Scientific Realism in Particle Physics: A Causal Approach 9783110354409, 9783110354393

Citation preview

Matthias Egg Scientific Realism in Particle Physics

Epistemische Studien

Schriften zur Erkenntnis- und Wissenschaftstheorie Herausgegeben von/Edited by Michael Esfeld, Stephan Hartmann, Albert Newen

Band 29

Matthias Egg

Scientific Realism in Particle Physics

A Causal Approach

ISBN 978-3-11-035439-3 e-ISBN 978-3-11-035440-9 ISSN 2198-1884 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http://dnb.dnb.de. © 2014 Walter de Gruyter Inc., Boston/Berlin Printing: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Preface This is a revised version of my PhD thesis entitled Causal Explanations and Scienti�c Realism in Particle Physics, which I defended in October 2012 at the University of Lausanne. The revisions are based on various comments I received after completing the thesis, especially the detailed reports by the members of my thesis committee. They also re�ect some recent developments in the literature on the subject published during the past year. It is not easy to pin down exactly when I started working on this book, because the project gradually developed out of a general interest in the topic of scienti�c realism. A �rst encounter with the issue dates from more than ten years ago, when I read Willard Van Orman Quine’s Two Dogmas of Empiricism for my minor in philosophy. Majoring in physics, I was particularly struck by Quine’s (1951, 41) statement that “physical objects are conceptually imported into the situation . . . as irreducible posits comparable, epistemologically, to the gods of Homer”. Though I did not pursue the subject further at the time, I felt that some day I should think more deeply about what it is that di�erentiates physics from ancient greek mythology. Some years later, having completed my physics degree, I took the opportunity to expand my philosophical knowledge in a rather unsystematic way. In this process, I again stumbled upon the scienti�c realism issue, for example in the preface to Immanuel Kant’s Critique of Pure Reason, where Kant draws a contrast between those areas of inquiry in which we are “merely groping about in the dark” and those moving forward “with that undeviating certainty which characterises the progress of science” (Kant 1787, B VII; translation 1855, xxiv). My desire for a deeper understanding of this contrast motivated me to look more closely at what contemporary philosophy had to say about science. It was then that I came across two books which were to become the starting point of the present work, namely Ian Hacking’s Representing and Intervening and Nancy Cartwright’s How the Laws of Physics Lie. The experimentally oriented version of scienti�c realism that Hacking and Cartwright defended seemed to me very plausible and worthy of further development. So in early 2007, I made a �rst attempt to get funding for a dissertation project pursuing that aim. In retrospect, it is not surprising that the attempt was unsuccessful, because my conception of what such a project would have to include in terms of argumentative work was at that time still rather rudimentary. But the topic had by then captivated me so much that giving up was no longer an option. During the year that followed, I worked towards remedying the de�ciencies of the �rst project submission and I launched a second try in 2008, this

VI � Preface time with more success. As a result, I can now gratefully acknowledge that the present work was supported by the Forschungskredit of the University of Zurich from August 2008 to February 2010. From 2010 onwards, I have been working as a research and teaching assistant at the University of Lausanne. This has proved to be an ideal environment for continued re�ection on scienti�c realism and for the completion of my doctoral work. In particular, it has sharpened my awareness of the challenges posed to scienti�c realism by quantum mechanics, as is now re�ected in the �nal part of the book. I am indebted to many people who supported me during the lengthy process just described. With respect to the �rst phase of this process, I owe special gratitude to Peter Schulthess, my thesis supervisor at the University of Zurich. Without his support and advice, I would have had neither the courage nor the competence to embark on this project. He also deserves credit for putting me in contact with Michael Esfeld and for facilitating a very smooth transition when the opportunity came for me to continue my work under the latter’s supervision. It has been a great privilege to work with Michael, and I cannot imagine how poor and confused my thesis would be without his encouraging comments, his incisive criticism, and his thought-provoking questions. I am also grateful to Anjan Chakravartty, Nicolas Gisin, and Mauricio Suárez for their willingness to serve as members of my thesis committee and for many helpful comments and discussions. Various other people have contributed to the quality of this work by responding to my questions, by commenting on drafts or talks, or by (sometimes unknowingly) providing me with new ideas in the course of discussions. I certainly do not remember the origin of every such contribution, so I apologize in advance for the incompleteness of the following list. Nonetheless, I would like to thank at least those contributors I do remember, namely Christof Aegerter, Andreas Allemann, Stefan Baumann, Walter Blum, Katherine Brading, Karim Bschir, Kai Büttner, Michaël Comte, Johannes Corrodi, Detlef Dürr, Jonathan Erhardt, Brigitte Falkenburg, Helmut Fink, Mathias Frisch, Hanjo Glock, Reto Gubelmann, Marion Hämmerli, Mario Hubert, Volker Jantzen, Philip Kitcher, Meinard Kuhlmann, Dominique Kuenzle, Raphaël Künstler, Vincent Lam, Dustin Lazarovici, Dennis Lehmkuhl, Tracy Lupher, David Lüthi, Tim Maudlin, Vilem Mudroch, Paul Näger, Wolfgang Pietsch, Tim Räz, Johannes Röhl, Dominic Roser, Christian Sachse, Laura Saller, Raphael Scholl, Charles Sebens, Norman Sieroka, Michael Sollberger, Patrice Soom, Jakob Sprickerhof, Kyle Stanford, Christian Steiner, Ulrich Straumann, Stevie Turkington, Antonio Vassallo, Sebastian Weiner, Adrian Wüthrich, and Chris Wüthrich. Several chapters of the present work have also bene�ted from comments by anonymous referees. Apart from the reviewers involved in the publications mentioned in the following paragraph, I should thank two referees for the journal Syn-

Preface � VII

these for helpful comments concerning chapter 3, which, however, remained unpublished until now. Chapters 4 and 5 are based on “Causal Warrant for Realism about Particle Physics”, Journal for General Philosophy of Science 43: 1124–1135, © 2012 Springer Science+Business Media. Chapter 6 is based on “Expanding Our Grasp: Causal Knowledge and the Problem of Unconceived Alternatives”, forthcoming in The British Journal for the Philosophy of Science, © Oxford University Press. Parts of section 7.1 are based on “Causal Realism in the Context of Bell-type Experiments” in T. Sauer and A. Wüthrich (eds.), New Vistas on Old Problems: Recent Approaches to the Foundations of Quantum Mechanics: 139–148, © 2013 Max Planck Research Library for the History and Development of Knowledge, Edition Open Access. The second part of section 7.3 is based on “The Foundational Signi�cance of Leggett’s Non-local Hidden-Variable Theories”, Foundations of Physics 43: 872–880, © 2013 Springer Science+Business Media. Chapter 8 is based on “Delayed-Choice Experiments and the Metaphysics of Entanglement”, Foundations of Physics 43: 1124– 1135, © 2013 Springer Science+Business Media. Looking beyond the professional realm, I gratefully acknowledge yet other sources of invaluable support. Above all, I recognize my complete dependence on the grace of my Lord Jesus Christ. To Him I owe everything that enabled me to write this book. Immeasurable thanks are also due to my wife Ursina for her unfailing love and for keeping me on track with much needed encouragement, intellectual stimulation, and excellent cooking. I thank my parents, my parentsin-law, my brother, my sister, and their families for constantly supporting me in what must, at times, have seemed to them a rather obscure endeavor. My father’s cousin Milo deserves special mention for his ongoing interest in my work and for all the inspiring scienti�c chats through the years. Finally, I would like to dedicate this work to the one person who got involved with it without being given a choice. Our elder son Nicolas was born in autumn 2010, and he has spent a signi�cant part of his early life sitting on my lap while I was trying to type some philosophical thoughts into my computer. Little darling, I hope you enjoyed these moments as much as I did, and I thank you for keeping me in contact with reality when all my mind was occupied with realism.

