in particular cases,
and considering the history of science in the light of what we know now
(presentism), does not imply denigrating or ignoring the success of aban-
doned or rival theories or shallow judgment of historical actors. As discussed
earlier, Gale (1968) and Koertege (1969) are very clear about the successes
of phlogiston theory in the course of their explanations of why ultimately it
had to go. It is question-​begging against realism to insist that the history of
science should be explained without reference to what we now know about
the world, and that internalist and presentist explanations of theory-​choice
are ruled out. From the realist perspective, historiography that does not take
account of current scientific knowledge is likely to mislead us about the his-
tory of science and to beg the question against all forms of realism. What is so
compelling about the problem of theory change is that it works from within
realist historiography. Solving the problem requires being realistic about the
history of science and the limitations of scientific progress, while celebrating
its success.

Acknowledgments

Many thanks to Alexander Bird, Geoff Blumenthal, Steven French, Stephan

Lewandowsky, Tim Lyons, Naomi Oreskes, Stathis Psillos, Don Ross, Peter
Vickers, and the participants of the Durham Selective Realism conference.

References
Alvargonzález, D. (2013), “Is the History of Science Essentially Whiggish?” History of
Science 51 (1), pp. 85–​99.
Berenstain, N, and Ladyman, J. (2012), “Ontic Structural Realism and Modality” in
Landry, E. & Rickles, D. (eds.) Structural Realism: Structure, Object and Causality, The
Western Ontario Series in Philosophy of Science 77, Dordrecht: Springer, pp. 149–​168.
Blumenthal, G., and Ladyman J. (2017), “The Development of Problems within the
Phlogiston Theories, 1766–​1791.” Foundations of Chemistry 19 (3), pp. 241–​280.
Blumenthal, G., and Ladyman, J. (2018), “Theory Comparison and Choice in Chemistry,
1766–​1791.” Foundations of Chemistry 20 (3), pp. 169–​189.
Cartwright, N. (1983), How the Laws of Physics Lie, Oxford: Oxford University Press.
 Structure not Selection 255

Chakravartty, A. (2017), Scientific Ontology: Integrating Naturalised Metaphysics and

Voluntarist Epistemology, New York: Oxford University Press.
Chang, H. (2012), Is Water H2O? Evidence, Realism and Pluralism, Boston Studies in the
Philosophy and History of Science, Dordrecht: Springer.
Churchland, P., and Hooker, C. (eds.). (1985), Images of Science: Essays on Realism
and Empiricism, (with a Reply from Bas C. van Fraassen), Chicago: University of
Chicago Press.
Dennett, D. (1991), “Real Patterns.” Journal of Philosophy 88 (1), pp. 27–​51. doi:10.2307/​
2027085
Esfeld, M., and Lam, V. (2008), “Moderate Structural Realism about Space-​Time.”
Synthese 160, pp. 27–​46.
French, S. (2014), The Structure of the World: Metaphysics and Representation,
Oxford: Oxford University Press.
French, S., and Ladyman, J. (2003), “Remodelling Structural Realism: Quantum Physics
and the Metaphysics of Structure.” Synthese 136 (1), pp. 31–​56
French, S., and Ladyman, J. (2011), “In Defence of Ontic Structural Realism” ’ in Bokulich,
A. & Bokulich, P. (eds.) Boston Studies in the Philosophy of Science 281, Scientific
Structuralism, Dordrecht: Springer, pp. 25–​42.
Gale, G. (1968), “Phlogiston Revisited: Explanatory Models and Conceptual Change.”
Chemistry 41, pp. 16–​20.
Koertege, N. (1969), The General Correspondence Principle: A Study of Relations Between
Scientific Theories, Doctoral Thesis, University of London.
Kuhn, T.S. (1962), The Structure of Scientific Revolutions, Chicago: University of
Chicago Press.
Kuhn, T.S. (1977), The Essential Tension: Selected Studies in Scientific Tradition and
Change, Chicago: University of Chicago Press.
Ladyman, J. (1998), “What is Structural Realism?,” Studies in History and Philosophy of
Science 29 (3), pp. 409–​424.
Ladyman, J. (2002), Understanding Philosophy of Science, London: Routledge.
Ladyman, J. (2011), “Structural Realism versus Standard Scientific Realism: The Case of
Phlogiston and Dephlogisticated Air.” Synthese 180 (2), pp. 87–​101.
Ladyman, J. (2015), “Are There Individuals in Physics, and If So, What Are They?” in Guay,
A. and Pradeu, T. (eds.) Individuals across the Sciences, Oxford: Oxford University
Press, pp. 193–​206.
Ladyman, J. (2016), “The Foundations of Structuralism and the Metaphysics of Relations”
in Marmodoro, A. and Yates, D. (eds.) The Metaphysics of Relations, Oxford: Oxford
University Press, pp. 177–​197.
Ladyman, J. (2017), “An Apology for Naturalised Metaphysics” in Slater, M. and Yudell,
Z. (eds.) Metaphysics and the Philosophy of Science: New Essays, New York: Oxford
University Press, pp, 141–​162.
Ladyman, J. (2019), “Structuralists of the World Unite.” Studies in History and Philosophy
of Science 74, pp. 1–​3.
Ladyman J. and Ross, D., Spurrett, D., and Collier, J.G. (2007), Every Thing Must
Go: Metaphysics Naturalised, New York: Oxford University Press.
Laudan, L. (1977), Progress and its Problems: Toward a Theory of Scientific Growth,
Berkeley: University of California Press.
Laudan, L. (1984), Science and Values: The Aims of Science and their Role in Scientific
Debate, Berkeley: University of California Press.
256 Contemporary Scientific Realism

Mach, E. (1911), History and Root of the Principle of Conservation of Energy,

Chicago: Open Court.
Papineau, D. (ed.). (1996), The Philosophy of Science, Oxford: Oxford University Press.
Poincaré, H. (1905/​1952), Science and Hypothesis, New York: Dover.
Post, H.R. (1971), “Correspondence, Invariance and Heuristics: In Praise of Conservative
Induction.” Studies in History and Philosophy of Science 2, pp. 213–​255.
Psillos, S. (1999), How Science Tracks Truth, London: Routledge.
Putnam, H. (1978), Meaning and the Moral Sciences, London: Routledge.
Rosenberg, A. (2011), The Atheist’s Guide to Reality: Enjoying Life Without Illusions,
New York: W.W. Norton.
Saunders, S. (1993), “To What Physics Corresponds” in French, S. and Kamminga, H.
(eds.), Correspondence, Invariance and Heuristics, Dordrecht: Kluwer Academic
Publishers, pp. 295–​325.
Schurtz, G. (2009), “When Empirical Success Implies Theoretical Reference: A Structural
Correspondence Theorem.” British Journal for the Philosophy of Science 60, pp. 101–​133.
Shapin, S., and Shaffer, S. (1985), Leviathan and the Air-​Pump: Hobbes, Boyle and the
Experimental Life, Princeton: Princeton University Press.
Stein, H. (1989), “Yes, but . . . Some Skeptical Remarks on Realism and Anti‐Realism.”
Dialectica 43, pp. 47–​65.
Van Fraassen, B.C. (2006), “Structure: Its Shadow and Substance.” The British Journal for
the Philosophy of Science 57, pp. 275–​307.
Wallace, David. (2011), “Taking Particle Physics Seriously: A Critique of the Algebraic
Approach to Quantum Field Theory.” Studies in History and Philosophy of Science Part
B: Studies in History and Philosophy of Modern Physics 42 (2), pp. 116–​125.
Wallace, D. (2015), The Emergent Multiverse, Oxford: Oxford University Press.
Westfall, R. (1967), “Hooke and the Law of Universal Gravitation: A Reappraisal of a
Reappraisal.” British Journal for the History of Science 3 (3), pp. 245–​261.
Worrall, J. (1989), “Structural Realism: The Best of Both Worlds?” Dialectica 43 (1–​2), pp.
99–​124.
 12
The Case of the Consumption Function
Structural Realism in Macroeconomics
Jennifer S. Jhun

Department of Philosophy
Duke University
[email protected]

12.1 Introduction

Consider the aggregate consumption function. In its barest form:

C t = f ( Yt )

Ct is current aggregate consumption and Yt is current aggregate income. In the

United States, final consumption expenditure of gross domestic product (GDP)
is somewhere around 70%. It is one of the simplest macroeconomic relationships.
But what does the consumption function represent, if it represents anything
at all?
The traditional realist claims that our best theories (and models) say (approxi-
mately) true things about the world and refer to real entities that exist. But closer
inspection of the conceptual and historical background of the consumption
function will cast doubt on whether economics is the kind of discipline apt for
a realist interpretation. After all, it’s a truism that the central posits of economic
theory are often idealizations.1
This paper is an attempt to make space for realism, though perhaps not
the traditional kind. I argue that the underpinnings of the modern consump-
tion function—​namely the Euler equation—​are not meant to correspond to

1 This is actually true of other sciences as well, including physics.

Jennifer S. Jhun, The Case of the Consumption Function In: Contemporary Scientific Realism.
Edited by: Timothy D. Lyons and Peter Vickers, Oxford University Press. © Oxford University Press 2021.
DOI: 10.1093/​oso/​9780190946814.003.0012
258 Contemporary Scientific Realism

any particular things in the real world. Rather, they are methodological posits
that, in and of themselves, do not state truths; but they help the scientist un-
cover the truth about the way the world is. In particular, I’ll argue that such
posits help us discover scale-​dependent structural features of the economy
that are real.
We begin in section 12.2 with a (very!) stylized reconstruction of the history
of the consumption function, starting with Keynes (1936) and ending with
Hall’s (1978) response to the so-​called Lucas Critique. Hall’s insight was to ex-
plicitly incorporate agent expectations into the formulation of the consump-
tion function; in this way, he could avoid the criticism that agents would alter
their behavior in anticipation of policy changes and therefore change the very
structure of the economy that the modeler attempted to capture (rendering
the whole project moot). This explicit incorporation of intertemporal choice
is what Chao (2003), inspired by Worrall’s structural realism, identifies as real
structure.
However, a number of authors have expressed skepticism about its empirical ad-
equacy. It seems as if actual consumer behavior suggests that the Euler equation, or
the hypotheses it helps scaffold, generally don’t hold. Some more investigative work
must be done if we want to account for actual economic behavior. Yet, the Euler
equation approach persists as a cornerstone of consumption modeling. In order to
make sense of what role it is playing in economists’ reasoning strategies, section 12.3
considers the empirical challenges that the Euler equation faces—​both by itself and
embedded in, for instance, the rational expectations permanent income hypothesis
proposed by Hall. The suggestion is that the consumption function—​and indeed,
the underlying Euler equation itself—​may not be the right object of the realist atti-
tude. We offer a proposal: the Euler equation partakes in a methodology that aims
to find out about real structure. That is, we can embed the Euler equation in a larger
realist project.
These considerations will help us identify a modest structural realism in
section 12.4 that respects that theoretical fixtures such as the Euler equation
in themselves do not represent real structure, and yet recognizes that they are
still used to uncover real structure. To hash out our position, we look to Lyons’s
(2005, 2011, 2017) axiological realism—​which maintains that the aim of the-
ories is to find truth (whether or not we actually do, and whether or not we
are justified in thinking we do) in order to formulate an epistemic counterpart.
This epistemic counterpart will depart, however, fundamentally from Lyons’s
vision in that it takes a pragmatic, perspectival view of truth. Finally, in sec-
tion 12.5, we argue that this version of realism has the benefit of sidestepping
some of Reiss’s (2012) criticisms of standard realist defenses against charges of
instrumentalism.
 The Case of the Consumption Function 259

12.2 A Brief History of the Consumption Function

The Keynesian formulation of the consumption function was conceived as a linear

(fairly invariant) relationship.2

C t = α + βYt

Here, α is a constant, β is the marginal propensity to consume, and Yt is disposable

income. This equation is supposed to represent, according to Keynes, “The funda-
mental psychological law . . . that men are disposed, as a rule and on the average,
to increase their consumption as their income increases, but not by as much as the
increase in their income” (1936, 96). Known as the absolute income hypothesis, this
formula implies that changes in consumption depend on changes in income. That
is, given increases in GDP, we expect some increased consumption and increased
savings.
According to this theory, average propensity to consume (APC) would be greater
than marginal propensity to consume (MPC), which is less than one per the funda-
mental psychological law. This implies that average propensity to consume would
decrease and average propensity to save would increase as income increased. So, for
instance, it predicts that post-​war savings would increase and that consumption would
decrease as government spending fell and the economy headed toward a recession.
However, empirical issues plagued the Keynesian consumption function.
As demonstrated by a series of papers by Kuznets and his colleagues, it became
clear that the function could not hold in the long run.3 For instance, there was a
contradiction between cross-​section (declining) and time-​series (constant av-
erage) data of the savings-​income ratio, which reported different amounts for
the average propensity to save. In 1942, Kuznets, who was then associate director
of the Bureau of Planning and Statistics of the War Production Board, showed
that even though there was a significant increase in per income capita from
1879 to 1928, the savings-​income ratio remained constant. Keynesian theory
had predicted that it would decrease.4 A 1946 publication demonstrated that

2 Technically, it is not Keynesian as in Keynes himself did not advocate for this (and certainly

not in the sense of the General Theory). People who considered themselves Keynesians presented
it in this way, but even for them it was a convenient tool rather than something to be unshakably
committed to.
3 Notably, in the long run it also looks linear through the origin. APC is more or less constant over

time, but not over the business cycle. That fact gave rise to an entire literature on the consumption
function from the 1940s to the 1970s. Friedman was working with Kuznets during this time, and the
original permanent income hypothesis can actually be found in a joint publication (1945).
4 He would document this long-​run constancy again in a 1952 paper.
260 Contemporary Scientific Realism

the time series ran contrary to “Keynesian” theory, and instead (as Chao 2003
documents):5

(1) MPC is less than APC in budget data and short-​run time-​series data but
is equal to APC in the long run; (2) APS [average propensity to save] and APC
did not rise secularly, and (3) private demand increased sharply and APS was
sharply lower than in the interwar period level. (85)6

Alongside Duesenberry’s (1949) relative income hypothesis and Modigliani and

Brumberg’s (1954) life-​cycle hypothesis models, Friedman (1957) too would
offer an alternative in response to these difficulties: the permanent income hy-
pothesis. Though he himself did not present a formalized version, the intuition
was as follows. Consumption would depend not just on the amount of my cur-
rent after-​tax income, but my estimation of the overall income I believe I’ll earn
in the future. In particular it’s going to depend on the permanent (long-​run)
rather than transitory income that I receive. As a forward-​looking agent, I try
to assess what future income might look like by forming adaptive expectations
based on my past experiences.7 And I should react more strongly to permanent
shocks than transitory ones. So most of my consumption is going to be based on
permanent income—​the average amount I expect to receive over the years in
the future—​rather than transitory income, like lottery winnings. Along the way
I learn adaptively—​sometimes my estimates will be off. A standard economics
textbook may present the general form of my consumption function in the fol-
lowing way.8 Let permanent income be:

5 And here I put Keynesian in scare-​quotes; Keynes himself in the second chapter of the General

Theory seems to gesture at something that looks like the permanent income hypothesis.
6 A number of other features cast doubt on the Keynesian formulation. For example, Molana

(1993) documents the following empirical challenges (assuming the consumption function takes the
form of Ct = α + βYt):
For instance: (i) Kuznets’ (1946) study illustrated that the APC did not contain a significant
trend, (ii) the cross section budget investigations by Brady and Friedman (1947) showed
that α was in fact positive with a tendency to shift upwards, (iii) Smithies’ (1945) time se-
ries study confirmed that the presence of a positive deterministic trend in a could not be
ruled out, and finally, (iv) estimates of β usually turned out to be lower than expected (see
Haavelmo, 1947) and the model persistently under predicted consumption. (338)
Cate (2013) further notes that the Keynesian account neglected the fact that consumption is not
only affected by changes in income but is also by wealth (572).
7 Modigliani’s life-​ cycle hypothesis of consumption is similarly based on forward-​ looking
expectations—​this is why such models are now called permanent income-​life cycle hypothesis
models. Both Keynes and Modigliani argued that consumers tend to smooth their consumption over
time, so consumption doesn’t depend so much on current income.
8 I note this because—​ as is unsurprising with any stylized history—​this is not Friedman’s
formulation.
 The Case of the Consumption Function 261

YtP = Yt −1 + q (YtP−1 − Yt −1 )

q is the fraction of last year’s error subtracted from Yt–​1 in order to estimate this
period’s permanent income, Ytp . It is the amount by which my estimated in-
come differs from my estimate last year. The consumption function tells us how
permanent consumption depends on permanent income (rather than current
income), i.e.

C Pt = kYtP

Therefore,

C Pt = k (1 − q ) Yt −1 + kqYtP−1

k is the marginal propensity to consume out of my permanent income. Subtract

my permanent income from my actual income, and the remainder is transi-
tory income. The idea is that people like to smooth their consumption behavior
over time, rather than letting it fluctuate in response to short-​term changes. The
permanent income hypothesis even resolves the earlier empirical puzzle that
long-​run and short-​run consumption habits seem to run in different directions—​
wealthier households save more, yet the savings rate stays constant over time. We
might, for instance, attribute those wealthier households with a large amount of
transitory income—​such as bonuses from work—​but poorer households tend to
have negative transitory income.
However, even with this adjustment, there remained a worry that we call the
Lucas Critique. Lucas (1976) had noted a more general problem. Even when
Keynesian models worked well as forecasting models, they were inappropriate
for assessing alternative policies—​ i.e. they could not accommodate policy
change because “any change in policy will systematically alter the structure of
econometric models” (41). Lucas’s proposed solution was that economists ought
to seek “deep” parameters invariant to policy changes, though this really meant
explicitly modeling consumer expectations, tastes, and technology. This involved
modeling fixed psychological features. Lucas and Sargent (1979) emphatically
criticize those models that did not as “incapable of providing reliable guidance in
formulating monetary, fiscal and other types of policy” (69).9

9 And even this is somewhat overdramatic. These models were only just being used when Lucas

criticized them, so they didn’t at this time have a historical record of being the spectacular failures
they were made out to be.
262 Contemporary Scientific Realism

Robert Hall (1990) would try to get around this problem by explicitly incor-
porating optimization behavior into his formalism, locating structure in the un-
derlying decision-​making mechanism.10

Although Lucas was scornful of existing econometric policy evaluation models,

his message was not completely destructive of all model-​building or empirical
research. There are structural relationships in the economy, but the consump-
tion function is not among them. For consumption, the structural relation, in-
variant to policy interventions and other shifts elsewhere in the economy, is the
intertemporal preference ordering. (135)

Aligning Friedman’s insights with Lucas’s, Hall proposed extending the per-
manent income hypothesis account to incorporate rational expectations, i.e.
assumptions about belief formation, directly into the model. The permanent in-
come hypothesis tells us that if permanent income were known (deterministic),
consumption would remain the same over time. Yet, it doesn’t—​we have uncer-
tainty with respect to our income. Agents, according to rational expectations,
take into account all information in order to forecast the future. Supposing that
our agents are utility maximizers, the Euler equation encodes rational expecta-
tions as governing intertemporal choice. In the case of income uncertainty, it
manifests as the following first order condition on the agent’s decision-​making
problem.

E t u ′ ( c t+1 ) = (1 + d ) / (1+ r ) u ′ ( c t )

Here, c is the agent’s consumption (indexed to a particular time), d is the subjec-

tive time discount rate, r is the real interest rate, and E represents mathematical
expectations. According to this formula, there are no reallocations of goods such
that the agent’s utility at the margin can be improved. In the case the utility func-
tion is quadratic and d = r, then

Ec t+1 = c t

Then by mathematical (rational) expectations:

c t+1 = c t + e t+1

10 As a note, Friedman did consider uncertainty, but not formally.

 The Case of the Consumption Function 263

That is, consumption tomorrow will be the same as today with some error.
Consumption follows a random walk, and any actual changes are unpredict-
able and random. If I’m forward looking, I’ll anticipate changes in my income by
smoothing consumption over time (but given that actual changes are random,
I do not anticipate substantial changes in income).

12.3 The Question of Realism

12.3.1 Reductionism and Microfoundations

One can read the Lucas Critique as an urge to pursue a reductionist project, and
interpret the ubiquitous representative agent as a first-​step maneuver in such a pro-
ject. As economics progresses, we will get ever closer to a model of the individual
agents in the economy rather than aggregates.11 So, as a first approximation, we
treat the consumption function as reflecting the behavior of a macroeconomic-​
sized representative agent. Perhaps we can move next to (aggregate) heterogeneous
agents, and so on. According to this picture, aggregates are just a shorthand way of
talking about groups of individuals. Macroeconomics, thus, talks about real things
insofar as the entities in question are really inhabitants of a microeconomic on-
tology. Ultimately, what we really need to do is find and then put all household
consumption functions together in some appropriate way.
There’s something suspect about this particular way Lucas’s criticism has been
borne out. This is because it strikes me as an unpromising strategy to pursue a re-
ductionist project along these lines. If macroeconomic aggregates are reducible
to talk of microeconomic entities—​the microfoundations project—​as the Lucas
Critique seems to require, then it seems like macroeconomic features are not re-
ally autonomous from microeconomic ones. So, if we take the Lucas Critique se-
riously, then to be a realist about macroeconomics implies that we are realist only
insofar as there are microfoundations for our theory.
The difficulty is that there is often no straightforward relationship between the
microscale and the macroscale of the economy. In fact, formal theorems such
as the 1970s Sonneschein-​Mantel-​Debreu (SMD) theorem—​also known as the
“anything goes” theorem—​emphasize that there is no such correspondence be-
tween the configurations of individual agents and behavior at the macroscale.
In the context of excess demand functions, the SMD theorem itself states that
under standard assumptions on agents, no interesting macroscopic properties
are guaranteed to emerge from the population. That is, they don’t give rise to any

11 For a critical overview, see Hoover (2015).

264 Contemporary Scientific Realism

particular equilibrium vector, meaning that properties at the microscale do not

transfer to the macroscale and there can be an arbitrary number of equilibria. We
could assign all agents a downward sloping demand curve, and yet we may not
see that behavior in the aggregate.
We should worry that similar difficulties will arise in the case of household
versus aggregate consumption. This was exactly the problem that arose during
the years after World War II; the usual practice in economic forecasting was
to put together several different consumption functions for different classes of
goods and simply add them together (Bronfenbrenner 1948, 318). Using these as
major parts of econometric models, economists would offer policy advice with
disappointing results.12 So facts about the macroscale cannot simply be deduced
from facts about goings-​on at the more microscale, indicating that at least former
kinds of facts are in some sense autonomous from the latter kind—​facts about
aggregate features may be independent of the sum of facts about individual ones,
though of course the former depend on the latter’s existing in the first place.13
But, the Federal Reserve thinks about the national economy in aggregate
terms; for instance, expansionary money policy, which increases the money
supply, is thought to be accompanied by an increase in GDP as well as an in-
crease in spending. Policy is aimed at the aggregate and is expected to bring
about certain effects. Without a realist reading of economics, interventions of
this kind simply don’t make much sense. But what kind of realist reading is the
appropriate one?

12.3.2 Worrall’s Structural Realism and Chao’s Realism

about Structure

One of the most popular—​if not the most popular—​account of realism is struc-
tural realism. The paradigmatic defense of structural realism comes from
Worrall (1989), who argues that:

It would be a miracle, a coincidence on a near cosmic scale, if a theory made as

many correct empirical predictions as, say, the general theory of relativity or
the photon theory of light without what that theory says about the fundamental
structure of the universe being correct or “essentially” or “basically” correct.
But we shouldn’t accept miracles, not at any rate if there is a non-​miraculous
alternative. If what these theories say is going on “behind” the phenomena is

12
See Hart (1946).
13
Note that even aggregation is not uniformly problematic, nor is it always problematic. Nominal
GDP is an unproblematic aggregate that consists in just exact aggregation of the values of final goods.
 The Case of the Consumption Function 265

indeed true or “approximately true” then it is no wonder that they get the phe-
nomena right. So it is plausible to conclude that presently accepted theories are
indeed “essentially” correct. (101)

Known as the No Miracles Argument, it insists that theories would not be as suc-
cessful as they are unless something had gone right—​and this something, says
Worrall, is structure. One way of interpreting the position (and the usual way
to interpret Worrall) is in an epistemic sense, i.e. a claim about what we can have
knowledge of, where this is typically “the (preserved) mathematical structure of
our theories” (Morganti 2011, 1166).14
In what follows I mean structure quite loosely. For one, I am here interested in
causal structure, where structure consists of modal relationships that satisfy an
interventionist conception of causation á la Hoover (2001, 2011) or Woodward
(2005). Toggle one thing, and something else changes as a result. That relation-
ship must be (to some extent) invariant: “When a relationship is invariant under
at least some interventions . . . it is potentially usable in the sense that . . . if an
intervention on X were to occur, this would be a way of manipulating or con-
trolling the value of Y” (Woodward 2005, 16).15 Something that partakes in a
causal relationship is part of the causal structure of a system, and articulating a
bit of causal structure will involve articulating the relata in a difference-​making
relationship (that is apt for counterfactual analysis). One important disclaimer,
however, is that the way I am thinking about structure will generalize from the
way that structure is typically understood by economists. Economists, at least
post–​ Lucas Critique, consider “deep parameters”—​ those that are invariant
under policy interventions—​as structural features, and structure itself is defined
as a system of equations.16 Unlike the Cowles era economists, I am not using it

14 For the most part, the distinction doesn’t bear much on the suggestion I’m trying to make in this

paper, because (spoilers) the account I turn to is pragmatist in nature; for more on the distinction, see
Ladyman (1998). Ross’s (2008) ontic structural realism claims that “The basic objects of economic
theory are optimization problems” (741). Objects, on this view, are “heuristics, bookkeeping devices
that help investigators manipulate partial models of reality so as to stay focused on common regions
of measurement from one probe to the next.” What really matters, say, in the case of economic games,
are that they are “mathematical structures, networks of relationships whose relata are distinguished
as such by the mathematical representations of the structures in question—​just like quarks and
bosons.”
15 One implication is that insofar as I am thinking about structure, I am really only concerned

with causal structure. I am here agnostic as to whether and what other kinds of structure there are.
I should note here that I am not using causation in the strictly Woodwardian sense either, who iden-
tifies causal relationships with a particular kind of relationship between variables. Despite this, causal
relationships we think of as holding in the world may very well on our account be the kinds of things
we try to capture with structural equations.
16 And even how and what exactly gets considered structure is going to go hand-​in-​hand with

the particular kind of methodology a school uses—​for example, London School of Economics
style econometrics and Cowles Commission style econometric methodology proceed in different
directions, disagreeing on how theory and data bear on modeling. And the New Classical approach
that takes rational expectations as foundational, too, is relying upon a different methodology.
266 Contemporary Scientific Realism

in order to draw a contrast with what is known in econometrics as reduced-​form

equations.
Chao (2007), inspired by the semantic view of theories, proposes what he calls
realism about structure. This account supplements Worrall’s account by distin-
guishing explicitly between structure and non-​structure in theories. The test case
is the consumption function; Chao identifies intertemporal choice (encapsulated
by the Euler equation) as the underlying structure. The structure that we regard
as real, Chao tells us, consists in the mathematical equations that are preserved as
theory develops. The mathematical equations that represent the abstract struc-
ture of a system related to their concrete empirical counterpart models up to iso-
morphism. The empirical models and the mathematical equations are related
via a representation theorem, which assures us that we have singled out relevant
structure in our model. Empirical structure remains invariant to disturbances.
Analogously, the mathematical equations that capture its structure are invariant
to transformations (such as when there is theory development).

[W]‌hen Worrall observes that in the history of science the mathematical equa-
tions of the new and old theories are invariant (isomorphism among theoretical
models), given that in the successful or mature scientific theories, the theoret-
ical model flourishingly represents a feature of the world (isomorphism be-
tween theoretical and empirical models), we can conclude that the represented
empirical models are also invariant under the same type of transformation as
the theoretical one . . .
In this interpretation the represented empirical model is invariant and there-
fore can be regarded as structure. (239)

These “invariant relations are structural and real.” (240)

The story is an optimistic one; it is possible to think about economics in re-
alist terms. But then the story takes an odd turn: “we accept a theory if it yields
a model containing the right structure. If a model is not supported by empirical
data when the model is considered as containing the true structure, we do not
reject the model but instead construct a new model with the same true structure”
(240). That is, in the face of recalcitrant experience, we stay committed to the
Euler equation as if it were part of a Lakatosian hard core or Kuhnian paradigm;
we are wedded to a strong a priori belief in it.
I do not disagree that the Fisherian intertemporal framework has survived
over time as consumption theory has evolved. But I want to suggest that we can
diverge from or at least reinterpret Chao with respect to the claim that the Euler
equation is real. For us, the Euler equation is not what is left once we have cleaned
up our mistaken suppositions in a model or a theory. It serves a different meth-
odological purpose.
 The Case of the Consumption Function 267

Now, the Euler equation itself doesn’t determine consumption. While it

doesn’t refute the Euler condition itself, from a practical standpoint, we might be
a bit unsettled if it systematically turned out that the Euler equation approach—​
such as that embodied in the permanent income hypothesis—​fails. And it does
seem that the permanent income hypothesis does not always describe consumer
behavior very well. As Blinder et al. (1985) remark:

. . . the research done to date has not supported the econometric restrictions im-
plied by the Euler equation approach. Nor has further investigation validated
the hypothesis that the response of consumption to income (henceforth, Y)
reflects only the usefulness of current Y in predicting future Y. Instead, research
typically finds “excess sensitivity” to current income. (467)

Flavin’s (1985) noteworthy study found that modeling income time series follows
an autoregressive–​moving-​average (ARMA) process, and also that consump-
tion exhibits excess sensitivity: “The empirical results indicate that the observed
sensitivity of consumption to current income is greater than is warranted by the
permanent income-​life cycle hypothesis” (976). Because of the statistical discrep-
ancy between the null hypothesis and the data, she proposes liquidity constraints
on household consumption to explain why it is that consumption is excessively
sensitive to changes in income.17
There are ways to use posits like the random walk equation, or even the un-
derlying Euler equation, without assuming that there must be some straightfor-
ward correspondence (or failure as such) to something in the real world in order
to help us get a grip on the economy. Consider Campbell and Mankiw’s (1989,
1990, 1991) papers that aimed to show that rational expectations did not cor-
rectly capture macroeconomic aggregate behavior. Given the aggregate macro-
economic data, they “propose a simple, alternative characterization of the time
series data . . . [that] the data are best viewed as generated not by a single forward-​
looking consumer but by two types of consumers” (1989, 185). About half of
these consumers behave like rational agents, but the other half follow a rule of
thumb, consuming their current income. (Both, as a note, optimize their utility
preferences.) This alternative hypothesis is “an economically important devia-
tion from the permanent income hypothesis” (187).
Campbell and Mankiw embed the permanent income hypothesis model
within a generalized model that allows for a fraction of income to accrue to
forward-​looking agents and the rest to those who consume their permanent

17 Later in 1985 she notes that “The null hypothesis in this empirical literature typically consists of

the joint hypothesis that 1) agents’ expectations are formed rationally, 2) desired consumption is de-
termined by permanent income, and 3) capital markets are ‘perfect’ ” (117–​118). There she diagnoses
excess sensitivity as a failure of the third assumption.
268 Contemporary Scientific Realism

income. While according to the traditional approach, the change in consump-

tion should simply be an error, the generalized model is:

∆C t = ∆C1t + ∆C 2t = λ∆Yt + (1 − λ)ε t

They can directly test the permanent income hypothesis by setting λ to zero. And
in fact, their estimate of λ—​that portion of the population that consumes cur-
rent income—​is about 0.5 (195). The macroscopic phenomenon of smoothness
is explained by the fact that aggregate consumption “is a ‘diversified portfolio’ of
the consumption of two groups of agents” (210).
The point of this exercise, which again does not discard the original consump-
tion formalism associated with the random walk hypothesis, is not a straightfor-
ward case of starting with an idealized model and gradually de-​idealizing it to
make it more realistic. If real behavior deviates from the idealized benchmark
(like one that assumes a single representative agent), however, we should seek
out what factors explain that deviation. The Euler construction is a tool that helps
us erect and navigate a larger, more complex framework, making space for po-
tentially relevant structural information (though not every and all pieces of in-
formation) that makes a difference to a system’s behavior. And for Mankiw and
Campbell in particular, the structural information about their population of in-
terest is how much of that income accrues to people who abide by rational ex-
pectations as opposed to those who consume their current income. They choose
to look at the distribution of agent kinds in the population. That is, theirs is a
project that tries to capture what we might call scale-​dominant behavior.18 This
is behavior that is characteristic of a system at a particular scale, where particular
structural characteristics will be salient there.19

18 Wilson (2017), in the context of representative volume element (RVE) methods in physics,

points out that complex behavior can be managed by parsing it out into separate sub-​models, each
“assigned the comparatively circumscribed duty of capturing only the central physical processes nor-
mally witnessed at its characteristic scale length . . . each localized model renders descriptive jus-
tice only to the dominant behaviors it normally encounters” (19). And indeed, as Batterman (2013)
emphasizes, “Many systems, say a steel girder, manifest radically different, dominant behaviors at
different length scales. At the scale of meters, we are interested in its bending properties, its buckling
strength, etc. At the scale of nanometers or smaller, it is composed of many atoms, and features of in-
terest include lattice properties, ionic bonding strengths, etc.” (255).
19 In order to check whether the results were robust, the authors carry out a battery of tests to

supplement their instrumental variables method. These include Monte Carlo methods to examine
the small-​sample distribution of the test statistics in order to check for the small-​sample bias. The
Monte Carlo experiment deploys their framework with adjustable parameters for population distri-
bution and aims to produce artificial data with similar structural—​in this case statistical—​properties
(such as that log-​income and consumption are integrated processes and consumption and income
are cointegrated). In addition, they explore various generalizations of their model to see if there are
other candidate explanations.
 The Case of the Consumption Function 269

Again, however (and as Chao points out), one might object that it is only the
random walk hypothesis that is in danger, rather than the Euler equation ap-
proach itself, as Chao notes: “Anomalies may reject the permanent income hy-
pothesis or the life-​cycle hypothesis, but they only motivate modelers to modify
the Euler equations instead of abandoning them” (243). For example, instead of
discarding the Euler equation, discrepancies in behavior are used to modify it.20
One might suggest that this only indicates that one or more of the assumptions
going into the random walk hypothesis itself must be problematic, but it need
not also imply that the Euler equation (the intertemporal preference ordering)
is problematic. We have something of an underdetermination problem; the con-
sumption equation’s failure might be due to any number of things, such as wrong
assumptions about preferences (like that of constant relative risk aversion, which
is quite common).
But there are reasons to question the viability of the Euler equation itself.
Carroll (2001) questions whether we are able to estimate the (log-​linearized)
Euler equations at all. Canzoneri et al (2007) compare the interest rates predicted
by the Euler equation against the money market interest rate to find out that they
are negatively correlated. Though they start out with preferences that satisfy con-
stant relative risk aversion, they do try out different utility functions. Estrella
and Fuhrer (1999) claim that “evidence that shows that some forward-​looking
models from the recent literature may be less stable—​more susceptible to the
Lucas critique—​than their better-​fitting backward-​looking counterparts,” (4) a
count against the Euler approach. Blinder and Deaton (1985) have their “doubts
about the wisdom of modeling aggregate consumption as the interior solution
to a single individual’s optimization problem in adjacent periods, but in any case
think it fair to say that the research done to date has not supported the econo-
metric restrictions implied by the Euler equation approach” (467). There are at
least two reasons for this: corner solutions due to liquidity constraints and aggre-
gation problems. For instance, at least at the aggregate level, the Euler equation
seems suspicious.21 Attanasio (1999) finds that “The Euler equations for (non-​
durable) consumption . . . are, for most specifications of preferences, non-​linear.
As they refer to individual households, their aggregation is problematic” (781).
Elsewhere, Hvranek’s (2015) meta-​analysis finds that on average, the estimated

20 For another interesting case study, see Zeldes (1989) and Runkle (1991), who estimated versions

of the Euler equation with different results. Attanasio (1999) reports that Zeldes (1989) “splits the
sample according to the wealth held and finds that the rate of growth of consumption is related with
the lagged level of income for the low wealth sample. The same result does not hold for the high
wealth sample. Zeldes interprets this result as evidence of binding liquidity constraints for a large
fraction of the population” (790). That is, the low-​wealth sample violates the standard Euler equation
while the high-​wealth sample does not. On the other hand, Runkle (1991) finds no such evidence.
21 Deaton (1992) specifies that “if tastes are the same, the Euler conditions will aggregate perfectly

if three conditions hold: (a) people live for ever, (b) felicity functions are quadratic (or the time in-
terval is very short), and (c) individuals know all the aggregate information” (167).
270 Contemporary Scientific Realism

elasticity of intertemporal substitution is zero—​too low, given the way the Euler
equation is usually specified (consumption should be sensitive to variations in
the real interest rate).22,23
It’s doubtful that even individual agents optimize in this way, and when they
do it may be limited in scope—​after all, agents are often thought to be “myopic”
in nature. (Of course, we could always restrict the behavior to a finite number
of periods, rather than an infinite horizon, which is entirely consistent with the
Euler approach.) Even if we were only considering individuals as the kinds of
agents that intertemporally optimize, aggregation problems notwithstanding,
this very well may be false except for fairly short time horizons. One substantial
task would then be to assess over what temporal scale intertemporal optimiza-
tion holds. So it seems a bit odd to say that the Euler equation is the product of
our uncovering some deep structure that is actually out there in the world.
Work since Hall (1978) has focused on extending the baseline Euler model
in various directions. For instance, one might incorporate habit-​formation in
consumption behavior, which might explain the lagged behavior of consump-
tion in response to interest rate or follow Campbell and Mankiw and incorpo-
rate hand-​to-​mouth consumers. Throughout all this, as Chao notes, the Euler
construction still stands. Now, rather than taking the Euler construction itself
to be real, however, I’m more inclined to say that it helps find out what is real. I’d
argue that stipulating the Euler equation is a special case of a story about equilib-
rium reasoning that I have told elsewhere.24 Equilibrium conditions, constraints
on behavior, do not in themselves state truths about the way the world is—​but
they do help yield interesting, informative insights about the world. That is, equi-
librium conditions are methodological tools rather than straightforward claims
about or corresponding to states of affairs. For instance, equilibrium conditions
like the Euler equation function more like the ceteris paribus laws we see com-
monly in economics. And while it may hold somewhere, and across some range
of circumstances, it need not hold universally.
Here’s what I mean. For example, we typically do not think “An increase in the
supply of a good, ceteris paribus, leads to an increase in price of that good” is in-
variant in the sense that it holds everywhere. We do, however, use it to figure out
in which pockets of the economy in which it does hold, so it will have to be in-
variant over some range of circumstances for some market over some time scale.
If our local widget economy is somewhat (causally) isolated from the rest of the
nationwide economy, we could apply ceteris paribus reasoning in order to think
about what would happen under certain kinds of shocks or interventions. I think

22 A finding consistent with Yogo (20045).

23 Readers should also note that there is work on sensitivity at the micro rather than at the aggre-
gate level. See Shea (1995), Parker (1999), and Hall and Miskin (1982).
24 See Jhun (2018).
 The Case of the Consumption Function 271

the Euler equation works in somewhat the same way—​it’s not that it, itself, is
invariant—​but rather, we use it to locate where relationships can be treated as
invariant, which means picking out a relevant scale of interest. When it fails,
we seek reasons as to why actual behavior diverges. And these relationships of
interest are often causal, in that they are apt for an interventionist difference-​
making interpretation.
In particular, using the Euler equation means trying to see where, and
how well, it fits. To take an example, when Mumtaz and Surico (2011) ad-
just parameters to fit the data over time, what they are noting is where it is
and in what form that the Euler equation holds and doesn’t hold, as “the
parameters of the aggregate consumption Euler equation may vary over the
business cycle” (5). One of their self-​imposed tasks is to show that “periods of
conditionally low (high) consumption correspond to periods of below-​trend
(above-​trend) consumption” (12). What this helps highlight is the partic-
ular scale of the system we should be paying attention to. So one thing that
Mumtaz and Surico’s account does is make salient particular time scales (and
salient behavioral properties at those scales) as relevant to the investigation.
For instance, they identify recession periods in the 1970s, 1981, 1992, and
2008 as periods during which agents tend to be more backward-​looking and
consumption tends to be less sensitive to the real interest rate (13). Similarly,
we can use the Euler equation to investigate whether agents are subject to bor-
rowing constraints; strictly speaking this doesn’t mean that the Euler equation
should be thrown out, but that it applies within a certain scope. Rather than
merely pointing out that agents don’t optimize over time, we say something
about the extent to which they do. That is, I can tell an analogous story here
about the role of the Euler equation as I did with that of the random walk
hypothesis in the Campbell and Mankiw case. These equations are fixtures
that help us identify particular scales of interest and the behaviors that appear
dominant at those scales—​scale-​dominant behaviors.
Let me use an analogy from materials engineering to concretize what I mean
by scale-​dominant behavior. Though this reductionist attitude is now less pop-
ular, one might think that the right way to determine the behavior of a system
is first to accumulate all the details at a micro-​level. After doing so, we will be
able to infer or deduce information about what goes on at the macro-​level. So a
bottom-​up methodology would aim to achieve a picture of the whole system (be
it physical or social) first by gleaning information about its constituents and then
somehow putting all that information together.
But this approach is unsatisfactory in engineering contexts. For instance,
one everyday kind of problem we might worry about is how to manufacture
materials in order to construct sturdy buildings that can sustain some wear and
tear. Consider a steel beam:
272 Contemporary Scientific Realism

If we engaged in a purely bottom-​up lattice view about steel, paying attention

only to the structure for the pure crystal lattice, then we would get completely
wrong estimates for its total energy, for its average density, and for its elastic
properties. The relevant Hamiltonians require terms that simply do not appear
at the smallest scales. (Batterman 2013, 268)

What’s also important in this case is structural detail at the mesoscale, where steel’s
granular structure becomes apparent—​say at 30 nanometers. These boundaries
between grains affect the way molecular bonds break and reform when external
force is exerted on the beam. So, this structure that manifests at the mesoscale
contributes to the system’s macroscale behavior.
We can even tell a similar story about the permanent income hypothesis.
Campbell and Mankiw’s search for relevant population parameters (or Mumtaz
and Surico looking to see patterns of forward-​looking behavior) is akin to
looking for details about grain in the steel beam. And like granularity, popula-
tion distribution is a mesoscale detail that we can treat as real as goings-​on at
any other scale. But we had to do a little bit of work first—​figuring out where the
Euler equation could plausibly fit in as successfully describing behavior. Rather
than describing the structure of the economy-​at-​large, the Euler consumption
equation might characterize the behavior of some members of the economy over
some (but not the infinite) time horizon. The particular configuration of these
members might be something that matters. Distributional features capture ad-
ditional relevant structure that affects the macro-​behavior of a system; this is
additional information we need to understand this system’s behavior. This in-
formation also explains why it is that actual population behavior departs from
the idealized case with a homogeneous population where everybody consumes
current income.
Explaining the behavior of complex systems seems to require treating
features that appear as salient at particular scales as real parts of a real struc-
ture, because they are the kinds of things that figure into causal relationships.
But our treating economic structure as real is not the same as Worrall’s structural
realism; we are not realists about the random walk hypothesis, or even about
the Euler equation. The Euler equation plays a different role for us. It does not
figure as a theoretical posit that we understand as true, and thus is preserved over
time as the theory progresses. Rather, the Euler equation helps us orient our-
selves in a larger framework; it allows us to establish a vantage point from which
to look at a particular system of interest and get a grip on the relevant causal
relationships involved. We can think of investigation at a particular scale to be
investigation from a particular perspective of that system. To accommodate this,
we need to be perspectival realists, which, according to Hoover (2011), is “predi-
cated on the belief that models are used to assert true general claims about causal
 The Case of the Consumption Function 273

relationships . . . Such a realism is compatible with models viewing the world

from different perspectives . . . The truth that we seek is what the world is actually
like when seen this way” (6).25
For the most part, economic (causal) structure that accounts for deviations
in a system’s behavior from, say, the idealized benchmark case constitutes real
structure. This structure will appear as salient or dominant at a particular char-
acteristic scale. It must fit into a larger framework that tells a coherent story about
why it is that a complex system behaves the way it does, and different properties
and behaviors may appear as dominant at different scales. For instance, we can
tell a coherent story about steel grain size and beam elasticity—​it is fundamen-
tally a story that has to make reference to different scale lengths. In an analo-
gous sense, we can tell a coherent story about the economy that in part refers to
(say) population distribution, excess sensitivity, or even the time scales at which
agents do (or don’t) appear to act in a forward-​looking manner.26,27

12.4 Modest (Pragmatic and Perspectival)

Structural Realism

The kind of realism I favor is modest; it doesn’t even claim that a theory marches
steadily toward a unique true theory even in the limit. All that it requires is that
at least sometimes, we’re justified in thinking that we are getting things right.28
I propose that the perspectival realism we endorse corresponds to a pragmatic
interpretation of Lyons’s axiological realism, capturing what is intuitively attrac-
tive about the view in addition to seamlessly providing an epistemic counterpart
to his position.
Lyons (2005) suggests that “science is interested, not in true statements, per
se, but a certain type of true statements, those that are manifested as true” (174).
These statements partake in theory complexes—​groups of theories—​and are
such that they have bearing on our world of experience; they can manifest as true.