Contents Preface � V Part I

The Recent Debate on Scienti�c Realism

� �.� �.�

Scienti�c Realism and Its Relation to Common Sense � 3 Key Arguments Surrounding Scienti�c Realism � 4 Scienti�c Realism without Common Sense Realism? � 13

� �.� �.� �.�

Entity Realism � 19 From Theories to Entities � 20 Is Manipulability an Adequate Criterion of Reality? � 23 From Manipulation to Explanation � 28

� �.� �.� �.�

NOA and the Vices of the Realism Debate � 33 In Defense of Interpretation � 33 The Principle of Fairness � 40 NOA, Entity Realism, and the Homely Line � 44

Part II Causal Realism � �.� �.� �.� �.�

Causal vs. Theoretical Warrant � 49 Criterion 1: Non-redundancy � 50 Criterion 2: Material Inference � 54 Criterion 3: Empirical Adequacy � 59 Causal Realism’s Advantages over Entity Realism � 62

� �.� �.�

Causal Warrant for the Neutrino: A Case Study � 67 Bohr and Pauli on Beta Decay � 67 The Impact of Fermi’s Theory and the Need for Direct Detection � 69 The Detection of the Neutrino by Reines and Cowan � 72

�.� � �.� �.� �.�

The Problem of Unconceived Alternatives � 79 Previous Attempts to Undermine the PUA � 80 Causal Knowledge as a Criterion for the Realist � 85 Causal Realism, Unconceived Alternatives, and the Atomic Hypothesis � 92

X � Contents Part III The Quantum Challenge � �.� �.� �.� � �.� �.� �.� � �.� �.� �.� �.�

Causal Realism in the Context of Bell-Type Experiments � 105 Bell-Experiments: Causal Warrant for Superluminal Causation � 105 Causal Realism and Underdetermination in Quantum Mechanics � 114 Some Experimental Constraints on Explanations for EPR � 122 Delayed-Choice Experiments and the Metaphysics of Entanglement � 137 Delayed Choice in the Double-Slit Experiment � 139 The Quantum Eraser � 140 Delayed-Choice Entanglement Swapping � 145 Particle Physics without Particles? On Causal Realism in Quantum Field Theory � 149 Against Localizability: Malament’s Theorem and Its Generalizations � 149 Against Countability: Unruh E�ect and Interacting Fields � 154 Defending Localizability and Countability � 161 Concluding Remarks on Realism, Fundamentalism, and QFT � 169

Bibliography � 175 Index � 187

�

Part I: The Recent Debate on Scienti�c Realism “You reproach me with unbelief: ‘You see, but you don’t believe.’ But, my friend, I am not alone in that, all of us there are stirred up now, and it all comes from your science. While there were still just atoms, �ve senses, four elements, well, then it all still stayed together anyhow. They had atoms in the ancient world, too. But when we found out that you had discovered your ‘chemical molecule,’ and ‘protoplasm,’ and devil knows what else—then we put our tails between our legs.” Fyodor Dostoevsky, The Brothers Karamazov

1 Scienti�c Realism and Its Relation to Common Sense In early 1909, Hans Geiger and Ernest Marsden, under the direction of Ernest Rutherford, performed one of the most famous experiments in the history of physics. It involved the scattering of alpha particles (products of radioactive decay consisting of two protons and two neutrons) on a thin gold foil, and it showed that some of the alpha particles were scattered by surprisingly large angles: It was expected that the particles would pass through the foil more or less unde�ected, but in fact some of them almost �ew back to where they came from. Rutherford’s later comment on this result is nearly as famous as the experiment itself: It was quite the most incredible event that has ever happened to me in my life. It was almost as incredible as if you �red a 15-inch shell at a piece of tissue paper and it came back and hit you. (quoted in Andrade 1964, 111)

The scienti�c impact of this “incredible event” was that it led Rutherford to postulate the atomic nucleus, but this is not the point here. What I would like to highlight is the following remarkable aspect of the analogy drawn in the above quotation: While Rutherford compares the gold foil—a visible, tangible object—to something as fragile as a tissue paper, he speaks of a 15-inch shell to illustrate the role of an alpha particle—an invisible, intangible object, which no one had even suspected to exist until only a few years prior to the Geiger-Marsden experiment. Rutherford’s metaphor reveals his deep conviction that these elusive entities were just as real as the gold foil with which they interacted. In the philosophical debate, such a conviction is called scienti�c realism. The aim of this chapter is to add some precision to this rough characterization and to introduce the central arguments employed in the debate on scienti�c realism. A recurring theme in the expository section 1.1 is that scienti�c realists and their opponents seem to agree on a certain realism about the objects of common sense. Section 1.2 will then show that this common presupposition is in tension with a particular way to defend scienti�c realism. In the course of resolving this tension, I will introduce a distinction which will be important in many of the later chapters, namely the distinction between the real and the fundamental.

4 � Scienti�c Realism and Its Relation to Common Sense

�.� Key Arguments Surrounding Scienti�c Realism Scienti�c Realism as an Extrapolation of Common Sense Realism Although scienti�c realism is a philosophical position, it is highly unlikely that Rutherford employed any philosophical re�ections to arrive at his realism about alpha particles. Rather, having performed experiments with such particles for some years, it was natural for him to believe in their reality. In the introduction to an encyclopedia entry on scienti�c realism, Richard Boyd (2010, sec. 1) re�ects on this somewhat unphilosophical character of realism: What requires explanation is why [scienti�c realism] is a philosophical position rather than just a common sense one. Consider, for example, tropical �sh realism—the doctrine that there really are tropical �sh; that the little books you buy about them at pet stores tend to get it approximately right about their appearance, behavior, food and temperature requirements, etc.; and that the �sh have these properties largely independently of our theories about them. That’s a pretty clear doctrine, but it’s so commonsensical that it doesn’t seem to have any particular philosophical import. Why is the analogous doctrine about science a philosophical doctrine? The answer is that—setting aside skepticism about the external world—there are no philosophical arguments against tropical �sh realism, whereas important philosophical challenges have been raised against scienti�c realism. The dimensions of scienti�c realism, understood as a philosophical position, have been largely determined by the responses scienti�c realists have o�ered to these challenges.

Boyd’s way of introducing scienti�c realism by comparing it with a commonsensical position such as tropical �sh realism has two advantages: First, it makes clear what it takes to defend scienti�c realism philosophically, namely a reply to the challenges put forth by antirealists. Second, it gives us a hint about what is not needed in a philosophical defense of scienti�c realism, namely a reply to skeptical challenges which would threaten tropical �sh realism just as much as scienti�c realism. Boyd speaks of “setting aside skepticism about the external world”, but there are in fact quite a few other philosophical worries that the tropical �sh realist needs to set aside, for example the idea that, although there is an external world, its containing such things as tropical �sh depends on our possessing the concept “tropical �sh”. The point is that, for debating scienti�c realism, we can set all these worries aside, because the antirealist accepts realism about tropical �sh and the like. In other words, the debate on scienti�c realism can only start once we have agreed to assume common sense realism, the doctrine that the objects of common sense exist independently of our beliefs about them and that they actually have (more or less) the properties these beliefs attribute to them. This is not

Key Arguments Surrounding Scienti�c Realism

� 5

to say that common sense realism is an indisputable truth, but the dispute about it falls outside the scope of the debate on scienti�c realism. By treating common sense realism as an unquestioned presupposition of the debate itself, I may not be doing full justice to all extant versions of scienti�c realism; some scienti�c realists will want to defend their position without relying on common sense realism. Later in this chapter I will give reasons why I do not �nd this strategy very promising, at least as far as the project I am interested in is concerned. For the moment, I just wish to make the more basic point that I’m not even sure I understand what scienti�c realism claims without relying in some way on common sense realism. The reader will have noticed that I have not given a de�nition or an explication of the term “real”, nor will I try to do so, because the notion is so basic that it cannot be explained in more fundamental terms. All that can be done to elucidate the notion is to give some commonsensical examples of real entities, such as gold foils and tropical �sh. Indeed, if someone does not already understand what it means for such familiar entities to be real, I shall not even try to speak to him about the reality of alpha particles. The claim that common sense realism is a shared presupposition of the realist and the antirealist also excludes some forms of antirealism from consideration. One example is social constructivism, the view that scienti�c �ndings are social constructs rather than discoveries about an independent reality. An adherent of this view will probably regard not only science but also common sense as socially determined, and he will therefore oppose common sense realism along with scienti�c realism. But then, I do not see much bene�t in turning to the scienti�c context for discussing social constructivism. It seems much simpler to discuss the issue in the context of common sense, rather than in the context of science, which is, after all, one of the more complex human activities. If the constructivist can refute common sense realism by showing that common sense entities are nothing but social constructs, the refutation of scienti�c realism will come along for free (insofar as I am right about the dependence of scienti�c realism on common sense realism). Since I am speci�cally interested in the scienti�c context, I will therefore not further discuss versions of antirealism about science which involve an antirealism about common sense.¹

1 Accordingly, I will henceforth use the term “antirealist” to denote any opponent of scienti�c realism who is not opposed to common sense realism. In contrast, opponents to both common sense realism and scienti�c realism will be called “skeptic”. “Realist” will usually mean “scienti�c realist”, unless otherwise stated. Whenever the need arises, I will specify what exactly the realist is a realist about.

6 � Scienti�c Realism and Its Relation to Common Sense There is, however, an antirealist position which I do not want to exclude from consideration, though it does not fully incorporate common sense realism: The constructive empiricist does not believe in the reality of what cannot be observed by the unaided human senses, so there are arguably some common sense entities (e.g., laser beams, food calories, or unemployment rates) about which he is not a realist. Nevertheless, there is a substantive class of entities about which he agrees with the common sense realist, namely the class of observable common sense entities (such as gold foils and tropical �sh). Furthermore, although some unobservable entities are recognized by common sense (as the above examples show), most such entities appear in scienti�c contexts, which makes constructive empiricism a proper subject for philosophy of science. I will therefore treat constructive empiricism as an antirealism in my sense of the term, tacitly admitting that the common sense realism it incorporates is not a complete one, but one restricted to observable common sense entities. But the signi�cance of common sense realism is not exhausted by the fact that it is presupposed in the debate and is therefore crucial for even formulating scienti�c realism; it also plays a signi�cant role in defending it. Put in terms of the two examples we have considered so far, the underlying intuition is that whoever is a realist about tropical �sh should also be a realist about alpha particles, because the two cases are relevantly similar, or so the scienti�c realist claims. One argument for scienti�c realism which is particularly explicit about the importance of this intuition is Philip Kitcher’s (2001) Galilean strategy. It derives its name from an epistemological problem confronting Galileo Galilei when he claimed that the newly invented telescope revealed surprising astronomical phenomena, which contradicted the traditional belief that heavenly bodies (beyond the moon) were immutable. Even Galilei’s opponents had to admit that the telescope was a reliable instrument, allowing one to see objects too far away to be discerned by the unaided eye, but they denied that this reliability was universal: There was a well-recognized distinction between the sublunary world and the [heavenly] sphere,² and critics of the telescope . . . denied the legitimacy of extrapolating from reliability on earth to reliability in the heavens. (Kitcher 2001, 174)

Just as Galilei extrapolated from the telescope’s reliability on earth to its reliability when pointed at planets and stars, so the scienti�c realist extrapolates from realism in the domain of common sense to realism in the scienti�c domain. And just as Galilei justi�ed his extrapolation by arguing that the celestial/terrestrial dis-

2 The original reads “terrestrial sphere”, which I can only interpret as a slip of the pen.

Key Arguments Surrounding Scienti�c Realism

� 7

tinction is irrelevant to the reliability of the telescope, so the realist can justify his extrapolation by denying the relevancy of the di�erences between common sense and science for the arguments supporting realism. We will encounter this strategy in many arguments for scienti�c realism, although its extrapolative character is not always made as explicit as in Kitcher’s account.³ For the moment, I focus on the best-known argument in favor of scienti�c realism, the so-called miracle argument (or no-miracles argument). Hilary Putnam (1975, 73) famously captured its essence in the slogan: “The positive argument for realism is that it is the only philosophy that doesn’t make the success of science a miracle”. I suggest that the intuition underlying this claim has to do with the extrapolation just described. To see how, let us look at a passage in which Kitcher makes this intuition explicit: Consider a paradigm situation in which we observe . . . another person who is responding to an environment that we can also observe. . . . This subject may form representations that are either inaccurate or accurate, for there are entities independent of her of which they may not be correct. Further, we appreciate the force of the suggestion that our subject’s successes in responding to and shaping her environment would be inexplicable unless some of her representations were accurate. If we are observing a scientist, a particle physicist or a molecular geneticist, say, we see her predicting and intervening in the observable realm on the basis of representations purporting to characterize much more elusive entities, and we suppose that the successes of her ventures in forecast and control probably depend on the accuracy of her representations of those entities. (Kitcher 2001, 154–155)

The reason why we associate success in the scienti�c domain with some kind of realism about scienti�c theories (Kitcher’s “representations purporting to characterize much more elusive entities”) is that a similar association holds in everyday contexts: People couldn’t be systematically successful in their actions unless some of their beliefs about their surroundings were true. To return to the above example, the tropical �sh in my aquarium would not �ourish if the little book I

3 For completeness, I mention that Kitcher’s (2001) defense of realism contains more than an extrapolation from common sense realism to scienti�c realism. In fact, Kitcher’s starting point is not common sense realism, but a more modest position he calls “Natural Epistemological Attitude” (153). This attitude is not a realism at all, but only the assumption that people sometimes form representations of entities that are independent of them. To arrive at what he calls “real realism”, Kitcher does not only perform the extrapolation discussed above, but also a second extrapolation, saying that people’s representations of independent entities do not depend on their being observed by anyone. As a result of this extrapolation, we “envisage a world of entities independent not just of each but of all of us, a world that we represent more or less accurately” (155), which amounts to something like common sense realism. For a more detailed discussion of Kitcher’s real realism, see Egg (2009).

8 � Scienti�c Realism and Its Relation to Common Sense bought at the pet store misinformed me about their needs. Since the antirealist accepts this kind of connection between successful actions and (approximately) true beliefs in the realm of common sense, the realist can legitimately ask him why he does not accept the same connection in the scienti�c realm. Many of the arguments against scienti�c realism can be seen as an attempt to answer that question.

Di�erent Dimensions of Scienti�c Realism Before elaborating on the arguments against scienti�c realism, let us get a clearer account of what a scienti�c realist is committed to. There are many di�erent characterizations of this commitment on the market, but one that has gained fairly widespread acceptance in recent years is Stathis Psillos’s (1999, xix) view that scienti�c realism incorporates three di�erent theses (see also Chakravartty 2011, sec. 1.2). I quote them here in the compact form given in Psillos (2005, 385): The Metaphysical Thesis: The world has a de�nite and mind-independent structure. The Semantic Thesis: Scienti�c theories should be taken at face-value. The Epistemic Thesis: Mature and predictively successful scienti�c theories are (approximately) true of the world. The entities posited by them, or, at any rate, entities very similar to those posited, inhabit the world.