25 That is, economic structures are structures relative to perspective (scale). Our account is also

different from another attractive account of realism, selective realism. First, perspectival realism gives
us something to make sense of theories synchronically—​the attitude that we have at any given time
for a theory under consideration. On the other hand, selective realism is a diachronic account—​it
explains something about a persistent element of a theory (or theories) over time.
26 It is within this larger framework that we can ask further questions such as: Is wealth a factor

that we should have included in our macroeconomic analysis? Is income inequality a structural fea-
ture that matters? And so on. See, for work along these lines, Carroll, Slacalek, and Tokuoka (2014).
27 But articulating multi-​scale relationships is a tricky matter in itself. Bursten (2018) argues in

the case of nanoscience that articulating the relationship between scales may require conceptual
maneuvers that resist traditional categorization (e.g. reductionist ones). I would like to leave space,
for economics, to take advantage of these oddball conceptual maneuvers, though this may lead to
further complications for thinking about realism down the line.
28 Note here a sympathy with something like Mäki’s (2009) realistic realism.
274 Contemporary Scientific Realism

Lyons (2011) has called these statements, as a subclass of true statements, those
that are “experientially concretized” truths (XT statements): “true statements
whose truth is made to deductively impact, is deductively pushed to and enters
into, documented reports of specific experiences” (329).29 True statements are
candidates for—​but not all are—​XT statements, whose truth is experientially
concretized as true. And while actual XT statements are required to be true
in order to qualify as being so, we do not necessarily think ourselves to have
achieved truth per se. That is, despite the fact that that they inform the articula-
tion of or imply some claim that corresponds to an experience:

Crucially, the experiential concretization of XT statements is non-​

epistemic: While XT statements are such that their truth is deductively pushed
down to and enters into such documented reports, no claim is being made here
that the fact of this relation to documented reports informs us of the truth of XT
statements. (329)

The emphasis here is on the pursuit of truth, rather than justification for believing
we have achieved it. That is, we can be committed to seeking truth and yet be ag-
nostic about whether we have achieved it or not.
Macroeconomics can be considered a complex of theories and models.
Statements such as “the proportion of income accrued to forward-​looking agents
is λ” have implications for what we expect to see in the macroeconomic aggre-
gate. The Euler equations are such that (given some additional context) they
imply particular empirical consequences that may or may not hold.30
However, fixtures such as the Euler equations cannot be XT statements ac-
cording to Lyons’s account. And fundamental theoretical posits like the Euler
equation on its own are not even really treated as genuine candidates for truth.
Some may simply assume that they are outright false because they’re often con-
sidered idealizations. After all, the Euler equation approach is probably not a
good way of describing aggregate macroeconomic behavior. It is not even a good
way to think about individual behavior (people often simply fail at intertemporal
utility maximization, never mind that they are not infinitely lived agents
anyway). So in order to make Lyons’s insights relevant to economics, I propose
that we step away from the foundational conception of theory complexes as lying
on a bedrock of (manifest) truth from which we can infer other (manifest) truths
and think of theoretical posits as doing something other than serving as such

29 In his earlier work he tells us that a truth is manifest when “it has bearing on, makes a difference

to, the world of experience and is transmitted to tested predictions derived from the complex” (2005,
174). Such truth is “not manifested as true to us but in the consequences derived from the theory and
in the world of our experience” (177).
30 See, for this position, Hoover (2009).
 The Case of the Consumption Function 275

ground-​level manifest truths. The Euler equation has a shaping or framing role,
not a grounding one.
The Euler equation, in and of itself, is not literally true (nor, arguably, is it
false by itself either), so it cannot be an XT statement.31 But we use it to formu-
late claims that we do compare against empirical data to determine whether it
explains or describes our data well. And if we take Lyons’s account seriously, we’d
actually have what he calls an “evident XT-​deficiency” where “it is evident that
non-​XT statements are present in the complex” (2011, 330).
But it’s consistent to claim that the Euler equation doesn’t usually hold of any-
thing by itself and yet also claim that this approach is part of a methodology that
helps us find out what’s true about the world, by allowing us to anchor ourselves
in a particular economic context. Used wisely, I think such theoretical posits em-
body strategies that help us find manifest (or experientially concretized) truths—​
truths that make a difference to our world of experience. The way it is doing so,
then, can’t be hashed out in terms of deductive informativeness (however way we
spell that out). Such non-​XT statements figure as central to theory. For example,
Campbell and Mankiw do not discard the Euler equation approach from their
toolbox in lieu of a more realistic equation. Nor is it an idealization that simply
needs to be filled out with more detail in order to account for empirical data.
They use it, in fact, to construct a larger generalized framework around it. It re-
mains central to their reasoning, and such statements are never quite eradicated
from the science. But their primary role is not to deductively impute truth.32
Chao was right to emphasize the persistence of the Euler equation. I am
suggesting that the role of such equations is methodological rather than repre-
sentational; they serve to locate and articulate structure and serve as diagnostic
tools. They are fixtures that are central to analysis. When it comes to complex sys-
tems, we had to identify scale-​dominant behavior in order to diagnose the ways

31 If anything, I would argue that the Euler equation has a role in shaping the complex of XT

statements, so that its role is prior to the question of whether or not a particular model represents
phenomena accurately. To put it in Lyons’s terms, we could understand the Euler equation as an in-
strument that even helps “remedy evident XT-​deficiencies by increasing the number—​and/​or the
extent, degree, or exactitude of the experiential concretization—​of XT statements” (2011, 330).
32 Lyons (2017) anticipates the kind of suggestion that I’m making.

[P]‌atently false posits can serve toward “bringing the truth” of high level posits “down” to
statements that describe experiences. Once novel predictions are derived, even from false
hypotheses, and new data is gathered, then, on this view, those mediating blatantly false
constituents previously deployed are eliminated in want of increasing the experientially
concretized truth of the accepted system. (3213)
So false posits can have a central role, making the distinction between our positions a subtle matter.
But for us the posit is never really eliminated. This is because (for instance) Mankiw and Campbell do
not attain their results on the basis of false assumptions; rather, the posit fully serves as a foundation
for the model they eventually articulate, so it is in effect built into the final product they offer. But
I think ultimately that the difference between our positions isn’t so much of a disagreement so much
as we are simply pointing at the nuanced guises that posits take in scientific reasoning.
276 Contemporary Scientific Realism

the actual behavior of a system deviates from the idealized model. We might start
an enquiry by examining whether the Euler equation fits our population-​wide
data; when it fails to do so, we must figure out where else (if anywhere) it can have
explanatory or descriptive power—​where those causal relationships stipulated
by the equation actually hold. The Euler equation is more like a navigational tool
(that helps us answer questions such as, where does it hold, and where does it
fail to hold? Why not?). And the more we are able to diagnose different kinds of
deviations from the ideal and able to tell when that benchmark is or isn’t the ap-
propriate one to use, the more our science progresses. In a way, economists too
are in the business of seeking and maximizing manifest truth.
Readers will note that this view, given its endorsement of Hoover’s perspec-
tival realism, also shares kinship with Chang’s (2016) pragmatic realism, where
the pragmatic refers to the practical aspect of activity:

I define (pragmatist) coherence as a harmonious fitting-​together of actions

that leads to the successful achievement of one’s aims. Such coherence may
be exhibited in something as simple as the correct coordination of bodily
movements needed in riding a bicycle . . . or something as complex as the
successful integration of a range of material technologies and various ab-
stract theories in the operation of the global positioning system (GPS). A co-
herent epistemic activity achieves its aim well, and avoids performative
self-​contradiction. (112–​113)

With this notion of coherence in hand, Chang (2017) formulates his coherence
theory of reality as follows: “a putative entity should be considered real if it is
employed in a coherent epistemic activity that relies on its existence and its basic
properties (by which we identify it)” (15).
So for the economist, a coherent activity might be implementing a bit of policy,
which in turn requires conducting a bit of counterfactual analysis in a model.
Policymaking would require an economist to take seriously how behavior at the
macroscale might change in response to changes at other scales. I might want
to know how overall consumption changes when the underlying distribution of
agents changes. That is, exercises of counterfactual reasoning about complex sys-
tems behavior rely on the existence of a (multiscale) framework. While some-
thing like the Euler equation need not be considered real, it partakes in framing a
coherent epistemic project—​this coherent multiscale framework where goings-​
on at different scales must coordinate with one another. The scale-​specific
structure that we discover in our investigation, insofar as it partakes in such a
framework, is what is real. Grasping the causal structure of the world is partly a
matter of trying to figure out how fundamental equations like the Euler equation
fit in and help articulate such a framework.
 The Case of the Consumption Function 277

This modest realism is nothing new. What I am proposing here is that such
a realism supplies what’s missing in Lyons’s axiological account—​the epistemic
oomph of what is experientially concretized as true. Experientially concret-
ized truth can be legitimate, plain old truth that we can have at hand—​there’s
nothing that is truth per se in the pragmatic account apart from what is expe-
rientially concretized. Positing the Euler equation is the beginning of an enter-
prise that is open to the possibility that multiple perspectives of a system have to
be taken into account, and that the parcels of information we obtain from dif-
ferent perspectives must cooperate with one another. While how we place those
perspectives (and they very well may be separate models) into a coherent frame-
work is a question for another time, that complex is one—​insofar as it allows us
to successfully conduct something like counterfactual analyses—​that is apt for a
realist interpretation.
A final note. I have been deliberate in my noncommittal attitude to whether
or not the realism I care about is epistemic or ontic. This is because for the most
part, the distinction doesn’t bear much on the suggestion I’m trying to make.
Structural realism is, in both senses, a positive and a negative thesis. It is a posi-
tive thesis about what we can know, and what there is, and by parity about what
we do not know, and what there isn’t. My thesis is simply that there is structure
that we can be realists about, namely causal structure, and that the Euler equa-
tion is one tool that economists use to uncover it. And given the pragmatic na-
ture of our discussion, in that our knowledge of the economy involves how to
do things, the simple answer is: I am a realist both about what we can know and
about what there is at the same time, though I do not commit here to the neg-
ative thesis that that’s all we can know or that that’s all that there is. The realist
attitude outlined here simply tries to get right how economists actually go about
doing things, insofar as we conceive of economics as a policy science. This prag-
matic view emphasizes the coherence of the multi-​scale causal structure, where
structure occupies a fairly fundamental place in economists’ thinking. At least,
given our investigation of economic methodology, we shouldn’t take as granted
the priority of (knowledge of) individuals over structure.

12.5 Against Reiss Against Realism

A few remarks before I conclude. I don’t consider the Euler equation itself to
be an approximation of the truth, nor something that is a pit stop for a model
that’s developing toward a future truer version of itself. The Euler equation aids
us in finding the truth, but in itself is not representational—​so it is not an ap-
proximation of anything. If the Euler equation fails to “fit the data” it may be for
a number of different reasons; that the population is heterogeneous is a causally
278 Contemporary Scientific Realism

relevant feature about the structure of the population of interest. That is, the
Euler equation’s purpose is more diagnostic than representational.
Because this realism looks somewhat different from standard forms of re-
alism, which are grounded on correspondence views of truth, the defenses that
it can provide will look a bit different, too. In this section, we consider Reiss’s
(2012) criticism of realism; he argues that instrumentalism has at least three
advantages over it. I argue that we can sidestep those criticisms.

The first criticism: “Once we have usefulness, truth is redundant” (370).

The redundancy shows up in our account in an automatic kind of way. If useful-

ness is what we mean by pragmatic coherence, then truth trivially follows be-
cause we conceived of truth in terms of being able to undertake coherent activity.
So truth is built in to our notion of success rather than something vestigial.

The second criticism: “There is something disturbing about causal

structure” (372).

Reiss worries that “if we seek predictive success, we don’t necessarily want to
build causal models” (373). I do not claim anywhere that predictive success is the
mark of a successful model, nor does a model need be an accurate representation
in order to be true. Reiss maintains that “a model is not predictively successful
qua representing causal structure” (373–​374). In fact, I agree! Reduced-​form
models can be perfectly suitable for prediction in some cases. And this is because
a good reduced form model is, in fact, the collapsed form of a structural model.
So the relevant structure is incorporated into the reduced-​form construction.33
And in order to use a model for diagnostic purposes, á la counterfactual rea-
soning, it ought to be causal.
Reiss’s larger worry is that “even when we are able to establish causality, to
make a causal model practically useful, additional facts about the represented
relations have to be discovered. But once we’ve discovered the additional facts, it
is irrelevant whether the relation at hand is causal or not” (374). That is, the fact
that a relationship is causal is irrelevant as to whether it is successful or not. Given
that diagnostic success for us is conceived in terms of possible (and possibly ac-
tual!) interventions, this seems implausible. If the structure of the economy is
something that counts as a difference-​maker to its overall behavior, then I con-
sider that a causal contribution.

33 Every structural model gives rise to different reduced forms, and there might be observational

equivalence between some of them, but there remain legitimate questions about which one is a good
one. These questions are going to ask which ones have got the structure right (where structure is un-
derstood in my sense!).
 The Case of the Consumption Function 279

The third criticism: “It’s better to do what one can than to chase rainbows” (375).

The final worry is that “building causal models has enormous informational
requirements” (375), and so a different heuristic that one might use is what is
known as Marschak’s Maxim. Marschak (1953) observed that for many questions
of policy analysis there is no need to identify fully specified models that are in-
variant to whole classes of policies. If the researcher instead focuses on particular
interventions, all that may be needed are combinations of subsets of the struc-
tural parameters—​those required to identify the effect of the policy intervention.
There’s nothing stopping a structural realist of my stripe from abiding by
Marschak’s Maxim.34 Think of the analogy with materials. We don’t need all
the information about all the atoms in the steel beam—​we do, however, need to
know about grain and how that granular structure affects macroscale behavior.35
The engineer’s version of Marschak’s Maxim would tell us to focus our attentions
at this scale. Something similar is the case in economics. It’s not necessary to
uncover all the individual preferences of individuals to get a sense of what con-
sumption in the aggregate looks like or all the causes that might be at play in
order to assess the effect of a particular disturbance to a system. We may need
to know something about the distribution of preferences, and finding this infor-
mation is a matter of finding the right perspective—​that is, the appropriate scale.
And insofar as it is specific to scale and partakes in a multi-​scale framework that
enables us to undertake counterfactual exercises, I call that structural and real.

12.6 Conclusion

I’ve argued thus far for a modest structural realism that accommodates at least
some of the ways that economists actually deploy idealization in their search for
the real causes underlying dynamic behavior. Making sense of these practices
necessitates our moving away from correspondence notions of truth to coherence
ones; aligning our realism with Lyons’s axiological realism cannot be achieved
otherwise. But a pragmatic spin on realism allows us to recognize the central
methodological role that idealizations have to play in obtaining (in particular

34 In Marschak’s 1953 Cowles paper, he actually describes what can be recognized as the Lucas

Critique. One way in which they differ is that Marschak is not concerned with expectations in the
way Lucas is. Marschak’s point, nonetheless, is that we need invariance—​but the notion of being in-
variant is relative (as nothing is invariant to everything)—​something that Lucas himself seems to
accept! This theme reaches back to Haavelmo and Frisch (see Aldrich 1989).
35 And the range of circumstances under which a particular granular configuration will persist; for

instance, with extremely high temperatures we can induce grain growth, which in turn affects macro-
scale properties by making steel more elastic.
280 Contemporary Scientific Realism

causal) knowledge. It is not the Euler equation that is real but the structure that
the Euler equation helps us uncover and use to explain economic behavior.

Acknowledgments

I am grateful to audience members at Duke’s Center for the History of Political

Economy for their helpful discussion on an earlier draft of this paper. In addition,
I am especially indebted to Hsiang-​Ke Chao, Kevin Hoover, Timothy Lyons, and
an anonymous referee for their extensive and invaluable comments.

References
Aldrich, J. (1989). Autonomy. Oxford Economic Papers, 41(1), 15–​34.
Attanasio O. P. (1999). Consumption. In, Taylor, J. B., & Woodford, M. (eds.) Handbook of
Macroeconomics, Vol. 1B. Amsterdam: Elsevier Sci, pp. 741–​812.
Batterman, R.W. (2013). The Tyranny of Scales. In, The Oxford Handbook of Philosophy of
Physics. Oxford: Oxford University Press, pp. 255–​286.
Blinder, A. S., Deaton, A., Hall, R. E., & Hubbard, R. G. (1985). The time series consump-
tion function revisited. Brookings Papers on Economic Activity, 1985(2), 465–​521.
Brady, D. S., & Friedman, R. D. (1947). Savings and the income distribution. In, Studies
in Income and Wealth, Vol. 10. National Bureau of Economic Research, pp. 247–​265.
Bronfenbrenner, M. (1948). The consumption function controversy. Southern Economic
Journal, 14(3), 304–​320.
Bursten, J. (2018). Conceptual strategies and inter-​theory relations: The case of nano-
scale cracks. Studies in History and Philosophy of Science Part B: Studies in History and
Philosophy of Modern Physics, 62, 158–​165.
Campbell, J.Y. and Mankiw, N.G. (1991). The response of consumption to income: A
cross-​country investigation. European Economic Review, 35(4), 723–​756.
Campbell, J. Y., & Mankiw, N. G. (1990). Permanent income, current income, and con-
sumption. Journal of Business & Economic Statistics, 8(3), 265–​279
Campbell, J. Y., & Mankiw, N. G. (1989). International evidence on the persistence of eco-
nomic fluctuations. Journal of Monetary Economics, 23(2), 319–​333.
Canzoneri, M. B., Cumby, R. E., & Diba, B. T. (2007). Euler equations and money market
interest rates: A challenge for monetary policy models. Journal of Monetary Economics,
54(7), 1863–​1881.
Carroll, C. D. (2001). Death to the log-​linearized consumption Euler equation! (And
very poor health to the second-​order approximation). Advances in Macroeconomics,
1(1), 1–​38.
Carroll, C. D., Slacalek, J., & Tokuoka, K. (2014). The distribution of wealth and the
MPC: Implications of new European data. American Economic Review, 104(5), 107–​11.
Cate, T. (ed.). (2013). An encyclopedia of Keynesian economics. Edward Elgar Publishing.
Chang, H. (2017). Operational Coherence as the Source of Truth. Proceedings of the
Aristotelian Society, 117(2), 103–​122.
 The Case of the Consumption Function 281

Chang, H. (2016). Pragmatic realism. Journal of Humanities of Valparaiso, 0(8), 107–​122.

Chao, H. K. (2007). A structure of the consumption function. Journal of Economic
Methodology, 14(2), 227–​248.
Chao, H. K. (2003). Milton Friedman and the emergence of the permanent income hypo-
thesis. History of Political Economy, 35(1), 77–​104.
Deaton, A. (1992). Understanding consumption. Oxford University Press.
Dennett, D. (1991). Real patterns. Journal of Philosophy, 88: 27–​51.
Duesenberry, J. S. (1949). Income, saving and the theory of consumption behavior.
Cambridge, Mass: Harvard University Press.
Estrella, A., & Fuhrer, J. C. (1999). Are “Deep” Parameters Stable? The Lucas Critique as
an Empirical Hypothesis, Working Papers 99-​4, Federal Reserve Bank of Boston; paper
available at: fmwww.bc.edu/​cef99/​papers/​efpaper.pdf
Estrella, A., & Fuhrer, J. C. (2002). Dynamic inconsistencies: Counterfactual implications
of a class of rational-​ expectations models. American Economic Review, 92(4),
1013–​1028.
Ferber, R. (1953). A study of aggregate consumption functions. New York: National Bureau
of Economic Research.
Flavin, M. (1985). Excess sensitivity of consumption to current income: Liquidity
constraints of myopia?. Canadian Journal of Economics, 18(1), 117–​136.
Friedman, M. (1957). The Permanent Income Hypothesis. In, Friedman, M. (ed.) A Theory
of the Consumption Function. Princeton, NJ: Princeton University Press, pp. 20–​37.
Friedman, M., and Kuznets, S. (1945). Income from independent professional practice.
New York: National Bureau of Economic Research.
Haavelmo, T. (1947). Methods of measuring the marginal propensity to consume. Journal
of the American Statistical Association, 42(237), 105–​122.
Haavelmo, T. (1944). The probability approach in econometrics. Econometrica: Journal of
the Econometric Society, Supplement (July 1944), iii–​115.
Hall, R. (1978). Stochastic implications of the life cycle-​ permanent income hypo-
thesis: theory and evidence. Journal of Political Economy, 86(6), 971–​987.
Hall, R. E. (1990). The rational consumer: theory and evidence. Cambridge, MA: MIT
Press. Chapter 7: ‘Survey of research on the random walk of consumption’, pp. 133-​158.
Hall R., & Mishkin F. S. (1982). The sensitivity of consumption to transitory in-
come: estimates from panel data on households. Econometrica, 50, 461–​481.
Harker, D. (2012). How to split a theory: Defending selective realism and convergence
without proximity. The British Journal for the Philosophy of Science, 64(1), 79–​106.
Hart, A. G. (1946). National Budgets and National Policy: A Rejoinder. The American
Economic Review, 36(4), 632–​636.
Heckman, J. J., & Vytlacil, E. J. (2007). Econometric evaluation of social programs, part
I: Causal models, structural models and econometric policy evaluation. Handbook of
econometrics, 6, 4779–​4874.
Hendry, D. F. & Mizon, G. E. (2005). Forecasting in the presence of structural breaks
and policy regime shifts. In, Andrews, D. W., and Stock, J. H. (eds.) Identification and
Inference for Econometric Models: Essays in Honor of Thomas Rothenberg. Cambridge,
UK. Cambridge University Press, pp. 480–​502.
Hoover, K. D. (2015) Reductionism in economics: Intentionality and eschatological jus-
tification in the microfoundations of macroeconomics. Philosophy of Science, 82(4),
689–​711.
282 Contemporary Scientific Realism

Hoover, K. (2011). Counterfactuals and Causal Structure. In, Illari, P. M., Russo, F., &
Williamson, J. (eds.) Causality in the Sciences. Oxford: Oxford University Press, pp.
338–​360.
Hoover, K. (2009). Microfoundations and the Ontology of Macroeconomics. In,
Kincaid, H., & Ross, D. (eds.) Oxford Handbook of the Philosophy of Economic Science.
Oxford: Oxford University Press, pp. 386–​409.
Hoover, K. (2001). Is Macroeconomics for real? In, Maki, U. (ed.) The Economic World
View: Studies in the Ontology of Economics. Cambridge: Cambridge University Press,
pp. 225–​45.
Havranek, T. (2015). Measuring intertemporal substitution: The importance of method
choices and selective reporting. Journal of the European Economic Association, 13(6),
1180–​1204.
Jhun, J. S. (forthcoming). Economics, equilibrium methods, and multi-​scale modeling.
Erkenntnis. https://​doi.org/​10.1007/​s10670-​019-​00113-​6.
Jhun, J. S. (2018). What’s the point of ceteris paribus? or, how to understand supply and
demand curves. Philosophy of Science, 85(2), 271–​292.
Keynes, J.M. (1936) The General Theory of Employment, Interest and Money.
Cambridge: Macmillan Cambridge University Press.
Kuznets, S. (1952). Proportion of capital formation to national product. The American
Economic Review, 42(2), 507–​526.
Kuznets, S. (1942). Uses of national income in peace and war. New York: National Bureau
of Economic Research.
Ladyman, J. (1998). What is structural realism?. Studies in History and Philosophy of
Science, 29(3), 409–​424.
Lucas, R. E. (1976). Econometric policy evaluation: A critique. In, Brunner, K., and
Meltzer, A. H. (eds.) The Phillips Curve and Labour Markets. Amsterdam: North
Holland, pp. 19–​46.
Lyons, T. D. (2005). Toward a purely axiological scientific realism. Erkenntnis, 63(2),
167–​204.
Lyons T.D. (2011). The problem of deep competitors and the pursuit of epistemically uto-
pian truths. Journal of General Philosophy of Science 42, 317–​0338.
Lyons, T. D. (2017). Epistemic selectivity, historical threats, and the non-​epistemic tenets
of scientific realism. Synthese, 194(9), 3203–​3219.
Lucas, R. E., & Sargent T. J. (1979). After Keynesian macroeconomics. Federal Reserve
Bank of Minneapolis Quarterly Review, 3(2), 1–​16.
Mäki, U. (2009). Realistic realism about unrealistic models. In, Kincaid, Harold, & Ross,
Don (eds.) The Oxford Handbook of Philosophy of Economics. New York: Oxford
University Press, pp. 68–​98.
Mankiw, N. G., Rotemberg, J. J., & Summers, L. H. (1985). Intertemporal substitution in
macroeconomics. The Quarterly Journal of Economics, 100(1), 225–​251.
Marschak, J. (1953). Economic measurements for policy and prediction. In, Hood, W.,
& Koopmans, T. (eds.) Studies in Econometric Method. New York and London: John
Wiley and Sons, 1–​26.
Modigliani, G., & Brumberg, R. (1954). Utility analysis and the consumption function: An
interpretation of cross-​section data. In Kurihara, K. K. (ed.) Post-​Keynesian economics.
New Brunswick, NJ: Rutgers University Press, pp. 388–​436.
Molana, H. (1993). The role of income in the consumption function: A review of on-​going
developments. Scottish Journal of Political Economy, 40(3), 335–​352.
 The Case of the Consumption Function 283

Morganti, M. (2011). Is there a compelling argument for Ontic Structural Realism?

Philosophy of Science, 78(5), 1165–​1176.
Mumtaz, H., & Surico, P. (2011). Estimating the aggregate consumption Euler Equation
with state-​dependent parameters (No. 8233). CEPR Discussion Papers.
Orcutt, G. H., & Roy, A. D. (1949). A bibliography of the consumption function. Cambridge
University, Dept. of Applied Economics.
Parker, J. A. (1999). The reaction of household consumption to predictable changes in so-
cial security taxes. American Economic Review, 89, 959–​73
Reiss, J. (2012). Idealization and the aims of economics: Three cheers for instrumentalism.
Economics & Philosophy, 28(3), 363–​383.
Ross, D. (2008). Ontic structural realism and economics. Philosophy of Science, 75(5),
732–​743.
Runkle, D. E. (1991). Liquidity constraints and the permanent-​income hypothesis: evi-
dence from panel data. Journal of Monetary Economics 27, 73–​98.
Shea, J. (1995). Union contracts and the life-​ cycle permanent income hypothesis.
American Economic Review, 85, 186–​200
Smithies, A. (1945). Forecasting postwar demand. Econometrica, 13, 1–​14.
Wilson, M. (2017). Physics avoidance: And other essays in conceptual strategy.
Oxford: Oxford University Press.
Worrall, J. (1989). Structural realism: The best of both worlds? Dialectica, 43(1–​2),
99–​124.
Woodward, J. (2005). Making things happen: A theory of causal explanation.
New York: Oxford University Press.
Yogo, M. (2004). Estimating the elasticity of intertemporal substitution when instruments
are weak. Review of Economics and Statistics, 86(3), 797–​810.
Zeldes, S. P. (1989). Consumption and liquidity constraints: an empirical investigation.
Journal of political economy, 97(2), 305–​346.
 13
We Think, They Thought
A Critique of the Pessimistic Meta-​Meta-​Induction
Ludwig Fahrbach

University of Duesseldorf
[email protected]

13.1 Introduction

The aim of this paper is to analyze and assess a certain argument used by antirealists
in the scientific realism debate, the pessimistic meta-​meta-​induction (two “meta”s).
Let scientific realism be the view that our current successful theories are probably
approximately true. Antirealists challenge realism by presenting the pessimistic
meta-​induction, PMI (one “meta”), according to which many theories in the past
of science were also successful for a while, but were refuted later on. Typically the
very first response by scientific realists to the PMI is that science has improved a
lot since the times of the past refuted theories and that these improvements block
the PMI and save realism.1 Antirealists then often reply that past realists could have
said the same thing, namely that science has improved a lot in the same manner as
realists claim for the recent past, but those improvements didn’t help past realists,
as the subsequent theory refutations show; hence, the recent improvements like-
wise don’t help current realists to block the PMI and to save realism, so the realists’
defense against the PMI fails. It is this argument by antirealists which is the focus of
this paper. I call it the pessimistic meta-​meta-​induction, PMMI.2 I will analyze the
PMMI and attempt to show that it does not succeed as a reply to the realist’s defense.

1 Stathis Psillos calls this response “the Privilege-​for-​current-​theories strategy.” (Psillos 2018, Sect.

2.8). Mario Alai calls it the “discontinuity strategy”: “there are marked differences between past and
present science,” which imply that “pessimism about past theories can[not] be extended to current
ones.” (Alai 2017, p. 3271/​67).
2 The term “pessimistic meta-​meta-​induction” is also used by Devitt (1991, p. 163) and by Psillos (2018,

Sect. 2.8). The PMMI is expounded by Wray (2013, 2018, Ch. 6) to whom this paper is a rejoinder, and Psillos
(2018). A related version directed at the realist defense based on the improvement of scientific methods is
criticized by Devitt (1991, p. 163/​4). See also Doppelt (2014, p. 286), Müller (2015), Stanford (2015, p. 3),
Alai (2017, p. 3282), Vickers (2018, p. 49), Rowbottom (2019, p. 476), and Frost-​Arnold (2018, p. 3).

Ludwig Fahrbach, We Think, They Thought In: Contemporary Scientific Realism. Edited by: Timothy D. Lyons and Peter
Vickers, Oxford University Press. © Oxford University Press 2021. DOI: 10.1093/​oso/​9780190946814.003.0013
 We Think, They Thought 285

The defense of realism just presented, which invokes recent improvements in

science to block the PMI and to which the PMMI is a reply, is currently not the
most prominent realist response to the PMI; rather that is selective realism. This
paper is not about selective realism, however, so I will only briefly note some
problems of selective realism, which are meant to show that it is not unreason-
able to explore alternative responses to the PMI.
Selective realism typically consists of two claims: Theories are required to
enjoy novel predictive success, and novel predictive success of a theory only
allows us to infer some sort of partial truth of the theory which concerns specific
types of parts of a theory, such as structure, entities, or success-​conferring parts.3
It has become quite clear that many past refuted theories also enjoyed novel pre-
dictive success, hence the defense of selective realism has to rely primarily on the
respective notion of partial truth.4 Such notions have likewise been extensively
discussed, but one problem has not received much attention.
Selective realism implies that currently accepted scientific theories contain
specific types of false parts which can be identified today and which can be ex-
pected to be replaced in future theory changes. However, selective realists have
rarely made any effort to actually determine those false parts of current theories.
They should do so. They should spread out into the scientific world, examine
the numerous statements accepted as scientific fact today, and try to find the
false ones. An important task would be the examination of scientific textbooks
containing what scientists take to be established scientific knowledge in discip-
lines such as chemistry, biology, geology, and so on. Selective realists should
check all statements in the textbooks, determine the parts that are idle, or about
non-​entities, or about non-​structure (nature, substance, content, interpreta-
tion, . . .), inform the textbook authors of their findings, and urge the correction
or deletion of the problematic statements. Of course, if they actually did so, they
would meet with surprise and resistance from scientists who deem the estab-
lished scientific knowledge to be confirmed by compelling empirical evidence
and would not be prepared to rewrite their textbooks.5
What this shows is that selective realism is highly revisionary. It is in deep
conflict with the judgments of practicing scientists. The conflict is not just the-
oretical, but may also concern practical matters such as decisions about the di-
rection of future research and concrete applications of scientific knowledge in
modern society. These short remarks have to be worked out in detail. Still, if one

3 For a list of versions of (selective) realism see Lyons (2017, p. 3215). For an overview of structural

realism see Frigg and Votsis (2011).

4 See, for example, Lyons (2002, 2006, 2017) and Vickers (2013, 2016, footnotes 6 and 10). Novelty

of predictions is here understood in the usual way, as use-​novelty: The prediction is not used in the
construction of the theory.
5 In Section 13.11 I will hint at some reasons why I think scientists are right on this matter.
286 Contemporary Scientific Realism

is bothered by the revisionary nature of selective realism, as I am, then one will
find it worthwhile to explore alternative responses to the PMI.
The dialectic in which the PMMI arises has to be developed with some care
(Sections 13.2 to 13.5, and 13.7). In Section 13.2 I define scientific realism and
present the PMI. In Section 13.3 I analyze the PMI. Then I discuss the realist’s
defense against the PMI, which consists in claiming that science has improved
a lot since the times of the refuted theories, and formulate three assumptions
on which the defense is based (Section 13.4). Finally Section 13.5 introduces
the PMMI. I analyze it, and discuss some objections (Section 13.6). Then comes
the central part of the paper. I show how the assessment of the PMMI is related
to the assessment of the PMI, and present my objections to the PMMI (Section
13.7 to 13.9). Section 13.10 deals with the charge of ad hocness. In Section 13.11
I briefly discuss how the PMI and the PMMI fare, if it is assumed that there has
been a large increase in degrees of success in the recent past. A short conclusion
summarizes the argument against the PMMI.
The dialectic can be developed in a number of different ways. Obviously
I cannot discuss every possible branch of the dialectic, but have to make some
choices and assumptions. Some assumptions constitute, or rely on, strong
simplifications. For example, as will become apparent later on, I rely on a strongly
idealized picture of the history of science. The assumptions and simplifications
are always meant to serve the goal of understanding and assessing the PMMI.
Thus I won’t necessarily use the most prominent or plausible versions of the
positions and arguments in the realism debate, but ones that are useful and con-
venient for the purposes of this goal.

13.2 Realism and the PMI

In this Section I set up the dialectic between the realist and the antirealist. Let
scientific realism be defined as the position that our current successful theories
are probably approximately true. (In the following I will omit “probably” and
“approximately.”) Examples of such theories are plate tectonics, the heliocen-
tric system, and the theory of evolution. There are many versions of scientific
realism, of course, but this definition is a fairly standard one. A theory enjoys
empirical success iff it has made sufficiently many sufficiently significant true
predictions and no important false predictions, in other words, iff it has passed
sufficiently many sufficiently severe tests and not failed any important tests. This
is not very precise, but will do for our purposes.
Scientific realists support their position with the No-​Miracles Argument
(NMA), according to which the success of our current successful theories would
be a miracle if they were false (Putnam 1978, Smart 1963, p. 39). Although this
 We Think, They Thought 287

argument is usually spelled out as an inference to the best explanation, I will only
work with the basic intuition of the NMA. Antirealists usually do not accept the
NMA, but considering challenges to the NMA other than the PMI would lead us
too far afield.
Antirealists attack realism by means of the PMI. I begin with a version of the
PMI that highlights the commonalities between the PMI and the PMMI, and
therefore looks a bit different from the ways it is usually presented. This version
starts with the observation that many realists in the past said the same thing as
today’s realists do: They also believed that their successful theories were true,
supporting their realism with arguments resembling the NMA. For example,
Clavius, Kepler, and Whewell defended realism about their theories by alluding
to the empirical success of their theories (Musgrave 1988, p. 229/​30). However,
so the attack goes, many of their successful theories were later refuted. For ex-
ample, the Ptolemaic system endorsed by Clavius was refuted, and Kepler relied
on all sorts of later-​abandoned assumptions about the sun and the planets (Lyons
2006). All theories just mentioned (heliocentric system, theory of evolution,
plate tectonics) had precursors that were successful to some degree and held
by many scientists to be true, but were later replaced.6 Thus, past realists were
proven wrong about their realism. It follows that today’s realists will probably
share the same fate as past realists, implying that realism about current successful
theories is likewise wrong. So, this is the PMI. In the next section we will see
that it is not essentially different from more familiar versions of the PMI. Wray
captures the spirit of the PMI nicely:

The Pessimistic Induction asks us to see ourselves as similar to the scientists of

the past. We are not to be Whigs or deluded, assuming that we are not prone to
make the same sorts of mistakes that they were prone to make. . . . The pessi-
mistic induction is . . . designed to aid us in recognizing the similarities between
our predicament and the predicament of our predecessors. (2018, p. 95/​6)

There is another objection to realism. Realism as just defined refers only to pre-
sent, but not to past successful theories. The reason is obviously that past suc-
cessful theories were often refuted, hence cannot be included in the scope of
realism. This invites the objection that this is not a good reason to restrict the
scope of realism to current successful theories. The realist has to offer an inde-
pendent reason for the restriction, or else the restriction is ad hoc, only serving

6 Other examples famously include the phlogiston theory, the caloric theory of heat, the ether

theory of light, and some more (see Laudan 1981 and the references in footnote 4). Note that the suc-
cessful refuted theories do not necessarily refer to unobservables, rather some refer chiefly or exclu-
sively to observables (Stanford 2006, Ch. 2, Fahrbach 2011, 2017, Sect. 3.3). For this reason the PMI is
also a threat to realism about theories of the latter kind.
288 Contemporary Scientific Realism

the purpose of making realism compatible with the existence of past refutations.7
This objection differs from the PMI since it does not use past refuted theories to
undermine realism directly, instead it only uses them to point out a weakness in
the definition of realism, namely an unprincipled restriction of the scope of re-
alism. I will discuss the charge of ad hocness in Section 13.10.
The PMI as just presented is an instance of a general argumentative strategy
on the antirealist’s side that has the following form: The target of each instance
of the strategy is some claim or piece of reasoning endorsed by the realist which
refers to current theories. The strategy consists in first observing that past realists
made, or could have made, the same claim (employed, or could have employed,
the same piece of reasoning) about past theories as the realist does about current
theories, and, second, pointing out that past realists were, or would have been,
proven wrong by later theory refutations or rejections. We are then invited to
conclude that the claim (or piece of reasoning) offered by today’s realist about
current theories is wrong as well. I will call any instance of this general strategy
an argument from the past. Different arguments from the past differ with respect
to which claim or piece of reasoning is inserted into the general form. The PMI
is obviously an argument from the past. It targets the realist claim that current
successful theories are true. Later we will encounter further arguments from the
past, in particular the PMMI.