Other authors give more detailed di�erentiations of the realist commitment. Howard Sankey (2008, 12–20), for example, distinguishes six core doctrines and a number of optional doctrines associated with scienti�c realism. It is not immediately obvious in how far Sankey’s realism actually di�ers from Psillos’s or if it is just a di�erent manner of presenting more or less the same position. Be that as it may, I will for now adopt Psillos’s three-dimensional characterization, because it is �ne-grained enough for my present purposes. But there are also those who think that Psillos’s picture already has too many dimensions. Let me brie�y discuss two proposals for a one-dimensional understanding of scienti�c realism. Michael Devitt (1997) sees realism (which in this case includes both common sense realism and scienti�c realism) as a metaphysical position, which should be clearly distinguished from any semantic or epistemic issues. Indeed, he considers it an outright aberration of the realism debate that semantic and epistemic issues have played such a central role in it (Devitt 1991). However, I do not think that the contradiction between Psillos’s and Devitt’s formulation of realism is as strong as it might appear from this initial sketch. To see why, notice �rst that Devitt’s (metaphysical) formulation of realism is signi�cantly richer than Psillos’s “metaphysical thesis”:

Key Arguments Surrounding Scienti�c Realism

� 9

Realism Tokens of most current common-sense and scienti�c physical types objectively exist independently of the mental. (Devitt 1997, 23)

In particular, Devitt’s realism incorporates a thesis which �gures in the epistemic part of Psillos’s realism, asserting the existence of (some of) the entities posited by science.� Furthermore, recall Boyd’s observation (quoted above) that the different dimensions of scienti�c realism have been largely determined by the responses scienti�c realists have o�ered to antirealist challenges. With this in mind, a rapprochement between Psillos and Devitt seems achievable. On the one hand, Psillos would probably agree that scienti�c realism is �rst and foremost a metaphysical thesis, that is, a thesis about what there is in the world and what it is like, rather than a thesis about the meaning of scienti�c theories or about scienti�c knowledge. On the other hand, Devitt surely does not deny that there are di�erent ways of disputing this metaphysical thesis, and (as will be discussed in the next subsection) this is how semantic and epistemic considerations enter the picture. Another way of reducing scienti�c realism to a one-dimensional a�air is to understand it as nothing more than a doctrine about the aim of science. Such a purely axiological realism is defended by Timothy Lyons (2005), who explicitly cuts o� the epistemic thesis from his formulation of scienti�c realism. (I take it that scienti�c realism thereby also loses its metaphysical component, while I am not quite sure what happens to the semantic dimension.) Whatever virtues such a conception of scienti�c realism might have in general, it is of no interest for the kind of question I am going to discuss in the following chapters. As the title of this book indicates, I am concerned with realism in the context of a speci�c branch of science, namely particle physics. This branch of science claims to have discovered some real entities and to have found out some truths about them, and the question will be whether such claims are justi�ed. This presupposes a scienti�c realism which does not merely claim that science aims at true statements about real entities, but also claims that it has, to a certain extent, actually achieved this aim. Otherwise, scienti�c realism would be compatible with antirealism about all contemporary science, and there could not be a substantial debate about realism in (present-day) particle physics.

4 Psillos (2005, 396) concedes that this renders the designation “epistemic thesis” somewhat inappropriate: “Perhaps it was unfortunate that I called the last dimension of scienti�c realism ‘epistemic’. I was carried away by sceptical anti-realist attacks on realism. I would now call it: the factualist thesis.”

10 � Scienti�c Realism and Its Relation to Common Sense The Most Influential Arguments against Scienti�c Realism I mentioned above that the di�erent ways in which one might dispute scienti�c realism lead to semantic and epistemic considerations, even if one starts with a purely metaphysical conception of realism. To see how this comes about, let us consider the following example of a scienti�c statement: “In a β+ decay, a positron and a neutrino are produced.” The scienti�c realist takes this statement to be true in the sense that there really are such processes as β+ decays and that they really do produce such objects as positrons and neutrinos. But there is a variety of possible reactions to this statement which avoid realism about such entities. 1. One may simply refuse to accept the statement in any sense. I will not further consider this option, because such a radical anti-science attitude would be at odds with the very project of doing philosophy of science. 2. One can accept the statement, but not believe it to be true. This option comes in two variants: (a) The statement may be taken not to have a truth value at all. The main representative of this view is the instrumentalist, who thinks of scienti�c statements only as instruments to derive predictions, not as truth-valued descriptions of any reality. (b) The statement may be taken to have a truth value, but it may be doubted that we can know it. This kind of partial skepticism in turn takes di�erent forms (to be discussed below), depending on what motivates the idea that the truth of scienti�c statements lies outside the limits of our knowledge. 3. The statement may be accepted as true, but not taken at face value. This is the approach of reductive empiricism, which is closely related to instrumentalism (option 2a), but di�ers from it by admitting that the statement is about something (albeit not about positrons etc.). The idea is that although the statement purports to refer to unobservable entities, it should really be understood as referring to what can be observed under particular circumstances. 4. Finally, one can take the statement at face value and accept it as true, but understand truth in a non-realistic way. Under this interpretation, what makes the statement true is not some fact about β+ decay, but the statement’s ful�lling some epistemic condition (e.g., warranted assertibility). This rough sketch of di�erent kinds of antirealism su�ces to show how defending scienti�c realism inevitably becomes a multi-dimensional activity. In confronting antirealism of the types 2a and 3, semantic considerations about the meaning of scienti�c claims are crucial. Type 2b antirealism can only be addressed by attending to the epistemic issue of how scienti�c claims can be justi�ed. And dealing

Key Arguments Surrounding Scienti�c Realism

� 11

with option 4 requires the realist to enter into a debate about truth (which itself has metaphysical, semantic, and epistemic aspects). However, not all dimensions are equally important at all times. While the debate in the early twentieth century was very much concerned with the semantic issues raised by instrumentalism and reductive empiricismreductive empiricism, the more recent debate has centered around the epistemic concerns expressed in option 2b. In fact, it seems to me that the clearest way to characterize the beginning of “the recent debate” is by reference to two highly in�uential publications, both of which contain type 2b attacks on scienti�c realism: Bas van Fraassen’s Scienti�c Image (1980) and Larry Laudan’s Confutation of Convergent Realism (1981). The other dimensions have, of course, not been completely absent from the recent debate (as will be seen in section 3.1, where I will brie�y address some arguments of type 4), but they have not been nearly as important in shaping the course of the debate in the past three decades. I will not discuss van Fraassen’s and Laudan’s arguments in detail, because they have already been widely discussed in the literature. Nevertheless, some key points deserve mention, not only because later developments depend on them, but also because they give substance to the above claim that the legitimacy of the realist’s extrapolation from common sense to science is the central issue of the debate. When introducing option 2 above, I spoke of accepting a statement without believing it to be true, which may sound somewhat strange. Van Fraassen (1980, 12) gives a precise meaning to this distinction by de�ning acceptance of a theory in terms of its empirical adequacy: Science aims to give us theories which are empirically adequate; and acceptance of a theory involves as belief only that it is empirically adequate. This is the statement of the anti-realist position I advocate; I shall call it constructive empiricism.

The concept of empirical adequacy di�ers from the concept of truth by restricting the truth content of a theory to the domain of what is observable: “A theory is empirically adequate exactly if what it says about the observable things and events in this world, is true—exactly if it ‘saves the phenomena’.” (ibid.)� Van Fraassen’s constructive empiricism is therefore a partial skepticism in the sense that it questions our justi�cation for any belief which goes beyond what is observable by the unaided human senses. More precisely, van Fraassen holds that the realist’s be-

5 Van Fraassen calls this a “preliminary explication” of empirical adequacy. His more complete explication (van Fraassen 1980, chap. 3) relies heavily on the semantic approach to scienti�c theories, which is not relevant for the present discussion.