13.3 Analyzing the PMI

I will now provide a detailed analysis of the PMI. This will pave the way for the
later analysis of the PMMI. The analysis consists of two remarks and a procedure
to simplify the PMI.
First remark. The PMI starts from the observation that many past realists said
the same thing as current realists do: they endorsed realism and supported it with
arguments like the NMA. The PMI goes on to state that past realists were proven
wrong by subsequent theory refutations and that today’s realists will share the
same fate as past realists. Now, past realists obviously didn’t say the very same
thing as current realists; rather they referred to different theories, namely past
successful theories. Furthermore, there is an inductive step at the end of the PMI
which carries us from the statement that past realists were wrong about past re-
alism to the conclusion that current realists are wrong about current realism.
This inductive step is obscured when the PMI states that past realists said “the

7 The charge of ad hocness has several variants, see Stanford (2006, p. 10), Devitt (2011, p. 292),

Vickers (2018, p. 52), and Fahrbach (2017).

 We Think, They Thought 289

same thing” as present realists, and present realists “will share the same fate” as
past realists. The inductive step will be examined more closely in Section 13.7.
Second remark. The claim of the PMI that past realists were shown to be
wrong by subsequent theory refutations should not be taken to mean that past
realists were unjustified or irrational when they endorsed realism for their the-
ories. After all, the relevant refutations could not be known by past realists
at their respective times, because they occurred at respective later times.
Generally, the conditions which determine whether a position held by some
people at some time is justified or not should be knowable by those people
at that time, and should not depend on empirical information which only
becomes available later on.8 Hence, the PMI should not be understood to be
talking about the justifiedness of the assertions of past realists, but about the
truth or falsity of those assertions. We learn empirically, using information ac-
quired later on, that their assertions were false. Whether or not their assertions
where justified can be left open. The PMI then invites us to infer that current
realism is also false, and this is intended to imply that endorsing current re-
alism is unjustified.
Let us now simplify the PMI. First, referring to present and past realists and
their respective assertions makes the argument more vivid by turning it into a
story about the misfortunes of real people, but this is not essential to the argu-
ment. Rather we can focus on the contents of the assertions and just talk about
theories and the properties of theories such as success, truth, and falsity. Then
the PMI is the following inference:9

Many past successful theories were refuted. (Premise)

Many past realisms are false. (Intermediate conclusion)
Current realism is false. (Final conclusion)

Second, the intermediate conclusion is likewise not essential to the argument

and can be removed. The refuted theories are the “negative thing,” the source
of all trouble, hence occur in the premise. Take them away and there is no argu-
ment. In contrast, the intermediate conclusion is just that: an intermediary for
transmitting the trouble to the final conclusion (the falsity of current realism).
Hence, instead of first using past refutations to confute past realisms and then

8 I use an internalistic notion of justification.

9 Single underlines indicate a deductive inference, double underlines indicate an inductive infer-
ence. Unless stated otherwise I assume that the inductive inferences we discuss are strong enough for
detachment, i.e., the premises justify believing the conclusion. In the intermediate conclusion, “past
realisms” is plural, since past theories were successful and accepted at different times. I mostly ignore
this complication.
290 Contemporary Scientific Realism

inferring therefrom that present realism is false, we can use past refutations to
undermine present realism directly. Then the PMI becomes:

Many past successful theories were refuted.

Current realism is false.

This is a more familiar version of the PMI.10 I will use it from now on. Note
that the two simplifications mean that it is not essential to the PMI that we “see
ourselves as similar to the scientists of the past” or that we “recognize the sim-
ilarities between our predicament and the predicament of our predecessors”
(Wray 2013).
If both the PMI and the NMA are still operative at the end of the dialectic,
i.e., neither is taken to be defeated by some argument or other, then they have
to be balanced with each other to reach a final verdict on realism. To simplify
the discussion, I will assume that the PMI always trumps the NMA.11 Hence as
long as the PMI is operative during the dialectic, the NMA can be ignored. If
the PMI is not operative at the end of the dialectic, e.g., judged to be blocked
by the improvements of science, then the realist may invoke the NMA to sup-
port realism, but doing so still faces the charge of ad hocness, which I discuss in
Section 13.10.
Both simplifications may also be applied to other arguments from the past.
Recall that arguments from the past are instances of the following general
strategy: The aim is to undermine a certain target claim made by a current re-
alist about current theories by noting that past realists made (or could have
made) the same claim about past theories, but were (or would have been) shown
wrong by later theory refutations. The conclusion we are meant to draw is that
the target claim is wrong as well. The first simplification consists in abstracting
from present and past realists and their respective assertions, and just working
with the contents of the assertions, which refer to theories and their properties.
The second simplification cuts out the intermediate conclusion that the claim
about past theories was undermined by the ensuing refutations. Instead past
refutations (possibly combined with further information) are used to undermine

10 Current realism is the claim that current successful theories are probably true. Hence the ne-

gation of current realism, which is the conclusion of the PMI, can be taken to be the statement that
many current successful theories are false. Then the PMI is the inference from “Many past successful
theories were refuted” to “Many current successful theories are false.”
11 This assumption is actually quite strong. The NMA is arguably equivalent to, or reducible to, a

first-​order argument about the support of theories by observation (see e.g., Bird 1998), while the PMI
is a second-​order argument (on the meta-​level), and one might hold that first-​order arguments are
generally stronger than second-​order arguments, or at least not that easily trumped by the latter. The
whole dialectic can be easily modified to accommodate a different judgement about the balancing of
the two arguments, although things get a bit more complicated.
 We Think, They Thought 291

the present realist’s claim about present theories directly. Later we will apply both
simplifications to the PMMI.

13.4 The Defense of the Realist

Realists have developed a number of responses to the PMI. One response is se-
lective realism, on which I commented earlier. In this paper I focus on another
defense, the one that gives rise to the PMMI. This defense claims that science has
improved a lot since the times of the refuted theories, and these improvements
block the PMI. There are several versions of this defense, which differ with re-
spect to the precise nature of the invoked improvements. One version claims that
the methods of science have improved.12 Other versions hold that current theo-
ries are better than past theories in this or that respect, e.g., have fewer rival the-
ories, have more “explanatory success,” are more mature, or have more empirical
success.13 I will use the last version, involving the notion of empirical success.
Then the realist defense against the PMI can be taken to make two claims: First,
present theories are significantly more successful than past refuted theories, and
second, this difference in success between present and past theories blocks the
PMI. Both claims will be explained over the course of the paper.
The realist’s defense is based on three assumptions. I will adopt them without
much discussion, because they are quite plausible, and defending them is not
our topic here. The first assumption is that empirical success comes in degrees.
Earlier I defined a theory to be successful if it makes sufficiently many sufficiently
significant true predictions and only negligible false ones. Accordingly we can
understand the degree of success of a theory to be determined by the number
and quality of true predictions of the theory, in other words, by the number and
severity of the empirical tests the theory has passed. I use the notion of degree of
success because the realism debate is usually framed with the notion of success,
but one could just as well use other, closely related notions from confirmation
theory, such as the empirical support enjoyed by a theory from empirical evi-
dence, or some notion of confirmation of a theory by the empirical evidence.14
The notion of degree of success may be identified with some probabilistic notion

12 “[N]‌ot only are scientists learning more and more about the world, but also . . . they are learning

more and more about how to find out about the world; there is an improvement in methodology.”
(Devitt 1991 p. 163). See also Devitt (2007, p. 287, 2011, Sect. 3), Roush (2009), and Wray (2018,
Ch. 6).
13 For fewer rival theories (“attrition”) see Ruhmkorff (2013, 412), for “explanatory success” see

Doppelt (2007), for maturity see Vickers (2018, 49), for empirical success see, e.g., Bird (2007, p. 80),
Devitt (2011, p. 292), and Fahrbach (2017). It is an interesting question how the different versions are
related to each other, but I won’t discuss that here.
14 Compare Vickers (2013, Sect. 3) and Vickers (2019, Sect. 4).
292 Contemporary Scientific Realism

such as the likelihood ratio Pr(O|T)/​Pr(O|¬ T) or the posterior Pr(T|O), but we

need not discuss such proposals here.
Generally the degree of success of a theory at some time15 depends on the
computing power available at the time (needed to produce predictions from
quantitative theories) and the amount and quality of the observations gathered
by scientists until that time, where the quality of the observations consists
in good-​making features such as diversity, specificity, precision, and so on.16
A theory that makes non-​negligible false predictions still counts as refuted.
The second assumption is that the accepted scientific theories have by and
large become continuously more successful over the history of science (except
when they were refuted). This is very plausible, because the ability to produce
predictions from theories has generally been growing due to more computing
power, better mathematical techniques to solve equations, and so on, and the
amount and quality of observations and experimental results has generally been
growing continuously due to better scientific instruments, more scientists, better
data-​gathering techniques, and so on. Observations were generally only lost or
forgotten when better observations became available. Given more and better
predictions from theories and more and better observations, tests of theories can
be more severe and diverse, leading to higher degrees of success for the theories
that pass the tests.
In addition to the amount and quality of the empirical evidence, the degree
of success of a theory could be taken to depend on further factors such as the
intrinsic properties of the theory, like simplicity and scope, and the nature of
the relation between the theory and the empirical evidence (entailment or prob-
abilistic relations). However, I will ignore these other factors here, because past
and present theories mostly don’t differ very much in these regards, certainly not
enough to be of help for scientific realism. If anything, later theories are often less
simple, more general, more precise, and more probabilified than earlier theories.
For example, quantum mechanics and general relativity are considerably more
complicated than Newtonian mechanics and Newtonian gravitational theory
(although the losses in simplicity and so on are usually far outweighed by the
gains in empirical support from true predictions, or so the realist defense has to
claim).
The third assumption is that different theories, for instance from different sci-
entific areas (with disjoint domains), can be compared with respect to degree of
success, at least roughly. Statements such as “The theory of evolution enjoys a

15 Strictly speaking, degrees of success are not ascribed to theories simpliciter, but to temporal

stages of theories.
16 If one thinks that novel predictive success comes in degrees (Vickers 2013, p. 196, 2016, Alai

2014, p. 298, 312), then one may construct a version of the realist defense against the PMI that relies
on degrees of novel predictive success. Such a defense also faces a PMMI-​like argument.
 We Think, They Thought 293

higher degree of success than phlogiston theory” are meaningful and justifiable.
The third assumption may be deemed especially plausible if the notion of de-
gree of success is identified with some probabilistic notion such as the likelihood
ratio. If one does not accept that theories from different specialties can be com-
pared with respect to degree of success, or only accepts comparisons inside some
discipline or subdiscipline of science, then one may pursue the whole discussion
at a more local level.
The increase in success over the history of science has obviously been far from
uniform. Different scientific areas started life at different times, and have been
developing at different rates. Hence, the theories accepted by scientists today
occupy a broad range of degrees of success. Some areas, especially in the nat-
ural sciences, have produced theories with enormous success, while many other
areas are stuck with theories at medium or lower levels of success. The realist
defense against the PMI as I am developing it here focuses on the theories with
the highest degrees of success. These are “our current best theories.” They consti-
tute the easiest kind of case for defending realism. Once secured, such a realism
may serve as a bridgehead for advancing to realist positions for harder cases, i.e.,
to develop appropriate (possibly weaker) realist positions for theories with me-
dium and lower degrees of success.
The realist’s defense against the PMI can now be taken to consist of two
claims: Our current best theories are significantly more successful than practi-
cally all refuted theories, and this difference in success blocks the PMI. The first
claim presupposes a way to delineate the set of “our current best theories,” a no-
tion of difference in success between theories, and an explanation of the term
“significant.” However these notions are understood, it is clear that the first
claim is a substantial empirical claim requiring support from the history of sci-
ence. I will return to all these issues later. The second claim (that the difference
in success blocks the PMI) means that the difference in success prevents past
refutations from undermining realism about our current best theories, i.e., the
inductive step of the PMI does not go through. Given the realist’s defense we can
finally formulate the PMMI.

13.5 The PMMI

Here is the PMMI:

Realists in the past could have reasoned in the same way as today’s realist does
in response to the PMI: “Our current best theories are significantly more suc-
cessful than past refuted theories, and this difference blocks the PMI.” But look
what subsequently happened, many of their theories were refuted. Hence, the
294 Contemporary Scientific Realism

reasoning of today’s realist fails, and the PMI is not blocked by the increase in
success.

Most people seem to find the PMMI to be intuitively quite compelling.17 However,
getting clear about its actual import is not so easy. To simplify the discussion let us
fix on a specific date at which the imagined realists of the past do their reasoning.
One plausible such time is the year 1900, at least for physics, because Newtonian
mechanics and the Newtonian theory of gravitation had been quite successful and
stable until around 1900, so that Max Planck famously got the advice, “in this field
[physics], almost everything is already discovered, and all that remains is to fill a
few unimportant holes” (Lightman 2005, p. 8). Then the imagined realists from
1900 don’t argue against our PMI, or PMItoday, but against a version of the PMI for
the year 1900, call it PMI1900. This leads to a second rendering of the PMMI:

Realists in 1900 could have reasoned in the same way as today’s realist does
in response to the PMItoday: “Our current best theories are significantly more
successful than past refuted theories, and this difference blocks the PMI1900.”
This reasoning by realists in 1900 would have been proven wrong by subse-
quent theory refutations. Hence the reasoning of today’s realist in response to
the PMItoday also fails and the PMItoday is not blocked by the increase in success.

The second rendering makes it clear that the realists of 1900 would not have
reasoned in exactly the same way as the realist does today, only in an analogous
way, namely at a lower level of success and targeting a different PMI. The PMMI
is obviously an argument from the past, i.e., an instance of the general antire-
alist strategy mentioned earlier: A piece of reasoning about current theories is
attacked by transferring it to the past and noting that the past version of the rea-
soning is undermined by subsequent theory changes.
The PMMI states that the reasoning of the imagined realists in 1900 “would
have been proven wrong” by subsequent refutations. This should not be taken
to mean that their reasoning would have been unjustified or irrational. The
refutations which proved their reasoning wrong occurred subsequently, hence
were not known in 1900. Therefore, the PMMI should not be understood as talking
about the justifiedness of the PMI1900, but about the reliability of the PMI1900.18 The
PMMI then invites us to infer that the PMItoday is likewise a reliable inference.19

17 Most references in footnote 3 seem to endorse the PMMI. Exceptions include Devitt and Psillos.
18 For our purposes we can define an inductive inference to be reliable, if its conclusion is true
given true premises. An inductive inference understood as a type is reliable, if the conclusions of its
instances are mostly true when its premises are true.
19 I assume that the statement that a given version of the PMI is not blocked by the increase in suc-

cess is equivalent to the statement that it goes through (is reliable).

 We Think, They Thought 295

We may also describe the PMMI as follows: Using historical information acquired
after 1900, we learn empirically which kind of meta-​inductive inference, pessi-
mistic or optimistic or neither, is reliable in 1900; we find that a pessimistic meta-​
induction is reliable and project the finding to the present. In other words, the
PMMI uses historical information to calibrate the meta-​inductive inference for
1900, and uses the result to calibrate the meta-​inductive inference for today. Note
that the conclusion of the PMMI is intended to mean that the realist’s assertion
today, that the PMItoday is blocked by the increase in success, is unjustified.
Like the PMI, the PMMI can be simplified in two steps. First, let the
PMMI refer to theories and their properties rather than to imagined people
and their assertions. Then the premise of the PMMI marshals the historical
evidence and can be taken to state that the past of science exhibits a certain
pattern, namely 〈refutations before 1900, successes until 1900, refutations
after 1900〉 (Figure 13.1). This premise supports the intermediate conclu-
sion that the PMI1900 is a reliable inference (is not blocked by the increase in
success until 1900), where the premise of the PMI1900 refers to the first and
second component of the pattern, and the conclusion of the PMI1900 refers to
the third component of the pattern. The intermediate conclusion supports
the final conclusion that the PMItoday is a reliable inference (is not blocked by
the recent increase in success).
Second, we can remove the intermediate conclusion from the PMMI. Then the
PMMI looks like this:

The past of science exhibits the pattern 〈refutations before 1900, successes
until 1900, refutations after 1900〉.                                   
The PMItoday is a reliable inference.

PMMI

PMI1900 PMItoday
refutations successes b1 refutations successes b2 refutations
l1 l2

1900 today

Figure 13.1 The pessimistic meta-​meta-​induction, PMMI. The depiction of the

history of science is strongly idealized, serving the goal of understanding the PMMI.
l1, l2, b1, and b2, the lengths and breadths of the two rectangles dubbed “successes,” are
explained in Section 13.9.
296 Contemporary Scientific Realism

So, instead of using the pattern of the premise to first calibrate the meta-​inductive
inference for 1900 and then using the result to calibrate the meta-​inductive in-
ference for today, we use the pattern of the premise to calibrate directly the meta-​
inductive inference for today. If this contraction of the PMMI is rejected, the
discussion that follows gets a bit more complicated, but yields essentially the same
conclusions.
Figure 13.1 invites a quick recapitulation of the whole discussion so far. The
realist is impressed by the successes of our current best theories (the second rec-
tangle from the right) and wants to infer realism about these theories. The antire-
alist is impressed by the previous refutations and recommends the projection of
theory failure to the future instead (the three rectangles to the right).20 The realist
replies that degrees of success have increased from past to present, blocking the
projection of theory failure. The antirealist responds that the whole situation al-
ready occurred in the past (the three rectangles to the left): First, past successes
also suggested realism about past theories; second, previous theory refutations
suggested the projection of theory failure instead; third, past realists could have
replied that the projection of theory failure is blocked by the increase in suc-
cess; and fourth, they would have been proven wrong by subsequent refutations,
showing that in 1900 the projection of theory failure would have been the right
thing to do. It follows (big curved arrow) that the projection of theory failure
goes through today. Unsurprisingly the big curved arrow will be a topic later.

13.6 The Double-​Counting of Evidence

In this section I discuss an objection to the PMMI which is not related to my

main critique, but is interesting and puzzling in its own right. Moreover, it
highlights further similarities between the PMI and the PMMI. The objection is
that endorsing both the PMI and the PMMI violates the principle that no piece
of evidence should be used twice in the assessment of a statement. The empirical
evidence about the occurrence of theory refutations in the history of science is
used twice over, for the premise of the PMI, and for the premise of the PMMI
which uses the pattern 〈refutations before 1900, successes until 1900, refutations
after 1900〉 to support the PMI. Using empirical information twice in this way is
fallacious, or so the objection goes.21

20 We can assume that science will go on in the same way in the future as in the past. This implies

that the conclusion of the PMI that realism is false is equivalent to the statement that many current
best theories will be refuted in the future, and similarly for the PMI1900.
21 The principle against the double-​counting of evidence may be seen as the complement of the

principle of total evidence, according to which every piece of evidence relevant for the truth of a state-
ment should contribute to our epistemic assessment of the statement. Together the two principles
 We Think, They Thought 297

One might respond to the objection by pointing out that using empirical evi-
dence twice is legitimate in this case, because it is used for two different purposes.
The existence of refuted theories is used, first, for the premise of the PMI, and
second, to support via the PMMI that the inference of the PMI is reliable, and
these are two different purposes. However, one may retort that using empirical
information twice in this manner is still dubious. One should not use the same
empirical information both as a premise for an inference and to help determine
what to infer from this premise (to calibrate the inference). What is more, both
purposes serve the same final purpose, namely to assess the conclusion of the in-
ference, in our case the conclusion of the PMI that current realism is false, so it is
true after all that one piece of evidence is used twice for the same purpose.
I want to offer two considerations indicating that this kind of double-​counting
may not be illegitimate in the case of the PMMI. First, the objection of double-​
counting of evidence can also be directed against endorsing both the PMI and
antirealism. Take antirealism to be the inference from the success of current suc-
cessful theories to their falsity.22 Let the PMI be the inference from the existence
of successful-​but-​refuted theories to the conclusion that antirealism understood
as an inference is reliable. (To simplify things I use a somewhat stronger ver-
sion of the PMI here than in the rest of the paper.) Then the observations that
are used to show that current successful theories are indeed successful—​which
is the premise of antirealism understood as an inference—​are also used both to
show that the successful-​but-​refuted theories were successful and to refute those
theories—​which is the premise of the PMI. Hence, these observations are used
both for the premise of antirealism and to determine via the PMI what to infer
from the premise of antirealism. They are used twice for the same final purpose,
namely to assess current successful theories (in the conclusion of antirealism un-
derstood as an inference). If we reject the PMMI for the reason that it uses the
same empirical evidence for both the premise of an inference and to determine
what to infer from the premise, then we should also reject the PMI for the same
reason. Given the intuitive plausibility of the PMI, this is a high price to pay.
Second, a version of the PMMI can be constructed that avoids the double-​
counting of evidence. Divide the set of scientific areas randomly into two subsets

assert that when assessing a statement every relevant piece of evidence should count at least once and
no more than once.

22 This sounds like an unduly strong version of antirealism, but it is not, if one assumes ordinary

Bayesian probabilities for example. The priors of typical scientific theories are very low, very near to
zero, because scientific theories are typically quite general and have numerous non-​negligible rivals.
Favorable observations eliminate many rivals, increasing the probabilities of the theories, but the
question is by how much. Antirealism can be taken to be the position that current successful theories
should still be expected to have many non-​negligible rivals not eliminated by the observations, hence
their posteriors, although higher than the priors, are still near zero, hence our current successful the-
ories should be judged to be false.
298 Contemporary Scientific Realism

of equal size. Use the first subset to run the PMI, i.e., use the theory refutations in
the histories of the scientific areas of this subset for the premise of the PMI and
let the conclusion of the PMI refer to the current theories of this subset. Use the
second subset for the premise of the PMMI, i.e., to calibrate the meta-​inductive
inference for the PMI. So, the premise of the PMMI refers to patterns such as
〈refutations before 1900, success until 1900, refutations after 1900〉 in the his-
tories of the second subset, and the conclusion of the PMMI states that the PMI,
which runs in the first subset, is reliable.
One may object that this solution comes at the price of weakening the prem-
ises of the PMI and the PMMI: the strength of both premises is halved, intuitively
speaking. However, this is not the case. The relevant quantity of both premises
is a frequency, the frequency of refutations in the two sets, i.e., the ratio of the
number of refuted theories in the respective periods of time to the number of
scientific areas. If the set of scientific areas is cut in half, the ratios will not change
much, because both the numerators (the number of refuted theories in the re-
spective periods of time) and the denominators (the number of scientific areas)
are cut in half. Hence, this proposal might work, and I will not hold the double-​
counting of evidence against the PMMI. However, in order to keep the discus-
sion simple let us put dividing all scientific areas into two subsets right out of
our minds.

13.7 Back to the PMI

Before assessing the PMMI, we have to take a closer look at the assessment of the
PMI. This is the aim of this section. The aim is not to reach an actual assessment
of the PMI, e.g., to argue on the realist’s behalf that the PMI is blocked; I only
want to get clear about which issues need to be addressed when assessing the
PMI, without actually addressing them. At the end of the section we will see how
the PMMI arises in this context.
The PMI of 13.3 states that many past successful theories were later refuted,
therefore current realism is false. It is an induction: its premise refers to once-​
successful theories and its conclusion refers to current successful theories. It uses
a binary notion of success where each theory at a given time is either successful
or is not.23 Then the induction proceeds inside the set of successful theories and
seems quite plausible. However, we now operate under the three assumptions
of the realist’s defense, in particular that empirical success is graded and that it

23 Most of the literature on the PMI starting with Laudan (1981) relies on a binary notion of (novel

predictive) success.
 We Think, They Thought 299

has increased over the history of science. Given these assumptions, assessing the
PMI becomes much more difficult.
At this point I will make a fourth assumption. It states that there is a reason-
able definition of a global difference in success between past refuted and pre-
sent best theories such that this difference is non-​zero. This assumption is quite
vague, but will do for our purposes.24 It is a substantial assumption about the
history of science, for which the realist has to provide empirical support, but the
realist defense is making a considerably stronger claim anyway, namely asserting
a “significant” difference in success between the two sets of theories, on which
more later. The antirealist may attack the fourth assumption, of course. She may
argue, for example, that a substantial number of refuted theories enjoyed degrees
of success comparable to the theories of the set of our current best theories, on
any reasonable definition of this set. This would imply, she could argue, that, cor-
rectly understood, the global difference in success between past refuted and cur-
rent best theories is zero, and the realist’s defense against the PMI fails. However,
this would amount to an objection to the realist’s defense that is different from
the PMMI, hence will not be pursued here.
The fourth assumption (that there is a non-​zero global difference in success
between past refuted and present best theories) has the important consequence
that the inductive step of the PMI is an extrapolation. It extrapolates the occur-
rence of theory failure along degrees of success from past to present levels of
success. It is then not obvious that the extrapolation goes through, that theory
failure can be extrapolated in this way.
The realist may reject the extrapolation of theory failure along degrees of
success altogether. He may argue that from the existence of refuted theories at
a given level of success one can only infer something about the assessment of
theories at that level of success, not about any higher levels of success. This would
mean that the PMI is blocked by any non-​zero global difference in success. The
antirealist may counter this move by arguing that theory failure can be extrapo-
lated along degrees of success, and this may then be the debate between the two
sides. However, I don’t want to pursue this branch of the dialectic; rather I will
assume that the realist grants that extrapolations of theory failure are possible
in principle. The question is only whether the extrapolation goes through in the
particular case at hand.

24 The global difference in success between past refuted and present best theories can be defined

in a number of ways. For example, start from today’s most successful theory overall, then descend
degress of success until sufficiently many theories for a set of “current best theories” have been
gathered. Determine the lowest degree of success of the theories in this set. Below that, determine
the highest degree of success above which there are only a negligible number of later refuted theories.
The difference between these two degrees of success may be taken to define the global difference in
success between past refuted and present best theories.
300 Contemporary Scientific Realism

Whether this extrapolation goes through will depend on a number of factors.

For example, it may depend on the frequency of past refutations, on the severity
of past refutations (whether past refuted theories were completely false or only
partly false), and on further aspects of the distribution of theory stability and
theory change over degrees of success. To simplify the discussion, I will focus on
just one factor: the global difference in success between past refuted and current
best theories. Then it is plausible that the bigger the global difference, the smaller
the bearing of past refutations on the assessment of our current best theories.
Let us call a global difference in success “small” if it allows the extrapolation of
theory failure, and “big” if it blocks the extrapolation of theory failure.
It is not hard to see that big differences in success are possible. Imagine a fu-
ture scenario in which science will keep growing and degrees of success keep
increasing in similar ways in the next 100 years as they did in the past of science.
Imagine that scientific data, instruments, techniques, computing power, and so
on, will keep improving at the same pace as they have done until now. Then the
extrapolation of theory failure from our past to the levels of success of the year
2100 is surely very implausible. If the increase in success expected for the next
100 years doesn’t seem high enough to block the extrapolation, then surely there
are increases in degrees of success, 1000 years or 1 million years from now, that
would block the extrapolation.
It is also worth considering the maximal level of empirical success that is in
principle possible for scientific theories at which they accord with all possible
empirical evidence. If there are any false theories at that level of success in some
scientific field, then we have underdetermination of theories by all possible evi-
dence in that field. Hence, if past refutations supported the existence of false the-
ories at such a maximal level of success, then they would support the occurrence
of underdetermination of theories by all possible evidence. However, as far as
I am aware, nobody in the literature on underdetermination has argued in this
way to support underdetermination, using a PMI-​like argument referring to past
refutations. The arguments offered in the literature are of a quite different sort,
either using artificially constructed empirically equivalent rivals to theories, or
referring to concrete examples of incompatible empirical equivalent theories in
fundamental physics.25 This indicates that nobody thinks that past refutations
can be extrapolated that far along degrees of success.
The future scenario case and the underdetermination case show that mere
quantitative differences in success can suffice to block the PMI. Such differences
presumably count as “fundamental” differences in the sense of Wray:

25 See for example Norton (2008).

 We Think, They Thought 301

[I]‌f realists are to blunt the threat of the Pessimistic Induction, they must iden-
tify some significant difference between today’s theories and past theories.
Without an argument to the effect that there is a fundamental difference be-
tween the theories we currently accept and the once successful theories we have
since rejected, we have little reason to believe that today’s theories will not end
up on the pile of ruins to which Poincaré drew attention. (2018, p. 93, emphasis
in original)

We have established that there are differences in success for which the extrapo-
lation of theory failure is highly implausible. However, we are interested in the
difference between past and present levels of success and today’s PMI. The ques-
tion is whether this difference is small, so as to allow the extrapolation of theory
failure to present levels of success, or big, so as to block the extrapolation. As the
discussion so far already suggests, this is a hard question. To answer it we have
to engage with difficult empirical and epistemological issues. We have to find
some way to actually measure or estimate the degrees of success of past refuted
and current successful theories, at least roughly. Then we have to find a suitable
delineation of the set of present best theories. Then we have to settle on a suitable
definition of global difference in success between past and present best theories,
and have to determine this difference. Finally we have to formulate and defend
a judgment concerning whether the global difference is small or big, whether it
allows or blocks the extrapolation of theory failure. It seems that we can only as-
sess the PMI, if we tackle these difficult issues.26
The realist defense against the PMI as presented here does not engage with the
difficult issues. It just claims that the global difference in success between past
and present theories is “significant” (meaning “big,” i.e., capable of blocking the
PMI), without telling us how the difference was determined, and why it is “signif-
icant.” Hence, the realist has some work to do here. He has to engage with the dif-
ficult epistemological and empirical issues to defend his claim that the difference
is “significant.” What about the antirealist? She wants the PMI to succeed. Hence,
she has to show that the difference in success between past and present theories
is small, i.e., does not block the extrapolation of theory failure. To show this she,
too, has to engage with the difficult epistemological and empirical issues. As long
as she does not do that, she, too, has some unfinished business—​or so it seems.
It is at this point that the antirealist may offer the PMMI. The PMMI supports
that the PMI goes through, that is its conclusion, hence it promises to give
the antirealist an advantage, possibly decisive, over the realist in the debate

26 And I didn’t even mention the other factors that may be relevant for deciding whether theory

failure should be extrapolated to current levels of success, and the NMA which I assume for the sake
of simplicity to be trumped by the PMI.
302 Contemporary Scientific Realism

concerning the PMI. What is more, it avoids the difficult epistemological and
empirical issues of the PMI, because it looks at the past of science to establish
the reliability of the PMI. But I will now argue that despite first appearances, the
PMMI does not succeed in giving the antirealist an advantage in the debate.

13.8 First Objection to the PMMI

I will offer two objections to the PMMI. The first concerns the inference, the second
the premise of the PMMI. The first objection focuses on the inductive step at the
end of the PMMI (the curved arrow in Figure 13.1). Given the fourth assumption,
that past refuted and present best theories are separated by a non-​zero global dif-
ference in degrees of success, the inductive step of the PMMI is an extrapolation
from past to present levels of success. It extrapolates the pattern 〈refutations before
1900, successes until 1900, refutations after 1900〉 to the present to support the PMI.
Then the PMMI faces problems very similar to those faced by the PMI. It is not ob-
vious that the extrapolation goes through; it depends on the size of the difference
between past and present levels of success (as well as on other factors, which I ignore
here). Let us call the difference “small” if the extrapolation of the PMMI succeeds
and “big” otherwise. Once again the very existence of big differences can be backed
by considering future scenarios in which science will develop in a similar manner
in the next 100 or 1000 years as it did up to now, producing an extremely big in-
crease in success. For such an increase it is highly plausible that the extrapolation of
the PMMI fails. The question, though, is whether the increase in success so far has
been small or big. Once again this is a hard question. To find an answer, we have to
engage with difficult empirical and epistemological issues: we have to determine the
degrees of success of the involved theories, define and determine the global differ-
ence, and formulate and defend a judgment whether the global difference in success
is small or big, i.e., whether the extrapolation of the PMMI is plausible or not. Only
by engaging with these issues can we assess the PMMI, or so it seems.
The antirealist may evade the difficult issues of the PMMI in the same way
as she evaded those of the PMI: She may present a PMMMI. What then looms
is an infinite hierarchy PMI, PMMI, . . . PMnI, . . ., in which the assessment of
each PMnI faces difficult epistemological and empirical issues, which the anti-
realist evades by offering another argument from the past, the PMn+1I, which
invokes the past of science to establish the reliability of the PMnI. In this way the
antirealist avoids the difficult epistemological and empirical problems at every
level, but it is a bit unclear whether such an infinite hierarchy of arguments is
legitimate in principle.27 Anyway, it does not succeed for practical reasons: The

27 A related infinite hierarchy of arguments and counterarguments arises as follows: The realist
 We Think, They Thought 303

premise of the PMI refers to the occurrence of past refutations. The premise of
the PMMI is more complicated, invoking patterns such as 〈refutations before
1900, successes until 1900, refutations after 1900〉. The premise of the PMMMI
refers to even more complicated patterns such as

〈〈refutations before 1850, successes until 1850, refutations between 1850

and 1900〉,
〈refutations between 1850 and 1900, successes until 1900, refutations after
1900〉〉.

(The conclusion of the PMMMI is the statement that the inference of the PMMI,
the curved arrow in Figure 13.1, goes through.) As n increases, the premises
of the PMnI become more and more complex, presupposing more and more
episodes of theory refutation. But the number of such episodes in the history of
science is finite. Hence, for some level N, the antirealist will not be able to rise to
the level N+1. It follows that she has to face the difficult empirical and epistemo-
logical issues at some level.28 She may choose the level, say level n, with n ≤ N, but
she cannot avoid the difficult empirical and epistemological issues altogether;
at some point she has to engage with them. What is more, it does not seem ad-
visable to go up the hierarchy of PMnI’s very far, because the complexity of the
premises grows exponentially as n increases, making it ever harder for the anti-
realist to show that the premise of the chosen PMnI is actually true (compare the
next section). So let us return to the assessment of the PMMI.
My first objection to the PMMI is then as follows. The PMMI promised to
help the antirealist to gain an advantage in the debate over the PMI. It seemed to
show that the extrapolation of the PMI goes through by invoking patterns such
as 〈refutations before 1900, successes until 1900, refutations after 1900〉, while
avoiding the difficult epistemological and empirical issues arising in the assess-
ment of the PMI. But we just saw that the inductive step of the PMMI is also
an extrapolation along degrees of success, whose assessment faces very similar
problems as the assessment of the extrapolation of the PMI. Thus, invoking the
PMMI just trades the problems of the PMI for very similar problems that are at

objects to each PMnI by arguing that it is blocked by the increase in success, and the antirealist
responds with the PMn+1I according to which realists in the past could have objected to the past ver-
sion of the PMnI in the same way, namely that it is blocked by the increase in success, but would have
been proven wrong by later theory refutations. This infinite hierarchy has the same problem as the
one in the text, namely the exponentially increasing complexity of the premises of the PMnIs.

28 Note that the antirealist only has to show for one level n that the PMnI succeeds, because the con-

clusion of the PMnI is that the PMn-​1I succeeds, which has the conclusion that the PMn-​2I succeeds,
and so on, until the PMMI shows that the PMI succeeds, which has the conclusion that realism is
false. In contrast, the realist has to argue that all PMnIs fail.
304 Contemporary Scientific Realism

least as hard as those of the PMI. It follows that the PMMI does not succeed in
helping the antirealist to gain an advantage in the debate over the PMI.

13.9 Second Objection to the PMMI

For the second objection, consider the PMMI in its unsimplified form. It starts
by claiming that people in the past, e.g., in 1900, could have reasoned in the same
way as today’s realist does. They, too, could have said: “Our current best theo-
ries are significantly more successful than past refuted theories, and this dif-
ference blocks the PMI1900.” This beginning of the PMMI presupposes that the
situation in 1900 was actually similar in relevant respects to the situation today.
That is a substantial claim, which needs empirical support from the history of
science. Likewise for the simplified PMMI, whose premise refers to the pattern
〈refutations before 1900, successes until 1900, refutations after 1900〉. It is a sub-
stantial empirical claim that the first and second component of the pattern are
sufficiently similar in relevant respects to the corresponding situations in the re-
cent past. More generally, the PMMI requires the premise that there were times
in the history of science in which the situation was sufficiently similar in relevant
respects to the situation today.29
Which respects are relevant for the similarity between the past and present
situations? I will examine two respects, the global difference in success between
the respective sets of best theories and the respective sets of refuted theories, and
the number of the respective scientific fields. Then the PMMI requires that there
were times in the history of science in which the global difference in success
and the number of relevant fields were both comparable to the corresponding
global difference and number in the recent past. Figure 13.1 makes clear what
I mean: the two rectangles dubbed “successes” should have roughly similar
lengths l1 and l2 (which represent the respective global differences in success)
and roughly similar breadths b1 and b2 (which represent the respective num-
bers of scientific fields with best theories). If, on the other hand, the lengths and
breadths of successes in the past were never similar to those today, for example, if
l1 was always much smaller than l2, and b1 was always much smaller than b2, then
the premise of the PMMI is never true.30
We have identified a substantial empirical assumption of the PMMI
about the existence of periods of time relevantly similar to the recent past.

29 I want to thank Ioannis Votsis for a number of challenging comments at this and other portions

of my argument.
30 If at some time l is smaller than l , but b is bigger than b , and some suitable function of l and
1 2 1 2 1
b1 such as their product l1∙b1 is roughly the same as that of l2 and b2, then the two situations may also
justifiably count as similar.
 We Think, They Thought 305

Earlier, we took 1900 to be a suitable time, for the reason that central the-
ories of physics had been quite successful and stable for a while up until
around 1900. But the realist may argue that this is only then-​f undamental
physics; the simultaneous stability of theories in recent times has been much
broader, encompassing the best theories of a far bigger number of scien-
tific fields (mostly from the natural sciences) than just then-​fundamental
physics. The antirealist either has to show that theory stability occurred on
a much broader scale in 1900 than just then-​f undamental physics, or she has
to present some other historical periods at which the increase in success and
the breadth of stability was comparable to today. Once again we find that
the antirealist, if she wants to hold on to the PMMI, has some unfinished
business.
It is worth noting that both objections may also be directed against other
arguments from the past. Recall that arguments from the past aim to undermine
a target claim made by a realist about current theories by noting that past realists
could have made the same claim about past theories, but would have been
proven wrong by later theory refutations. One objection consists in the challenge
to show that there actually were situations in the past of science that were suffi-
ciently similar in relevant respects to the situation today. The antirealist cannot
just claim that this is so, but has to provide adequate evidence that such situations
really occurred in the past. The other objection is that there is one respect in
which all past situations are dissimilar to the present situation, namely with re-
spect to degrees of success. If the fourth assumption (about a non-​zero global dif-
ference between past and current best theories) is correct, then the inductive step
of any argument from the past will be an extrapolation from past lower to current
higher degrees of success. Whether the extrapolation is plausible will depend on
several factors, such as the global difference in success between past and present.
The antirealist has to show that the factors are such as to allow the extrapolation
of theory failure.

13.10 The Charge of Ad Hocness

In this section I discuss the charge of ad hocness. In my definition of realism

the scope of realism is restricted to current successful theories, at the exclusion
of past successful theories. The obvious reason is that past theories were often
refuted. The charge of ad hocness states that defining realism in this way is ad
hoc, merely serving the purpose of making realism compatible with the exist-
ence of past refutations. The realist has to offer independent reasons why the def-
inition of realism excludes past successful theories. To simplify the discussion in
this section, I will set the PMI aside, ignoring the question whether the existence
306 Contemporary Scientific Realism

of past refutations can be used to attack realism about current theories. Only the
seemingly arbitrary restriction of realism to current theories is at issue.
The realist may offer a similar response to the charge of ad hocness as to the
PMI: our current best theories are more successful than past refuted theories,
and this difference in success provides an independent reason to treat past theo-
ries differently from our current best theories in the definition of realism. To this
move Stanford responds: “More success gives no reason to believe that we have
now crossed over some kind of threshold . . . such that these predictive powers
[of scientific theories] are now finally substantial enough [to accept them as
true]” (2009, 384 footnote, emphasis added). At this point we need to reintro-
duce the NMA intuition, as it is generally taken to be the source of support for
realism. Then it is plausible that the NMA intuition gets stronger as degrees of
success increase: the higher the success of a theory, the bigger the “miracle” if the
theory were false, and the higher the probability that the theory is true. It follows
that Stanford’s claim that more success gives us no reason to believe that we have
crossed over the threshold for truth is false. Rather, more success gives us some
reason.31 The question is, only, whether the reason is strong enough to support
realism as defined above.
At other places Stanford’s wording is a bit more cautious. In his (2015), he says
“little reason”:

Of course there are always important differences between each successive gen-
eration of theories (including our own) and their historical predecessors in a
given domain of scientific inquiry, but there seems little reason to think that
such differences are sufficiently categorical to warrant the conviction that con-
temporary scientific theories have now finally managed to more-​or-​less sort
things out at last, given that the same inference as applied to earlier theories,
predicated on the salient advances and advantages of those theories over their
predecessors, has turned out to be so repeatedly and reliably mistaken. (p. 3,
emphasis added)

“Little reason” is more cautious than “no reason,” but it is still clearly meant to
imply “not enough reason.” To support this claim, Stanford merely offers an ar-
gument from the past (starting with “given that” in the quote), which runs into
the problems noted earlier.32 He does not offer any first order reasons concerning

31 Even if the NMA intuition is not granted, more empirical success of a theory means that the

theory has passed more tests, which generally implies, on standard accounts of confirmation such
as Bayesianism, that the theory is incrementally confirmed. Hence, more generally, such additional
success offers some reason to think that we have crossed the threshold for truth.
32 It is not clear whether one should interpret Stanford’s argument from the past as the PMI, the

PMMI, or a further argument from the past concerning only ad hocness. Maybe it is a combination of
these. In any case, it runs into the problems noted earlier.
 We Think, They Thought 307

the quantity and quality of the empirical evidence, and the strength of the empir-
ical support for our current best theories. Hence, concerning first order reasons,
his claim is as unprincipled as the opposite claim of the realist that present levels
of success suffice for the NMA to support realism. Whether either claim is justi-
fied, or no claim at all can be justified, surely depends on how big the increase in
success from past to present has actually been. If it has been small, then Stanford’s
claim may be justified and the charge of ad hocness may be plausible, but if it has
been big, then not. Consider once again the future scenario in which degrees of
success keep growing for another 1000 years at the same rate as they have done so
far. In such a scenario the NMA would surely give us more than “little reason” to
believe that we have crossed over the threshold for truth.
But the question is, of course, whether current levels of success are high
enough for the NMA to establish realism. The realist answers this question in
the affirmative, but according to the charge of ad hocness the affirmative answer
requires a justification, which the realist has not yet given. To find a justification,
the realist must engage with difficult empirical and epistemological issues: he has
to determine and compare the degrees of success of past and present best the-
ories and formulate and defend judgments based on the NMA intuition about
the plausibility of restricting realism to present levels of success at the expense of
past levels of success. So, the realist finds himself in a situation not unlike with re-
spect to the PMI as detailed earlier. In the next section I will hint at one possible
way how the realist may proceed in this situation.33

13.11 The Big Increase

We saw at several points that the debate can only progress if we engage with diffi-
cult empirical and epistemological issues, namely compare the degrees of success
of past and present theories and formulate and defend judgments about the plau-
sibility of the PMI, the PMMI, and the charge of ad hocness.
Actually some such work has already been done.34 It strongly suggests that our
current best theories have enjoyed a very large increase in degree of success in
the recent past and swim in a sea of evidence today. One reason is, very roughly,
that the amount of scientific research in general has been growing at very high
rates in the recent history of science, which has led to huge improvements in

33 Alternatively, the realist may search for some deep epistemic difference between past refuted

and current best theories. However, I don’t think that there is such a magic bullet, which scientists
are unaware of, which is deeply buried in scientific practice, requiring the analytic digging powers of
philosophers for its unearthing. To my mind, there is just ordinary empirical evidence and ordinary
empirical success, but with large differences between different theories and different times, as I indi-
cate in the next section.
34 See, for example, Doppelt (2007, 2014), Park (2011), Mizrahi (2013), and Fahrbach (2017).
308 Contemporary Scientific Realism

computing power and the amount and quality of empirical evidence. This has
translated into ever-​increasing degrees of success for the best theories. While
acquiring their tremendous success, practically all of them have been entirely
stable. There are no convincing examples of theories that enjoyed comparable
successes and were later refuted. At several points in the discussion we imagined
scenarios in which science develops in the same manner in the next few centu-
ries as it has done so far. The work just mentioned shows that the state of science
today is not so far from what we imagined in those scenarios. (Thus, Figure 13.1
offers a totally distorted view of the history of science.) Much more work needs
to be done to develop this approach further. Here I will just assume that it is gen-
erally correct, and confine myself to noting the implications for the assessment of
the PMI, the PMMI, and the charge of ad hocness.
First, the large increase in success and simultaneous stability of our current best
theories implies that the extrapolation of theory failure from past to present levels of
success is highly implausible. Our current best theories are far more successful than
any past refuted theories; hence the existence of the latter should have no bearing on
our assessments of the former. So the difference in success indeed blocks the PMI.
Second, the large difference between past and present levels of success also
blocks the PMMI. It prevents the extrapolation of patterns like 〈refutations be-
fore 1900, successes until 1900, refutations after 1900〉 to the present. What is
more, it is highly doubtful that the premise of the PMMI can be made true. To
make it true we have to find times in the history of science in which the situa-
tion was sufficiently similar to the recent past, in which the then-​best theories
were similar in number and made comparable gains in success as in the recent
past. However, the recent past is quite unprecedented in this regard. No other
period comes anywhere close. (Again the picture of the history of science as
presented by Figure 13.1 is much distorted.) Thus, the PMMI does not succeed
in supporting the PMI and undermining the realist’s defense against the PMI.35
Third, the tremendous success of our current best theories makes the appli-
cation of the full NMA intuition to these theories very plausible: their success
would indeed be a miracle if they were false. By contrast, the success of past theo-
ries, whether refuted later on or not, does not suffice to trigger the full NMA intu-
ition: their success was far more moderate, and it was not such a big miracle when
they occasionally turned out to be false. So, the NMA fully supports the infer-
ence to truth for our current best theories, but does so only rather weakly for past

35 Similar remarks apply to the PMnIs with n > 2. Note that, the bigger and broader the recent

increase in success, the more the PMI faces competition from an optimistic meta-​induction, OMI,
which uses the recent stability of our current best theories to project the stability of these theories
into the future, and likewise for higher-​level optimistic meta-​inductions, OMnIs.
 We Think, They Thought 309

theories.36 It follows that the realist’s restriction of realism to current and future
highest levels of success is not ad hoc but supported by good reasons. What is
more, the big increase in success in recent times means that the realist need not
commit himself to a precise threshold beyond which success suffices for truth,
but can be very vague in this matter. All he needs to make plausible is that the
threshold is somewhere in the large interval of degrees of success between past
and present, and the NMA is able to do that. Thus, some of the difficult empirical
and epistemological issues turn out to be not so difficult after all, but are readily
solvable, because the recent increase in success has been so enormous.

13.12 Conclusion

Let me summarize my critique of the PMMI. (I ignore the previous section.)

According to the PMI, we should extrapolate theory failure from past to present
levels of success, thereby undermining realism about present theories. The re-
alist reacts to the PMI by claiming that our current best theories are significantly
more successful than past refuted theories, blocking the PMI. The antirealist
denies this. She claims that, despite the difference in success, past refutations are
severe enough to allow the extrapolation of theory failure to go through. In this
situation, either the realist is right and the PMI is blocked by the increase in suc-
cess, or the antirealist is right and the PMI succeeds. It then seems that the matter
can only be decided if the two sides engage with difficult empirical and episte-
mological issues: they have to determine and compare the degrees of success
of past refuted and current best theories and formulate and defend judgments
concerning the plausibility of the extrapolation of theory failure and the under-
mining of realism.
At this point the antirealist presents the PMMI in the hope of gaining an
advantage in the debate. The PMMI seems to avoid the difficult empirical and
epistemological issues while implying that the PMI succeeds. It states that past
realists could have reasoned in the same way as today’s realist just did, namely
that their theories were significantly more successful than earlier refuted the-
ories, blocking the PMI of their day, but this didn’t prevent further refutations
from happening; hence the defense of today’s realist fails and the PMI is not
blocked by the increase in success. We analyzed the PMMI and saw that it faces

36 Most of past refuted theories were not completely false, but contained a substantial amount of

truth (judged from today), as selective realists have shown in numerous case studies. Furthermore
many theories have been completely stable for centuries now, such as the heliocentric system, oxygen
chemistry, the reduction of temperature to mean kinetic energy, and so on (for more examples see
Bird 2007, 73).
310 Contemporary Scientific Realism

very similar problems as the PMI. It also involves an extrapolation from past to
present levels of success, which gives rise to empirical and epistemological is-
sues very similar to those of the PMI: determining the degrees of success of the
involved theories and reaching and justifying judgments concerning the plau-
sibility of the extrapolation. So, the PMMI trades one set of problems, those of
the PMI, for another set of very similar problems (as well as some additional
problems). Therefore, the PMMI does not succeed in helping the antirealist to
gain an advantage in the debate over the PMI. The upshot is that both realists and
antirealists can only make progress in the debate if they engage with the difficult
empirical and epistemological issues.

Acknowledgments

For discussion and support I would like to thank Claus Beisbart, Christopher
von Bülow, Matthias Egg, Christian Feldbacher, Berna Kilinc, Tim Lyons, Sam
Ruhmkorff, Yafeng Shan, Corina Strößner, Paul Thorn, Peter Vickers, Ioannis Votsis,
Brad Wray, two anonymous referees, and audiences in Lausanne and Durham.

References
Alai, M. (2014). “Novel Predictions and the No Miracle Argument.” Erkenntnis, 79(2),
297–​326.
Alai, M. (2017). “Resisting the Historical Objections to Realism: Is Doppelt’s a Viable
Solution?” Synthese, 194(9), 3267–​3290.
Bird, A. (1998). Philosophy of Science. Montreal & Kingston: McGill-​ Queen’s
University Press.
Bird, A. (2007). “What is Scientific Progress?” Noûs, 41(1), 64–​89.
Devitt, M. (1991). Realism and Truth. 2nd edn. Oxford: Basil Blackwell.
Devitt, M. (2007). “Scientific Realism,” in: Frank Jackson and Michael Smith (eds) The
Oxford Handbook of Contemporary Philosophy. Oxford: Oxford University Press,
767–​791.
Devitt, M. (2011). “Are Unconceived Alternatives a Problem for Scientific Realism?”
Journal for the General Philosophy of Science, 42(2), 285–​293.
Doppelt, G. (2007). “Reconstructing Scientific Realism to Rebut the Pessimistic Meta-​
Induction.” Philosophy of Science, 74, 96–​118.
Doppelt, G. (2014). “Best Theory Scientific Realism.” European Journal for Philosophy of
Science, 4(2), 271–​291.
Fahrbach, L. (2011). “How the Growth of Science Ended Theory Change.” Synthese,
180(2), 139–​155
Fahrbach, L. (2017). “Scientific Revolutions and the Explosion of Scientific Evidence.”
Synthese, 194(12), 5039–​5072.
Frigg, R., & Votsis, I. (2011). “Everything You Always Wanted to Know about Structural
Realism but Were Afraid to Ask.” European Journal for Philosophy of Science, 1(2),
227–​276.
 We Think, They Thought 311

Frost-​Arnold, G. (2018). “How to be a Historically Motivated Antirealist: The problem of

Misleading Evidence.” (ms)
Laudan, L. (1981). “A Refutation of Convergent Realism.” Philosophy of Science,
48(March), 19–​49.
Lightman, Alan P. (2005). The Discoveries: Great Breakthroughs in Twentieth-​Century
Science, Including the Original Papers. Toronto: Alfred A. Knopf Canada.
Lyons, T. D. (2002). “Scientific Realism and the Pessimistic Meta-​Modus Tollens,” in: S.
Clarke and T. D. Lyons (eds.) Recent Themes in the Philosophy of Science: Scientific
Realism and Commonsense. Dordrecht: Kluwer, 63–​90.
Lyons, T. D. (2006). “Scientific Realism and the Stratagema de Divide et Impera.” The
British Journal for the Philosophy of Science, 57(3), 537–​560.
Lyons, T. D. (2017). “Epistemic Selectivity, Historical Threats, and the Non-​Epistemic
Tenets of Scientific Realism.” Synthese, 194(9), 3203–​3219.
Mizrahi, M. (2013). “The Pessimistic Induction: A Bad Argument Gone Too Far.”
Synthese, 1–​18.
Musgrave, A. (1988). “The Ultimate Argument for Scientific Realism,” in: Robert Nola
(ed.) Relativism and Realism in Science. Dordrecht: Springer, 229–​252.
Müller, F. (2015). “The Pessimistic Meta-​ induction: Obsolete Through Scientific
Progress?” International Studies in the Philosophy of Science, 29(4), 393–​412.
Norton, J. (2008). “Must Evidence Underdetermine Theory?” in: M. Carrier, D. Howard,
and J. Kourany (eds.) The Challenge of the Social and the Pressure of Practice: Science
and Values Revisited. Pittsburgh: University of Pittsburgh Press, 17–​44.
Park, S. (2011). “A Confutation of the Pessimistic Induction.” Journal for the General
Philosophy of Science, 42(1), 75–​84.
Psillos, S. (2018). “Realism and Theory Change in Science.” The Stanford Encyclopedia of
Philosophy (Summer 2018 Edition), Edward N. Zalta (ed.), URL = <https://​plato.stan-
ford.edu/​archives/​sum2018/​entries/​realism-​theory-​change/​>.
Putnam, H. (1978). Meaning and the Moral Sciences. Boston: Routledge and Kegan Paul.
Roush S. (2009). “Optimism about the Pessimistic Induction,” in: P. D. Magnus and M.
Busch (eds.) New Waves in Philosophy of Science. Palgrave MacMillan, 29–​58.
Ruhmkorff, S. (2013). “Global and Local Pessimistic Meta-​Inductions.” International
Studies in the Philosophy of Science, 27(4), 409–​428.
Rowbottom, D. P. (2019). “Scientific Realism: What It Is, the Contemporary Debate, and
New Directions.” Synthese, 196(2), 451–​484.
Smart, J. J. C. (1963). Philosophy and Scientific Realism. London: Routledge and
Kegan Paul.
Stanford, P. K. (2006). Exceeding Our Grasp: Science, History, and the Problem of
Unconceived Alternatives. Oxford University Press.
Stanford, P. K. (2015). “‘Atoms Exist’ Is Probably True, and Other Facts That Should Not
Comfort Scientific Realists.” Journal of Philosophy, 112(8), 397–​416.
Vickers, P. (2013). “A Confrontation of Convergent Realism.” Philosophy of Science, 80(2),
189–​211.
Vickers, P. (2018). “Historical Challenges to Realism,” in: Juha Saatsi (ed.) The Routledge
Handbook of Scientific Realism. London: Routledge, 48–​60.
Vickers, P. (2019). “Towards a Realistic Success-​to-​Truth Inference for Scientific Realism.”
Synthese, 196(2), 571–​585.
Wray, K. B. (2013). “The Pessimistic Induction and the Exponential Growth of Science
Reassessed.” Synthese, 190(18), 4321–​4330.
Wray, K. B. (2018). Resisting Scientific Realism. Cambridge: Cambridge University Press.
 14
The Paradox of Infinite Limits
A Realist Response
Patricia Palacios1 and Giovanni Valente2

1Department of Philosophy

University of Salzburg
[email protected]
2Department of Mathematics

Politecnico di Milano
[email protected]

14.1 Introduction

Scientific realism is a central topic in philosophy of science. Although there are

many different formulations of this concept in the literature, most scientific
realists are committed to the idea that we have good reason to believe that the
content of our best scientific theories, regarding both observable and unob-
servable aspects of the world, is true or at least approximately true. According
to most realists, scientific realism involves a semantic dimension, according to
which one is committed to a literal interpretation of scientific claims about the
world (Chakravartty 2017). Perhaps the strongest argument in favor of scientific
realism is the “no miracles argument,” (NMA) which asserts that the success of
our well-​established scientific theories would be a miracle if the content of such
theories were not true or at least approximately true (Putnam 1975; Boyd 1983).
The notion of “approximate truth” plays an important role in current approaches
to scientific realism, since it is widely held, even by realists, that our best scien-
tific theories are likely false, strictly speaking. Important technical work has been
developed to make the notion of “approximate truth” precise and we will address
part of this work in the present paper.
As plausible as it is, a scientific realist position faces various difficulties that
have cast doubt on the no miracles argument and have motivated an anti-​
realist attitude toward our most successful theories (van Fraassen 1980; Rosen
1994). One of the most outstanding challenges for scientific realism is the use

Patricia Palacios and Giovanni Valente, The Paradox of Infinite Limits In: Contemporary Scientific Realism.
Edited by: Timothy D. Lyons and Peter Vickers, Oxford University Press. © Oxford University Press 2021.
DOI: 10.1093/​oso/​9780190946814.003.0014
 The Paradox of Infinite Limits 313

of idealizations in scientific theories, which involves the assumption of fictional

systems that are intended to resemble the real-​world systems we are interested in
(Godfrey-​Smith 2009). An important example of idealizations in physics is the
use of “infinite idealizations,” which involves the introduction of infinite systems
that can be constructed by means of mathematical limits that are invoked to ex-
plain the behavior of target systems, notwithstanding the fact that the latter are
considered to be finite according to our most successful background theories.1
The problem with infinite idealizations, as it has been presented in the litera-
ture, is that it apparently leads one to what we refer to as the “Paradox of Infinite
Limits,” which poses a challenge to scientific realism. Informally, the intended
paradox can be formulated as follows. On the one hand, a scientific realist must
believe that real physical systems are finite, as it is indeed suggested by some of
our most successful background theories such as the atomic theory of matter
and general relativity. On the other hand, she must believe that the content of
scientific theories invoking infinite idealizations is true or approximatively true
insofar as they are indispensable to recover empirically correct results. Allegedly,
that calls scientific realism into question.2
This paradox was first introduced by Callender (2001) in the context of phase
transitions and then further discussed, for instance, by Butterfield (2011),
Norton (2012), and Shech (2013).3 Recently, interesting attempts to generalize
the problem for scientific realism have been made by Baron (2019), Liu (2019),
and Shech (2019). However, the extent to which such proposals provide a defi-
nite solution is still open. Indeed, it remains unclear under what conditions the
assumption of infinite limits leads one to paradoxes of this form. The present
paper aims to offer a general formulation of the Paradox of Infinite Limits. In the
attempt to show how it can be resolved, we elaborate a taxonomy of the different
uses of infinite limits in physics, which is partially based on distinctions made by
Norton (2012) and Godfrey-​Smith (2009). Specifically, we point out that when
being understood as approximations and abstractions, infinite limits do not
pose any substantial problems to scientific realism. Yet, when they give rise to

1 It is important to point out that one can also “construct” infinite systems without taking any

limits such as with the assumption of an infinitely long cylinder or two dimensional systems.
Although such examples deserve attention and have been discussed in the philosophical literature
(e.g. Earman 2017; Shech 2018), we will restrict our analysis to infinite idealizations resulting from
the use of infinite limits.
2 For completeness, let us clarify that in the literature one can find at least three different versions

of scientific realism: Explanatorianism, Entity realism and Structural realism (Chakravartty 2017).
In this paper we consider a general characterization of scientific realism, leaving the question of how
these particular versions can deal with infinite idealizations for future work. However, we point out
that the Paradox of Infinite Limits is especially problematic for Explanatorianism (Kitcher 1993;
Psillos 1999), which recommends a realist commitment with respect to those parts of our best theo-
ries that are indispensable to explaining empirical success.
3 See also (Fletcher, Palacios, Ruetsche, and Shech 2019) for a recent collection of papers on infi-

nite idealizations in science.

314 Contemporary Scientific Realism

infinite idealizations they can actually lead one to the Paradox of Infinite Limits,
depending on whether the idealization is regarded as essential for the explana-
tion of the physical phenomenon under investigation. We then argue that, even
in the case of essential idealizations, there are ways of coping with the alleged
incompatibility between infinite idealizations and scientific realism, which ulti-
mately rely on empirical considerations.
We organize the paper as follows. In Section 14.2, we distinguish between
idealizations, approximations, and abstractions broadly constructed. This dis-
tinction is partially based on Norton’s (2012) distinction between idealizations
and approximations and on Godfrey-​ Smith’s (2009) distinction between
idealizations and abstractions. In the following section, we formulate the general
Paradox of Infinite Limits and we explain in greater detail the sense in which it
raises a challenge for scientific realism. In particular, in sub-​section 14.3.3 we
state the condition of Empirical Correctness in precise terms, and then we pro-
ceed to analyze the various possible understandings of the use of infinite limits
in physics. Specifically, in Section 14.4 we develop the concept of approxima-
tion and argue that it is not problematic from a scientific realist perspective by
means of concrete examples: in fact, the paradox does not arise in the case of
approximations without idealizations, whereas it can be readily disarmed in the
case of idealizations yielding approximations. In Section 14.5, we address the use
of infinite limits as essential idealizations, explaining why that would lead the
scientific realist to a paradox. Yet, by focusing on the controversial example of
first-​order phase transitions, we show that there is available a procedure to dis-
pense the infinite idealization “on the way to the limit,” thereby avoiding a con-
tradiction with the claim that real target systems undergoing the phenomenon
to be explained are finite. Finally, in the last section, 14.6, we address the case of
continuous phase transitions to illustrate the use of infinite limits as abstractions,
and we also explain why these limits would not raise any conflict with scientific
realism.

14.2 Idealizations, Approximations, and Abstractions

14.2.1 Preliminary Concepts

In the scientific practice it is ubiquitous to, so to speak, “modify” the systems

encountered in the world with the goal of making our theories computationally
tractable, pedagogically accessible, or explanatorily rich. The representation of
real planes as frictionless planes in which objects can move uniformly and per-
petually is a prototypical example of this practice, in which real systems are mod-
ified with the purpose of making the theory manageable. Looking at the specific
 The Paradox of Infinite Limits 315

ways in which real systems are represented in scientific theories, one can distin-
guish at least between three different strategies labeled by philosophers of sci-
ence as idealizations, approximations, and abstractions.
Although there is no consensus about the nature of scientific idealizations,
most philosophers agree that idealizations involve a misrepresentation of a real
system (i.e. the target system) driven by pragmatic concerns, such as mathe-
matical tractability. Recent accounts provide a more precise characterization of
idealizations by arguing that they always refer to imaginary systems considered
to be analogues of the real-​world systems of interest. For example, Godfrey-​
Smith (2009, p. 47) says “I will treat [idealizations] as equivalent to imagining the
existence of a fictional thing that is similar to the real-​world object we are inter-
ested in.” In a similar vein, Norton (2012, p. 209) defines an idealization as “a real
or fictitious system, distinct from the target system, some of whose properties
provide an inexact description of some aspects of the target system.” According
to these authors, there will be an idealization only when there is reference to a
novel (fictional) system, which has properties that resemble the properties of
real-​world systems. In other words, when the misrepresentation of the proper-
ties of a target system coincide with the exact properties of a fictional system.
An example of idealizations is the frictionless plane just mentioned. This fic-
tional system was firstly introduced by Galileo to derive the equations of motion
of an object moving down an inclined plane. Although no such planes exist in
reality, they have proven to be extremely useful to predict the behavior of real
world systems. Why can these fictional systems explain the behavior of con-
crete systems observed in the world, and how can we justify their use from a
scientific realist perspective? The standard justification for idealizations of this
kind is that they provide approximately true descriptions of real world systems,
where approximation to truth is simply understood as a relation of similarity be-
tween the properties of fictional systems and the properties of real world systems
(Godfrey-​Smith 2009). It is also believed that these idealizations are dispensable,
in the sense that it is possible in principle to de-​idealize the theory by systemati-
cally eliminating distortions and by adding back to the model details of concrete
systems (McMullin 1985).4
This characterization of idealizations puts us in position to distinguish
them from other strategies of theory construction such as approximations and
abstractions. For instance, according to Norton (2012), approximations are in-
exact descriptions of certain properties of the target system, which are given in
terms of propositions expressed in the language of a theory, that do not need to
correspond to the relevant properties of some other fictional system. In fact, the

4 This kind of idealization corresponds to what has been called “Galilean idealization” (e.g.

McMullin 1985; Weisberg 2007)).

316 Contemporary Scientific Realism

use of approximations does not require one to make reference to any new system
different from the target systems. In this sense, one can say that approximations
involve distortions or misrepresentations of the target system but, differently
from the case of idealizations, these distortions are merely propositional. For
Norton (2012), idealizations can be demoted to approximations by discarding
the idealizing system and extracting the inexact description, but the inverse
promotion will not always succeed. However, we will see that there seem to be
cases of essential idealizations in which idealizations cannot be easily demoted
to approximations.
In recent years, the topic of essential idealizations has generated a great deal
of excitement among philosophers of science. In particular, it has been argued
that infinite idealizations arising from mathematical limits that are ineliminable
cannot provide approximations of realistic systems, because the latter exhibit be-
havior that is qualitatively different from the behavior exhibited by the idealized
infinite systems (e.g. Batterman 2005). We will examine some possible cases of
ineliminable idealizations in Section 14.5, but before doing so, let us mention
some examples of approximations and abstractions.
Let us recall that an approximation is an inexact description of the target
system that does not (necessarily) involve the introduction of a fictional system.
This means, there is no appeal to fictional system in which the inexact properties
of the target system are true. A good example of an approximation without ideal-
ization also mentioned by Norton (2014) is the case of a mass falling in a weakly
resisting medium. As we know, the speed v of a falling mass at time t is given by:

dv / dt = g − kv ,

where g is the gravitational acceleration and k the friction coefficient. The speed
of the mass at time t, as it starts from rest, is given in terms of the Taylor expan-
sion series by:

v(t ) = ( g / k )(1 − exp(−kt )) = gt − gkt 2 / 2 + gk 2t 3 / 6 − ...

If we assume that the friction coefficient is small, we can approximate the pre-
vious expression by taking only the first term of the series expansion:

v(t )= gt

When we use this expression to describe the behavior of a real mass falling in a
resisting medium, we do not intend to give a literal description of the situation,
 The Paradox of Infinite Limits 317

but rather to give a good approximation of it. This strategy allows us to simplify
problems that may be otherwise intractable. It is important to note, however, that
one can promote this approximation to an idealization by introducing a fictional
system in which a body falls under the same gravity in a vacuum, so that the fall
is described exactly by v(t) = gt (Norton 2014).
So understood, idealizations can also be distinguished from abstractions.
According to GodfreySmith (2009), an abstraction is the act of “leaving things
out while still giving a literally true description of the target system” (p. 48). In
contrast to idealizations, abstractions do not intend to state claims that are lit-
erally false and do not make reference to fictional systems.5 Instead, they in-
volve the omission of a truth by leaving out features considered to be irrelevant.
Abstractions also differ from approximations in that the former do not involve
propositional misrepresentations of the target system whereas the latter do.

14.2.2 Approximate Truth in the Context of Idealizations,

Approximations, and Abstractions

We will argue next that all the forms of inaccurate representations mentioned in
the previous section, i.e. idealizations, approximation, and abstractions, can be
made compatible with the more relaxed realist framework that accepts the con-
tent of scientific theories to be at least approximately true. However, in order to
arrive at that conclusion, we need to offer a more precise definition of the notion
of “approximate truth.” Chakravartty (2010) distinguishes between three kinds
of approaches for approximate truth in the standard philosophical literature: the
verisimilitude approach, the possible-​world approach, and the type hierarchy
approach. The verisimilitude approach, elaborated by Popper (1972), consists
in comparing the true and false consequences of different theories. In the pos-
sible word approach, which is meant to be an improvement of the verisimilitude
approach, the truth-​likeness is calculated by means of a function that measures
a mathematical “distance” between the actual world and the possible worlds in
which the theory is strictly correct (Tichy 1976; Oddie 1986) so that one can gen-
erate an ordering of theories with respect to truth-​likeness. Finally, in the type hi-
erarchy approach, truth-​likeness is calculated in terms of similarity relationships
between nodes that represent concepts or things in the word (Aronson 1990).
As Chakravartty (2010) points out, the problem that all these approaches have
in common is that they do not pay attention to the different ways in which scien-
tific representations give inaccurate account of the target systems. This is an im-
portant limitation of these approaches, because the notion of approximate truth

5 A similar view is defended by Jones (2005)

318 Contemporary Scientific Realism

is best understood differently in different circumstances, especially in cases of

idealizations yielding approximations, essential idealizations, approximations
and abstractions.
Let us discuss first the notion of approximate truth in the case of idealizations.
As said, idealizations involve a misrepresentation of a real system (i.e. the target
system) by means of introducing an imaginary system considered to be an ana-
logue of the real-​world systems of interest. Here we follow (Chakravartty 2010,
p. 40) in considering that the adequate notion of approximate truth concerns
the degree to which this fictional system that successfully captures aspects of the
target system resembles a non-​idealized representation of that system. In some
cases, especially when the idealization is a limit case of the de-​idealized system,
the degree of resemblance can be specified mathematically so that one can even
quantify the degree of misrepresentation. All cases of idealizations yielding
approximations can be put in this category and will be discussed in greater detail
in the next sections. These cases do not represent a challenge for scientific re-
alism since the notion of approximate truth can be adequately quantified.
A similar notion of approximate truth can be given in the case of approxi-
mation that do not involve idealizations. The difference is that degree of resem-
blance should not be evaluated between the properties of a fictional system and
those of a target system, but rather between the misrepresentation of the prop-
erties of the target system and the actual properties of the real target system. In
cases like the example of a mass falling in a weakly resisting medium, we can
quantify the degree of misrepresentation by considering the terms of the series
expansion that have been taken into account. In fact, more accurate descriptions
will imply incorporating more terms of the series expansion. The more terms we
consider, the better the approximation of the real properties of the target system
will be. In this sense, approximations are straightforwardly compatible with re-
alism, in that it is possible to quantify the degree of misrepresentation.
The challenge for scientific realism comes instead from the possibility of es-
sential idealizations, which appear to be ineliminable to the explanation of a cer-
tain phenomenon and cannot be de-​idealized toward more faithful explanations.
In other words, they cannot be demoted to approximations. In these cases, there
do not seem to be a straightforward notion of approximate truth and the degree
of resemblance between the fictional system and the target system do not seem to
be easily quantifiable. We will discuss such cases in Section 14.5.
Finally, let us refer again to abstractions. As said earlier, abstractions omit
details that are considered to be irrelevant but do not involve any misrepresen-
tation of the target system. For example, in trying to describe the behavior of a
cannonball that has been fired on some particular day, there are innumerable
features that will not be taken into account such as the composition of the can-
nonball, its color, its temperature, or the mechanism by which the initial velocity
 The Paradox of Infinite Limits 319

is conferred to the ball. The scientific model that predicts where the cannonball
will land may include a number of distortions with respect to other properties
like the gravitational force, which is generally assumed to have the same magni-
tude and direction at all points of the trajectory. However, such a model does not
involve misrepresentation with respect to the properties that do not make a dif-
ference for the occurrence of the phenomena. In fact, the model is simply silent
about them, and hence it does not say anything false as regards these irrelevant
properties (Jones 2005). Although this strategy does not involve misrepresenta-
tion of certain properties it does give an inexact description of the target system
because it leaves out factors that are irrelevant for the behavior under considera-
tion. One then needs an appropriate notion of approximate truth in this case too.
The intended notion will differ from the one involved in cases of idealizations and
approximations in that there is no misrepresentation of properties. Chakravartty
(2010, p. 39) suggests an articulation of the notion of approximate truth qua
abstractions connected with the notion of comprehensiveness: “The greater the
number of factors built into the representation [i.e. the more comprehensive is
the description], the greater its approximate truth.” In so far as one can quantify
these factors, there will be a precise notion of approximate truth applying also to
the case of abstractions. It is important to note that sometimes abstractions may
be even crucial or essential for the explanation of certain classes of behavior.6 For
instance, if we want to explain the behavior of cannonballs in general, we should
not include details that are specific of a certain fire. This is possible just because
abstractions generally aid in the explanation of certain phenomena by enabling
us to account for some common behavior that is generated among disparate sys-
tems (see also Weisberg 2007).

14.3 The Paradox of Infinite Limits: A Challenge

for Scientific Realism?

Mathematical limits are vastly used in physics. For instance, in the statistical me-
chanical theory of phase transitions, it is assumed that the number of particles as
well as the volume of the system goes to infinity; similarly, in the ergodic theory
of equilibrium and in the explanation of reversibility in thermodynamics it is
assumed that time goes to infinity. Appealing to infinite limits has the technical
advantage that they render the formal account of physical phenomena more trac-
table. Furthermore, in some cases, like in the examples we just mentioned, it even

6 This account of abstraction is consistent with Cartwright’s (1994, p. 187) view, according to

which an abstraction is a mental operation, where we “strip away—​in our imagination—​all that is
irrelevant to the concerns of the moment to focus on some single property or set of properties, as if
they were separate.”
320 Contemporary Scientific Realism

appears as a necessary condition to offer a mathematical description of real target

systems in agreement with the empirical results. Nevertheless, when the variable
growing to infinity corresponds to a physical parameter, e.g. the number of mi-
croscopic constituents, taking the limit introduces an unrealistic assumption, at
least as long as real systems are believed to be finite. This raises many conceptual
questions. For instance: How can we explain the empirical success of models that
introduce such an unrealistic assumption? To what extent are models that use in-
finite limits compatible with scientific realism? In this section and the reminder
of this paper we will address these questions based on the previous distinction
between idealizations, approximations and abstractions.

14.3.1 A Taxonomy for Infinite Limits

The use of infinite limits has been frequently equated with the use of an infi-
nite idealization. However, we want to stress a distinction between different uses
of infinite limits as approximations, idealizations, and abstractions. In order to
draw such a distinction, we begin by setting up the formal framework within
which our discussion is cast. Let Sn represent a system characterized by some
physical parameter n, which may take on discrete or continuous values: in par-
ticular, it could denote the number N of molecules constituting a gas system, so
that the parameter takes on values in the natural numbers ; or, it could denote
the time t during which a physical process unfolds, so that the parameter takes
on values in the real numbers . As the variable n increases, one defines the fol-
lowing sequence of systems

{S1 , S2 , ..., Sn } n ∈, 

The limit of such a sequence for n → ∞ , if it exists, corresponds to the infinite

system S∞. Arguably, the latter is just a mathematical entity, and as such it would
represent a fictitious rather than a real system. If so, by recalling the content of the
previous section, taking the limit where the variable n goes to infinity gives rise
to an idealization. We can thereby characterize the limit system S∞ as an infinite
idealization.
Now, limits can also be used as approximations without idealizations.
Consider again the above sequence of systems {S1, S2, ..., Sn}, we can also define
the following sequence of functions representing a putative physical quantity f:

{ f1 , f 2 , ..., fn } n ∈, 
 The Paradox of Infinite Limits 321

where the notation fn indicates the relevant property possessed by each finite
system Sn, so that fn := f S n . Here, some care should be taken when evaluating
the infinite limit n → ∞ . In fact, as Butterfield (2011) suggested, there is a crucial
difference between “the limit of a sequence of functions” and “what is true at that
limit”: that is, respectively,

• (i) the limit f∞ of the sequence {f1, f2, ..., fn}n of functions, and
• (ii) the function fS associated with the limit system S∞.
∞

More to the point, if the parameter n takes on values on the natural num-
bers, (i) obtains when one adjoints infinity to the set , and hence the
limit f ∞ := limn→∞ fn is defined as the function being the last element of
this sequence with n ∈  ∪ {∞}. In this case, there is no reference to the in-
finite system S∞, so that using the limit in this sense does not lead to an ide-
alization. To the contrary, (ii) depends exactly on how the limit system is
constructed. Indeed, the limit function fS represents the relevant quantity
∞
possessed by the infinite system S∞: as such, it tells us just what is true at the
limit. It should be emphasized, though, that its existence is contingent upon
the type of convergence one adopts: this point will be useful for our discus-
sion in Section 14.5.
The proposed distinction becomes less abstract if one casts it in terms of
values of quantities. In fact, supplying numerical values is right what enables us
to directly check whether or not the expected results turn out to be empirically
correct. So, given that the putative function f takes on values v(f), one should
consider yet another sequence, namely the sequence of values

{v( f1 ), v( f 2 ), ..., v( f n )} n ∈, 

Of course, the actual value of each function fn depends on the state sn in which the
system Sn is: in fact, there is also a sequence of states {s1, s2, ..., sn} implicitly under-
stood along with the sequence of systems.
As above, one needs to distinguish between (i) the limit of the sequence of
values of the function f, i.e. limn→∞ v ( fn ) , and (ii) the value v(fS ) of the natural
∞
limit function computed when the limit system S∞ is in the limit state s∞.
Before addressing the issue whether, and how, it is possible to dispense from
the infinity system in the explanation of some physical phenomenon, let us
conclude this section by noting that the formal setting we have just presented
enables us to sharply distinguish between different ways to characterize the
use of infinite limits in physics. The proposed taxonomy identifies three main
types, that is:
322 Contemporary Scientific Realism

1. Approximations without idealizations, where (ii) is not well defined, or (i) is

empirically correct but (ii) is not;
2. Idealizations yielding approximations, where (i) and (ii) are well defined
and equal; and
3. Essential idealizations, where (i) and (ii) are well defined but are not equal,
and (ii) rather than (i) is empirically correct.

Beside these cases, one can add to this taxonomy the use of infinite limits as
abstractions, which does not directly follow from such a scheme: in fact, as we
will see in greater detail in Section 14.6, these are cases in which the variable
n does not represent any physical parameter of the target system. For example,
the parameter can represent the number of times that one has to apply a trans-
formation that successively coarse grains the system. Here the appeal to an in-
finite limit is merely instrumental in that it allows us to find fixed points in a
topological space.
In the rest of the paper, we evaluate how each type of infinite limit fares against
the so-​called Paradox of Infinite Limits and its consequences for scientific re-
alism, which we present next.

14.3.2 The Paradox of Infinite Limits

Suppose that in order to represent a physical phenomenon P we define a system

of the form Sn, then the Paradox of Infinite Limits can be essentially formulated
as a combination of the following statements:

• (I) Finiteness of Real Systems: If Sn represents a real system, then the variable
n corresponding to some physical parameter cannot take on infinite values.
• (II) Indispensability of the Limit System: The explanation of the phenom-
enon P can only be given by means of claims about an infinite system S∞
constructed in the limit n → ∞.
• (III) Enhanced Indispensability Argument (EIA): If a claim plays an indis-
pensable role in the explanation of a phenomenon P we ought to believe in
its existence.

The ostensive problem can be further articulated and made more precise
when dealing with the description of particular phenomena, as we will see in
greater detail for the Paradox of Reversible Processes and the Paradox of Phase
Transitions in Sections 14.4 and 14.5, respectively. But the tension between these
statements captures the core idea of the paradox. Intuitively, the fact that real
systems are finite in the sense expressed by statement (I) can be understood as
 The Paradox of Infinite Limits 323

a basic desideratum of scientific realism. Indeed, according to our most suc-

cessful theories, such as the atomic theory of matter, real gases contain only a
finite, albeit extremely large, number N of molecules, just as real thermody-
namical processes, even when being very slow, always take a finite amount of
time t to complete. However, if one commits to the reality of an infinite system
as demanded by statement (II) together with statement (III), then one infringes
on such a basic desideratum. In fact, a seeming contradiction with statement
(I) arises insofar as appealing to the infinite idealization S∞ proves necessary. In
their strongest form, claims that taking the limit is indispensable are backed by
no-​go theorems, established within the formalism of a given theory, to the effect
that certain features of P cannot be recovered unless n is infinite. A threat to sci-
entific realism can then be mounted when these results are coupled with state-
ment (III), which Baker (2009) called the Enhanced Indispensability Argument
(EIA). According to the EIA, we ought to rationally believe in the existence of
claims (e.g. mathematical entities or idealizations) that play an indispensable
role in our best scientific theories (cfr. Shech 2019) for a discussion of EIA in
relation to infinite idealizations). In fact, it follows that, if the best explanation
available for the physical phenomenon P requires one to take the limit n → ∞,
owing to EIA one is bound to ontologically commit to the infinite system S∞,
even though statement (I) entails that the latter cannot correspond to any real
system. A realist stance toward our best physical theories is thus endangered by
the Paradox of Infinite Limits.
In order to resolve the problem, one should jettison one of the three statements
that jointly engender the ostensive paradox, at least in the form they have been
presented here. The first statement seems rather uncontroversial, and as such it
is hard to give it up. Indeed, the finiteness of a real system Sn involved in the
phenomenon P to be explained is grounded in empirical considerations, so long
as the variable n denotes a physical parameter. Furthermore, in some cases the
truth of statement (I) is granted by the content of successful background the-
ories: for example, the number of molecules in a thermal system being finite
is presupposed by the atomic theory of matter itself and general relativity (see
Baron 2019). Hence, insofar as the variable n represents a physical parameter as
in statement (I) in the paradox, the culprit must be traced back to statement (II)
or statement (III). Our own strategy will consist in giving up statement (II), but
let us first review some attempts to block statement (III).7
Baron (2016) casts doubts on statement (III) by offering a helpful character-
ization of the sense in which idealizations can be regarded as indispensable to
the best explanation of a phenomenon. According to him, although idealizations

7 See (Colyvan 2001), (Baker 2005; Baker 2009) and (Baron 2016) for a careful discussion of the

indispensability arguments.
324 Contemporary Scientific Realism

may be indeed indispensable to the purported explanation, they do not carry

explanatory load (understood as counterfactual dependence between the ideali-
zation and the effect) and therefore should not be reified. However, this strategy
is questionable in the case of infinite idealizations, since the latter can have ex-
planatory load, at least in the sense that we will expose in Section 14.5. In Baron
(2019), he adopts an alternative strategy by distinguishing between construc-
tive indispensability and substantive indispensability: given a mathematical
entity that is explanatorily indispensable to our current best scientific theories,
according to the former notion there is reason to believe that such a claim can
actually be dispensed, although we do not know how to do it yet; according to
the latter notion, instead, there is no reason to suppose that the claim can ever
be dispensed. In other words, in the first case, indispensability is just a contin-
gent matter, whereas in the second case it is an unavoidable fact. So, under this
understanding, if an infinite limit that appears as necessary to explain a phys-
ical phenomenon P is constructively rather than substantively indispensable, the
EIA does not apply with sufficient cogency to commit one to the reality of the
infinite system constructed in the limit, thereby disarming the implications of
the Paradox of Infinite Limits for scientific realism. The crucial question then
becomes how to determine whether one is dealing with an instance of substan-
tive indispensability or an instance of constructive indispensability. Although
the exact answer can only be given on a case-​by-​case basis, there are two major
strategies that one may adopt, according to Baron. One approach, which Baron
himself favors, is grounded on the notion of coherence. Specifically, one ought
to test the alleged indispensability of an infinite limit against other background
scientific theories: if the claim under test is not consistent with other accepted
theories, then there is reason to suppose that it is only constructively indispen-
sable. For instance, when a gas system is contained in a finite region the limit for
the number N of molecules going to infinity is at odds with the atomic theory of
matter; moreover, Baron argues, it fails to cohere with the general theory of rel-
ativity in that the total mass of the molecules would become infinite despite the
volume remaining finite, and hence the density would become infinite thereby
giving rise to a black hole in the region where the gas is confined, which is not
really observed in real-​life phenomena. Although plausible, this solution seems
to beg the question, since the problem that we are trying to solve concerns pre-
cisely the inconsistency between well-​established background theories and
mathematical entities used to explain certain phenomena. For us, the motiva-
tion for accepting statement (III) comes from the acceptance of the inference-​
to-​the-​best-​explanation. Indeed, the best argument for scientific realism is the
no miracles argument, which maintains that we ought to believe that abstract
and theoretical claims about existing entities postulated by our most successful
 The Paradox of Infinite Limits 325

theories are (at least approximately) true because otherwise their success would
be a miracle. Underlying this argument is the inference-​to-​the-​best-​explanation
(Boyd 1983). By using a similar reasoning, if the best explanation available of a
certain phenomenon P involves assuming an infinite idealization and there is no
known way to de-​idealize the model without losing explanatory power, then we
should commit to the existence of such an idealization for the same reasons that
ground scientific realism. This a particularly important for Explanationanists
(Kitcher 1993; Psillos 1999), since they recommend holding a realist attitude to-
ward entities that are indispensable to the explanation of certain phenomena.8
Therefore, we claim, a more satisfactory solution to the Paradox of Infinite
Limits comes from the rejection of statement (II), rather than statement
(III). To make our point, we need a criterion of empirical correctness that we
introduce next.