12 � Scienti�c Realism and Its Relation to Common Sense lief in unobservable entities is not necessary for an adequate account of science: “[Constructive empiricism] makes better sense of science, and of scienti�c activity, than realism does and does so without in�ationary metaphysics” (73). The constructive empiricist thus resists the invitation to extrapolate from common sense realism to scienti�c realism, because he sees the limits of observability as limiting also the justi�ability of the inferences which generate realism. In response, realists have argued against placing such prime epistemic importance on the limits of human perception (Psillos 1999, chap. 9; Kitcher 2001, sec. 6). I will add some remarks on the signi�cance of the observable/unobservable distinction for the realism debate in section 3.2. Laudan (1981) gives a di�erent reason for distrusting the realist’s conviction that the success of science argues for the truth of scienti�c theories. He looks at the history of science and claims that many successful scienti�c theories of the past would now be regarded as false in the sense that their central terms did not refer to any real entities: I daresay that for every highly successful theory in the past of science which we now believe to be a genuinely referring theory, one could �nd half a dozen once successful theories which we now regard as substantially non-referring. (Laudan 1981, 35)

In support of this claim, Laudan lists some examples of once successful theories now believed to be false, and he alleges that the list “could be extended ad nauseam” (33). Laudan’s argument is usually called pessimistic (meta-)induction, which may not be the most accurate name, given that Laudan does not explicitly use any inductive reasoning (Lyons 2002 therefore calls the argument pessimistic meta-modus tollens). The most widely employed strategies by realists to avoid such pessimism consist in denying that the theories on Laudan’s list were actually successful (in some speci�ed sense of “success”) or in maintaining that at least some of them could still be regarded as approximately true or partially referring (Kitcher 1993, 140–149; Psillos 1999, chap. 5). The degree to which these strategies succeed is still a matter of debate, and I will discuss it in some detail in chapter 6. To complete this quick tour of the antirealist’s argumentative arsenal, I should mention the underdetermination of theory by evidence. This is the idea that every �nite set of empirical data is compatible with in�nitely many, mutually incompatible theories accounting for that data, so we are never justi�ed in singling out any one of these theories as true. The argument from underdetermination plays some role in van Fraassen (1980, chap. 3), and it can be combined with a Laudan-type historical argument to yield the kind of contemporary antirealism to be discussed in chapter 6, but I take it to be of limited importance as an independent argument

Scienti�c Realism without Common Sense Realism?

�

13

against scienti�c realism. The reason for this is that in its general form, underdetermination is just as threatening to common sense realism as it is to scienti�c realism (see section 4.1 for more details). However, there are some cases of underdetermination associated with speci�c scienti�c theories, and in this sense, underdetermination is relevant for scienti�c realism. One of the most serious of these cases will be discussed in section 7.2.

�.� Scienti�c Realism without Common Sense Realism? Stanford and Psillos on Common Sense Realism In the previous section, I described scienti�c realism as the result of an extrapolation which takes its starting point from common sense realism. This presupposes a continuity between common sense and science, and one may doubt whether such a continuity exists. In this section, I will address an argument which seems to support this kind of doubt, because it seems to defend scienti�c realism at the expense of common sense realism. The context of the argument is Psillos’s critique of Kyle Stanford’s (2006) antirealism. The details of Stanford’s position do not matter here (they will be discussed in chapter 6). For present purposes, the following summary of its central tenets will su�ce. Stanford sees large parts of science as a�ected by what he calls the problem of unconceived alternatives (PUA). This means that we should expect our present scienti�c theories to be replaced by radically di�erent (as yet unconceived) alternative theories in the future. Consequently, we should take an instrumentalist, rather than realist attitude towards them. But not all of our knowledge is subject to the PUA. In particular, common sense realism is not harmed by it. Hence we see that Stanford’s approach precisely follows the antirealist pattern described in section 1.1: Since the PUA a�ects scienti�c claims in a way it does not a�ect the claims of common sense, we have reasons to doubt the truth of the former, but not of the latter. Psillos (2009, 79) disagrees: “Common sense is not a theory towards which we can have a stance of strict and literal belief. Many terms and predicates we use in our commonsensical description of the world are vague and imprecise.” Furthermore, he denies that common sense is immune to the PUA: “The biggest— at some point unconceived—alternative to the hypothesis of common bodies is science itself . . . science is not just piled upon common sense. It adds to it and it corrects it” (79–80). These statements may not yet amount to a denial of common sense realism. That we do not strictly believe the commonsensical description of the world does

14 � Scienti�c Realism and Its Relation to Common Sense not mean we do not believe it at all. Perhaps Psillos just invites us to regard the claims of common sense as approximately true, and this would still be a form of common sense realism. But if that was his idea, it is hard to see how it could serve as an argument against Stanford: The plausibility of Stanford’s antirealism, which combines common sense realism with an instrumentalism about higher level theories does not hinge on whether the former is spelled out in terms of strict and literal belief or in terms of approximate truth. (After all, even scienti�c realism is usually formulated in terms of approximate truth, as we have seen in section 1.1.) Furthermore, Psillos’s claim that common sense is a�ected by the PUA seems to imply that commonsensical beliefs cannot be approximately true. For if a hypothesis can still count as approximately true after it has been replaced by one of its (previously unconceived) alternatives, there is not really a problem of unconceived alternatives at all. All this seems to suggest that Psillos does not accept common sense realism.

Disentangling Realism from Fundamentalism Before investigating more deeply whether this is in fact Psillos’s position, I want to explore in some detail one problematic consequence of rejecting common sense realism. If we deny common sense realism on the grounds that common sense is corrected by science, then we should also abstain from realism about any scienti�c theory which is likely to be corrected by another scienti�c theory, and it is unclear whether realism about any of our current theories would survive such a restrictive policy. To better understand this problem, notice �rst that the notion of being corrected can be read in a diachronic as well as in a synchronic sense. The �rst reading refers to the temporal development of scienti�c knowledge, which involves the replacement of earlier beliefs by later, more accurate ones. The second reading is concerned with coexisting theories, one of which is taken to provide a more accurate picture of reality than the other, such that, although the latter has not been replaced by the former, it is, at least in principle, taken to be reducible to it. If realism were only to be adopted with respect to theories which are not likely to be corrected in any of these two senses, the only theory worthy of a realist’s commitment would be the �nal, fundamental theory (assuming that there is ever going

Scienti�c Realism without Common Sense Realism?

� 15

to be such a thing). Carl Hoefer (2003, 1411) gives a particularly clear statement of the resulting fundamentalism:� The ultimate set of mathematical laws . . . is meant to be uni�ed, consistent, coherent, and of clear applicability to any real situation. . . . It is these laws that the fundamentalist believes in, not the halfway houses we have managed to construct to date.

In section 1.1, I argued that whoever has a philosophical interest in actual scienti�c theories should not rest content with a purely axiological conception of scienti�c realism, because such a realism is silent about what science has achieved up to now. Insofar as the fundamentalist conception of scienti�c realism introduced here avoids any commitment to what present science tells us, it exhibits precisely the same weakness. But perhaps we do not have to be as pessimistic about current science as the above quotation seems to suggest. Indeed, labeling current scienti�c theories as “halfway houses” does not stop Hoefer from being strongly committed to at least some parts of present science: The primary argument for fundamentalism, not yet mentioned, is this: we all believe, with very good reason, that things in the physical world are all composed of a few basic types of particles: electrons, protons, neutrons, and photons, mostly, along with a tiny amount of more esoteric particle kinds. We know that these tiny things are puzzling in various ways, and they cannot be thought of as Newtonian-style billiard balls moving on smooth trajectories under the in�uence of purely local force �elds. Nevertheless, they are here to stay. Whatever radical changes future physics may bring, it is not really conceivable that, à la phlogiston, these entities will vanish without a trace and come to be seen as embarrassing errors with no correlate or counterparts in the True Physics. (Hoefer 2003, 1410)

To say that electrons, protons etc. “are here to stay” is to assert that our beliefs about these entities are not subject to what I above called correction in the diachronic sense. As the following chapters will show, I fully agree with this claim, which I shall express by saying that these entities are real. But this raises the question whether they are fundamental, that is, whether their place “in the True Physics” will be such that statements about them are not subject to correction in the synchronic sense. Chapter 9 will discuss some powerful arguments for answering this question in the negative (see also Teller 2004, 436–438). Although these arguments are not unassailable, the least one can say is that the question whether

6 The context of Hoefer’s paper is slightly di�erent from what I am discussing here; he is not concerned with common sense realism, but is criticizing Nancy Cartwright’s (1999) ontological pluralism. However, Hoefer’s arguments transfer nicely to the present context, because common sense realism seems to be very much part of ontological pluralism.