14.3.3 The Condition of Empirical Correctness

Before spelling out our intended criterion for correctness, it should be stressed
that the concept of explanatory indispensability heavily depends on what one
means by scientific explanation, which is a huge and much debated topic in
philosophy of science.9 For our purposes, we restrict ourselves to what we con-
ceive as a minimal requirement for a good explanation of some physical phe-
nomenon P, namely that one recovers empirically correct results. Arguably,
this yields a necessary condition in that, if an account fails to agree, at least
approximatively, with the observed data then it cannot be said to explain P
(whether it yields also a sufficient condition is less straightforward to estab-
lish, but our argument here does not really need that much). The relevant data
are given by the values of physical quantities of interest, like energy, position,
momentum, spin, etc., corresponding to properties of the physical system in-
volved in the phenomenon. The proposed condition for empirical correctness
is as follows:

8 Shech (2013, p. 1177) makes a similar point, when he says:

Insofar as arguments like the “no miracles argument” and “inference to best explana-
tion” are cogent and give us good reason to believe the assertions of our best scientific ac-
counts, including those about fundamental laws and unobservable entities, then in the
case of accounts appealing to [essential idealizations], these arguments can be used via an
Indispensability Argument to reduce the realist position to absurdity.
9 See Woodward (2014) for an excellent review of the different approaches to scientific explana-

tion in the philosophical literature.

326 Contemporary Scientific Realism

Empirical Correctness: Let D be the observed data relative to a physical quan-

tity represented by the function f, which characterizes the physical phenom-
enon P: then, the system Sn recovers empirically correct results for f just in
case v(fn) ≈ D, in the sense that there exists an arbitrarily chosen real number
ε > 0 such that |v(fn) − D| < ε.

Let us explain the content of this definition. Note that one cannot expect
that empirical data can be sharply determined, in that observations always
involve some margin of error. Hence, our condition is formulated in such a
way to allow for degrees of inexactness: in fact, it only requires that the value
of f be approximately equal, rather than exactly equal, to the data D, that
is v( fn )≈ D for some given n. Deciding the degrees of inexactness that one
may tolerate is ultimately a pragmatic matter, and that is why the number ε
is left unfixed in the definition: yet, by keeping ε very small, one ensures that
the results produced by system Sn are empirically correct to a pretty good
approximation.
Infinite limits can be used in the explanation of the physical phenomenon
P when empirical correctness is satisfied by the infinite system S∞ for some
physically relevant function f. Accordingly, one has v( f s∞ ) ≈ D, meaning that
the value at the limit, namely (ii) the function fS associated with the limit
∞
system S∞, is the same, or at least approximatively the same, as the observed
data D. More to the point, infinite idealizations are regarded as indispensable
to the explanation of P insomuch as no finite system Sn can yield empirically
correct results for f. That is, even though the parameter n grows while still
remaining less than ∞, the values taken on by fn fail to provide an approxi-
mation of the observed data for some sufficiently small ε. To put it in tech-
nical terms, this alleged indispensability of the infinite idealization typically
manifests itself in the cases in which the limit is singular, and hence (i) the
limit limn→∞ v ( fn ) of the sequence of values does not coincide with (ii) the
value v(fS ) yielded by the limit system. (Butterfield (2011) actually makes
∞
a similar point when discussing the emergence of certain properties at the
limit).
Armed with this stated condition of empirical correctness, we now proceed
to show how infinite idealizations can be actually dispensed to the explanation
of physical phenomena that seem to require one to take the limit n → ∞, thereby
allowing one to give up statement (II) in the Paradox of Infinite Limits. In fact,
such a condition grounds the use of limits as approximations of the properties
of real target systems. As we argue later, that is the basis to demonstrate that
cases of idealizations yielding approximations as well as the more controversial
cases of essential idealizations do not pose a threat to scientific realism.
 The Paradox of Infinite Limits 327

14.4 Approximations with and without Idealizations

In Section 14.2 we gave a general characterization of the concept of approximations

as inexact descriptions of target systems. Here, we can make this idea more pre-
cise thanks to the framework presented in previous section, according to which
the relevant properties of a target system corresponding to physical quantities are
represented by mathematical functions yielding numerical values: in fact, the very
condition of empirical correctness rests on the possibility that such values match
the observed data with a certain margin of accuracy. An approximation can there-
fore be defined as a formal description of some specific property of the target
system that, despite being inexact, puts one in a position to recover empirically
correct results. The choice of what properties are relevant to the purported descrip-
tion as well as the degrees of inexactness being allowed are dictated by pragmatic
considerations regarding the physical phenomenon to be explained. Such an un-
derstanding of approximations is particularly suitable for our discussion of the
use of mathematical limits, the more so because it is probably the most common
attitude physicists tend to take on in their scientific practice. For, recall that in
order to assuage the worries concerning scientific realism posed by the Paradox
of Infinite Limits one ought to find a way to dispense from the infinite idealization
introduced in statement (II) of the paradox. Interpreting infinite limits in terms of
approximations, as contemplated by the first two classes listed in our proposed tax-
onomy, enables us to do so. Let us discuss both scenarios in general terms and then
apply our reasoning to particular examples.
The scenario in which the limit n → ∞ gives rise to an idealization yielding
approximations may appear complicated in that it features a fictitious infinite
system S∞ that gives an inexact description of the real target system, and as such it
effectively satisfies the condition of empirical correctness. However, given that in
this case the limits (i) and (ii) in Butterfield’s earlier distinction coincide, there is
a strategy to dispense the infinite idealization that is readily available. Specifically,
one can show that, as the variable n grows, for some finite value n0 < ∞ the beha-
vior of the corresponding system Sn is approximatively the same as the behavior
0
of the limit system S∞, thereby recovering empirically correct results for the rele-
vant properties “on the way to the limit” rather than at the limit. More precisely,
n0 is supposed to be the actual value characterizing the real target system that
undergoes the physical phenomenon P to be explained: so, if the value of the
function fSn is sufficiently close to the limit value of fS , one does not need to on-
0 ∞
tologically commit to S∞ in order to fulfill empirical correctness. One can thus,
in the same way as we presented earlier, dispense the infinite idealization to the
explanation of P, in full compliance with statement (I) of the paradox asserting
the finiteness of real systems. If idealizations yield approximations, though, one
328 Contemporary Scientific Realism

may still wonder why one should appeal to a mathematical limit in the first place.
In other words, one may ask, how can one justify the use of an infinite limit to
describe the target system? In order to answer this question, Butterfield (2011)
puts forward what he calls a Straightforward Justification, which is based on two
features that limits enjoy, namely mathematical convenience and empirical ad-
equacy. Regarding the first feature, taking the limit often enables one to ignore
some degrees of freedom that complicate the calculations, and so infinite sys-
tems, when they exist, turn out to be more tractable than finite systems for which
the parameter n is very large.10 As for empirical adequacy, arguably that assures
that the values obtained at the limit are close enough to the real values. Hence,
while the empirically correct results are given by the values of the function
computed for the actual n0, namely the values of the real target system, taking
the limit for n growing to infinity still proves adequate within some acceptable
margin of approximation. Thus, based upon these two desirable features pro-
posed by Butterfield, one has both pragmatic and empirical reasons to justify
the use of the infinite limit. That reinforces our claim that one does not need to
commit to the reality of the infinite idealization S∞, contrary to what is entailed
by statements (II) and (III) in the Paradox of Infinite Limits.
The other scenario, namely the case in which taking the infinite limit n → ∞
yields an approximation without idealization, is more straightforward to deal
with. On the basis of our taxonomy, it occurs when the limit (i) rather than (ii)
yields empirically correct results or (ii) is not well defined at all for some phys-
ically significant function f: as a consequence, the infinite system S∞ does not
satisfy empirical correctness. It follows that it cannot be even used to explain the
relevant phenomenon. Indeed, an approximation without idealization should
be understood as a misrepresentation of the target system whose properties are
given an inexact description in terms of the limit f∞ of the sequence {f1, f2, ...., fn}
of functions representing the relevant physical quantities, rather than in terms
of the properties of a fictional limit system. Accordingly, it would not make
any sense to reify the infinite idealization, which means that statement (II) in
the Paradox of Infinite Limits does not hold, and hence there cannot arise any
ensuing problem for scientific realism. Very recently, Norton (2012) recognized
that sometimes infinite limits are used precisely as approximations without
idealizations. For him, there are two sufficient conditions under which an ap-
proximation cannot be promoted to the status of idealization, which are closely
related to the formal conditions we stated earlier: that is, (1) the limit system

10 In addition, Palacios (2018) points out that it does not suffice to show that the behavior that

arises in the limit arises already for a large value of the parameter n, but one also needs to show that it
arises for realistic values of n0. This latter requirement ought to be empirically grounded and is meant
to assure that the lower level theory that results from a limiting operation is empirically correct and
therefore capable of describing realistic behavior.
 The Paradox of Infinite Limits 329

does not exist in the sense that it is paradoxical, and (2) the limit system has
properties that are inadequate for the idealization in that they do not match with
the properties of the finite target system. The following quotation illustrates his
proposed criterion for the failure of idealizations:

Another type of limit used in thermodynamics cannot be used to create

idealizations. Its limiting processes are beset with pathologies so that it either
yields no limit system or yields one with properties unsuited for an idealization.
(Norton 2012, p. 13)

Norton’s goal is to show that, on the basis of these two conditions, some infinite
limits that are regarded as idealizations in the literature do not deserve to be ele-
vated to such a status and should in fact be demoted to mere approximations, as
it happens in many examples in which mathematical limits are employed in ther-
modynamics and statistical mechanics.11 In his view, an example in which an in-
finite limit does not give rise to a suitable idealization due to condition (2) is the
account of phase transitions in the thermodynamical limit within classical sta-
tistical mechanics that we will address in the next section. Another controversial
case that Norton claims to be an instance of failure of idealization due to condi-
tion (1) is the paradox of thermodynamically reversible processes arising in the
infinite-​time limit, which has recently drawn attention in the philosophical lit-
erature (Norton 2014; Norton 2016; Valente 2019). Let us discuss this case next.
Reversible processes are conceived as sequences of equilibrium states through
which a thermodynamical system progressively passes in the course of time.
They are typically constructed by means of the infinite-​time limit, and as such
they are interpreted as processes taking place infinitely slowly. They owe their
name to the fact that, ideally, they could be traversed in both directions of time.
However, reversible processes are fictitious processes: indeed, thermodynam-
ical processes just take a finite amount of time t to complete, even if they pro-
ceed quasi-​statically, and they can only unfold in one temporal direction but not
in the reversed one. So, at best, taking the infinite-​time limit t → ∞ can yield
approximations of real processes. The question that interests us is whether or not
the limit processes can also be intended as infinite idealizations. According to
Norton, reversible processes do not deserve to be elevated to such a status since
they display contradictory properties. More to the point, he submits that they are

11 To be sure, Norton concedes that, under certain circumstances, approximations can actually

be promoted to idealizations. For instance, in his (2014) paper Norton presents the case of the law of
ideal gases in thermodynamics as an example of an approximation which gives rise to an ideal system
that can as well serve as an idealization. In fact, an ideal gas system is defined as constituted by a large
number of non-​interacting, spatially localized particles, and as such it can describe, at least approxi-
mately, the behavior of very rarefied real gases under appropriate circumstances.
330 Contemporary Scientific Realism

plagued by a paradox: for, on the one hand, (I) thermodynamical processes are
driven by a non-​equilibrium imbalance of driving forces, which is necessary in
order to enact the transition of the system from one state to another; on the other
hand, though, (II) a reversible process amounts to a sequence of equilibrium
states, for which there must be no imbalance of driving forces. By connecting the
driving forces enacting the process with its duration in time, Norton’s Paradox
of Reversible Processes becomes somewhat similar to the general Paradox of
Infinite Limits. For instance, when a fixed quantity Q of heat is exchanged be-
tween two bodies in thermal contact, the driving forces are given by the tempera-
ture difference ∆T between the bodies. So, if the latter is equal to zero, the process
of heat transfer cannot take place. Now, by assuming Fourier law Q = − k ∆T ∆t
(with k being the heat transfer coefficient), the amount of heat that passes from
the hotter to the colder body is also proportional to the time ∆t that the process
would take to complete. As a consequence, if one tries to construct a reversible
process by letting time go to infinity, the temperature difference must vanish, as
prescribed by statement (II), but then no heat would be exchanged between the
two bodies, thereby raising a contradiction with statement (I). Based on such a
paradox, Norton argues that reversible processes fail to be idealizations.
However, Valente (2019) objected to this conclusion, by offering a way to cir-
cumvent the alleged contradiction. As he pointed out, Norton’s paradox arises
due to the misconception that reversible processes should be treated as ac-
tual thermodynamical processes. Instead, they ought to be regarded as mere
mathematical constructions that are introduced to apply infinitesimal calculus
to thermodynamics. In fact, formally, they simply correspond to continuous
curves in the space of equilibrium states of a thermal system: along these curves
one can calculate the exact integrals of state-​functions, such as entropy, and
then compute the values of other physical quantities of interest, like heat and
work. For this reason, as it was observed by the mathematical physicist Tatjana
Ehrenfest-​Afanassjewa (1956) in her book on the foundations of thermody-
namics, reversible processes should better be called “quasi-​processes,” so as to
avoid the misunderstanding that they would correspond to actual processes,
which could even be reversed. Accordingly, while statement (II) holds by def-
inition, it would be a mistake to ascribe to reversible processes the properties
required by statement (I): that is, contrary to what happens during real ther-
modynamical processes, there cannot be any non-​equilibrium imbalance of
forces moving a thermal system from one state of equilibrium to another along
the continuous curve representing a quasi-​process. As a result, the ostensive
paradox disappears, and hence one may as well regard reversible processes as
idealizations, in accordance with Norton’s own criterion. Notice that here, dif-
ferently from the Paradox of Infinite Limits, in order to disarm the apparent
 The Paradox of Infinite Limits 331

contradiction we need to drop statement (I) rather than statement (II): the
reason is that what is to be explained in this case is just the idealized object
constructed in the limit. The connection with real thermodynamical processes
can then be given thanks to the notion of approximations. To illustrate this idea,
let us again refer to Ehrenfest-​Afanassjewa’s own work. After warning that,
strictly speaking, infinitely slow processes cannot exist (cfr. p. 11), she went on
to argue as follows:

In order to make a quasi-​process [i.e. a reversible process] open to experi-

mental investigation or just to connect it with experiments in thought, one
has to conceive it as approximated by quasi-​static processes [i.e. extremely
slow processes]. These are in turn real processes, if also idealized. (Ehrenfest-​
Afanassjewa 1956, p. 56)

Thus, even though an infinite-​time limit process has no counterpart in reality, in

practice it can serve to describe, albeit inexactly, the relevant properties of ther-
modynamical processes unfolding very slowly. To put it in terms of our char-
acterization of idealizations yielding approximations: once a certain margin of
accuracy is stipulated, the fictitious quasi-​process constructed in the limit t → ∞
gives approximatively true descriptions of real physical quantities, such as heat
and work exchanged during real processes that take a large yet finite time t0 to
complete.
So, although Norton’s distinction between approximations and idealizations
may be fruitful, the extent to which the examples he addresses are genuine
cases of approximations without idealizations is still subject to debate. In our
opinion, a less controversial case where one can have approximations without
idealization is the reduction of the classical equations of motion to relativistic
equations in the Newtonian limit. As is well known, in the theory of special rel-
ativity physical quantities such as the energy and momentum of a moving body
with an invariant mass m0 can be expressed in terms of the so-​called Lorentz
factor γ:

E = γ m0 c 2

p = γ m0 v

Arguably, when the Lorentz factor goes to 1, one can recover the values of the
classical counterparts of these quantities defined, respectively, as E = m0 c 2 and
p = m0 v . Since the Lorentz factor takes on the form
332 Contemporary Scientific Realism

1
γ (v ) = ,
1 − (v / c )
2

2
v
one can make this expression go to 1 in the limit   → 0, where c is the speed of
c
light and v the velocity of a moving body in a given inertial frame. There are dif-
ferent possible ways of interpreting this limit. First, one can interpret it as taking
the limit of a sequence of systems in which the velocity v goes to zero. The limit
will then give rise to an idealization, namely to a fictional system in which the
value of the velocity for all bodies is null in the given inertial frame. Nonetheless,
understanding the limit in this sense has the problem that, e.g. the momentum
p, which depends exactly on the velocity of the bodies, will be always zero. This
naturally means that the quantities defined in the limit system will not provide
a good approximation for the behavior of restless objects, in which momentum
is different from zero. Therefore, the purported idealization would not serve to
describe the behavior of moving objects. Likewise, a similar problem arises if
one interprets the Newtonian limit as taking the limit of a sequence of systems in
which c goes to infinity. In this case, the limit system will be an imaginary system
in which the speed of light is infinite. In this system, quantities such as the ki-
netic energy that are defined as a function of the speed of light c will also go to
infinity, thereby failing to give an approximation of the actual kinetic energy of
real moving bodies, which is instead finite.
On the contrary, a much more suitable interpretation of the Newtonian limit
2
v
 c  → 0 should not be given in terms of fictional systems, but rather as a mere
 
approximation of the velocity of bodies that are moving slowly compared to the
speed of light. This can be seen more clearly by noticing that the Lorentz factor γ
can be expanded into a Taylor series:

2n 2 4 6
1 ∞
 v  n  2k − 1  1 v 3 v 5  v
γ (v ) = = ∑  ∏   = 1 + 2  c  + 8  c  + 16  c  + ...
1 − (v / c )
2
n=0  c  k = 1  2k 

Accordingly, if one considers only the first term of this Taylor expansion, one will
recover the exact values of the classical quantities of interest, and this will consti-
tute just an approximation of the velocity v of the objects for which v << c. Taking
more terms into account will give results that are more accurate from the rela-
tivistic point of view, but that depart from the classical values. It is important to
 The Paradox of Infinite Limits 333

stress that the thus-​described process of approximation is analogous to the one

employed to account for the behavior of a mass falling in a weakly resisting me-
dium presented in Section 14.2, and therefore it lends itself to the same interpre-
tation with respect to the notion of approximate truth. In fact, limits of this kind
should be understood as a mere misrepresentation of the values of certain prop-
erties of the target system, which will gradually disappear if one takes more terms
of the series expansion into account. Note that here, like in all other examples
of approximations without idealization, one does not posit any fictional system
that would yield empirically adequate results, and hence the Paradox of Infinite
Limits does not arise. Since there are no infinite systems involved, the procedure
we have just outlined offers a straightforward way to make the process of taking
the Newtonian limit compatible with a form of scientific realism that allows for
theories to be approximately true.
In the last analysis, using mathematical limits to provide approximations
of properties of real finite systems enables one to dispense limit systems from
the explanation of the relevant physical phenomena. Indeed, as we have argued
throughout the present section, despite the fact that letting some physical pa-
rameter grow to infinity is tantamount to introducing a prima facie unrealistic
assumption, cases of approximations without idealizations and even cases of
idealizations yielding approximations should not pose any worry to a scientific
realist. A more outstanding challenge is instead raised by cases in which an infi-
nite limit is used as an essential idealization, to which we now turn.

14.5 Essential Idealizations

The claim that there are essential idealizations has been put forward by some
authors, especially Batterman (e.g. 2001, 2005, 2011), in reference to singular
limits, whereby empirically adequate results are obtained just for the infinite
system and not for finite systems. In terms of Butterfield’s distinction, this appar-
ently mysterious case occurs when, given a function f representing some physical
quantity, (i) the limit of the sequence of values and (ii) what is true at the limit for
n → ∞ differ from each other, but only (ii) is empirically adequate (these cases
correspond to essential idealizations according to the taxonomy presented in
Section 14.3.1). So, if the variable n growing to infinity represents some physical
parameter, it means that the limit system would not yield an approximation of the
relevant property of the target systems, which are instead finite. An infinite ide-
alization is then deemed as essential in that, arguably, it is only the limit system
S∞ that allows one to explain the physical phenomenon for which the function
f is relevant. So, if one further assumes EIA, then one falls into the Paradox of
Infinite Limits, hence threatening scientific realism. In this section we present a
334 Contemporary Scientific Realism

possible solution to this paradox that casts doubt on the “essential character” of
the idealization, which is partially based on Butterfield’s (2011) results.

14.5.1 Resolving the Paradox for First-​Order Phase Transitions

Phase transitions are sudden transformations of a thermal system from one state
into another, occurring for instance when some material changes from solid to
liquid state due to an increase of temperature. That is a much debated case study
where, according to indispensabilists like Batterman, there arises an essential
idealization when taking the so-​called thermodynamic limit. Informally, the ar-
gument goes as follows. According to thermodynamics, which deals with the be-
havior of thermal systems from a macroscopic point of view, phase transitions
occur when the function representing the derivatives of the free energy is discon-
tinuous. Allegedly, such a discontinuity matches with the observed data, since
sudden transitions from one phase to another appear to take place abruptly. Yet,
when attempting to recover the same phenomena within statistical mechanics,
which describes thermal systems at the microscopic level as being composed by
a very large number N of molecules, one faces a technical impossibility: that is,
if N is finite, the function representing the derivative of the free energy remains
continuous no matter how large N is. Instead, one can recover the sought-​after
discontinuity by taking the thermodynamical limit, which prescribes that both
the number of molecules N and the volume V of the system go to infinity while
keeping its density fixed (Goldenfeld 1992). This motivates the persistent atti-
tude among physics, such as Kadanoff (2009), to emphasize the importance of
infinite systems to explain the phenomenon of phase transitions:

Phase transitions cannot occur in finite systems, phase transitions are solely a
property of infinite systems. (p. 7)

Accordingly, it seems that one is bound to assume the infinite system S∞

constructed in the thermodynamical limit as being indispensable in order to
account for thermodynamical phase transitions within statistical mechanics.
Hence, the thus-​defined infinite idealization is supposed to be essential. If so,
though, the Paradox of Phase Transitions would present itself.
In the philosophical literature, the paradox was originally proposed by
Callender (2001). In his formulation, a contradiction arises due to the joint com-
bination of four conditions: (1) Phase transitions are governed by classical statis-
tical mechanics, (2) real systems have finite N, (3) phase transitions occur when
the partition function has a discontinuity, and (4) real systems display phase
 The Paradox of Infinite Limits 335

transitions. Let us stress that this is just a special case of the general Paradox of
Infinite Limits. In fact, Callender’s conditions can be explicitly connected with
the three statements presented in Section 14.3.1. For, condition (2) is tanta-
mount to statement (i) expressing the basic desideratum for scientific realism
that real systems are finite, whereas condition (4) assures that the phenomenon
to be explained, namely phase transitions, occurs for such systems, which is
captured by statement (ii). On the other hand, condition (3) requires empirical
correctness to be satisfied for the partition function being discontinuous: yet, as
discussed earlier, if one tries to explain the phenomenon within the framework
of statistical mechanics, in accordance with condition (1), then one is bound
to reify the thermodynamical limit, which gives rise to an infinite idealization
proving indispensable, exactly as our statement (iii) dictates. Other formulations
of the paradox have been put forward in the literature, which differ from the one
just presented only regarding how the allegedly conflicting conditions are stated,
but they all basically agree on the content of the problem. There is also a variety
of proposed solutions. Callender himself suggests that one should not take “too
seriously” the theory that prompts one to introduce a discontinuity in the de-
scription of phase transitions, namely thermodynamics: accordingly, condition
(3) can be dropped, which means that one does not need to appeal to the thermo-
dynamical limit and therefore denies statement (ii) of our formulation of the par-
adox. Shech (2013), on the other hand, suggests that one can resolve the paradox
by noticing that the terms related to the existence of infinite systems do not refer
to concrete physical systems but just to mathematical objects, which do not carry
any ontological significance. However, as he correctly recognizes, this cannot be
the attitude of scientific realists, who are interested in our abstract scientific ac-
counts getting something right about the real world. Alternatively, Liu (2019)
suggests that the alleged contradiction disappears if one adopts a form of con-
textual realism, whereby realist claims should be evaluated relative to anchoring
assumptions formulated within the background theory: in particular, the claim
that the number of molecules grows to infinity can be regarded as true in the
context of a microscopic theory holding that condensed matter is continuous.
However, the jury is still out as to whether these proposals effectively dissolve the
paradox of phase transitions.
Instead, our preferred strategy to elude the paradox and thus salvage sci-
entific realism goes along the lines of Butterfield’s (2011) own dissolution
of the mystery of singular limits, which is endorsed, at least in connection
with classical phase transitions, by Menon and Callender (2013) as well as by
Norton (2014) and Palacios (2019). It develops into two steps: first of all, one
ought to make a careful choice of the physical quantities to work with, so as
to avoid those quantities for which the infinite idealization appears essential
336 Contemporary Scientific Realism

in that the limit is singular; then, one employs the notion of approximation
to show that for the selected quantities empirically correct results obtain on
the way to the limit, without having to commit to the reality of the infinite
system. Thus, the underlying idea of the proposed strategy is that of focusing
on just the properties that are relevant for the behavior that we intend to
explain, instead of requiring that all properties of the finite target system
extend smoothly to the limit system. More to the point, Butterfield observes
that the alleged mystery of singular limits is simply a consequence of looking
at functions that do not give information regarding the behavior of real sys-
tems as N increases toward infinity. For instance, if one restricts one’s atten-
tion only to the fact that some function f be discontinuous, like in the case
of the derivative of the free energy phase, one would lose insight of what
happens for very large but finite N, since the expected discontinuity obtains
only at the limit “ N = ∞ .” To the contrary, one ought to turn one’s attention
to physical quantities represented by different functions, so that the corre-
sponding properties of the actual target system SN are approximated by the
0
properties of S∞. In other words, even though it appears to be an essential
idealization for some quantities, the limit system is an idealization yielding
approximations for other properties that one regards as physically salient for
the phenomenon to be explained. In this way the desired behavior is recov-
ered on the way to the limit, and hence one can dispense from the infinite
idealization, by demonstrating that this apparent “essential idealization” is a
case of idealizations yielding approximations.
In order to make this proposal more concrete, let us look at a specific ex-
ample of first-​order phase transitions, that is a ferromagnet at sub-​critical
temperature. The Ising model portrays a ferromagnet as a chain of N spins,
wherein a physical quantity called magnetization is represented as a func-
tion of the applied magnetic field. At very low temperatures, there are two
possible phases available: a state where all spins are oriented in the up-​
direction, for which the magnetization takes on the value +1; and a state
where all spins are oriented in the down-​direction, for which the magneti-
zation takes on the value −1. When the sub-​critical temperature is reached,
even if the applied magnetic field is null, one observes an abrupt flip from
one phase to the other. That is formally captured by the magnetization func-
tion being discontinuous. Arguably, one can describe this phenomenon just
in case one takes the thermodynamical limit whereby the number N of spins
grows to infinity. Butterfield’s proposed strategy can then be illustrated by
means of a toy model. He defines a sequence of real-​valued functions {g1, g2,
..., gN} with argument x belonging to the real numbers , which take on the
following form:
 The Paradox of Infinite Limits 337

 1
 −1 iff x ≤ − N
 1 1
g N ( x ) : = Nx iff − ≤ x ≤
 N N
+1 iff x ≥ 1
 N

All such functions are continuous, in that they remain equal to –​1 until x = –​1/​N
and then they start to grow linearly with gradient N up to the value +1 for x = 1/​N,
after which they constant again. However, when one takes the limit for N → ∞, the
resulting function is no more continuous: in fact, the limit is given by

 −1 iff x < 0

g ∞ ( x ) =  0 iff x = 0
+1 iff x > 0


and hence it exhibits a discontinuity at point x = 0. Concretely, in the example

of ferromagnetism, the behavior of the functions gN’s mimics the magnetiza-
tion function for a chain of N spins, with the argument x representing the ap-
plied magnetic field. A singular limit arises if one introduces a binary function
f N :  ∪ {∞} → {0,1} whose numerical values are determined by whether gN is
continuous or not, i.e.,

 1 iff g N continuous
f N := 
 0 iff g N discontinuous

Here, in accordance with the condition for essential idealizations stated in

Section 14.3, when N goes to infinity one distinguishes between two possible
values that are quite different: that is, (i) the value 1 taken on by all the functions
in the sequence {f1, f2, ..., fN} for finite N, which results from the fact that all fN are
continuous; and (ii) the value 0 taken on by the function fS evaluated on the
∞
limit system S∞, which results from the fact that g∞ is discontinuous. So, if one
focuses on the function fN, the values for finite N ’s will always be continuous
from the value of the quantity evaluated on the limit system, no matter how large
N is. In particular, no real system with a finite number N0 of molecules would
have values of fN that are approximated by the function fS . For Butterfield, that is
0 ∞
just what makes the limit N → ∞ seem mysterious.
338 Contemporary Scientific Realism

However, the mystery can be explained away if one looks at the beha-
vior of the functions gNs instead of the functions fN’s. In fact, as N grows, the
functions gN become more and more similar to the step function g∞, even
though, contrary to the latter, they remain continuous. Specifically, when N0
is extremely large, for most points x one has g N 0 (x ) = g ∞ (x ) , which is true in
particular when the argument is x = 0; furthermore, whenever the values of
these functions are not strictly equal, namely around the singular point x = 0,
they will still be close to each other, so that g N 0 (x ) ≈ g ∞ (x ) . Therefore, the
empirically correct values of the magnetization function can be recovered
by the properties of the real system SN without having to resort to the infi-
0
nite system S∞. In this way, the values obtained when taking the limit N → ∞
just yield approximations of the relevant property of the real target system.
More importantly, as a theoretical analysis also shows for realistic values of
N, the gradient in the derivatives of the free energy is sufficiently steep that
the difference in the limit values of the thermodynamic quantities as N → ∞
and realistic systems with finite N0 becomes negligibly small (Schmelzer and
Ulbricht 1987; Fisher and Berker 1982).
A straightforward justification similar to the one obtained in cases
of idealizations yielding approximations can now be given also for infi-
nite limits that appear to give rise to essential idealizations. In fact, the use
of the latter is justified based on the two desirable properties of mathemat-
ical convenience and empirical adequacy. Here, though, we would like
to emphasize a point that Butterfield does not develop, namely the fact
that the choice of a certain topology over the other determines different
degrees of empirical adequacy. For this purpose, let us consider the func-
tion fn indexed by the parameter n that maps the independent variable x
onto the real numbers: accordingly, the notation fn(x) indicates the value
that the function takes on for each x in the domain, where in concrete phys-
ical cases the variable x would represent the possible states of the system
under investigation. A natural topology is induced by the following type of
convergence:

Pointwise Convergence: The sequence { f (x)}

n n
converges pointwise to
f ∞ (x ) if, given any x in  and given any ε > 0 , there exists a natural number
( )
n0 ε, x such that | fn (x ) − f ∞ (x )|< ε for every n > n0 ( ε , x ).

Translated into our framework, the fact that the sequence of functions { fn (x )}n
converges pointwise to the function f ∞ (x ) means that there is a real finite
system Sn for which the value of the relevant function is approximated by the
0
 The Paradox of Infinite Limits 339

value of the limit function for a given state x, that is fn 0 (x ) ≈ f ∞ (x ) . But one may
as well adopt a different type of convergence, that is:

Uniform Convergence: The sequence { f n ( x )} converges uniformly to f ∞ ( x )

if, given any ε > 0, there exists a natural number n0 ( ε ) such that, for any x in ,
| fn (x ) − f ∞ (x )|< ε for every n > n0 ( ε ).

Since in this case n0 depends only on ε, and not even on the variable x like in
the case of pointwise convergence, uniform convergence proves stronger
than the latter (in fact, it is strictly stronger in that one can show by means of
counterexamples that pointwise convergence does not imply uniform conver-
gence). Indeed, here one can choose a natural number n0 for which the sought-​
after approximation fn 0 (x ) ≈ f ∞ (x ) holds for all the possible states x. To the
contrary, pointwise convergence allows for exceptions, in the sense that once n0 is
fixed together with the margin of approximation determined by ε > 0 there may
be some state x for which | fn ( x ) − f ∞ ( x )|< ε does not hold for any n > n0 . The up-
shot of this analysis is that the standard hierarchy of convergence conditions for
functions representing physical quantities entails different degrees of accuracy
up to which empirical accuracy is satisfied. Hence, just as the precise notion of
approximation rests on what one means by the expression “sufficiently close,”
Butterfield’s straightforward justification of infinite limits ultimately depends on
the topology under which one takes the limits. In fact, in the explanation of first
order phase transitions the criterion of empirical adequacy is satisfied only in
its weak degrees of accuracy: for, it is just when one chooses the topology in-
duced by pointwise convergence, and not by the stronger uniform convergence,
that a sequence of continuous functions approaches a discontinuous function in
the limit.
This analysis shows that the property of empirical adequacy is sensitive
to the choice of topology. Thus, the extent to which one satisfies empirical
correctness, namely the condition that the value of a function representing a
given physical quantity is approximately equal to the empirical data, depends
on how the limit is taken. On this point it is worth making an important clari-
fication: the issue whether the use of the limit is justified and the issue whether
the use of the limit raises a threat to scientific realism, even though they have
a common root, they should be kept separate. The common root is that the
problem that, if the variable n represents a physical parameter in that, then
when n becomes infinite the limit system would not be real. The former issue
asks the question: idealizations arising in the limit in physical applications?
The second issue, instead, arises because it appears as if one violates statement
(I) of the Paradox of Infinite Limits, which is a basic desideratum for scientific
340 Contemporary Scientific Realism

realism. Butterfield’s two suggested properties, i.e. empirical adequacy and

mathematical tractability, helps one answer the first question. Arguably, the
use of an infinite limit is justified only if it is empirically adequate, that is if it
yields empirically correct results, at least approximately. Furthermore, if one
can recover empirically correct results also without taking the infinite limit but
the latter is mathematically more tractable, then one has a pragmatic reason
to favor the use of the unrealistic limit. That seems a prima facie reasonable
justification for one to use an infinite idealization for practical calculations
even though it does not, strictly speaking, represent the finite target system.12
But, the fact that for pragmatic purposes one is justified to use the limit in
physical applications is independent from the issue whether one is a scien-
tific realist or not. In fact, contrary to the issue of justification, this second
issue bears on whether or not one ontologically commits to the existence of
the limit system. When one wishes to give an explanation for some physical
phenomenon P, if the only way to recover empirical correct results is by as-
suming the limit system, the resulting infinite idealization appears explana-
torily indispensable, as prescribed by statement (II) in the Paradox of Infinite
Limits. Scientific realism is then called into question if one accepts the EIA,
namely statement (III) in the paradox, whereby the limit system is supposed
to exist. Thus, it would seem that a scientific realist is not just justified to use
the infinite limit, but also that she ought to be ontologically committed to it,
in conflict with statement (I). Our strategy to resolve the problem, at least for
the purported form of scientific realism that allows for approximate truth, is
to reject statement (II) in the paradox, in that empirical correct results can be
approximately obtained already on the way to the limit for the relevant phys-
ical quantities of interest.

14.5.2 Quantum Phase Transitions and Other Cases

of Essential Idealizations

Let us move on to address the question whether the purported strategy to cope
with the indispensability of infinite limits can be generalized to other cases where
there appear essential idealizations. Butterfield (2011) conjectures that the solu-
tion of the Paradox of Phase Transitions proposed in the classical context holds
generally. However, matters are less straightforward in other cases such as in the

12 For a complete investigation of this issue one should actually discuss in much greater details

than we can here the sense in which the limit system can be said to represent the target system. To
this extent, Shech (2014) put forward a distinction between epistemologically and ontologically
faithful representations. However, it goes beyond the scope of the present paper to survey the notion
of representation.
 The Paradox of Infinite Limits 341

description of phase transitions in quantum statistical mechanics. In fact, in this

case the indispensability of the thermodynamical limit seems to present itself in
a stronger form than in classical statistical mechanics. A rigorous description of
quantum phase transitions can be given within the algebraic approach to phys-
ical theories. In this framework, one represents a physical system by means of
an algebra of observables, that is the set of physical quantities representing its
observable properties. The possible states that the system can occupy are then
defined on such an algebra. In some cases of interest, in order to describe the rel-
evant physical quantities, instead of working directly with the algebra one needs
to refer to its representation into a concrete Hilbert space (the so-​called GNS rep-
resentation), which is induced by a chosen state. For instance, in the Ising model,
both the phase in which all spins in the chain are oriented in the up-​direction
and the phase in which all spins in the chain are oriented in the down-​direction
give rise to distinct concrete representations. Now, the familiar issue concerning
the indispensability of infinite idealizations arises since a quantum observable
representing magnetization can be rigorously defined only if one takes the ther-
modynamical limit. However, such an observable is state-​dependent, in the sense
that it is constructed only within the representation of one phase or, alternatively,
within the representation of the other phase. This fact marks a difference with
respect to the description of phase transitions in classical statistical mechanics.
Indeed, in the classical context the values taken on by the magnetization func-
tion depend on the particular state of the system, which is determined by the
applied magnetic field; instead, in the quantum context it is the magnetization
function itself, and not just its values, that depends on a particular state.
Furthermore, in the case of quantum phase transitions, there is an additional
problem arising since the representation induced by the state in which all spins
are induced by the state in which all spins are oriented in the down-​direction
are not unitarily equivalent. Without entering into too many technicalities, this
means that, if one defines the magnetization observable with respect to one phase
by taking the infinite limit N → ∞ , one cannot recover the values of magnetiza-
tion relative to the other phase (and vice versa), contrary to what happens for
finite N’s when the up and down representations always remain unitarily equiv-
alent. Arguably, the issue of unitary inequivalence of the phase representations
thus complicates the mystery of quantum phase transitions in the example of the
Ising model. It goes beyond the scope of the present paper to settle this entire
problem. For our purposes here, it is sufficient to point out that two contrasting
positions can be identified in the literature: on the one hand, there are authors
such as Liu and Emch (2005) and Ruetsche (2011), who claim that taking the
thermodynamical limit to account for quantum phase transitions is indispen-
sable; on the other hand, there are authors, most notably Landsman (2013) and
Fraser (2016), who take a deflationary view toward the limit. In this respect, the
342 Contemporary Scientific Realism

jury is still out as to whether or not phase transitions in quantum statistical me-
chanics constitute a case of essential idealizations where one can provide an ex-
planation of the phenomenon without committing to the reality of the infinite
system constructed in the thermodynamical limit.13
Be that as it may, there is a further point that is worth emphasizing re-
garding the general strategy to cope with other cases of essential idealizations,
besides classical phase transitions. As pointed out by Palacios (2018), it does
not suffice to demonstrate that approximately the same behavior that occurs
in the limit, also occurs “on the way to the limit,” but in addition we need
to demonstrate that it arises for realistic values of the parameter n that goes
to infinity. This latter condition is important because there are cases, such as
the ergodic approach to equilibrium, in which the expected behavior arises
for finite but unrealistic values of the parameter that goes to infinity. In more
detail, in the ergodic theory of equilibrium, one takes the infinite-​time limit
t → ∞ in order to assure that phase-​averages and time-​averages coincide.
Even in simple examples like a small sample of diluted hydrogen, though, one
can estimate that the desired behavior can occur for times t0 that are finite but
unimaginably longer than the age of universe. In such cases, even if we can ac-
tually demonstrate that the behavior arises for finite values of the parameter,
we would not have succeeded in demonstrating the empirical adequacy of the
theory.
To conclude, the analysis we have developed in the present section indicates
that the challenge posed by essential idealizations to scientific realism ought to
be resolved on a case-​by-​case basis, and that ultimately calls for empirical consid-
erations. In general, the strategy to dispense with the infinite-​limit system in the
explanation of a certain phenomenon involves a suitable selection of the phys-
ical quantities to focus on, so as to yield approximations of the relevant prop-
erties of the target system already “on the way to the limit,” where the degrees
of inexactness that one may accept depends on the particular situation at stake
and the choice of an adequate topology. Accordingly, as the example of classical
phase transitions shows, one can assuage worries concerning scientific realism.
In fact, if the idealization is not genuinely essential, one does not need to assume
statement (II) of the Paradox of Infinite Limits. Moreover, the limit values of the
relevant physical quantities can still give good, albeit inexact, descriptions of the
properties of realistic systems, which are supposed to be finite in accordance with
statement (I), and so in order to evade the paradox we do not even need to cast
doubts upon the validity of the enhanced indispensability argument contained
in statement (III). The Paradox of Infinite Limits can thus be avoided in spite of

13 Another example of apparent essential idealizations are continuous phase transitions. We will

address this case in Section 14.6.