16 � Scienti�c Realism and Its Relation to Common Sense electrons etc. are fundamental must remain open at the current stage of scienti�c inquiry. But once Hoefer is prepared to accept as real some non-fundamental entities, it seems that he also has to accept common sense realism, for his criterion of reality applies to tropical �sh and billiard balls just as much as it applies to electrons and protons: It is not really conceivable that future scienti�c developments will render these entities as obsolete as phlogiston.� The upshot of this is that we need to distinguish between two versions of fundamentalism. The moderate version just described accepts the reality of nonfundamental entities alongside the fundamental ones (insofar as beliefs about them are unlikely to be diachronically corrected in the future), and is therefore compatible with common sense realism. By contrast, refusal to be committed to any theory subject to synchronic correction leads to a more radical, eliminative version of fundamentalism. Eliminativism rejects the distinction between the real and the fundamental, and instead claims that only what is fundamental is real: There really are no such things as the bodies of common sense or the objects of non-fundamental science, because ultimately everything reduces to the fundamental constituents of the world (whatever these are). If we understood scienti�c realism in this eliminativist sense, it would again incur the problem of not being informative about actual present-day science, because we do not seem to be in a position to ascribe fundamental status to any entities so far discovered by our science.� Furthermore, such an eliminativism would commit what Musgrave (1996, 22) calls the “explaining-is-explaining-away fallacy”: From the fact that the properties and the behavior of some objects can be explained by the properties and the behavior of their constituents, it does not follow that these objects do not exist. These features speak against an eliminativist conception of scienti�c realism.

7 The fundamentalist might try to drive a wedge between his commitment to elementary particles and a commitment to common sense realism by pointing out that particle physics is somehow “closer to” the true, �nal theory than the beliefs of common sense. Something like this seems to be the idea of Sklar (2003), although quali�cations similar to the ones given in footnote 6 apply here as well. For a criticism of this strategy, see Teller (2004, sec. 5–8). 8 As noted above, this claim will be defended in chapter 9. Of course, the claim is absolutely trivial for any science other than fundamental physics, because these sciences do not even pretend to deal with fundamental entities. The fact that an eliminativist version of scienti�c realism excludes all special sciences from realist treatment is yet another reason against adopting this approach.

Scienti�c Realism without Common Sense Realism?

� 17

Psillos’s Argument Revisited I have argued that conjoining a defense of scienti�c realism with a rejection of common sense realism leads to an implausible eliminativist conception of scienti�c realism. So if it is really part of Psillos’s argument against Stanford to reject common sense realism, it seems that he must be committed to such an eliminativist realism, and we would expect him to defend this conception against the kind of criticism I have just voiced. Surprisingly, Psillos (2005) does just the opposite. That is, he explicitly (and convincingly) argues for a conceptual separation of scienti�c realism and fundamentalism (in both the moderate and the eliminative version), captured in the slogan “realism is about what is real and not about what is fundamentally real” (390). His reasons for doing so partly overlap with the above arguments against an eliminativist understanding of realism, but his emphasis is on the need to separate the question of reality from the question of fundamentality. We can be realists about a number of domains (or subject-matters) without necessarily taking a stance on independent metaphysical issues. . . . The issue of whether some entities are basic, or derivative, or irreducible, or sui generis, is a separate concern and needs to be addressed separately. In general, it will stem from other metaphysical commitments that one might have, e.g., a commitment to physicalism, or naturalism, or materialism, or pluralism. (391)

Psillos’s insistence on separating realism from fundamentalism creates a puzzle. If what I said above is correct, then the conjunction of scienti�c realism with a rejection of common sense realism implies a fundamentalist (even eliminativist) conception of scienti�c realism. Since Psillos opposes the consequent of this implication, and since he is certainly unwilling to give up scienti�c realism, it would be inconsistent for him to reject common sense realism. But that is precisely what he seems to be doing in his argument against Stanford. Here is my proposal on how to resolve the puzzle: Psillos’s argument against Stanford should not be understood as an argument against common sense realism, although it is easy to mistake it as such. Instead, I take Psillos to argue for the following conditional: (*) If the problem of unconceived alternatives (PUA) refutes scienti�c realism, then it also refutes common sense realism. In other words, Psillos’s statements quoted at the beginning of this section, which seemed to show his opposition to common sense realism, are asserted only conditionally on the acknowledgement (for the sake of the argument) that the PUA really is a problem for scienti�c realism.

18 � Scienti�c Realism and Its Relation to Common Sense This reinterpretation of Psillos’s argument not only has the advantage of avoiding a commitment to a fundamentalist conception of scienti�c realism, it also brings the argument back in line with the dialectic of the realism debate described in section 1.1: Despite appearances, Psillos and Stanford actually agree on common sense realism; this is the common ground from where the debate starts. And the key point of the debate is whether there is a relevant di�erence between common sense and science with respect to the PUA. If Stanford is right in saying that there is such a di�erence, then there is room for antirealism about science, even though one started with realism about common sense. Conversely, if Psillos is right about the conditional (*), then a simple application of modus tollens shows that the PUA does not support Stanford’s antirealism. The impact of the PUA on scienti�c realism will be discussed in detail in chapter 6. What I have been arguing here is that this discussion exempli�es very clearly the basic role of common sense realism in the debate on scienti�c realism, even if one focuses on an argument which �rst looked as if it disconnected scienti�c realism from common sense realism.

2 Entity Realism As remarked in section 1.1, antirealist challenges have played a major role in shaping the historical development of scienti�c realism. This is particularly so for the pessimistic induction, which prompted realists to become selective about which parts or aspects of science they want to be realists. These parts are then supposed to survive the scienti�c revolution associated with the replacement of one scienti�c theory by another. As a consequence, such a selective realism promises to be immune to the pessimistic induction. There are di�erent ways to be a selective realist. I already mentioned Kitcher (1993, 140–149) and Psillos (1999, chap. 5), who suggest the following strategy: Identify the theoretical constituents that contributed essentially to a theory’s success (in particular its success in making novel predictions) and then limit your realism to those constituents. Anjan Chakravartty (2011, sec. 2.3) calls this approach explanationism, because it is concerned with explaining the success of scienti�c theories. I will discuss some aspects of it in section 6.2. Another important selective realist response to the pessimistic induction (which I will not discuss in detail) is structural realism, introduced into the contemporary debate by John Worrall (1989). The structural realist claims that the structural content of successful theories is retained in theory change, while other theory parts (e.g., what a theory says about the nature of things) are likely to change radically when one theory is superseded by another. We should therefore only be realists about the structural claims of scienti�c theories. Simple as this idea may appear, it has given rise to a host of di�erent forms of structural realism, some of them epistemic (claiming that all we know about is structure), others ontic (claiming that structure is, in some sense, all there is). For recent surveys of this rapidly developing �eld, see Ladyman (2014) and Frigg and Votsis (2011). Explanationism and structural realism have in common that their focus in the search for something which survives scienti�c revolutions is on theories. By contrast, the approach to be discussed in this chapter focuses on experiments. It was initiated by Ian Hacking (1983) and Nancy Cartwright (1983) and goes by the name of entity realism or experimental realism (I will use the two terms interchangeably). The central idea of this kind of selective realism is to be realist about the entities with which we interact in experiments but not about the theories which describe them. Such a realism is compatible with what the pessimistic induction purports to show, namely that even our best scienti�c theories may turn out to be false. In the brief presentation just given, the three realist responses to the pessimistic induction (explanationism, structural realism, and experimental realism) seem to be incompatible with each other. But this may no longer be so on closer