 The Paradox of Infinite Limits 343

apparent essential idealizations emerging in the infinite limit, while endorsing a

form of scientific realism that allows for our best theories to be sufficiently close
to the truth. There now remains to discuss the last type of use of infinite limits,
namely as abstractions. As we will see, differently from the case studies we have
investigated so far, sometimes in such cases the variable n growing to infinity
does not represent a physical parameter, and hence in principle one may not sat-
isfy statement (I) in the paradox.

14.6 Infinite Limits as Abstractions

For the sake of completeness, in this last section we address a different role played
by mathematical limits that has been much less discussed in the philosoph-
ical literature than approximations and idealizations, namely the use of limits
as abstractions. We mentioned earlier that the thermodynamical limit, in which
the number of particles N as well as the volume V go to infinity, can be used
to recover the quantities that successfully describe phase transitions in thermo-
dynamics. Another important role of the thermodynamic limit is to enable the
removal of irrelevant contributions such as “surface” and “edge” effects. More
to the point, any finite lattice system will include contributions to the partition
function coming from the edges and surfaces, which may be considerably dif-
ferent from those coming from the center of the sample. Taking the thermody-
namical limit allows one to treat the system as a bulk, leaving out surface effects
that are mostly irrelevant for the behavior. In this sense, the thermodynamical
limit enables one to abstract away details that do not make a difference for the
phenomenon under investigation (see also Mainwood 2006; Butterfield 2011;
Jones 2006).
A more interesting case of infinite limits used as abstractions is given in
the context of continuous phase transitions. In contrast to first-​order phase
transitions that involve discontinuities in the derivatives of the free energy, in the
case of continuous phase transitions there are no discontinuities but rather there
are divergences in the response functions (e.g. specific heat, susceptibility for a
magnet, compressibility for a fluid). An example of a continuous phase transition
is the transition in magnetic materials from the phase featuring spontaneous
magnetization—​the ferromagnetic phase—​to the phase where the spontaneous
magnetization vanishes—​the paramagnetic phase. Continuous phase transitions
are also characterized by the divergence of a quantity called the correlation
length ξ, which measures the distance over which the particles are correlated.
A typical way of dealing with these long correlations is by introducing renor-
malization group methods, which are mathematical and conceptual tools that
consist in defining a transformation that successively coarse-​grains the effective
344 Contemporary Scientific Realism

degrees of freedom while keeping the partition function and the free energy (ap-
proximately) invariant. Palacios (2019) pointed out that such a process can be
interpreted as assuming an infinite limit, in which the number of iterations n of
the renormalization group transformation goes to infinity. An important aspect
of the infinite limit for n → ∞ is that, at each step of the sequence, irrelevant cou-
pling constants are abstracted away, so as to retain only factors that are relevant
for the behavior under description. That follows because, in this process, the par-
tition function and the free energy that determine the behavior at a phase tran-
sition remain (approximately) invariant. After an infinite number of iterations,
the sequence of systems may converge toward non-​trivial fixed points, where the
values of the coupling constants no longer change by applying the transforma-
tion. Note that the limit for the number n of iterations going to infinite does not
represent any physical parameter of the system under description, but rather the
number of transformations that one needs to apply by means of the renormaliza-
tion group procedure.
The beauty of renormalization group methods is that linearizing around fixed
points allows one to calculate the critical exponents of the power laws that de-
termine the behavior close to the transition. Furthermore, it allows one to give
an account of universality, namely the remarkable fact that physical systems as
heterogeneous as fluids and magnets exhibit the same behavior near a phase
transition (details in Goldenfeld 1992). There is good reason to consider the in-
finite iteration limit n → ∞ as an abstraction. In fact, in Section 14.2, we de-
fined abstraction as a process that consists in leaving some factors out without
misrepresenting the properties of the original system. In fact, when applying a
renormalization group transformation to the description of a given system, one
does exactly this: that is, at each stage n, one leaves out irrelevant details while re-
taining those factors that make a difference for the phenomenon to be explained.
To be sure, in doing that, one might have to use approximations; yet, one does
not introduce any idealization. Moreover, it is also important to point out that
the renormalization group process has the same epistemic role attributed to
abstractions, namely that it enables us to give an explanation of the common
behavior generated by systems that are diverse from each other at a fine-​grained
level, thereby accounting for universality.
In recent philosophical literature, the use of renormalization group methods
has often been considered as an example of essential idealizations, e.g. Batterman
(2011), Batterman (2017) and Morrison (2012). Indeed, Batterman says in a
footnote:

It seems to me that if one is going to hold that the use of the infinite limits is a
convenience, then one should be able to say how (even if inconveniently) one
might go about finding a fixed point of the RG transformation without infinite
 The Paradox of Infinite Limits 345

iterations. I have not seen any sketch of how this is to be done. The point is that
the fixed point, as just noted, determines the behavior of the flow in its neigh-
borhood. If we want to explain the universal behavior of finite but large systems
using the RG, then we need to find a fixed point and, to my knowledge, this
requires an infinite system. (2017, p. 571)

And, later on in the paper, he argues

Kadanoff ’s understanding of the new RG theory of critical phenomena reflects

a different conception of the role of asymptotics and infinities. The kind of
upscaling that leads to an understanding of the universal macroscopic beha-
vior of micro-​diverse systems is different than the upscaling provided in the
ensemble averaging of mean field theory. (p. 573)

However, we resist the conclusion that the appeal to renormalization group

theory gives rise to an essential idealization. For the purpose of our argument, it
is useful to distinguish between the roles played by the different limits involved
in this approach, namely the thermodynamical limit entailing N → ∞ for the
number of molecules and the limit n → ∞ for the number of iterations employed
in the use of renormalization group theory. As is well known, in order to define a
system with an infinite correlation length, one needs to take the thermodynam-
ical limit. Granted, the latter may perhaps be interpreted as an idealization, in the
sense that it gives rise to a fictional system with an infinite number of particles.
Nonetheless, as Palacios (2019) pointed out, this limit, which appears essential
to find the non-​trivial fixed points that explain critical phenomena, can be also
demoted to an approximation. In order to understand why, one should note that
the two limits involved in the explanation of critical phenomena, i.e. N → ∞
and n → ∞, do not commute with each other: indeed, taking the limit N → ∞
followed by the limit n → ∞ yields empirically correct results, whereas taking
the limit n → ∞ before N → ∞ yields results that are not empirically correct.
What Palacios emphasized is that for finite systems, i.e. when N < ∞, one should
not take the second infinite limit if one wants to obtain empirically correct results.
In fact, she showed—​based on the work done by Wilson and Kogut (1974)—​that
in finite systems one can find effective fixed point solutions that approximate the
desired behavior for a large but finite number n of iterations, which means that
one obtains empirically correct results “on the way” to the second limit n → ∞
as well as “on the way” to the thermodynamic limit N → ∞.
These results are relevant for our answer to the aforementioned question be-
cause they indicate that the Paradox of Infinite Limits can again be resolved by
resorting to the notion of approximations, like in the examples discussed in
the previous sections. Note, first of all, that the variable n corresponding to the
346 Contemporary Scientific Realism

number of iterations is not a physical parameter, hence in principle one may not
violate statement (I) even if one takes the limit n → ∞. However, since this limit
does not commute with the thermodynamical limit, taking the former implies
taking the latter limit too, and thus if n grows to infinity so does the number N
of particles in the gas, thereby violating statement (I). Instead, the resolution to
the paradox comes once again by rejecting statement (II), which says that the
explanation of the phenomenon P can only be given by means of claims about an
infinite system constructed in the limit n → ∞. In fact, the procedure outlined
earlier shows how the two limits N → ∞ and n → ∞ can be dispensed by means
of the notion of approximation: accordingly, one can satisfy the condition of em-
pirical correctness without committing to the limit system S∞. In light of this
analysis, we conclude that one has good reason to believe that renormalization
group methods, at least in the example we considered, do not pose by themselves
any challenge for a form of scientific realism allowing for approximate truth.

14.7 Conclusion

In this paper, we addressed the issue of whether or not the use of infinite limits
in physics raises a problem of compatibility with scientific realism. For this pur-
pose, we surveyed various physical examples where infinite limits are invoked
in order to describe real target systems, so as to offer a taxonomy of the dif-
ferent uses of infinite limits (Section 14.3). In particular, we distinguished be-
tween approximations that do not constitute idealizations, idealizations yielding
approximations, essential idealizations, and abstractions, which we then
discussed in greater details in the subsequent sections. We argued in Section 14.5
that a challenge for scientific realism arises just when infinite limits are intended
as essential idealizations, due to the fact that it appears as if empirically correct
results can be recovered only in the limit. However, if one commits to the limit
system in that it seems indispensable, one runs against the Paradox of Infinite
Limits since the limit system is infinite whereas real systems are necessarily fi-
nite. We then went on to suggest how the ensuing worries for scientific realism
can be assuaged in concrete examples, e.g. in the much debated case of classical
phase transitions. The strategy to do so is to show, without referring to the in-
finite system, that the limit values of some physical quantities of interest yield
approximations of the values obtained for realistic systems. That is of course
compatible with a form of scientific realism that allows for theories to be just suf-
ficiently close to the truth. In the same vein, as we explained in Section 14.4 and
Section 14.6 by means of physical examples such as that of thermodynamically
reversible processes and that of continuous phase transitions, understanding
 The Paradox of Infinite Limits 347

infinite limits as approximations that do not constitute idealizations or even as

abstractions, respectively, does not raise any further issue for scientific realism.

Acknowledgments

The authors thank two anonymous referees for helpful and constructive
comments. Giovanni Valente acknowledges financial support from the Italian
Ministry of Education, Universities and Research (MIUR) through the grant
n. 201743F9YE (PRIN 2017 project “From models to decisions”).

References
Aronson, J. (1990). Verisimilitude and types of hierarchies. Philosophical Topics 18, 5–​28.
Baker, A. (2005). Are there genuine mathematical explanations of physical phenomena?
Mind 114(454), 223–​238.
Baker, A. (2009). Mathematical explanation in science. British Journal for the Philosophy
of Science 60(3), 611–​633.
Baron, S. (2016). The explanatory dispensability of idealizations. Synthese 193(2),
365–​386.
Baron, S. (2019). Infinite lies and explanatory ties: Idealization in phase transitions.
Synthese 196, 1939–​1961.
Batterman, R. W. (2001). The devil in the details: Asymptotic reasoning in explanation, re-
duction, and emergence. New York: Oxford University Press.
Batterman, R. W. (2005). Critical phenomena and breaking drops: Infinite idealizations
in physics. Studies in History and Philosophy of Science Part B: Studies in History and
Modern Physics 36(2), 225–​244.
Batterman, R. W. (2011). Emergence, singularities and symmetry breaking. Foundations
of Physics 41(6), 1031–​1050.
Batterman, R. W. (2017). Philosophical implications of Kadanoff ’s work on the renormal-
ization group. Journal of Statistical Physics 167(3–​4), 559–​574.
Boyd, R. (1983). On the current status of the issue of scientific realism. Erkenntnis
19(1–​3), 45–​90.
Butterfield, J. (2011). Less is different: Emergence and reduction reconciled. Foundations
of Physics 41(6), 1065–​1135.
Callender, C. (2001). Taking thermodynamics too seriously. Studies in History and
Philosophy of Science Part B: Studies in History and Philosophy of Modern Physics 32(4),
539–​553.
Cartwright, N. (1994). Nature’s capacities and their measurement. New York: Oxford
University Press.
Chakravartty, A. (2010). Truth and Representation in Science: Two Inspirations from
Art. In R. Frigg and M. Hunter (Eds.), Beyond Mimesis and Convention: Representation
in Art and Science, pp. 33–​ 50, Boston Studies in the Philosophy of Science.
Dordrecht: Springer.
348 Contemporary Scientific Realism

Chakravartty, A. (2017). Scientific Realism. In Edward N. Zalta (Ed.), Stanford

Encyclopedia of Philosophy (Summer 2017 Edition): available in: https://​plato.stanford.
edu/​archives/​sum2017/​entries/​scientific-​realism/​.
Colyvan, M. (2001). The indispensability of mathematics. New York: Oxford
University Press.
Earman, J. (2019). The role of idealizations in the Aharonov-​Bohm effect. Synthese 196(5),
1991–​2019.
Ehrenfest-​Afanassjewa, T. (1956). Die Grundlagen der Thermodynamik. Leiden: E.J.Brill.
Fisher, M. E., and A. N. Berker (1982). Scaling for first-​order phase transitions in thermo-
dynamic and finite systems. Physical Review B 26(5), 2507.
Fletcher, S., P. Palacios, L. Ruetsche, and E. Shech (2019). Special issue: Infinite
idealizations in science. Synthese 196(5), 1657–​2019.
Fraser, J. (2016). Spontaneous symmetry breaking in finite systems. Philosophy of
Science 83(4).
Godfrey-​ Smith, P. (2009). Abstractions, idealizations, and evolutionary biology. In
Mapping the future of biology, pp. 47–​56. Springer, Dordrecht.
Goldenfeld, N. (1992). Lectures on phase transitions and the renormalization group.
Westview Press.
Jones, M. (2005). Idealization and abstraction: A framework. Poznan Studies in the
Philosophy of the Sciences and the Humanities 86(1), 173–​218.
Jones, N. (2006). Ineliminable idealizations, phase transitions, and irreversibility.
Ph.D. thesis, Ohio State University.
Kadanoff, L. P. (2009). More is the same: Phase transitions and mean field theories. Journal
of Statistical Physics 137(5–​6), 777.
Kitcher, P. (1993). The advancement of science: Science without legend, objectivity without
illusions. Oxford: Oxford University Press.
Landsman, N. P. (2013). Spontaneous symmetry breaking in quantum systems: Emergence
or reduction? Studies in History and Philosophy of Science Part B: Studies in History and
Modern Physics 44(4), 379–​394.
Liu, C. (2019). Infinite idealization and contextual realism. Synthese 196, 1885–​1918.
Liu, C. and G. Emch (2005). Explaining quantum spontaneous symmetry breaking.
Studies in History and Philosophy of Science Part B: Studies in History and Modern
Physics 36(1), 137–​163.
Mainwood, P. (2006). Phase transitions in finite systems. Ph.D. thesis, University of Oxford.
McMullin, E. (1985). Galilean idealization. Studies in History and Philosophy of Science
Part A 16(3), 247–​273.
Menon, T. and C. Callender (2013). Turn and face the strange . . . Ch-​ ch-​
changes: Philosophical questions raised by phase transitions. In R. Batterman (Ed.),
The Oxford Handbook of Philosophy of Physics, pp. 189–​223. New York: Oxford
University Press.
Morrison, M. (2012). Emergent physics and micro-​ontology. Philosophy of Science 79(1),
141–​166.
Norton, J. (2016). The impossible process: Thermodynamic reversibility. Studies in History
and Philosophy of Science Part B: Studies in History and Modern Physics 55(1), 43–​61.
Norton, J. D. (2012). Approximation and idealizations: Why the difference matters.
Philosophy of Science 79(2), 207–​232.
 The Paradox of Infinite Limits 349

Norton, J. D. (2014). Infinite idealizations. In M. C. Galavotti, E. Nemeth, and F. Stadler

(Eds.), European Philosophy of Science—​Philosophy of Science in Europe and the
Viennese Heritage, pp. 197–​210. Dordrecht: Springer.
Oddie, G. (1986). Likeness to truth. Dordrecht: Reidel.
Palacios, P. (2018). Had we but world enough, and time . . . but we don‘t!: Justifying the
thermodynamic and infinite-​ time limits in statistical mechanics. Foundations of
Physics 5(48), 526–​541.
Palacios, P. (2019). Phase transitions: A challenge for intertheoretic reduction? Philosophy
of Science 86(4), 612–​640.
Popper, K. (1972). Conjectures and refutations: The Growth of Knowledge. 4th ed.
London: Routledge & Kegan Paul.
Psillos, S. (1999). Scientific realism: How science tracks truth. London: Routledge.
Putnam, H. (1975). Mathematics, matter and method. Cambridge: Cambridge
University Press.
Rosen, G. (1994). What is constructive empiricism? Philosophical Studies 74(2), 146–​178.
Ruetsche, L. (2011). Interpreting quantum theories. Oxford: Oxford University Press.
Schmelzer, J., and H. Ulbricht (1987). Thermodynamics of finite systems and the ki-
netics of first-​order phase transitions. Journal of Colloid and Interface Science 117(2),
325–​338.
Shech, E. (2013). What is the paradox of phase transitions. Philosophy of Science 80(5),
1170–​1180.
Shech, E. (2014). Scientific misrepresentation and guides to ontology: The need for repre-
sentational code and contents. Synthese 192, 3463–​3485.
Shech, E. (2019). Philosophical issues concerning phase transitions and
anyons: Emergence, reduction, and explanatory fictions. Erkenntnis 84, 585–​615.
Shech, E. (2019). Infinitesimal idealization, easy road nominalism, and fractional
quantum statistics. Synthese 196, 1963–​1990 forthcoming.
Tichy, P. (1976). Verisimilitude redefined. British Journal for the Philosophy of Science
27(1), 25–​42.
Valente, G. (2019). On the paradox of reversible processes in thermodynamics. Synthese
196(5), 1761–​1781.
van Fraassen, B. C. (1980). The scientific image. Oxford: Oxford University Press.
Weisberg, M. (2007). Three kinds of idealization. The Journal of Philosophy 104(12),
639–​659.
Wilson, K., and J. Kogut (1974). The renormalization group and the ε expansion. Physics
Reports 12(2), 75–​199.
Woodward, J. (2014). Scientific explanation. Stanford Encyclopedia of Philosophy, avail-
able in: https://​plato.stanford.edu/​entries/​scientific-​explanation/​.
 15
Realist Representations of Particles
The Standard Model, Top Down and Bottom Up
Anjan Chakravartty

Department of Philosophy
University of Miami
[email protected]

It is the sense in which Tycho and Kepler do not observe the same thing which
must be grasped if one is to understand disagreements within microphysics.
Fundamental physics is primarily a search for intelligibility—​it is a philosophy
of matter. Only secondarily is it a search for objects and facts (though the two
endeavors are as hand and glove). Microphysicists seek new modes of concep-
tual organization. If that can be done the finding of new entities will follow.
Norwood Russell Hanson (1965/​1958, pp. 18–​19)

15.1 Fixing the Content of Realism: Reference

and Description

Scientific realism is commonly understood as the idea that our best scientific
theories, read literally as descriptions of a mind-​independent world, afford
knowledge of their subject matters independently of the question of whether
they are detectable with the unaided senses or, in some cases, detectable at all.
It is a staple of the field of history and philosophy of science to wonder whether
any such prescription for interpreting theories (and models and other scientific
representations; I will take this as read henceforth) is plausible given the history
of theory change in specific domains of the sciences. A lot of ink has been spilled
on the question of whether, or under what circumstances, a realist interpreta-
tion of theories is reasonable. Antirealists of various kinds have argued that given
the lessons of changing descriptions of targets of scientific interest over time,
adhering to realism is something of a fool’s errand. Conversely, realists of var-
ious kinds—​often referred to as selective realists—​have sought to identify some

Anjan Chakravartty, Realist Representations of Particles In: Contemporary Scientific Realism.

Edited by: Timothy D. Lyons and Peter Vickers, Oxford University Press. © Oxford University Press 2021.
DOI: 10.1093/​oso/​9780190946814.003.0015
 Realist Representations of Particles 351

principled part of theories regarding which there has been continuity across
theory change in the past, thus fostering the reasonableness of expectations of
continuity in the future.
My focus in this essay is not the historical framing of these particular debates
about realism per se, but rather a key feature of them that amounts to a more
general problematic for realism. The shared strategy among selective realists for
dealing with descriptive discontinuity across historical theory change has been,
unsurprisingly, selectivity in what they take to be correct about a theory, pro-
posed in discussions of how certain claims have had or do have greater epistemic
warrant than others, such that continuity regarding these claims may then serve
as a bulwark for realism even while discontinuity rules more generally. Hence the
now familiar maneuver of associating realism with only certain parts of theories,
such as those involved in making successful novel predictions, or concerning ex-
perimental entities or mathematical structures. In each case we find concomitant
arguments about the typical preservation of the relevant parts of theories across
theory change, both as a reading of history and as a promise for the future.
Here emerges the key feature of debates surrounding the shared strategy of
selective realism on which I will focus. The hope of selective realism is that less
is more. By associating realist commitments with less, the hope is that it will be-
come easier to defend—​indeed, that it will amount to a plausible epistemology of
science. It is by no means easy, however, to know how much is enough. Consider,
in connection with any given theory, an imagined spectrum of epistemic
commitments one might make regarding its content. At one end of the spectrum
one believes almost nothing; at the other end, one believes everything the theory
states or suggests. Arguably, if realism is purchased at the cost of believing al-
most nothing, it is largely empty; if instead realism is made more substantial by
licensing ever greater quantities of substantive belief, it runs an ever greater risk
of (for example) falling prey to concerns arising from theory change. The realist,
then, in any given case, must perform a kind of balancing act appropriate to that
case. Let me label this challenge the realist tightrope. On one side, there is the
temptation to affirm less and less, and on the other, the temptation to affirm more
and more. Giving in to either of these temptations may spell disaster, but it is no
easy feat to get the balance just right.
The potential benefit to realism of walking the tightrope is wide-​ranging, in
that it is relevant to both historical and ahistorical defenses of the position. As
noted, if the realist were able to get the balance just right in some particular do-
main of science, she might then be in a position to furnish a narrative of con-
tinuity of warranted belief across theory change in that domain, past, present,
and future. But the tightrope is something that must be walked not only in con-
nection with historical lineages of theories, but also in connection with any
given theory, for it is often a challenge to work out how any one theory should be
352 Contemporary Scientific Realism

interpreted in a realist way. As Jones (1991) notes, it can be rather unclear how
best to articulate the subject matters of theories in physics (as per his examples),
where different interpretations can amount to different explanatory frameworks,
each suggesting a different ontology. This he presents as a challenge to realism,
which “envisions mature science as populating the world with a clearly defined
and described set of objects, properties, and processes, and progressing by steady
refinement of the descriptions and consequent clarification of the referential
taxonomy to a full blown correspondence with the natural order” (p. 186). In
this characterization of realism we catch a glimpse of how the tightrope must be
walked both synchronically and diachronically.
How shall we understand the spectrum of commitment, from thinner to more
substantial? I take it that an understanding of this is implicit in most discussions
of realism; indeed, it is implicit in Jones’s characterization. The thinnest possible
realist commitment is to the mere existence of something, which we capture
by speaking of successful reference. We say that the term “ribonucleic acid” or
“black hole” refers, which is (typically) shorthand for saying that it refers deter-
minately to something in the world. From here, commitment becomes increas-
ingly substantial as realists assert descriptions of the properties and relations of
these things. A very substantial commitment may involve asserting all of the
descriptions comprising or entailed by a theory, but it is often expressed other-
wise, not by believing everything—​often not in the cards in any case, since the-
ories often contain known idealizations and approximations—​but by asserting
increasingly detailed descriptions of whatever the realist does, in fact, en-
dorse. For example, an entity realist might describe the natures of the entities
she endorses in terms of certain causally efficacious properties. A structural re-
alist might describe the natures of these same entities in terms of certain struc-
tural relations. One can imagine yet further, finer-​grained descriptions of the
natures of these properties and structures. On the thin end of the spectrum we
have bare reference, and on the other end, ever more comprehensive or detailed
descriptions of the natures of the referents.
With this understanding in hand, a clearer picture of the realist tightrope
emerges. Believing as little as possible and thereby asserting successful refer-
ence alone, which might seem a more defensible position than believing sig-
nificantly more, might also seem to run the risk of rendering realism empty of
much content.1 The more one believes, however, the wider one opens the door to
both the epistemic peril of believing things that may be weeded out as theories
develop and improve, and the metaphysical peril of defending finer-​and finer-​
grained descriptions of the subject matters of realist commitment, in virtue of

1 Stanford 2015 argues that it would make realism into something that is, if not empty, so weak

that antirealists need not dispute it. I will return to this contention in section 15.5.
 Realist Representations of Particles 353

the inevitably and increasingly abstruse concepts and objections to which meta-
physical theorizing is prone. In some cases the epistemic and metaphysical perils
come together: in these cases, succumbing to the temptation to articulate the
natures of the referents of a theory in some fine-​grained detail yields descriptions
that may become outmoded by subsequent developments in the relevant science.
An examination of precisely this sort of case, to which I will turn now, forms the
backbone of what follows.
The Standard Model of particle physics, one of the landmark achievements of
twentieth-​century science, itemizes a taxonomy of subatomic particles along with
their properties and interactions. Beyond mere reference to these entities, how-
ever, their nature has been subject to realist wonderment and debate throughout
the history of theorizing in this domain. In ways that are well known and which
I will consider momentarily, the particles of the Standard Model are radically
unlike what could be imagined in classical physics—​thus providing an example
of how descriptions of the physical and metaphysical natures of something con-
ceived in connection with earlier theorizing would have to be relinquished in
light of subsequent theorizing. Just what the natures of these things enumer-
ated by the Standard Model are, however, is still far from clear. Indeed, if re-
alist attempts to characterize them are any guide, there is no consensus at all.
Advocates of different forms of selective realism, for example, have characterized
them in very different ways. And all the while, realists and antirealists alike have
suggested that in the absence of some unique characterization, realism is unten-
able. After considering various possibilities for thinking about the identity and
individuality of particles, for instance, van Fraassen (1991, p. 480) concludes that
we should say “good-​bye to metaphysics.” Ladyman (1998, p. 420) holds that tol-
erating metaphysical ambiguity regarding these issues would amount to a merely
“ersatz form of realism.”
In earlier work I have offered judgments that might be construed as echoing
these kinds of sentiments. “One cannot fully appreciate what it might mean to
be a realist until one has a clear picture of what one is being invited to be a re-
alist about” (Chakravartty 2007, p. 26). But having a clear picture is compatible,
I submit, with a number of different and defensible understandings of how best
to walk the realist tightrope in any given case. In the remainder of this essay I en-
deavor to explain why this is so, taking particles as a case study. In the next section
I briefly substantiate the contention that many questions regarding the natures
of particles are still very much up for grabs in contemporary physics and phi-
losophy of physics with a synopsis of a handful of the conceptual conundrums
surrounding them. In sections 15.3 and 15.4, I examine, respectively, what I de-
scribe as the two main approaches to thinking about the natures of particles and
their properties, which have in turn shaped varieties of selective realism—​“top-​
down” approaches, emphasizing formal, mathematical descriptions furnished
354 Contemporary Scientific Realism

by theory, and “bottom-​up” approaches, emphasizing causal interactions and

manipulations at the heart of experiment. In the final section, I argue that re-
alism is a commitment that can be shared, defensibly, by those who subscribe to
different and even conflicting conceptions of the natures of particles.

15.2 Searching for a Realist Interpretation of “Particles”

The electron is the veritable poster child of scientific realist commitment. Most
proponents of both realism and antirealism (rightly or wrongly) take various facts
about observable phenomena as uncontroversial, but seriously contest the status of the
unobservable. Though theories at every “level” of description—​from social and psy-
chological phenomena to biological and chemical phenomena through to the subject
matters of physics—​all theorize about putatively unobservable objects, events, pro-
cesses, and properties, realists often cite subatomic particles as a shining example of
entities conducive to realism. On the one hand, given the mind-​boggling success of the
uses to which we have put twentieth-​century theories concerning them, not least in
a host of startlingly effective technologies from computing to telecommunications to
medical imaging, this may not seem surprising. On the other hand, perhaps it should
be a cause for concern after all, because even a cursory foray into our best attempts to
grasp the natures of particles and particle behavior are fraught with conceptual diffi-
culties, and our best scientific theories are far from transparent on this particular score.
Trouble rears its head at the start with the term “particle.” What is a particle
in this domain? There is, of course, a classical conception of what a particle is,
which is easily graspable and conceptually undemanding, relatively speaking,
but this conception is simply inapplicable at atomic and subatomic scales in light
of twentieth-​century developments in theory and experiment. Classical particles
are solid entities that can be envisioned colliding with and recoiling from one
another in the ways that billiard balls appear to behave, phenomenologically. The
natures of particles described by the Standard Model are not in this way intelli-
gible. They appear to behave in the manner of discrete entities in some contexts
but like continuous, wave-​like entities in others. It is unclear whether all of their
properties are well defined at all times, though we can ostensibly detect them and
measure their values under certain conditions. As intimated earlier, there is con-
troversy as to whether they can be regarded in any compelling way as individ-
uals; if they can, it is certainly not in the way we commonly think about identity
conditions and individuation in classical contexts, in terms of differential pro-
perty ascription and allowing for their re-​identification over time.2 The natures

2 Arguably, one may overstate this last point. Cf. Saunders 2006, p. 61: “there are many classical

objects (shadows, droplets of water, patches of colour) that likewise may not be identifiable over time.”
 Realist Representations of Particles 355

of particles conceived today may become enmeshed over arbitrarily large spatial
distances—​as per quantum entanglement—​in ways not previously conceived.
All of this suggests the strangeness of particles from a classical point of view,
but this should not be inimical to realism all by itself—​if one adopts a thinner
conception of realism about particles framed in terms of reference. It is in the na-
ture of theoretical development that sometimes the features of target systems of
scientific interest that come to be viewed in a new light will appear strange from
the point of view of what came before. This by itself suggests only a common and
understandable propensity to regard the unfamiliar as strange. The challenge to
realism here is no mere strangeness en passant, but the fact that in this case in
particular, our attempts to interpret our theories in more substantial ways, so
as to make their content intelligible to ourselves, have resulted in a great deal of
unsettled debate and lasting conceptual puzzlement. Taking this as a basis for re-
alism regarding a more substantial conception of particles, an antirealist might
be forgiven for wondering whether an argument here against realism is in fact re-
quired, since it would appear that collectively, realists cannot themselves decide
what they should believe.
Consider, for example, the question of the basic ontological category to which
particles belong. There is a long history here of being flummoxed, even upon
careful consideration of the relevant physics, regarding what this might be. The
term particle is commonly associated with the notion of objecthood, but we have
already noted that particles cannot be objects if “objecthood” is allowed to carry
classical connotations. In quantum field theory, particles are often described as
modes of excitation of a quantum field. This does not sound very object-​like in
any traditional sense, which leads many to claim that particles are not objects
after all. Yet even physicists who observe that with the advent of quantum field
theory the ontology of the quantum realm might be thought of in terms of
fields rather than particles are happy to talk about particles at will. This suggests
that they either regard particle-​talk as merely elliptical for states of fields, or
that they do in fact regard particles as non-​classical objects of some sort, per-
haps standing in some sort of dependence relation to fields. Generally, there is
nothing like a precise specification of a basic ontology to be found, thus leaving
the answer to the question of an appropriate assignment of category ambiguous.
Here one might think that philosophers would lend a hand, but a sampling of the
views of philosophers merely reveals the trading of ambiguity for transparent
disagreements.3

3 The sample to follow is by no means comprehensive, given the long history of differing interpret-

ations, but representative of some of the most recent literature. Not everyone cited is a realist, neces-
sarily, but all are attempting to clarify the relevant ontology.
356 Contemporary Scientific Realism

For example, Jantzen (2011) considers the permutation invariance of

particles: representations of a physical state of particles of the same type that
interchange the particles are taken to represent the same state. This is fatal, he
thinks, to a particle ontology, where particles of a type are discrete objects that
share state-​independent properties and have state-​dependent monadic proper-
ties, conceived as “approximately independent” (p. 42) of the properties of other
things. Is it obvious, however, that objects need have precisely these features?
Necessary conditions for objecthood are themselves up for grabs. Thus, while
Bain (2011) acknowledges the prevalent view that relativistic quantum field the-
ories are not amenable to particle interpretations if this requires that particles
be localizable and countable (as expressed in terms of local and unique total
number operators, which these theories do not support; see Fraser 2008), he
argues that since the theorems on which this conclusion is based do not hold
for non-​relativistic quantum field theories, the characterization of localizability
and countability at issue here must depend on classical features of these latter
theories (specifically, regarding the structure of absolute spacetimes) that do not
apply to the former theories. This suggests that the characterization of objects
assumed here is inapplicable to non-​relativistic quantum field theories, which
then leaves open the possibility of a different conception of particles (or localiz-
ability and countability) in this different theoretical context.
No doubt the possibility of retooling our concept of objecthood in such a
way as to admit particles will not appeal to everyone—​perhaps it is too much
of a promissory note on which to rest a substantial realism. In that case, per-
haps a field interpretation of “particle” ontology is the way to go. Like many
others, however, Baker (2009) and Bigaj (2018) hold that while particle inter-
pretations of quantum field theory are problematic, the same is true of field
interpretations. Perhaps we could simply stop worrying about what ontolog-
ical category particles inhabit and instead satisfy ourselves with a clear under-
standing of the nature of their properties. Alas, further difficulties await, for
the precise natures of these properties are themselves elusive. Consider the
property of spin, which is one of a handful whose values are viewed as neces-
sary and jointly sufficient for classifying particles into their respective kinds.
What is spin? It is very difficult to say. Spin is usually described as a “sort
of ” internal or intrinsic angular momentum; the scare quotes are essential
to the description, because there is no analogue of this property in everyday
experience or otherwise familiar terms that allows for a more “visualizable,”
physical, dynamical (despite the term “spin” connoting some sort of rotation)
understanding of it. Spin is causally exploited in many technologies including
microscopy (more on which in section 15.3), and the Standard Model gives us
a mathematical framework for discussing it (more on which in section 15.4),
but it is difficult to say what it is.
 Realist Representations of Particles 357

At the very least, in the absence of an intuitive grasp of the natures of prop-
erties of the sort just suggested, perhaps we could say something about them in
terms that philosophers find perspicuous. Are these properties monadic, dyadic,
or polyadic? Are they intrinsic or extrinsic or essentially relational? Mass is com-
monly cited as an exemplar of an intrinsic property, but Bauer (2011) thinks that
it is extrinsic, since it is “grounded” in and thus ontologically dependent on the
Higgs field. French and McKenzie (2012) contend that there are no fundamental
intrinsic properties by means of an argument appealing to gauge theory,4 which
is integral to the Standard Model, but Livanios (2012) does not find this argu-
ment compelling. Lyre (2012, p. 170) maintains that properties such as mass,
charge, and spin are “structurally derived intrinsic properties,” which suggests
something of a hybrid, intrinsic-​extrinsic nature. And in some cases an exami-
nation of the natures of these properties brings us full circle, back to a consider-
ation of the ontological category of things that best corresponds to particle-​talk,
as when Berghofer (2018) holds that the relevant properties are, in fact, in-
trinsic and non-​relational, but features of fields, not particles, and when Muller
(2015, p. 201) contends that we would be better off with a new conception of
objects: particles, he argues, are “relationals”; “objects that can be discerned by
means of relations only and not by properties.”
The purpose of the preceding whirlwind tour has not been to suggest that
progress cannot be made on questions surrounding the ontological natures of
particles. No doubt some and perhaps many of the issues disputed in the pre-
ceding discussion may ultimately be resolved in ways that produce a measure of
consensus. The point here is a different one. If, in order that realism be a tenable
epistemic attitude to adopt in connection with the Standard Model, we were to
require a degree of communally sanctioned, fine-​grained clarity regarding de-
scription that could only follow from having resolved all such debates, this would
suggest a prima facie challenge to the very possibility of realism here and now. As
the brief glimpse into a number of contemporary debates just presented makes
plain, any attempt to clarify the ontology of the Standard Model quickly and in-
evitably draws one into contentious metaphysical discussions. In the following
two sections I will examine the two overarching approaches to prosecuting these
debates that have formed the basis of selective realist pronouncements regarding
the natures of particles and their properties, with the eventual goal of arguing for
a rapprochement between them qua realism—​one that clarifies how realism can
be a tenable, shared epistemic commitment even in cases where realists disagree
about details of description.

4 Offering a different argument, McKenzie 2016 takes this position with respect to various proper-

ties including mass, but excluding spin and parity.

358 Contemporary Scientific Realism

15.3 The Nature of Particles I: Top Down

As it happens, the two approaches to thinking about how best to interpret the
Standard Model I have in mind reflect a longstanding division of labor within
the community of physicists. On the one hand there is theoretical physics,
which views particles through the lens of formal, mathematical descriptions fur-
nished by theory, and on the other hand there is experimental physics, which
views particles through the lens of the sorts of detections and manipulations
of them that are part and parcel of laboratory practice. These communities of
scientists are not, of course, strictly isolated from one another; they must often
work together. Nevertheless, their approaches to the subject matter are of ne-
cessity shaped by the kinds of work they do. Corresponding to this rough di-
vision, in the philosophy of science, there are what I will refer to as “top-​down”
approaches to interpretation, which place primary emphasis on mathematical
descriptions of the properties and interactions of particles found in high theory
as a source of insight into the natures of particles in the world; on the flipside
there are what I will call “bottom-​up” approaches to interpreting the natures of
particles, which place primary emphasis on their behaviors in the trenches of
concrete interventions characteristic of experimental investigation. As we will
see, both approaches offer insight into the natures of particles, and both leave
important questions open.
Let us begin with the top-​down approach to thinking about particles. The
Standard Model provides a remarkably elegant description of fundamental
particles, their properties, and their (electromagnetic, weak, and strong)
interactions, neatly systematizing them by means of symmetry principles.
A symmetry is a transformation (or a group of transformations) of an en-
tity or a theory in which certain features of these things are unchanged. That
is, the relevant features are preserved or remain invariant under the trans-
formation. The kinds of transformations relevant to physical descriptions in-
clude translations in space or time or spacetime, reflections, rotations, boosts
of certain quantities such as velocity, and gauge transformations. When the
states of systems are related by symmetries, they have the same values of cer-
tain quantities or properties—​including those used to classify particles, such
as mass, charge, and spin. To take an everyday example, if one rotates a square
by 90 degrees, one gets back a square. “Squareness” is invariant under this
transformation, which maps the square onto itself. With the notion of a sym-
metry in hand we may define a symmetry group as a mathematical structure
comprising the set of all transformations that leave an entity unchanged to-
gether with the operation of composition of transformations on this set (satis-
fying the conditions: associativity; having an identity element; every element
having an inverse).
 Realist Representations of Particles 359

There are a variety of accounts of realism about particles that one might char-
acterize as top down. What they have in common is the (explicit or implicit)
operating principle that insight regarding the natures of particles should be in-
timately and exclusively connected to interpreting the mathematical formalism
I have just described. This is all we need to understand the natures of particles,
nothing more. There is nothing in my description of the top-​down approach that
suggests that it should provide exclusive insight into the natures of particles, but
in practice, this is how realists who take this approach proceed. To take this extra
step from adopting a top-​down approach to thinking, furthermore, that this is
our best or only legitimate source of insight into the natures of particles requires
some further motivation or argument. Let me now briefly consider a couple of
arguments of this sort, and for each suggest one of two things: either the top-​
down description of particles provided does not preclude supplementation with
bottom-​up description; or if it does, it is unclear why the top-​down characteri-
zation should be judged superior qua realism. Obviously, this will not amount
to a comprehensive survey of all possible arguments for an exclusive commit-
ment to the top-​down approach. Nonetheless, I take it to be suggestive of a plau-
sible general moral, that the necessity or irresistible appeal of this commitment
is unproven.
Motivating at least some realists who are exclusively committed to top-​down
characterizations of particles are desiderata such as descriptive or ontological
intelligibility or simplicity. In section 15.4 we will see in some detail how the
bottom-​up approach is typified by an ontologically robust understanding of the
causal or modal natures of the properties of particles, but for the time being it
will suffice to note that some realists who focus their attention on symmetries
hold that this focus alone is sufficient for understanding the natures of these
properties, thus precluding any “inflation” of our ontological commitments in
ways recommended by bottom-​up realists about particles. If the Standard Model
is simply interpreted as describing properties such as mass, charge, and spin as
invariants of certain symmetry groups, we might rest content with this purely
mathematical, theoretical apparatus for describing the natures of properties. On
this view it is unnecessary to appeal to the causal roles of things in order to iden-
tify or understand them. Armed with symmetries, we might then understand
the natures of the relevant properties without appealing to the notion of causal
features or roles at all, thus articulating realism in terms of a simpler ontological
picture.
Let us then consider whether the content of a realism about particles can
be provided solely through an examination of symmetries and invariants. It is
difficult to see how it could, given that questions about what realists justifiably
believe cannot be separated from matters of how evidence furnishes justifica-
tion. Detection and measurement are intimately connected to determining what
360 Contemporary Scientific Realism

things there are, in fact, in the world. Are properties thus conceived, as things
one might identify as existing or being exemplified in the physical world—​things
about which one might be a scientific realist—​identified independently of their
causal roles? Perhaps there are cases in which this happens, but the present case
does not seem like one. While there is no doubt that symmetry groups comprise
a beautiful framework for codifying particles and their properties, all that exam-
ining them can achieve in isolation is to generate descriptions of candidate enti-
ties that may then be put to the test of experimental detection. It is one thing to
describe the natures of some target of realist commitment in terms of the formal
or mathematical aspects of a theory, but generally, in order for descriptions
to have the sort of content required to support realism, they must be taken to
refer to some thing or things in the world, and establishing successful reference
requires more than the examination of a formalism. Some supplementation
seems necessary.
A nice illustration of this is furnished by permutation symmetry, which arose
earlier in section 15.2. Recall that in quantum theory, state representations of
particles of the same type in which the particles are interchanged do not count
as representing different states of affairs. An examination of the permutation
group yields certain “irreducible representations” corresponding to all of the
particles, the fermions and bosons, populating the Standard Model. However, in
addition to these fermionic and bosonic representations, there are also so-​called
“paraparticle representations,” and unlike fermions and bosons, paraparticles do
not appear to exist—​at least, not in the actual world subject to scientific realism.5
Thus, merely examining the mathematical formalism of the theory is insufficient
for the identification of entities to which realists should commit. To avoid being
misled about the ontology of the world, there would seem to be no substitute for
getting one’s hands dirty with the causal roles of properties in the context of ex-
perimental work using detectors, and this suggests that there may be something
to the thought that properties of particles have some sort of causal efficacy after
all, in virtue of which they are amenable to detection, measurement, manipu-
lation, and so on. But now we have entered the territory of the bottom-​up ap-
proach to understanding the natures of particles, to which we will return in the
following section.
Let us consider a second possible motivation for an exclusive reliance on
the top-​down approach for the purpose of illuminating particles. Perhaps the
boldest motivation yet proposed stems from a version of selective realism that
was designed specifically (in the first instance) to serve as an account befitting

5 The closest we have come to generating empirical evidence in this sphere is the detection of

paraparticle-​like states, though not paraparticles themselves, under very special conditions. For a
brief discussion, see Chakravartty 2019, p. 14.
 Realist Representations of Particles 361

fundamental physics: ontic structural realism. There are many variants of the
view, but generically, the common thread is what one might call a reversal of the
ontological priority traditionally associated with objects and properties relative
to their relations. In much traditional metaphysics, objects and/​or properties are
conceived as having forms of existence whereby their relations are in some way
derivative (and not vice versa). Some variants of ontic structural realism simply
boost the ontological “weight” of the relevant relations relative to their relata
such that they are all on a par, ontologically speaking. Others take the relations to
have greater ontological priority, and the most revisionary formulations do away
with objects and properties altogether, eliminating them in favor of structural re-
lations which are then viewed as ontologically subsistent in their own right, thus
constituting the concrete furniture of the world.6 If particles are conceived as
being entirely dependent on relations described by symmetries—​or stronger yet,
as epiphenomena of these relations—​it may well seem that a top-​down approach
to describing their natures should be sufficient.
As is true regarding any proposal for realism there are several aspects of this
view that one might seek to clarify, but perhaps the most fundamental concern
that has been raised is whether ontic structural realism can render intelligible the
idea that things described in purely mathematical terms—​such as symmetries
and invariants, which are standardly regarded as (at best) abstract entities—​can
be understood to constitute the world of the concrete. Merely stipulating that
some mathematical structures are subsistent appears to achieve no more than to
substitute the term “concrete” with “subsistent.” Something more is needed, and
no doubt with this in mind, advocates of the position sometimes explicate the
sense of concreteness or subsistence at issue by saying that the relevant structures
are causal or modal.7 Esfeld (2009, p. 180), for example, is explicit that on his
variant of ontic structural realism, “fundamental physical structures are causal
structures.” French (2014, p. 231) is clear that on his, “we should take laws and
symmetries—​and hence the structure of which these are features—​as inherently,
or primitively, modal,” and take this de re modality as serving the explanatory
functions commonly associated with attributions of causality, such as helping us
to explain what it means for something to be concrete.
If one goes this route, however, the first of our potential rationales for
favoring a top-​down approach to interpreting the natures of particles based
on the promise of a comparatively simple or streamlined ontology is ruined,
because it is difficult to see how a reification of symmetries and other mathe-
matical structures endowed with causal or modal efficacy should count as less

6 For a detailed and comprehensive exploration of the many variants, see Ladyman 2014/​2007.
7 Cf. Ben-​Menahem 2018, p. 14: “causal relations and constraints go beyond purely mathematical
constraints; they are (at least part of) what we add to mathematics to get physics.”
362 Contemporary Scientific Realism

ontologically inflationary than an understanding of particles based on an onto-

logically robust conception of causal or modal properties, as suggested on the
bottom-​up approach, to which we will turn next. There is no obvious reason to
think that Occam’s razor should point us toward the former and away from the
latter. As a guide to a description of the natures of particles that might satisfy the
aspiration to walk the realist tightrope by adding something substantial to an
otherwise spartan commitment to successful reference, the top-​down approach
furnishes a great deal, but not in so compelling a manner as to make it an irre-
sistible choice of interpretation, or an exclusive choice, for realists. One reason
for this, as I will now suggest, is that the natures of particles look significantly
different from the bottom up.

15.4 The Nature of Particles II: Bottom Up

From the point of view of detection, which is intimately linked to many of the
strongest cases that can be made for realism in specific instances, more abstract
descriptions of the properties of things are somewhat removed from the work
of physics. Where the focus of experimental work is the physical discernment
of interactions between particles and between particles and detectors, often
requiring extraordinarily precise adjustments and manipulations of both the ex-
perimental apparatus and the target entities under investigation, more precise
descriptions of concrete natures are necessary. It is here that the determinate
properties of particles, whose values are detected and manipulated in such work,
take center stage. This is not to say that group theoretic structures are irrelevant to
describing these properties, but simply that the descriptions afforded by symme-
tries and invariants are at a remove from the specificities of experimental work.
As Morganti (2013, p. 101) puts it, “[w]‌hen one focuses on invariants . . . one
moves at a high level of abstractness.” In contexts of experimentation and detec-
tion, it is necessary to move in the direction of more determinate description;
the specific values of mass, charge, etc. (pertaining to different particles) at issue
in these contexts are not given by descriptions of symmetries (cf. Wolff 2012,
p. 617).
In the realm of experiment it is what we can do that is our best guide to what
there is and what these things are like—​that is, to the ontology of our targets
of investigation. What we can do in the arena of particle physics is entirely de-
pendent on the precise values of the properties of particles. Since all of this
doing involves designing and engineering instruments to interact with those
parts of the world we aim to explore, and generating certain kinds of effects, it
is natural to describe it in terms of causal interactions, relations, and processes.
Thus it is no surprise that selective realists who take a bottom-​up approach to
 Realist Representations of Particles 363

understanding the natures of particles typically emphasize the roles of deter-

minate property values in causal interactions, relations, and processes. Perhaps
the most obvious (but not the only) example of a position taking this approach
is entity realism, wherein certain descriptions of the causal roles of entities are
interpreted as the basis of experimental work that ostensibly generates an ar-
gument for this form of realism. Some who take these descriptions seriously as
filling in our understandings of the natures of particles go further, giving more
detailed characterizations of the natures of their properties as being inherently
causal or modal, invoking conceptions of properties such as dispositions, pro-
pensities, and capacities—​types of properties whose natures comprise abilities to
do certain things.
Having seen just a moment ago how the top-​down approach lends itself to
increasingly detailed descriptions of the natures of particles, it may now be clear
(on the basis of the preceding paragraph) that the same is true here. It is all too
easy to drift from an initial question regarding the nature of some target of one’s
realism to further, deeper questions, and in attempting to answer these deeper
questions, giving ever more detailed descriptions—​all the while with little con-
cern for the realist tightrope. Just as one might, from the top down, begin by
thinking that the descriptive content of one’s realism should be informed by a
specification of the relevant symmetry groups, but then end up some way down
the road, after twists and turns of elaboration, advocating for reified mathemat-
ical structures imbued with primitive causality or modality, one may perform
analogous feats from the bottom up. One might begin by thinking that the causal
roles of certain properties associated with particles are central to the descriptive
content of one’s realism, and then through earnest inquiry find oneself some-
where down the road defending one or another specific conception of causation,
in just the way that some come to understand the natures of these properties as
inherently modal or dispositional. Mirroring the moral of section 15.3, let me
now suggest that bottom-​up characterizations do not preclude supplementation
with top-​down description. And in some cases, it is unclear why either should be
judged superior qua realism.
With a long history of empiricist concerns about the intelligibility of disposi-
tional properties in the background, fueled by concerns about the metaphysical
excesses of scholastic and neo-​Aristotelian philosophy more generally, the notion
that properties of particles should be understood dispositionally is unsurpris-
ingly controversial. A dispositional essentialist, for example, takes the natures of
the relevant properties to be exhausted by dispositions for certain kinds of beha-
vior: the identities of these properties—​their essences—​are dispositional. Could
we not simply deflate this scholastic-​sounding reference to essences? Consider
Livanios’s (2010, p. 301) query: “if the identity of the fundamental physical prop-
erties . . . can be provided via symmetry considerations, why can’t we claim that
364 Contemporary Scientific Realism

being invariant under the action of fundamental symmetries is an essential fea-

ture of the fundamental physical properties?” From the bottom up there is an im-
mediate reply: what is the “action” of the symmetries? Presumably the intention
here is not to claim that a mathematical description—​a linguistic entity—​is part
of the essence of something in the world. This would be to conflate descriptions
with that which they describe. Thus, the point must be that the symmetries are
themselves things in the world. They are part of the ontology of the world and
thus conceived, they are part of the essences of fundamental properties. But now
the view is sounding a lot like dispositional essentialism, and certainly no less
weighty as metaphysics!
Two points can be extrapolated from this brief illustration of how opposing
metaphysical sensibilities sometimes play out in realist interpretations of sci-
entific theories and models. In replying to the attempt to deflate the substantial
metaphysical claim inherent in their proposal for how to understand the prop-
erties of particles, dispositional essentialists need not reject the top-​down ap-
proach simpliciter, or broadly conceived, for as we have already acknowledged,
there is nothing in group theoretic descriptions of symmetries and invariants
that is incompatible with their view—​on the contrary. Rather, it is simply the case
that their entirely reasonable, bottom-​up preoccupations regarding realism have
not been well appreciated by their critics, which we noted earlier in terms of the
relatively general or abstract knowledge afforded by symmetry groups and, in
contrast, the utter centrality of the determinate values of properties in setting up,
generating, detecting, and recording the effects of causal interactions and pro-
cesses. A second point worth noting, though I will not detail it here in connec-
tion with this particular example, is how elaborate metaphysical proposals may
become on any approach to explicating realism. Under the guise of empiricist
or neo-​Humean rejections of metaphysical excess, some pots call kettles black.
Both pots and kettles, however, threaten to topple the realist off of her tightrope.
What, then, of the determinate property values of particles and their associ-
ated causal profiles? Perhaps the most serious concern about emphasizing these
aspects of particles for the purpose of describing their natures is the worry that
no such account can adequately explain certain constraints we find exhibited in
their behaviors. For example, in any closed system, the values of properties such
as mass-​energy, momentum, charge, and spin are conserved—​their totals re-
main constant. Emmy Noether proved in 1915 that for every continuous global
symmetry of the Lagrangian there is a conserved quantity (and vice versa). But
how might the causal efficacy of a particle, understood by means of an account
of the causal profiles associated with its properties, explain the conservation of
properties in an ensemble of particles (cf. Bird 2007, p. 213)? It is difficult to see
how the properties of a particle can be parlayed into a constraint on a collection of
particles, given that the relevant constraint must pertain to the collection, not to
 Realist Representations of Particles 365

any given particle. Similarly, consider principles of so-​called least action: for any
given system and a specification of some initial and final conditions, the evolu-
tion of the state of the system will minimize a quantity referred to as “action.” It is
difficult to see how the causal profiles associated with a particle and its properties
could somehow generate the minimization of action in systems more generally.
In order to explain constraints on behavior such as those expressed in prin-
ciples of conservation and least action in terms of the causal natures or profiles
of properties, it would seem we must think of these properties as belonging to
the systems to which these principles apply, and this is inevitably controver-
sial. Taking the dispositional variant of the bottom-​up approach as an illus-
tration once again, Harré (1986, p. 295) maintains that some dispositions may
be grounded in “properties of the universe itself ”—​a phenomenon he labels
“ultragrounding”—​attributing the idea to Mach’s discussion of inertia in the
context of Newtonian thought experiments. Imagine two globes connected by
a spring balance and rotating, alone in the universe. Newtonians held that there
would be a force tending to separate the globes, registered in the spring balance,
but on a Machian reading there is no reason to believe that the globes would have
inertia in an otherwise empty universe; it is better to think of the disposition to
resist acceleration as grounded in the universe itself. Bigelow, Ellis, and Lierse
(1992, pp. 384–​385; cf. Ellis 2005) go so far as to contend that the actual world
is a member of a natural kind whose essence includes various symmetry prin-
ciples, conservation laws, and so on. It is a short step from this to thinking that
the system-​level behaviors associated with these principles are properties of the
world—​a very large system indeed.
The standard objection to this family of speculations is that it is ad hoc.8
Granted, explaining constraints on the behaviors of systems of particles in terms
of properties of the entire world may seem, prima facie, rather convenient, es-
pecially in the absence of any independent motivation for the explanans. If one
takes a bottom-​up approach to understanding the natures of target systems of
scientific interest, however, it is simply a mistake to suggest that there is no in-
dependent motivation. Just as particles are investigated empirically in carefully
designed and executed experiments, systems of particles are likewise investi-
gated. It is an empirical fact, not a convenient fact, that certain kinds of systems
exhibit behaviors that conform to various principles of conservation and least
action. Having adopted a methodology of associating causal profiles with certain
types of particles and their characteristic properties, it is hardly an unmotivated
extension of this methodology to do likewise in connection with certain types of
systems, such as closed systems.

8 For recent discussion on both sides of this fence, see Smart & Thébault 2015 and Livanios 2018.
366 Contemporary Scientific Realism

Now, as it happens, the world itself is a system of this type. From the perspec-
tive of the bottom up, the attribution of a causal profile to it on the basis of a con-
sideration of empirical investigations into members of the type cannot be said to
be based on merely wishful speculation. And neither should it seem peculiar in
the era of quantum theory, in which systems are routinely viewed as having prop-
erties, such as entanglement, that cannot be reduced to the properties of their
parts. What may appear superficially as ad hoc speculation inevitably sounds
more credible when the metaphysical terminology in terms of which it is some-
times expressed (“natural kinds,” “essential properties”) is given a plausible in-
terpretation in the language of scientific description. Consider, for example, the
possibility entertained earlier that particles are in fact best understood as field
quanta. In that case the properties of systems suggested above would be proper-
ties of fields, and fields are global in the sense that they permeate the whole of the
world. Anyone moved by this description would then be in a position to ask yet
further questions about the natures of particles and their properties, depending
on whether one is a substantivalist about fields, or interprets the values of field
quantities as properties of spacetime points, or . . . . But let us stop here.
Having refined and extended a bottom-​up description of particles in such a
way as to answer a preeminent concern, let us once again inquire into how this
approach fares in comparison to its counterpart, top down. Here, once again,
it is very difficult to make a case one way or the other on the basis of some
imagined criteria of descriptive or ontological intelligibility or simplicity. On
a permissive enough conception of causation one may see it as an appropriate
descriptor of many different things. Ben-​Menahem (2018), for instance, applies
the label “causal” to any general constraint on change, where constraints deter-
mine what may happen, or is likely to happen, or what cannot happen. On such
a conception, symmetries, conservations laws, and variational principles (such
as the principle of least action) all qualify as causal. But are they causal in the
sense advocated by someone looking top down, as a primitive feature of certain
mathematical structures, or are they causal in the sense of, say, a dispositionalist
looking bottom up, where the (potential for) behaviors associated with the prop-
erties of various kinds of entities and systems determine their identities? And
does anything hang on this choice, from the point of view of defending realism?
In closing, let me attempt to shed some light on the latter question.

15.5 The Content of Realism

Redux: Anchoring Interpretation

I began this essay by citing a celebrated challenge to scientific realism. Given

that no one thinks that most scientific theories and models (including many
 Realist Representations of Particles 367

of our very best ones) are entirely correct, not merely in connection with the
idealizations and approximations we know of, but also in other ways we have
yet to discover, everyone appreciates that they will evolve over time as scientific
inquiry proceeds. Hence the various strategies found among realists, especially
selective realists, for identifying those aspects of theories and models that have
sufficient warrant to command realist commitment, both as a guide to interpre-
tation in the present and to reasonable expectations about what will survive into
the future. From this I distilled a more specific challenge to anyone hoping to be
successfully selective, which I called the realist tightrope: believing too little or
too much of a theory that is not entirely correct may well appear to spell trouble.
Believing too little—​in the limit, the bare reference of central terms, or claims re-
garding the existence of their referents—​may seem tenuous, but the more com-
prehensive or detailed the descriptions of such things one endorses, the more
one runs the risk of falling foul of future developments in the relevant science,
and/​or metaphysical objections to the increasingly fine-​grained natures pro-
posed. Where does the proper balance lie?
In the case of the Standard Model, walking the tightrope seems especially
fraught. Understanding the natures of the particles described by the theory has
always been difficult, and serious proposals for illuminating these natures have
inevitably required increasingly speculative and technical theorizing. Given this
state of affairs, it is hardly surprising that there is so much disagreement among
realists about how best to describe what particles are, exactly. On the surface this
may appear a victory for antirealism, for if there is nothing determinate here to
be found under the heading of “realism” to which all realists subscribe, but in-
stead a thousand splintered commitments to conflicting and (what will appear
to some as) increasingly esoteric interpretations, the camp of realism may look
more like a ball of confusion than anything endorsing a shared epistemic com-
mitment. This dismal portrait of the cognitive landscape of realism in connec-
tion with the Standard Model is, however, though perhaps understandable on
the basis of what we have seen, entirely misleading. I submit that there is in fact
something substantial to which all or most realists subscribe in the context of the
Standard Model, even as they debate how this commitment is best elaborated.
Here in conclusion I will attempt to explain how this can be so.
In order to understand how different and conflicting descriptive commitments
among realists regarding particles are compatible with a shared commitment
qua realism, we could do worse than to begin by looking at how scientists ap-
proach this area of physics. Here, just as Jones (1991, p. 191) notes in connection
with the analogous case of multiple candidate interpretations of quantum me-
chanics, one may reasonably worry about “the failure of any interpretation to
provide an ‘explanatorily satisfactory’ link between the mathematical formalism
and the world of laboratory experience.” He continues:
368 Contemporary Scientific Realism

The general approach of one interpretation may suit a physicist more than the
general approach of others, and he or she may spend some time adapting it to
issues that he or she thinks particularly important and developing arguments
as to why its lacunae are not devastating for its coherence. But every physicist
will admit that such allegiance is to some degree a matter of taste. No physicist
is unaware of competing interpretations, and none expects decisive evidence or
arguments for one against the others.

Analogously, in the case of the Standard Model, the challenges of connecting

the domain of abstract theorizing, conceived in terms of interpreting a math-
ematical formalism, and that of concrete experimentation, conceived in terms
of interpreting laboratory experience, have a basis in the work of physics, all of
which is mirrored in philosophers’ attempts to elaborate the natures of the phe-
nomena revealed by these practices. And as I will now contend, just as physicists
across these domains can be realists despite differences in how they characterize
their shared subject matter, philosophers of science can too.
To begin, note that terms like “physics” and even “fundamental physics” are
rather broad designators. This is true not only in the sense that there are a variety
of subareas of physics to which these terms are applied, but also in the sense that
even within subareas, different approaches to one and the same subject matter
can and often do take the form of highly disparate forms of scientific practice.
Galison’s (1997) detailed study of what he describes as the partly autonomous
subcultures of physics in the twentieth century—​experimenting; theorizing;
and instrument making—​furnishes a helpful and meticulous illustration. These
subcultures, he contends, are “intercalated” in that they constrain, guide, and in-
spire one another, but they also develop and function significantly independ-
ently of one another and are thus identifiable as separate subareas of research and
practice, with separate conferences, journals, and so on.
Most importantly for present purposes, the significant autonomy associated
with these different subareas generates significantly different understandings of
the subject matter. This is the source of Galison’s provocative adaptation of the
anthropological notion of a “trading zone”: “an intermediate domain in which
procedures could be coordinated locally even where broader meanings clashed”
(p. 46). Subcultures of physics do not associate precisely the same meanings with
the technical terms used in communication with one another: “Theorists and
experimenters, for example, can hammer out an agreement that a particular track
configuration found on a nuclear emulsion should be identified with an electron
and yet hold irreconcilable views about the properties of the electron, or about
philosophical interpretations of quantum field theory” (p. 46); when working
together, they set aside “the ‘deep’ and global ontological problems of what an
electron ‘really’ is” (p. 48). The upshot of a careful consideration of different
 Realist Representations of Particles 369

approaches to the physics of particles is thus clearly and immediately consequen-

tial for philosophers interested in questions of scientific knowledge: “the signifi-
cance of these partially separate lives is that—​once one abandons ‘observation’ or
‘theory’ as the basis for a univocal account—​no single narrative line can capture
the physics of the twentieth century, even within a single specialty” (p. 9).9
What’s good for the goose of particle physics, however, is good for the gander
of philosophy of particle physics. Indeed, the various projects of interpretation of
the natures of particles and their properties displayed in previous sections have
demonstrated just this. Reflecting the different conceptions of particles adopted
by physicists who approach them from the different vantages of mathematical
theorizing and experimental detection, philosophers often view the natures
and properties of particles in different ways, typically depending on the scien-
tific practices on which they are most focused or with which they are most con-
cerned. None of this all by itself is an argument for or against realism, but it does
shed crucial light on the question of what it means to be a realist in this domain,
if one is that way inclined. Just as scientists in different subareas of physics may
believe in electrons—​sharing an ontological commitment, but under different
descriptions (more precisely: partially different and overlapping descriptions)—​
so too may philosophers of science. Realism, in the limit, is a commitment to
the existence of something, to the idea that through theoretical descriptions
and/​or experimental detections, ideally both, we have picked out what Einstein,
Podolsky, and Rosen (1935) described so evocatively as an “element of reality.”
Triangulating, using our best tools of mathematical and causal investigation, we
have managed to pick something out in the world.
Thus we see how the challenge posed by the realist tightrope, with which we
began, is misleading. Realism about x does not face mortal danger on either side
by believing too much (believing increasingly refined descriptions of x) and
believing too little (simply believing in the existence of x). A supplemental meta-
phor is needed. From a realist perspective, successful reference is, in fact, all that
is required to anchor realism, and it is a shared judgment that such anchoring
has been achieved that unifies different sorts of realists about any given x. This
is compatible, of course, with further description rendering realist commitment
more substantial, with all the risk and reward this entails. To the extent that fur-
ther descriptions of the precise natures of things like particles are believed, the
anchor of reference is compatible with there being different species of realist
commitment (e.g., selective realisms), unified qua realism more broadly (as a

9 Galison 1997, pp. 833–​835, tells the story of how Sidney Drell and James Bjorken aspired to write

a book on quantum field theory in the early 1960s, but ended up producing two separate volumes, one
geared to experimentalists (concerned more with measurable quantities) and the other to theorists
(concerned more with formal properties of theories, such as symmetries and invariances). Some of
the differences between the volumes amounted to a “radical difference in the ontology” (p. 835).
370 Contemporary Scientific Realism

genus) by a commitment to shared reference. It is also compatible with com-

bining a high degree of confidence in our having picked something out in the
world with lesser degrees of confidence in some or all of the descriptions of the
nature of this thing elaborated in finer-​grained ways by different versions of re-
alism and in the metaphysics of science. As intimated earlier, talk of “particles”
is loose—​objects of some kind?; events?; some sort of hybrid?—​and likewise,
groups of cohering causal properties?; emergent or derivative features of an on-
tologically subsistent structure? Reference is the anchor.
Admittedly, the notion of anchoring is not by itself so comprehensive as to
yield determinate answers to further questions that realists are often pressed
to confront. Is a causal theory of reference best for anchoring? If so, the
commitments shared by different sorts of realists may sometimes prove maxi-
mally thin, though depending on the strength of the evidence they may prove
epistemically significant nonetheless. In many cases, as in the present case, a
causal-​descriptive or minimal descriptive theory (appealing to a shared subset of
descriptions) may be appropriate, since physicists and philosophers alike gener-
ally agree on a number of features of particles, their properties, and interactions,
with differences of interpretation emerging only in their finer-​grained proposals
for how best to understand the natures of these things. Should different species
of realists, imbued in different ways by top-​down and bottom-​up approaches to
particles, hold lower degrees of belief in their finer-​grained interpretations than
in the coarser descriptions they jointly affirm with others? If degrees of belief in
finer-​grained proposals are sufficiently low, this may suggest the wisdom of a
pragmatic pluralism of accounts; if they are sufficiently high, this may suggest an
agreement to disagree between different camps. Clearly, there is plenty of work
here left to do in grappling with these issues.
All of this said, it is nevertheless the case that while the intuitive pull of the
realist tightrope can be strong, it is properly resisted. Feeling the pull, Stanford
(2015) contends that claims about what exists or about which terms success-
fully refer are not at issue in debates about realism; instead, antirealist arguments
should be construed as targeting only scientific descriptions of the “fundamental
constitution and operation of various parts of the natural world.” One may nat-
urally wonder here about the relevant sense of “fundamental.” Is the intention
to target some special part of the spectrum of increasingly refined descriptions
of some focus of scientific investigation offered by some realists? This would be
puzzling: there is no obvious point at which these descriptions become “funda-
mental” and, in any case, different species of realism disagree about much de-
scription while still belonging to the genus. Perhaps instead, “fundamental” is
being used in the way familiar to us from accounts of ontological or explana-
tory reduction, in which some entities or phenomena are arguably “reducible” to
other, more fundamental ones. But this is likewise unpromising, even granting
 Realist Representations of Particles 371

the premise of reductionism, in the absence of some convincing argument to the

effect that less fundamental things should not be considered real. If it turns out
that superstring theory is true, then it will turn out that particles are modes of vi-
bration of strings, but it is at best unclear how this would make them any less real.
While many realists disagree about the natures of particles and go to great
lengths to explain, in conflicting ways, how such talk should be interpreted, they
are no less realist about particles. This suggests that realism simpliciter is some-
thing to which one may subscribe along a spectrum of descriptions, from the
minimal, as in the case of assertions of reference, to the most refined views of
metaphysical natures.10 Indeed, Stanford (2015, pp. 410–​411) acknowledges that
in some cases the sheer weight of theoretical and experimental evidence for the
existence of something (e.g., atoms) is so great that it is implausible to imagine
that future scientists will change their minds. Given that they may change their
minds about certain fundamental descriptions, however, and that it is dubious
that we are capable of predicting what subset of our current descriptions will be
retained in future, realism thus conceived would be “so weak that . . . no histori-
cist opponent will think it worthwhile to contend against it” (p. 416). It is all too
easy, though, to place this shoe on the other foot. If the evidence is sufficiently
strong as to indicate that we have successfully picked something out in the world,
this is music to realist ears—​and a justified expectation that this will be preserved
across theory change suggests that some significant portion of the theoretical
and experimental knowledge justifying this expectation will be preserved as
well, furnishing a basis for even more substantial conceptions of realism.
Let us take some final inspiration from those engaged in theorizing and
experimenting. Late in his life the great theoretician Werner Heisenberg
(1998/​1976) betrayed a striking ambivalence between top-​down and bottom-​
up approaches to characterizing the nature of matter: “The question, What
is an elementary particle? must find its answer primarily in experiment”;
“theory . . . cannot add much to this answer” (p. 211)—​but later he could not
resist adding, “The particles of modern physics are representations of symmetry
groups and to that extent they resemble the symmetrical bodies of Plato’s phi-
losophy” (p. 219). To return to earth once more from Plato’s heaven, nothing
smooths the way better than speaking to an experimentalist. Randal Ruchti is
part of the High Energy Physics Group at the University of Notre Dame which
participates in experiments at the Large Hadron Collider at CERN, for which
they developed a hand-​held detector that can be placed in high energy particle
beams to yield visual representations of particle interactions. The discourse of

10 Cf. Magnus 2012, p. 122: “Retail arguments [i.e., arguments stemming from evidence specific

to the case at hand] for believing in particular things can give us good reasons to believe that those
things exist on the basis of their connections to other things, while leaving questions of things’ funda-
mental nature either unmentioned or unresolved.”
372 Contemporary Scientific Realism

experimental particle physics is so rife with collisions, scattering, and detections

of “packets” of energy and momentum that realism about particles is a natural
default, but Ruchti is quick to add: “don’t ask me what they are!”

Acknowledgments

I would like to thank a number of colleagues for thoughts on an earlier version

of this paper, including Jonathan Bain, Steven French, Vassilis Livanios, Matteo
Morganti, Fred Muller, Kyle Stanford, and Peter Vickers, and audiences at the
Universities of Barcelona, Bergen, Indiana-​Purdue (Indianapolis), Lund, Miami,
Pittsburgh, and Roma Tre, and at the CSHPS annual conference, for stimulating
comments on various parts of this project.

References
Bain, J. 2011: “Quantum Field Theory in Classical Spacetimes and Particles,” Studies in
History and Philosophy of Modern Physics 42: 98–​106.
Baker, D. J. 2009: “Against Field Interpretations of Quantum Field Theory,” British Journal
for the Philosophy of Science 60: 585–​609.
Bauer, W. A. 2011: “An Argument for the Extrinsic Grounding of Mass,” Erkenntnis
74: 81–​99.
Ben-​Menahem, Y. 2018: Causation in Science. Princeton: Princeton University Press.
Berghofer, P. 2018: “Ontic Structural Realism and Quantum Field Theory: Are There
Intrinsic Properties at the Most Fundamental Level of Reality?” Studies in History and
Philosophy of Modern Physics 62: 176–​188.
Bigaj, T. 2018: “Are Field Quanta Real Objects? Some Remarks on the Ontology
of Quantum Field Theory,” Studies in History and Philosophy of Modern Physics
62: 145–​157.
Bigelow, J., B. Ellis, & C. Lierse 1992: “The World as One of a Kind: Natural Necessity and
Laws of Nature,” British Journal for the Philosophy of Science 43: 371–​388.
Bird, A. 2007: Nature’s Metaphysics: Laws and Properties. Oxford: Clarendon.
Chakravartty, A. 2007: A Metaphysics for Scientific Realism: Knowing the Unobservable.
Cambridge: Cambridge University Press.
Chakravartty, A. 2017: Scientific Ontology: Integrating Naturalized Metaphysics and
Voluntarist Epistemology. New York: Oxford University Press.
Chakravartty, A. 2019: “Physics, Metaphysics, Dispositions, and Symmetries—​ à la
French,” Studies in History and Philosophy of Science 74: 10–​15.
Einstein, A., B. Podolsky, & N. Rosen 1935: “Can Quantum-​Mechanical Description of
Physical Reality Be Considered Complete?” Physical Review 47: 777–​780.
Ellis, B. 2005: “Katzav on the Limitations of Essentialism,” Analysis 65: 90–​92.
Esfeld, M. 2009: “The Modal Nature of Structures in Ontic Structural Realism,”
International Studies in the Philosophy of Science 23: 179–​194.
Fraser, D. 2008: “The Fate of ‘Particles’ in Quantum Field Theories with Interactions,”
Studies in History and Philosophy of Modern Physics 39: 841–​859.
 Realist Representations of Particles 373

French, S. 2014: The Structure of the World: Metaphysics and Representation.

Oxford: Oxford University Press.
French, S., & K. McKenzie 2012: “Thinking Outside the (Tool)Box: Towards a More
Productive Engagement Between Metaphysics and Philosophy of Physics,” European
Journal of Analytic Philosophy 8: 43–​60.
Galison, P. 1997: Image and Logic: A Material Culture of Microphysics. Chicago: University
of Chicago Press.
Hanson, N. R. 1965/​ 1958: Patterns of Discovery: An Inquiry into the Conceptual
Foundations of Science. Cambridge: Cambridge University Press.
Harré, R. 1986: Varieties of Realism: A Rationale for the Natural Sciences. Oxford: Blackwell.
Heisenberg, W. 1998/​1976: “The Nature of Elementary Particles,” Physics Today (1976)
29: 32–​ 39. Reprinted in E. Castellani (ed., 1998), Interpreting Bodies: Classical
and Quantum Objects in Modern Physics, pp. 211–​ 222. Princeton: Princeton
University Press.
Jantzen, B. 2011: “An Awkward Symmetry: The Tension between Particle Ontologies and
Permutation Invariance,” Philosophy of Science 78: 39–​59.
Jones, R. 1991: “Realism About What?” Philosophy of Science 58: 185–​202.
Ladyman, J. 1998: “What is Structural Realism?” Studies in History and Philosophy of
Science 29: 409–​424.
Ladyman, J. 2014/​2007: “Structural Realism,” in E. N. Zalta (ed.), The Stanford
Encyclopedia of Philosophy, http://​ plato.stanford.edu/​ entries/​ structural-​realism/​
. Stanford: The Metaphysics Research Lab, Center for the Study of Language and
Information, Stanford University.
Livanios, V. 2010: “Symmetries, Dispositions and Essences,” Philosophical Studies
148: 295–​305.
Livanios, V. 2012: “Is There a (Compelling) Gauge-​Theoretic Argument against the
Intrinsicality of Fundamental Properties?” European Journal of Analytic Philosophy
8: 30–​38.
Livanios, V. 2018: “Hamilton’s Principle and Dispositional Essentialism: Friends or Foes?,”
Journal for General Philosophy of Science 49: 59–​71.
Lyre, H. 2012: “Structural Invariants, Structural Kinds, Structural Laws,” in D. Dieks, W.
J. Gonzalez, S. Hartmann, M. Stöltzner, & M. Weber (eds.), Probabilities, Laws, and
Structures, pp. 169–​181. Dordrecht: Springer.
Magnus, P. D. 2012: Scientific Enquiry and Natural Kinds: From Planets to Mallards.
London: Palgrave Macmillan.
McKenzie, K. 2016: “Looking Forward, Not Back: Supporting Structuralism in the
Present,” Studies in History and Philosophy of Science 59: 87–​94.
Morganti, M. 2013: Combining Science and Metaphysics: Contemporary Physics,
Conceptual Revision and Common Sense. New York: Palgrave Macmillan.
Muller, F. A. 2015: “The Rise of Relationals,” Mind 124: 201–​237.
Saunders, S. 2006: “Are Quantum Particles Objects?” Analysis 66: 52–​63.
Smart, B. T. H., & K. P. Y. Thébault 2015: “Dispositions and the Principle of Least Action
Revisited,” Analysis 75: 386–​395.
Stanford, P. K. 2015: “‘Atoms Exist’ is Probably True, and Other Facts that Should Not
Comfort Scientific Realists,” The Journal of Philosophy 112: 397–​416.
van Fraassen, B. C. 1991: Quantum Mechanics: An Empiricist View. Oxford: Clarendon.
Wolff, J. 2012: “Do Objects Depend on Structures?” British Journal for the Philosophy of
Science 63: 607–​625.
 Index

For the benefit of digital users, indexed terms that span two pages (e.g., 52–​53) may, on occasion,
appear on only one of those pages.
Figures are indicated by an italic f and tables by t following the page number.

absolute income hypothesis, 259 pessimistic meta-​induction, 11, 12–​13,

absolute thermometry, 59–​60 22, 29–​30
abstractions, 317 selective scientific realism, 130, 150, 155
approximate truth and, 317–​19 success and, 29
infinite limits as, 343–​46 theory change argument, 243–​44,
level of generality and, 216–​17, 229–​31, 245, 247–​48
235, 236 approximation, 315–​16
acidity approximate truth and, 318
Brønsted-​Lowry concept, 37 with idealization, 327–​33
Lewis concept, 37 idealizations yielding, 322, 327–​28
acidity, Lavoisier’s oxygen theory, 33–​52. See without idealization, 316–​17, 322, 327–​33
also boron discovery a priori principles, 72
complications, clarifications, objections, and artifact, 162–​63
replies, 45–​52 (see also under boron Asay, Jamin, 234–​35
discovery) Astronomiae physicae et geometricae elementa
false theory, 35–​38 (Gregory), 80
novel predictions, 38–​40 atomic model, Nicholson’s, 102–​4. See also
predictive success, 40–​41 Nicholson’s proto-​element theory
selective realism, challenge, 42–​45 atomic number, 104
ad hocness, pessimistic meta-​meta-​ atomic theory, Sommerfeld’s, 100–​1
induction, 305–​7 atomic weights of elements, Nicholson’s, 105–​
Afanassjewa, Tatiana, 330–​31 11, 107f, 108f, 110f
agnosticism, 83 Attanasio, O. P.l, 269–​70
air-​engine, Stirling, 59–​60, 64 August Kekulé, Friedrich, 132–​33
Aiton, E. J., 82–​83 average propensity to consume, 259, 260
Alai, Mario. See essential deployment Avogadro’s number, 160–​61, 176–​77, 206–​7
Alston, William, 152 axiological realism, 156, 258, 273, 277, 279–​80
Anaxagoras, 245
antirealism (antirealists), 1–​2, 33–​34, 216–​17, Bain, J., 356
350–​51. See also specific topics Baker, A., 322–​23
on pessimistic meta-​induction, 12–​13, 284 Baker, D. J., 356
anything goes theorem, 263–​64 Baldwin, P., 14
approximate truth, 34, 94, 229–​30, 312, 339–​40 Baron, S., 313–​14, 323–​25
definition, 317–​18 base-​rate fallacy, 152–​55
idealizations, approximations, abstractions Batterman, R. W., 333–​34, 344–​45
and, 317–​19 Bauer, W. A., 357
instrumentalists, 227–​28 Bayes’ theorem, 24–​25
miasma theory, 22, 29–​30 Ben-​Menahem, Y., 366
Newtonian limit, 332–​33 Berghofer, P., 357
Paradox of Infinite Limits, 345–​46 Bergmann, Michael, 152
376 Index