20 � Entity Realism inspection. For example, I will argue below that the most plausible reading of experimental realism relies on the causal explanation of experimental results, and this brings it close to explanationism. (It is precisely for this reason that I will have to check (in section 6.2) whether my enhanced version of experimental realism does not share the weaknesses of explanationism.) Similarly, although structural realism and entity realism are often portrayed as opposing views, Chakravartty (2007, 39–70) argues that it is actually the combination of the two which yields the most robust form of scienti�c realism.¹ The most widely discussed version of experimental realism is the one defended by Hacking. Section 2.1 will introduce this version and explore the connections to some related positions. Its centerpiece, Hacking’s argument from manipulability has been criticized for providing an inadequate criterion of reality, because being manipulable is, according to these critics, neither necessary nor su�cient for being real. I will address this �rst family of criticisms in section 2.2. Turning to a second family of criticisms, I will (in section 2.3) examine whether the experimental realist’s separation of entities from theories can be sustained. This inquiry will necessitate a shift of focus to Cartwright’s version of entity realism, which, as I will argue, is more defensible than Hacking’s.

�.� From Theories to Entities In the introduction to chapter 1, I used Rutherford’s belief in alpha particles to illustrate what scienti�c realism is. Objects like alpha particles, which cannot be observed by the unaided human senses, were traditionally called theoretical entities, but this term has now largely been replaced by the more appropriate name unobservable entities (van Fraassen 1980, 14). Hence, the Rutherford example depicted scienti�c realism as a commitment to the existence of (some) unobservable entities. On the other hand, the more precise account in section 1.1 repeatedly characterized scienti�c realism in terms of a commitment to the (approximate) truth of theories. I did not need to distinguish between these two characterizations, because they are usually taken to go together. The classic statement of this assumption is due to Wilfrid Sellars (1963, 91): “As I see it, to have good reason

1 However, as I will argue in sections 6.1 and 6.2, structural realism’s contribution to this combination, namely the notion of a “minimal interpretation” of (mathematical) structure, seems somewhat inert when it comes to defending realism against the pessimistic induction. It seems to me that what is really doing the work is the causal component contributed by entity realism.

From Theories to Entities

�

21

for holding a theory is ipso facto to have good reason for holding that the entities postulated by the theory exist.” The starting point for Hacking’s entity realism is the observation that these two commitments can come apart: “There are two kinds of scienti�c realism, one for theories, and one for entities” (1983, 26). Hacking then uses the example of the electron to show how realism of the second kind need not be accompanied by realism of the �rst kind: The scienti�c-entities version of [realism] says we have good reason to suppose that electrons exist, although no full-�edged description of electrons has any likelihood of being true. Our theories are constantly revised; for di�erent purposes we use di�erent and incompatible models of electrons which one does not think are literally true, but there are electrons, nonetheless. (27)

One of Hacking’s reasons to abstain from theory realism is obviously the pessimistic induction (“our theories are constantly revised”), another one has to do with the disunity of di�erent theoretical models of electrons. But how can he then assert that “there are electrons, nonetheless”? The reason is that he takes the strongest argument for realism about unobservable entities not to depend on realism about theories: Experimental work provides the strongest evidence for scienti�c realism. This is not because we test hypotheses about entities. It is because entities that in principle cannot be “observed” are regularly manipulated to produce new phenomena and to investigate other aspects of nature. They are tools, instruments not for thinking but for doing. (262)

The important contrast here is between thinking and doing (or, as the title of Hacking’s book expresses it, between representing and intervening). As long as only representation is concerned, even the most useful concepts (“tools for thinking”) can be interpreted non-realistically. But when it comes to intervention, Hacking’s suggestive idea is that something cannot be used as a tool unless it is real. And since, according to Hacking, successful intervention can occur in the absence of true representation, we can have entity realism without theory realism. I will further discuss this argument from manipulability in the next two sections, but I would now like to comment on some more general issues regarding Hacking’s entity realism. Let me �rst investigate how Hacking’s realism �ts into my discussion of di�erent dimensions of realism in section 1.1. The fact that entity realism is about the existence of unobservable entities has prompted some commentators to regard it as primarily or even exclusively metaphysical (Morrison 1990, 1; Elsamahi 1994, 174–175; Nola 2002, 1). This reading receives some initial support from a remark in the introduction to Hacking (1983), where Hacking claims that his book is not about logical or epistemological questions, but about

22 � Entity Realism metaphysical ones (1–2).² But the context of this remark shows that it is only a very speci�c cluster of epistemological questions which he wants to exclude from his investigation, namely the ones about the rationality of science raised by the writings of Thomas Kuhn (1962) and Paul Feyerabend (1975). When it comes to scienti�c realism, Hacking makes it very clear that he has no intention of separating claims about what we know from claims about reality. Interestingly, his reason for this is precisely the same as the one I gave in section 1.1 for rejecting a purely axiological realism: I run knowledge and reality together because the whole issue would be idle if we did not now have some entities that some of us think really do exist. If we were talking about some future scienti�c utopia I would withdraw from the discussion. (Hacking 1983, 28)

It is therefore not correct to say that Hacking’s realism privileges metaphysics at the expense of epistemology. (This may be di�erent for other entity realists; see below.) Mauricio Suárez (2008, 141) takes this one step further and attempts “to turn the presumed primacy of [metaphysical experimental realism] on its head, in order to defend experimental realism as only epistemology”. According to Suárez, the metaphysical experimental realist is making a mistake when he takes “x can be manipulated” to imply “x is real”. All it implies is that there is (a special kind of) warrant for “x is real”. These two conclusions are not equivalent, “for it is not part of the notion of warranted belief that warrant be infallible and the corresponding belief always true” (ibid.). Let us grant Suárez that warrant does not entail truth (although see Moon (2012) and references therein for arguments to the contrary). Would it not be strange for a realist to refrain from ever asserting a statement of the form “x is real”? Suárez agrees, and he consequently does not think of his position as a realism at all: “A more appropriate name for my views would be ‘the experimental attitude’, as among all contemporary epistemological views it is closer to the ‘neither-realism-nor-antirealism’ of Arthur Fine’s NOA” (139). I will express my dissatisfaction with this move in chapter 3. For the moment, I just want to comment on Suárez’s claim that the epistemic version of experimental realism does not provide grounds for its metaphysical version (161n8). I �nd this implausible, for reasons pointed out in section 1.1. The main arguments against scienti�c realism (including its metaphysical dimension) in the last three decades have been

2 Neither here nor in the chapter “What is scienti�c realism?” does Hacking explicitly mention the semantic dimension of scienti�c realism, although his book (in particular chap. 6) has some important things to say about semantics. Accordingly, the debate on the relative importance of the di�erent dimensions in Hacking’s realism has focused on the metaphysical and the epistemic dimension.

Is Manipulability an Adequate Criterion of Reality?