Berthollet, Claude Louis, 44 Campbell, Donald T., 122

Bigaj, T., 356 Campbell, J. Y., 267–​68, 270, 271, 272
Bigelow, J., 365 Canzoneri, M., 269–​70
biology, evolutionary, 121, 122, 128 Carnot, Sadi, 57–​58
Blinder, A. S., 267, 269–​70 heat-​engine, 58–​60, 61–​62, 67–​68
Blomstrand, Christian Wilhelm, 133–​35 heat-​pump, 59
Blomstrand-​Jørgensen chain theory, 130–​31, proto-​thermodynamics, 57
133–​35, 148–​49 Carnot cycle, 61, 64–​65
Blomstrand-​Jørgensen/​Werner Debate, 130–​ Carnot’s theory, 57–​59
31, 147–​48 Clausius’ modifications, 57–​58
Blumenthal, G., 253 concluding review, 67–​68
Bohr, Niels, 160 Thomson revival, 59–​60
Bohr’s atom and Nicholson’s proto-​element Carnot’s theory, James Thomson’s prediction
theory, 100, 109–​10, 117–​21, 122–​24, analysis, 61–​63
125, 127, 128–​29 methodological hesitation, 63–​64
essential deployment, 160, 201–​2, 207–​8 no miracle, 62–​63
Boltwood, Bertram, 160 quantitative calculation, 64–​66
boracic acid, 45–​47. See also boron discovery rationale behind, 60
as element, Lavoisier’s prediction, 47–​48 William Thomson’s figure for James’ engine’s
Born, Max, 174–​75 efficiency, 66–​67
Born adiabatic approximation, 171–​72, 173–​75 Carrier, M., 101
Born-​Oppenheimer approximation, 171–​ Carrington, Alan, Rotational Spectroscopy of
72, 173–​75 Diatomic Molecules, 175
boron discovery, 33–​52 Carroll, C. D., 269–​70
boracic acid, 45–​47 Cartesian natural philosophy, 72
boracic acid as element, Lavoisier’s Cartwright, Nancy, 248
prediction, 47–​48 catastrophism, 217–​18, 223–​24
boron as boracic radical, 50–​52 causal roles, 356, 359–​63, 364–​66, 369–​70
conclusion, 52 causal structure, 265–​66, 273, 276, 277, 278,
false-​but-​successful theories, 34–​35, 37, 48 353–​54, 356, 359–​63, 364–​66, 369–​70
false theory, 35–​38 certainty, moral, 77–​78
novel predictions, 38–​40 chain rule, 134–​35, 148
predictive success, 40–​41 chain theory, Blomstrand-​Jørgensen, 130–​31,
sample purity, 48–​50 133–​35, 148–​49
selective realism, challenge, 42–​45 Chakravartty, Anjan, 101, 250, 317–​19. See also
temporally novel predictive success, 33, 35 particles, realistic representations
boron trioxide, 40, 45–​47 Chang, Hasok, 36, 37, 50–​52, 101, 276
bottom up approach, Standard Model, 358–​62 Chao, Hsiang-​Ke, 258, 259–​60,
Boyd, Richard, 229–​30, 233 275–​76
Braggs, William Lawrence, 160 realism about structure, 264–​73
bridge principle, 73–​74, 86 Chen, Ruey-​Lin, 48–​49
Brønsted-​Lowry concept, 37 cholera
Brown, John, Rotational Spectroscopy of miasma theory (see miasma theory)
Diatomic Molecules, 175 Snow’s water-​borne theory, 17, 23, 26, 28–​29
Brownian motion, 159, 160–​61 Clapeyron, Émile, 65–​66
Brumberg, R., 260 Clausewitz, Carl von, 71, 94–​95
Butterfield, J., 313–​14, 327–​28, 333–​34, 335–​ Clausius, 3, 56–​58, 62–​63, 68
37, 340–​41 Clausius-​Clapeyron equation, 65–​66
Clavius, Christopher, 287
Callender, C., 233–​34, 313–​14, 334–​36 cobalt complexes, as chain-​like
caloric theory of heat, 2, 12–​13, 37–​38, 42, 93–​ structures, 133–​35
94, 221–​22, 243 combustion, phlogiston theory. See
essential deployment, 193–​94 phlogiston theory
 Index 377

confirmation, 233–​34, 235 Dalton, John

empirical, 139 essentiality condition, 192–​93
from laws, 24 Law of Multiple Proportions and Principle of
confirmation theory, 291–​92 Simplicit, 192–​93
“A Confutation of Convergent Realism” Davy, Humphry, 35, 40–​41, 44–​45, 46, 47, 48–​50, 52
(Laudan), 1 Deaton, A., 267, 269–​70
conservativism, theoretical, 226–​29 De Gravitatione (Newton), 78
constants Dennett, D., 252
molecular, molecular spectroscopy and stable deployment realism, 34, 183–​85, 186, 188, 199,
measurement, 163–​67, 166t 203, 205
Planck’s, 115f, 115–​16 derivation external posits, 41–​42, 147
Rydberg’s, 160 derivation internal posits, 41–​42, 147
constructive empiricism, 4, 233 Descartes, René
consumption function Cartesian natural philosophy, 72
aggregate consumption function basics, 257 Cartesian physics, 72–​73
axiological realism, 258, 273, 277, 279–​80 continuity thesis, 73–​74
Euler equation, 257–​58, 262–​63, 266–​67, hypothetico-​deductive model, 73, 82–​83
269–​71, 272–​73, 274–​80 as instrumentalist, 76–​77
fundamentals, 257–​58 on mechanical hypotheses, 72–​78
history, brief, 263–​73 bridge principle, 73–​74, 86
Keynesian formulation, 259 system of the world, 75–​76
empirical issues, 259–​61 vortices, fate, 77–​78
Lucas Critique and Hall, 261–​63 on motion
Lucas Critique, 261, 263, 265–​66, 269–​70 definitions, 75
Lucas Critique, Hall’s response, 258, 262–​63 vs. Tycho, 75–​76, 350
modest (pragmatic and perspectival) Principia, 72, 73, 75–​78
structural realism, 273–​77 vortex theory of planetary motion, Newton
realism question, 263–​73 on (see vortex theory of planetary
reductionism and motions, Newton on)
microfoundations, 263–​64 Descartes-​Newton relationship, 71. See also
Worrall’s structural realism and Chao’s Descartes, René; Newton, Isaac
realism about structure, 264–​73 description, scientific, 350–​54
Reiss, 277–​79 detection properties, 218
scale-​dominant behaviors, 268, 271, diatomic molecules
273, 275 molecular spectroscopy, 163–​64, 165–​70,
“Contemporary Scientific Realism and the 166t, 169f, 173–​78
Challenge from the History of Science” rotational, vibrational, and electronic energy
(AHRC), 6 states, 168–​69, 169f
continuity thesis, 73–​74 Dicken, Peter, 152–​54
Contra SRs, 149–​50, 151, 152, 156 Dictionary of Chemistry (Nicholson), 46
convergent realism, 1, 12, 30 disjunctive objection, 208–​9
coordination complexes, 134–​35, 139 divide et impera (divide and conquer strategy),
coordination theory, Werner’s, 130–​3 1, 1–​2, 20–​21, 101, 244
134–​3 5, 138–​3 9, 140, 142–​4 3, 147–​ Alais on, 244
49, 155 debate, 94–​95
Copernican Scholium, 86 by other means, 71, 93–​95
Cordero, A., 199, 203 Psillos against Laudan’s reductio, 184
core causal descriptions, 218, 222–​23 double case study, 71. See also evidence
Cotes, Roger, 79, 82, 83–​84 double-​counting evidence, pessimistic meta-​
Crookes, William, 103–​4 meta-​induction, 296–​98
Crosland, Maurice, 45–​46 Ducheyne, Steffen, 87
curve-​fitting, 23–​24 Duesenberry, J. S., 260
Cuvier, Georges, 217 Dynamical Theory of Gases, The (Jeans), 161
378 Index

Einstein, Albert, 369 quantum phase transitions and other

General Theory of Relativity, 197 cases, 340–​43
General Theory of Relativity, Newton’s theory essentiality
of gravity and, 89–​92 deployment realism, 185
Mercury, anomalous perihelion, 90, 91 improved criterion, escaping Lyons’s
retentionism, 89–​92 criticisms, 188–​92
special relativity, 92, 245–​46, 331–​33 Lyons’s criticism of Psillos’s criterion, 185–​88
elevation law, Farr’s, 15–​19, 18f, 19f, 20f, 22–​29 prospective detection, not possible, 199–​
eliminativism, 248–​49 204, 204f
Ellis, B., 365 ruling out Lyons’s purported
Emch, G., 341–​42 counterexamples, 192–​98
empirical confirmation/​disconfirmation, 139 caloric theory, 193–​94
empirical correctness, condition, 325–​26, 327–​ Dalton, 192–​93
29, 334–​35, 339–​40, 345–​46 Kepler, 194–​98
empiricism Mendeleev, 193
constructive, 4, 233 Estrella, A., 269–​70
structural, 241, 250 ether theory of light, 2, 243–​45
Enhanced Indispensability Argument, 322–​23 Euler equation, 257–​58, 262–​63, 266–​67, 269–​
entity realism, 352, 362–​63 71, 272–​73, 274–​80
epistemology, 143–​44, 151, 351 evidence, 70–​96
evolutionary, 122, 123, 126, 128 conclusions, 95–​96
Esfeld, M., 361 Descartes, on mechanical hypotheses, 72–​78
essential, 333–​43 bridge principle, 73–​74, 86
essential deployment, realism and, 183–​212 Descartes as instrumentalist, 76–​77
deployment realism, 183–​85 system of the world, 75–​76
essentiality, cannot be detected prospectively, theory, 72–​73
199–​204, 204f vortices, fate, 77–​78
essentiality, improved criterion, escaping divide et impera by other means, 71, 93–​95
Lyons’s criticisms, 188–​92 double-​counting, pessimistic meta-​meta-​
essentiality, Lyons’s criticism of Psillos’s induction, 296–​98
criterion, 185–​88 Einstein’s retentionism, 89–​92
essentiality condition, ruling out Lyons’s fundamentals, 70–​72
purported counterexamples, 192–​98 Newton’s deduction from
caloric theory, 193–​94 phenomena, 78–​89
Dalton, 192–​93 explanation, Newton’s account, 84–​86
Kepler, 194–​98 Huygens vs. Leibniz, 79–​82
Mendeleev, 193 natural philosophy, doing, 82–​84
Neptune, 198 problem with vortices, 79–​82
No Miracle Argument, 183, 184, 186, 188, quam proxime, 86–​89
191–​92, 194, 196–​97, 199, 203, 212 pessimistic induction, 70, 71, 93, 96
objections and answers, 208–​12 standards, changes, 70
disjunctive objection, 208–​9 evolutionary biology, 121, 122, 128
false may entail true objection, 208 evolutionary epistemology, 122, 123,
modus ponens objection, 212 126, 128
Ramsey sentence objection, 211–​12 experientially concretized truths, 273–​75
theoretician’s dilemma, objection 1, 209–​10 experimental individuation, 48–​50
theoretician’s dilemma, objection explanation, in science, Newton’s
2, 210–​11 account, 84–​86
Psillos’s condition, dispensability, 204–​8, Explanationist Defense of Realism, 12–​13
205f–​6f explanatory success, Nicholson’s proto-​element
essential idealizations, 316, 318, 322, 333–​43 theory, 99–​129. See also Nicholson’s
first-​order phase transitions, resolving proto-​element theory
paradox, 334–​40 Eyler, J. M., 14
 Index 379

Fahrbach, Ludwig, 203–​4. See also pessimistic General Scholium (Newton), 78, 82
meta-​meta-​induction General Theory of Relativity (General
False Basis Argument, 146–​49 Relativity), 89–​92, 197, 221–​22, 245,
false-​but-​successful theories, 34–​35, 37, 48 292, 313, 323
false may entail true objection, 208 Glymour, Clark, 159–​60
false theories. See also specific types Godfrey-​Smith, P., 58, 313–​14, 317
successful predictions, 56 (see also freezing gravity
point of water, pressure lowering) Huygens on, 79–​80, 81
Farr, William, 15–​19, 18f, 19f, 20f, 22 Leibniz on, 80–​81
objections, 22–​29 Newton’s theory, 81–​84, 85–​86
Feyerabend, Paul, 121, 246–​47 Einstein’s retentionism and General
Field, Robert, The Spectra and Dynamics of Theory of Relativity, 89–​92
Diatomic Molecules, 175 Gregory, David, Astronomiae physicae et
Fine, Arthur, 231–​32, 233–​34, 235 geometricae elementa, 80
first-​order phase transitions, 334–​40
Fisher, Ronald, 224–​25 Hall, Robert, 258, 262–​63, 270
Flavin, M. A., 267 Hamlin, C., 14–​15
Fraser, J., 341–​42 Hanson, Norwood Russell, 350
freezing point of water, pressure Harper, William, 85, 94–​95
lowering, 56–​68 Harré, R., 365
Carnot’s theory, 58–​59 Havranek, T., 269–​70
methodological hesitation, 63–​64 heat, caloric theory of. See caloric theory
Thomson’s revival, 59–​60, 61–​63 of heat
concluding review, 67–​68 heat-​engine, 58–​60, 61–​62, 67–​68
James Thomson’s prediction heat-​pump, 59–​60
analysis, 61–​63 Heilbron, John, 117–​18
methodological hesitation, 63–​64 Heisenberg, Werner, 128–​29, 371–​72
no miracle, 62–​63 Hempel, Carl, 209
quantitative calculation, 64–​66 Hendry, Robin, 47–​48
rationale behind, 60 Herzberg, Gerhard, Molecular Spectra and
William Thomson’s figure for James’ Molecular Structure, 163–​64, 166–​68,
engine’s efficiency, 66–​67 166t, 173–​74
no-​miracle argument, 56–​57, 62–​64, 67 Hesse, Mary, 1, 12–​13
French, Steven, 240–​41, 245–​51, 357, 361 Hicks, W. M., 117
Fresnel, Jean, 199–​200 historiography, of science, 253–​54
equations, 245–​46 Hooke, Robert, 253–​54
ether, 244–​45 Hoover, K., 265–​66, 272–​73
theory of light, Poisson white spot, 24–​25, Hoyningen-​Huene, P., 127
26–​27, 44–​45 Hricko, Jonathon. See boron discovery
Friedman, M., 260 Hutchison, Keith. See freezing point of water,
permanent income hypothesis, 258, 260–​61, pressure lowering
262, 267–​69, 272 Huygens, Christiaan, 79–​82
Fuhrer, J. C., 269–​70 Treatise on Light, 95
fundamentalism, 248 hypothetico-​deductive model, 73, 82–​83

Gale, George, 246–​47, 254 idealizations, 315

Galileo, 315 approximate truth and, 318
Galison, P., 368–​69 approximations with, 327–​33
Gay-​Lussac, Joseph Louis, 35, 40–​41, 44–​45, 47, approximations without, 316–​17,
48–​50, 52 322, 327–​33
Geiger, Hans, 160, 161 definition, 315
General Correspondence Principle, 245 example, 315
generality, particularism and, 229–​36 ineliminable, 316
380 Index

idealizations (cont.) Jantzen, B., 356

infinite, 313 (see also infinite limits, Jeans, James, 161
paradox of) Jhun, Jennifer. See consumption function
preliminary concepts, 314–​17 Jones, R., 351–​52, 367–​68
yielding approximations, 322, 327–​28 Jørgensen, Sophus Mads, 134–​35, 135f, 136f,
idealizations, essential, 316, 318, 333–​43 138f, 138–​39, 140
first-​order phase transitions, resolving Jørgensen/​Werner Debate, 134–​40, 142–​43
paradox, 334–​40 Joule, James, 57–​58, 66–​67
quantum phase transitions and other
cases, 340–​43 Kadanoff, L. P., 334, 345
idle posits, 147 Kauffman, George B., 134–​35, 140
incommensurability, 127 Kelvin, Lord (William Thomson), 57–​58, 105, 108–​9
individuation, experimental, 48–​50 Carnot’s theory, revival, 59–​60
ineliminable idealizations, 316 James Thomson’s engine’s efficiency, William’s
infinite idealizations, 313. See also infinite figure, 66–​67
limits, paradox of “Nineteenth Century Clouds over the
infinite limits, paradox of, 312–​47 Dynamical Theory of Heat and
abstractions, infinite limits as, 343–​46 Light,” 161–​62
approximate truth, 312, 332–​33, 339–​ Kepler, Johannes, 287
40, 345–​46 essentiality condition, 194–​98
approximations, with/​without orbital motion, 172–​73, 176
idealizations, 327–​33 Kepler’s laws, 89, 172–​73, 175, 176, 194–​98, 245,
challenge to scientific realism, 319–​26 287, 350
empirical correctness, condition, 325–​26 Huygens vs. Leibniz, 80–​81
Paradox of Infinite Limits, 322–​25 Newton on Descartes’ vortex hypothesis,
Enhanced Indispensability 77–​79, 87
Argument, 322–​25 Newton on Descartes’ vortex hypothesis,
essential idealizations, 316, 318, 322, 333–​43 account of explanation, 84–​85
first-​order phase transitions, resolving kinetic theory, 161–​62
paradox, 334–​40 Kitcher, Philip, 1–​2, 11, 13, 20–​21, 101, 123,
quantum phase transitions and other 184, 186
cases, 340–​43 Koertege, Noretta, 246–​47, 254
finiteness of real systems, 322–​23 Kogut, J., 345–​46
history, 313–​14 Kragh, H., 119
idealizations, approximations, and Kuhn, Thomas, 218–​19, 246–​47
abstractions, 314–​19 revolutions, 125–​28
approximate truth, 317–​19 revolution with evolution, 128–​29
preliminary concepts, 314–​17 Structure of Scientific Revolutions, 253
indispensability of limit system, 322–​23 Kuznets, S., 259–​60
no miracles argument, 56–​57, 312–​
13, 323–​25 Ladyman, James, 33–​52, 240–​41, 243–​44, 245–​
taxonomy, 320–​22 48, 249, 250–​51, 253, 353. See also boron
instrumentalism, 1–​2 discovery
Descartes, Rene, 76–​77 Landsman, N., P., 341–​42
realism, particularism, and, middle Laudan, Larry, 12–​13, 101, 184, 241–​42, 243
path, 216–​36 (see also realism, “A Confutation of Convergent Realism,” 1
instrumentalism, particularism, Lavoisier, Antoine-​Laurent, oxygen theory of
middle path) acidity, 33–​52. See also boron discovery
interpretation, anchoring, 366–​72 Lefebvre-​Brion, Hélène, The Spectra and
Ising model, 336, 340–​42 Dynamics of Diatomic Molecules, 175
Le Grand, H. E., 36
Jammer, Max, 118 Leibniz, Gottfried, 79–​82
Janiak, Andrew, 83 Tentamen de motuum coelestium causis, 80
 Index 381

Leplin, Jarrett, 186 Mendeleev, Dmitri

Le Verrier, Urbain, 88–​89 essentiality condition, 193
Lewis concept, 37 Periodic Law, 193
Lierse, C., 365 Menon, T., 335–​36
life-​cycle hypothesis, 260, 269 Mercator, Nicholaus, 172–​73
Ligand Field Theory, 147–​48 Mercury, anomalous perihelion, 90, 91
light meta-​induction, 184
ether theory, 2, 243–​45 meta-​modus tollens, 184, 199, 202, 203, 204
Fresnel’s theory, Poisson white spot, 24–​25, metaphysics, realistic, 250–​52
26–​27, 44–​45 method, scientific, 210, 253–​54
Liu, C., 313–​14, 334–​35, 341–​42 methodology, 70, 77–​78, 177, 232–​33, 253–​
Livanios, V., 357, 363–​64 54, 258, 271, 275, 277, 365. See also
Lockyer, Norman, 103–​4 specific types
Lorentz factor γ, 331–​33 miasma theory, 11–​30
Lucas, R. E., 261 approximate truth, 11, 22, 29–​30
Lucas Critique, 261, 263, 265–​66, 269–​70 vs. contagionist views, 14–​15
Hall’s response, 258, 262–​63 Farr, William, 15–​19, 18f, 19f, 20f, 22–​29
Lyell, Charles, 217 objections, 22–​29
Lyons, Timothy, 101, 131, 145–​46, 235–​36 pessimistic meta-​induction, 11–​12, 29
axiological realism, 156, 258, 273, pessimistic meta-​induction, realist
277, 279–​80 responses, 12–​13
Contra SRs, 149–​50, 151, 152, 156 problematic strategy and alternatives, 29–​30
on deployment realism, 184 realists and, 20–​22
on essentiality successor theory continuity, lack, 12
improved criterion, 188–​92 theory and its successes, 14–​19, 18f–​20f
Psillos’s criterion, 185–​88 use-​novel predictions, 12, 13, 15, 17, 19–​21,
False Basis Argument, 146–​49 22, 24–​25
No Instances Argument, 150–​56 use-​novel successful predictions, nonexistent
realist meta-​hypothesis, 145–​46, 149–​50, 151, central element, 12
152, 155–​56 microfoundations, 263–​64
Weak Evidence Argument, 149–​50 middle path, 216–​36. See also realism,
XT statements, 273–​75 instrumentalism, particularism
summary, 217–​22
Mach, Ernst, 241–​42 Miller, Richard, 234–​35
Maddy, Penelope, 219–​20, 232–​34 Millikan, Robert, 160
“Second Philosopher,” 233 Miyake, Teru. See molecular spectroscopy
Maddy/​Wilson Principle, 219–​21, 222, modal structure, 241, 245–​46, 247–​48, 250, 251,
223, 236 252, 359, 361–​64
Magnus, P. D., 233–​34 Modigliani, G., 260
Mankiw, N. G., 267–​68, 270, 271, 272 modus ponens objection, 212
marginal propensity to consume, 259–261 molecular constants
Marschak, J., 279 molecular spectroscopy and stable
Marschak’s Maxim, 279 measurement, 163–​67, 166t
mathematical limits, 319–​20 physically meaningful, 172
Maxwell, James, 199–​200 Molecular Spectra and Molecular Structure
Mayo, Deborah, 160–​61 (Herzberg), 163–​64, 166–​68,
McKenzie, K., 357 166t, 173–​74
mean-​free path, 162–​63 molecular spectroscopy, 159–​78
mechanical hypotheses, Descartes on, 72–​78 diatomic molecules, 163–​64, 165–​70, 166t,
bridge principle, 73–​74, 86 169f, 173–​78
system of the world, 75–​76 fundamentals, 159–​63
vortices, fate, 77–​78 history and origins, 162–​63
Meckel, J. F., 22, 24–​25 mean-​free path, 162–​63
382 Index

molecular spectroscopy (cont.) on Descartes’ vortex theory of planetary

molecular constants, stable measurement, motions, 79–​89
163–​67, 166t explanation, Newton’s account, 84–​86
molecular structure, access, 167–​73, 169f Huygens vs. Leibniz, 79–​82
realism, physical meaningfulness, natural philosophy, doing, 82–​84
and, 173–​78 problem with vortices, 79–​82
molecular structure quam proxime, 86–​89
Brownian motion, 159, 160–​61 General Scholium, 78, 82
kinetic theory, 161–​62 hypotheses non fingo, 83
study, 159–​63 Kepler’s laws, use, 194, 198
molecular structure theories, selective realism mechanics, 294
and truth-​transfer, 130–​57 planet existence, prediction, 183
19th c. chemistry and, 132–​33, 133f Principia, 78, 79, 84–​85, 86–​87, 172–​73
Blomstrand-​Jørgensen/​Werner Rule 4, 89
Debate, 130–​31 Rule III of Philosophy, 85–​86
cobalt complexes as chain-​like theory of gravity, 81–​84, 85–​86, 294
structures, 133–​35 theory of gravity, Einstein’s retentionism and
conclusion, 156–​57 General Theory of Relativity, 89–​92
fundamentals, 130–​31 Newton-​Einstein relationship, 71. See also
implications for scientific realism, 141–​56 Einstein, Albert; Newton, Isaac
False Basis Argument, 146–​49 Newtonian limit, 332–​33
No Instances Argument, 150–​56 Nichols, Amanda. See molecular structure
unconceived alternatives theories, selective realism and
problem, 141–​46 truth-​transfer
Weak Evidence Argument, 149–​50 Nicholson, John, 99
precipitation experiment, 135f–​ biography and early work, 99, 101–​2, 103f
39f, 135–​40 “The Spectrum of Nebulium,” 112–​13
moral certainty, 77–​78 Nicholson, William, Dictionary of Chemistry, 46
Morganti, M., 362 Nicholson’s proto-​element theory, explanatory
Morrison, Margaret, 139, 344 and predictive success, 99–​129
Moseley, H. G. J., 104 atomic weights, elements, 105–​11, 107f,
Muller, F. A., 357 108f, 110f
Mumtaz, H., 271, 272 conclusions, 129
John Nicholson and his early work, 102–​5,
Natural Ontological Attitude, 231–​32, 235–​36 103f, 104f
natural philosophy, 82–​84 Kuhn, revolution with evolution, 128–​29
nebula spectra, Nicholson on, 111f, 111–​ Kuhnian revolutions, 125–​28
13, 112f nebula spectra, 111f, 111–​13, 112f
Neptune Planck’s constant, 115f, 115–​16
discovery, 88–​89, 176, 184 planetary model of atom, 100–​1
essentiality condition, 198 preservative realism, 101–​2
Nernst, Walther, Theoretical Chemistry from proto-​atoms, 101, 103–​5, 104f, 107f
the Standpoint of Avogadro’s Rule and reactions to work of, 117–​21
Thermodynamics, 161 solar corona spectrum, calculations,
Newton, Isaac 114f, 114–​15
Copernican Scholium, 86 Sommerfeld’s atomic theory, 100–​1
deduction from phenomena, 78–​89 structural realism, 101–​2
explanation, Newton’s account, 84–​86 theory limitations and success, 121–​25
Huygens vs. Leibniz, 79–​82 Niiniluoto, Ilkka, 186
natural philosophy, doing, 82–​84 “Nineteenth Century Clouds over the
problem with vortices, 79–​82 Dynamical Theory of Heat and Light”
quam proxime, 86–​89 (Kelvin), 161–​62
De Gravitatione, 78 Noether, Emmy, 364–​65
 Index 383

No Final Theory Objection to the Problem of Papineau, David, 239–​40

Unconceived Alternatives, 144–​45 Paradox of Infinite Limits, 312–​47. See also
No Instances Argument, 150–​56 infinite limits, paradox of
no miracles argument/​miracle argument, 1, particle
33–​34, 56–​57, 154, 243, 265, 286–​87, definition, 354–​55
312–​13, 323–​25 ontological category, 355
essential deployment and deployment strangeness, 355
realism, 183, 184, 186, 188, 191–​92, 194, particle physics, 362–​66, 369, 372
196–​97, 199, 203, 212 particle physics, Standard Model, 249, 353. See
freezing point of water, pressure lowering, also particles, realistic representations
58, 62–​64, 67 (see also freezing point of bottom up approach, 358–​62
water, pressure lowering) top down approach, 358–​62
Norton, J. D., 313–​14, 315–​17, 328–​30, 331 particles, realistic representations, 350–​72
novel predictions. See also specific types causal roles, 353–​54, 356, 359–​63, 364–​
definition, 184 66, 369–​70
disjunctive objection, 209 interpretation, anchoring, 366–​72
essentiality, prospective nature, bottom up, 362–​66
determination, 201–​3 nature, top down, 358–​62
modus ponens objection, 212 realistic interpretation, 354–​57
Psillos on, 185 realist tightrope, 351–​54, 361–​62, 363, 364,
Psillos on, Lyons’s criticism, 186–​87 366–​67, 370–​71
improved criterion, 188, 190–​92 reference and description, 350–​54
Psillos’s condition, dispensability, 206–​8 spin, 100–​1, 325, 336–​37, 340–​42, 356–​57,
Ramsey sentence objection, 211 358, 359, 364–​65
theoretician’s dilemma, objection 1, 209–​10 particularism, 2
novel predictive success, 24–​25, 33–​35, 235–​ generality and, 229–​36
36, 285 realism, instrumentalism, and middle path,
Descartes’ vortices, 77–​78 216–​36 (see also realism, instrumentalism,
evolution by natural selection, 227–​28 particularism, middle path)
Laudan on, 243 Pelling, Myron, 14
Lavoisier’s oxygen theory of acidity, 35–​38, Penner, Myron. See molecular structure
40, 42, 43, 44, 45, 48, 52 (see also boron theories, selective realism and
discovery) truth-​transfer
Psillos’s formula, 239–​40 permanent income hypothesis, 258, 260–​61,
theory change arguments, 243 262, 267–​69, 272
Werner’s coordination theory, 147–​48 permutation symmetry, 356, 360
Perrin, Jean, 159–​61
observables, 22, 33–​34, 241–​42, 340–​41 Avogadro’s number, 160–​61, 176–​77, 206–​7
Occam’s razor, 188, 190 perspectival (structural) realism, 258, 272–​77
ontic structural realism, 240–​41, 245–​46, 247–​ pessimistic induction, 70, 71, 93, 96, 148–​49, 287
48, 250–​51, 360–​61 pessimistic meta-​induction, 199, 202, 203, 204,
ontological problems, for scientific 242, 298–​302
realism, 248–​50 analysis, 288–​91
Oppenheimer, Robert, 174–​75 antirealists on, 12–​13
Orion nebula spectrum, Nicholson on, approximate truth, 11, 22, 29–​30
111f, 111–​13 miasma theory, 11–​12, 20–​21, 29, 30 (see also
Oseen, Carl, 119–​20 miasma theory)
oxygen theory of acidity, Lavoisier’s, 33–​52. See realist responses, 11–​13
also acidity, Lavoisier’s oxygen theory realism and, 286–​88
realists on, 12–​13
Pais, Abraham, 117 realists on, defense, 291–​93
Palacios, Patricia, 335–​36, 342, 343–​44, 345. target, Explanationist Defense of
See also infinite limits Realism, 12–​13
384 Index

pessimistic meta-​meta-​induction, 284–​310. See temporally novel, 33

also pessimistic meta-​induction boron discovery and scientific
ad hocness charge, 305–​7 debate, 33, 35
antirealists on, 12–​13, 284 Meckel’s gill slits, 24–​25
definition and explanation, 293–​96, 295f use-​novel, 33
double-​counting evidence, 296–​98 preservative realism, 101–​2
fundamentals, 284–​86 pressure lowering freezing point of water,
objections 56–​68. See also freezing point of water,
first, 302–​4 pressure lowering
second, 304–​5 Principia, 78, 79, 83–​85, 86–​87, 172–​73
scientific realists on, 284 instrumentalism, 76–​77
selective realism, 285–​86 on mechanical hypotheses, 72
success, big increase, 307–​9 on motion, 75–​76
pessimistic modus tollens argument, 131 sensible bodies and insensible particles, 73
Peters, D., 199, 203 on vortices, 77–​78
phase transitions Problem of Unconceived Alternatives
first-​order, 334–​40 (Stanford), 131, 141–​46
quantum, 340–​43 Dicken on, 154
phlogiston theory, 2, 12–​13, 246–​48, 253, 254 No Final Theory Objection to the Problem of
physical meaningfulness, 173–​78 Unconceived Alternatives, 144–​45
molecular constants, 172 prospective identification, essentiality, 199–​204
molecular spectroscopy, 173–​78 proto-​atoms, 101, 103–​5, 107f. See also
representations, spectroscopy, 162–​63, Nicholson’s proto-​element theory
172, 173 proto-​element theory, Nicholson’s, predictive
Planck, Max, 160, 294 success and, 99–​129. See also
Planck’s constant, 164 Nicholson’s proto-​element theory
Nicholson’s proto-​element theory and, Psillos, Stathis, 1–​2, 11, 13, 21, 101, 139–​40, 141,
115f, 115–​16 152, 239–​40, 243, 247–​48, 250. See also
Podolsky, B., 369 divide et impera (divide and conquer
Poincaré, Henri, 241–​42, 245 strategy)
Poisson white spot, Fresnel’s theory of light, 24–​ Psillos, Stathis, essential deployment and
25, 26–​27, 44–​45 deployment realism, 184, 185. See also
Popper, K., 317–​18 deployment realism
possible-​world approach, 317–​18 essentiality conditions, dispensability, 204–​8,
Post, Heinz, 245 205f–​6f
pragmatic structural realism, 273–​77 essentiality criterion, Lyons’s
precipitation experiment, molecular structure, criticism, 185–​88
135f–​39f, 135–​40 on novel predictions, 185
predictions Lyons’s criticism, 186–​87
realists on, 27–​29 Lyons’s criticism, improved criterion,
use-​novel, 12, 13, 15, 17, 19–​21, 22, 188, 190–​92
24–​25, 33 Psillos’s condition, dispensability, 206–​8
predictive success, 33–​35 Putnam, Hilary, 218, 229–​30, 241–​42
Bayes’ theorem, 24–​25
Meckel’s gill slits, 24–​25 quam proxime, 86–​89
miasma theory, Farr’s use-​novel predictions quantum field theory, 100–​2, 125, 129, 174–​75,
and objections, 12, 17, 19–​29 176, 355–​56, 360, 366, 368–​69. See also
Nicholson’s proto-​element theory, 99–​129 particles, realistic representations
(see also Nicholson’s proto-​element quantum mechanics, 100–​1, 119, 128–​29, 144,
theory) 162, 163–​64, 227–​28, 245, 253, 292, 367
novel, impressive vs. unimpressive, 44
oxygen theory of acidity, Lavoisier’s, 40–​41 Raman spectroscopy, 169–​70
Snow, water-​borne cholera, 26 Ramsey sentence objection, 211–​12
 Index 385

Rankine, William, 56–​57, 63–​64 Kuhnian, 125–​200

Rathke, M. H., 24–​25 quantum, 128
Raydon, T. A. C., 127 uniformitarians, 217–​18
realism (realists), 1–​2, 216–​17. See also Rosen, N., 369
specific topics Rosenfeld, Leon, 118–​19
defense, 284–​85 Ross, Don, 240–​41, 243–​44, 246, 247–​48, 249,
miasma theory, 20–​22 250, 251
pessimistic meta-​induction, 11–​13, 286–​ Rotational Spectroscopy of Diatomic Molecules
88, 291–​93 (Brown and Carrington), 175
pessimistic meta-​meta-​induction, 284 Ruetsche, L., 341–​42
on predictions, 27–​29 Rutherford, Ernest, 102–​3, 105, 160, 161, 207–​8
Reiss and, 277–​79 Bohr’s letter on Nicholson to, 120
on structure, 264–​73 Rydberg’s constant, 160
on theory change, 20–​21
realism, consumption function, 263–​73 Saatsi, J., 222, 235
reductionism and microfoundations, 263–​64 Sankey, H., 186
Worrall’s structural realism and Chao’s Sargent, T. J., 261
realism about structure, 264–​73 scale-​dominant behaviors, 268, 271,
realism, essential deployment and, 183–​212. See 273, 275
also essential deployment, realism and scale relativity, of ontology, 241, 248, 249, 252
realism, instrumentalism, particularism, Scerri, Eric, 47
Middle Path, 216–​36 Schrödinger, Erwin, 128–​29
Middle Path explained, 217–​22 Schrödinger equation, 164
particularism and generality, 229–​36 Schurz, Gerhard, 247
remaining difference makes a science. See also specific topics
difference, 222–​36 historiography, 253–​54
scientific realism debate, recent, 216 Scientific Image, The (van Fraassen), 1
theoretical conservativism, 226–​29 scientific method, 210, 253–​54
realistic metaphysics, 250–​52 scientific realism, defined, 284, 350–​51
realist meta-​hypothesis, 145–​46, 149–​50, 151, scientific realism debate, 1–​3
152, 155–​56 caloric theory of heat, 2, 12–​13, 37, 42, 93–​94,
realist tightrope, 351–​54, 361–​62, 363, 364, 193, 221–​22, 243
366–​67, 370–​71 divide et impera, 1–​2
real patterns, 246–​47, 251–​52 ether theory of light, 2, 243–​45
reduction, 248 history, 1
reductionism, 263–​64 instrumentalists, 1–​2
reference, 350–​54, 355, 359–​60, 361–​62, 369–​ middle-​ground positions, 1–​2
70, 371 phlogiston theory of combustion, 2
Regnault, Henri Victor, 59–​60, 66–​67 realists vs. antirealists, 1–​2, 33–​34
Reiss, J., 277–​79 working parts of rejected theory, in successor
relative income hypothesis, 260 theory, 1–​2 (see also specific examples)
remaining difference that makes a scientific truth, 58
difference, 222–​36 selective realism, 34, 35, 223, 239–​41,
Renzi, B. A., 127 285–​86, 291, 351, 353–​54, 360–​61,
representation, scientific 369–​70
particles, realistic, 350–​72 (see also particles, challenge, 42–​45, 52
realistic representations) claims, 285
physical meaningfulness, spectroscopy, 162–​ false-​but-​successful theories, 34–​35, 37, 48
63, 172, 173 implications, 285
retentionism, Einstein’s, 89–​92 oxygen theory of acidity as challenge, 42–​45
reversibility, thermodynamics, 319–​20, 329–​31 pessimistic meta-​meta-​induction and, 285
revolution, 199, 222 as revisionary, 285–​86
Copernican, 199 structural realism vs., 241–​48
386 Index

selective realism (cont.) ontological problems, for scientific

truth-​transfer and, theories of molecular realism, 248–​50
structure, 130–​57 (see also molecular realistic metaphysics, 250–​52
structure theories, selective realism and selective realism problem, 239–​40
truth-​transfer) structural realism is not selective
selective scientific realism, 130–​31, 146, 154, realism, 241–​48
155, 156–​57, 218, 222–​23, 228 structural realism is not selective realism,
Blomstrand-​Jørgensen/​Werner Debate, 130–​31 theory change argument, 243
testing, historical example, 130 Worrall on structural realism, 239–​40
selectivity, 145, 351 Structure of Scientific Revolutions (Kuhn), 253
Shech, E., 313–​14, 334–​35 success. See also novel predictive success;
Sklar, Larry, 234–​35 predictive success; specific examples
Smart, J. J. C., 204, 218 and topics
Smith, George, 85, 87–​89, 91, 94–​95. See also approximate truth, 29
molecular spectroscopy empirical, 13
Snow, John, 17, 23, 26, 28–​29 false-​but-​successful theories, 34–​35, 37, 48
solar corona spectrum, Nicholson’s calculations, success-​to-​truth inference, 24, 28–​29
114f, 114–​15 Suppes, Patrick, 174
Sommerfeld, Arnold, 100–101, 192 Surico, P., 271, 272
Sonneschein-​Mantel-​Debreu theorem, 263–​64 system of the world, 75–​76
special relativity, 92, 245–​46, 331–​33
Spectra and Dynamics of Diatomic Molecules, temporally novel predictive success, 33
The (Lefebvre-​Brion and Field), 175 boron discovery and scientific debate, 33, 35
spectroscopy (see also boron discovery)
molecular, 159–​78 (see also molecular Meckel’s gill slits, 24–​25
spectroscopy) Tentamen de motuum coelestium causis
Raman, 169–​70 (Leibniz), 80
spin, 100–​1, 325, 336–​37, 340–​42, 356–​57, 358, Thénard, Louis Jacques, 35, 40–​41, 44–​45, 47,
359, 364–​65 48–​50, 52
Standard Model, particle physics, 249, 353. See Theoretical Chemistry from the Standpoint of
also particles, realistic representations Avogadro’s Rule and Thermodynamics
bottom up approach, 358–​62 (Nernst), 161
top down approach, 358–​62 theoretical conservatism, 226–​29
Stanford, P. Kyle, 91–​92, 123, 199, 201–​3, 286–​ theoretician’s dilemma
307, 370–​71 objection 1, 209–​10
steam engine efficiency, 58–​59 objection 2, 210–​11
Stein, Howard, 178 theory change, 13, 241–​43. See also specific types
Stirling air-​engine, 59–​60, 64 continuity between earlier and later theories,
Streete, Thomas, 172–​73 1–​2, 11, 13
structural empiricism, 241, 250 Einstein’s General Theory of Relativity, 89–​92
structural realism, 101–​2 ether theory of light, 2, 243–​45
pragmatic and perspectival, 273–​77 mathematical structure, 245–​46
vs. selective realism, 241–​48 realists, 20–​21
Worrall, John, 264–​73 structural realism is not selective realism, 243
structure thermodynamic limits, 329, 334, 343, 345
causal, 265–​66, 273, 276, 277, 278, 361 thermodynamics, 330–​31, 334–​35. See also
modal, 241, 245–​46, 247–​48, 250, 252 specific topics
realism about, 264–​73 freezing point of water, pressure lowering
structure, of nature, 218, 223 (see freezing point of water, pressure
structure not selection, 239–​54 lowering)
historiography of science, realism history, early, 56–​57
and, 253–​54 phase transitions, 343
ontic structural realism, 240–​41, 245–​46, Rankine’s, 56–​57, 63–​64
247–​48, 250–​51 reversibility, 319–​20, 329–​31
 Index 387

successful prediction from, using abandoned van’t Hoff, Jacob Henricus, 132
theory, 57–​58 verisimilitude approach, 317–​18
thermometry, absolute, 59–​60 Vickers, Peter, 44, 99–​101, 201, 235–​36. See also
Thomson, James, 132, 140 specific topics
engine efficiency, William Thomson’s derivation external vs. internal posits,
figure, 66–​67 41, 147
prediction on Farr and miasma theory, 22, 24–​25
analysis, 61–​63 on novel predictive success, 44
methodological hesitation, 63–​64 impressive vs. unimpressive, 44–​45
no miracle, 62–​63 on posits, 222
quantitative calculation, 64–​66 on Psillos and essential deployment, 188, 192,
rationale behind, 60 201–​2, 209
Thomson, Thomas, 44 on quantum theory and quantum
Thomson, William, 57–​58, 105, 108–​9 mechanics, 100
Carnot’s theory, revival, 59–​60 on Sommerfeld’s atomic theory, 100–​1
James Thomson’s engine’s efficiency, William’s vortex theory of planetary motions, Newton
figure, 66–​67 on, 79–​89
Thomson brothers, pressure lowers freezing explanation, Newton’s account, 84–​86
point of water novel prediction, 56–​68. Huygens vs. Leibniz, 79–​82
See also freezing point of water, pressure natural philosophy, doing, 82–​84
lowering problem with vortices, 79–​82
top down approach, Standard Model, 358–​62 quam proxime, 86–​89
Treatise on Light (Huygens), 95 vortices, fate of, 77–​78
triumphalism, 254 Votsis, I., 199, 203
truth-​likeness, 243, 317–​18
truth-​transfer, theories of molecular structure, water, pressure lowering freezing point, 56–​68.
selective realism and, 130–​57. See also See also freezing point of water, pressure
molecular structure theories, selective lowering
realism and truth-​transfer Weak Evidence Argument, 149–​50
Tulodziecki, Dana. See miasma theory Werner, Alfred, 136–​39, 137f, 140, 147–​48
Tycho, 75–​76, 350 Werner’s coordination theory, 130–​31,
type hierarchy approach, 317–​18 134–​35, 138–​39, 140, 142–​43, 147–​
49, 155
unconceived alternatives problem, 131, 141–​46 Whewell, William, 287
Dicken on, 154 Whiggism, 110, 253–​54
No Final Theory Objection to the Problem of Wilson, K., 345–​46
Unconceived Alternatives, 144–​45 Wilson, Mark, 219–​21, 222, 223, 236
underdetermination, 1, 239–​40, 241–​42, 253, Wilson, William, 117
269, 300 Wing, Vincent, 172–​73
uniformitarianism, 217–​19, 222, 236 Woodward, J., 265–​66
unobservables, 1, 12–​13, 20–​21, 33, 73–​74, 86, Worboys, M., 14
141, 197–​98, 212, 220, 233, 241–​42, working posits, 20–​21, 147, 148, 149, 152, 155–​
243–​44, 312, 354 56, 218, 222–​23, 228
use-​novel predictions, 12, 13, 15, 17, 19–​21, 22, Worrall, John, 11, 13, 20–​21, 22, 101–​2, 239–​
24–​25, 33 40, 264–​73
Wray, K. Brad, 126, 128, 287, 300–​1
valence, 132–​34, 135–​36, 139, 142–​43, 147–​48, Wright, John, 184
242–​43
Valente, Giovanni, 330–​31. See also XT statements, 273–​75
infinite limits
van den Broek, Antonius, 104 Yablo S., 188–​89
van Fraassen, Bas, 123, 159–​60, 161, 162–​63, 170,
173–​74, 176–​78, 232–​33, 241–​42, 250 Zenneck, Jonathan, 89–​90
Scientific Image, The, 1 Zewail, Ahmed, 175–​76