�

23

epistemic. Therefore, any successful defense of scienti�c realism nowadays has to address epistemology, and Hacking’s defense of realism is no exception. But this does not mean that such a defense only supports the epistemic dimension of realism. Rather, it supports scienti�c (entity) realism as a whole. To conclude this section, let me brie�y discuss the relations of Hacking’s realism with some other positions generally labeled as entity realisms. The most important connection, acknowledged in the very �rst sentence of Hacking (1983, vii), is to the ideas of Cartwright (1983). I will say more about this connection in section 2.3. Another author who expresses views very similar to Hacking’s is Ronald Giere (1988, chap. 5). The similarity is in fact so close that Giere feels obliged to emphasize that he developed his ideas before he saw Hacking’s book (288n2). The main di�erence between the two authors is probably that Hacking, as we saw above, disregards the issue of scienti�c rationality, while Giere is very much concerned with it (albeit in a highly critical attitude). Scienti�c realism therefore �gures less prominently in Giere’s book, which might explain why its in�uence on the development of experimental realism was small compared to the impact of Hacking’s work. (Giere’s work has had considerable in�uence in a di�erent part of the realism debate, concerning the role of scienti�c models. But this does not have much to do with experimental realism, since it is a matter of representation, not of intervention.) A di�erent strand of entity realism is defended by Michael Devitt (1997) and Brian Ellis (1987). We already saw in section 1.1 that Devitt prioritizes the metaphysical dimension of scienti�c realism. The same is true for Ellis, and this is the main di�erence between their realism and the one defended by Hacking and Cartwright. But there is also signi�cant overlap between these positions. In particular, both Ellis (1987, 56–58) and Devitt (1997, 111–113) endorse what I take to be a central idea of experimental realism (see section 2.3), namely that the success of science in some cases, but not in others, warrants belief in the truth of scienti�c claims, and that the di�erence between the two kinds of cases has to do with whether the claims are causal or not.

�.� Is Manipulability an Adequate Criterion of Reality? Many critics of entity realism have pointed out that it may have some counterintuitive consequences to tie the reality of hypothetical entities to our ability to manipulate them. On the one hand, there might be some real entities which we will never be able to manipulate. On the other hand, some entities which are taken to be manipulable may turn out not to exist. If this is so, then manipulability is

24 � Entity Realism neither a necessary nor a su�cient criterion of reality. I will argue that Hacking has an answer to the second of these issues, but that the �rst one necessitates a modi�cation of the criterion of manipulability.

Is It Necessary? In the �nal paragraph of his book, Hacking (1983, 275) confesses “a certain skepticism, about, say, black holes” and other “long lived theoretical entities, which don’t end up being manipulated”. Does he thereby advocate manipulability as a necessary criterion of reality? Not quite. The paragraph in question is preceded by the following statement: “The experimental argument for realism does not say that only experimenter’s objects exist” (ibid.). In a later paper, he elaborates: I did not say in my (1983) that an experimental argument is the only viable argument for scienti�c realism about unobserved entities. I said only that it is the most compelling one, and perhaps the only compelling one. (1989, 560–561)

Yet in that same paper, he restates his antirealism about entities which can neither be observed nor manipulated (e.g., gravitational lenses). It is therefore at least roughly correct to say that Hacking views manipulability as a necessary condition for realism about unobservable entities. The question then is whether such a restriction on realism is defensible. Antirealism about gravitational lenses may not be such an unpalatable consequence of experimental realism. However, this antirealism quickly spreads to many other scienti�c domains. Alan Gross (1990, 426) mentions “historical sciences like geology, cosmology, and evolutionary biology”. In these disciplines, “we must describe a series of events for which there are no witnesses, a series unavailable to experimentation. If Hacking’s criterion is applied, the events that evolutionary biology reconstructs will permanently lack reality” (427). This leaves the experimental realist with a dilemma: Either he bites the bullet and endorses antirealism about all these domains, thereby signi�cantly reducing the attractiveness of experimental realism, which initially promised to re�ect closely the commitments of scientists themselves. Or he accepts that manipulability is not necessary after all, and that there are other compelling arguments for realism. As we saw above, Hacking opts for the �rst horn of the dilemma as far as astrophysics is concerned, and Dudley Shapere is quick to point out that this leads to a con�ict between Hacking’s antirealism and the commitments implicit in the scienti�c enterprise:

Is Manipulability an Adequate Criterion of Reality? � 25 The fact is that scientists do build on what they have learned to make inferences even in cases where they cannot lay hands on the entities about which the inferences are made; they use what they have already found out—whether by interfering actively or passively observing—to probe further. Surely this is a most important aspect of the scienti�c enterprise. To disallow it is to truncate that enterprise for the sake of a contention which is unsupported by any real argument, scienti�c or otherwise—in short, a dogma. (Shapere 1993, 148)

In Shapere’s view, Hacking employs an over-restrictive reading of the term use when he restricts realism to entities which we can actively manipulate in order to interfere with something else. A more liberal (and more appropriate, according to Shapere) reading would extend realism to entities which we can use as tools of inquiry even though we cannot actively manipulate them. In this sense, we can, for example, use gravitational lenses for measuring cosmic distances, so we should be realists about them (146–147; see also Dorato 1988, 178–179). Hacking sometimes writes as if he could accept Shapere’s more liberal reading of use, particularly when he claims that we do not so much infer the existence of electrons from our manipulative success as presuppose their existence when we use them as tools of inquiry (Hacking 1983, 265). This presuppositional status of the entities in question is independent of whether we “use” them in Hacking’s narrow or in Shapere’s wide sense. While such a shift from manipulation to presupposition would save Hacking from the problems associated with the �rst horn of the above dilemma, it would considerably weaken his experimental argument for realism: Scientists have all kinds of presuppositions, and it is not clear why an inspection of them should be our best guide to reality.³ As I will argue in section 2.3, there is a more promising strategy to avoid the consequences of the �rst horn. The shift should not be from manipulation to presupposition, but from manipulation to causal explanation.

Is It Su�cient? While it is quite natural to question the necessity of Hacking’s criterion, it is not easy to see how manipulability could fail to be a su�cient condition for reality. The very notion of “manipulating something” seems to imply that this something must exist. Nevertheless, the su�ciency of the manipulability criterion has been questioned in two ways. The �rst one is mainly epistemic: If we knew that we are able to manipulate some entity x, we could indeed conclude that x is real, but in

3 In this context, Margaret Morrison (1990, sec. 4) speaks of a “transcendental turn” in the interpretation of Hacking’s argument, and she concludes (as I do) that this move is unsuccessful.

26 � Entity Realism fact we can at most believe that we have this ability. And this is not su�cient for the reality of x, for our belief might turn out to be false (Gross 1990). The second criticism is ontological: It claims that for some entities which can reasonably be regarded as manipulable, a rigorous ontological analysis shows that they should not be taken as real (Gelfert 2003). I think that the entity realist has a reply to these two criticisms. Let me address them in turn. Hacking (1983, 23) famously expresses his realism about electrons by means of the slogan “if you can spray them then they are real”. The phrase is catchy, but not unassailable. Gross (1990, 425) comments: “When we spray electrons, we say nothing criterial about their existence: we con�rm only the causal property we call spin”. In other words, manipulation may warrant belief in certain properties, but not a commitment to the entities supposed to possess these properties.� Accordingly, manipulability does not save entity realism from the pessimistic induction: When an old theory is discarded, its entities may be discarded along with it; if so, their causal properties, those still acknowledged by the new theory, will be redistributed in accordance with that theory. Some similar fate may well await the electron in some future science, its spin, mass, and charge reinterpreted and reassigned within a new ontology. (Gross 1990, 426)

Gross raises an important point here, namely that our knowledge of unobservable (causal) properties is more secure than our knowledge of unobservable entities. I will return to this in section 6.2, and limit myself here to a brief sketch of how Hacking can answer Gross’s objection. First, Hacking (1983, 271) admits that realism about entities does not immediately follow from realism about properties: Once upon a time it made good sense to doubt that there are electrons. Even after Thomson had measured the mass of his corpuscles, and Millikan their charge, doubt could have made sense. We needed to be sure that Millikan was measuring the same entity as Thomson.

But this does not mean that our realism must forever be con�ned to nothing but properties, as Hacking goes on to note: Things are di�erent now. The “direct” proof of electrons and the like is our ability to manipulate them using well-understood low-level causal properties. (274)

4 In the realism debate, “entities” usually refers to concrete particulars, such as objects, events, or processes, as opposed to abstract entities, such as numbers or properties. This is in contrast with a more general use of the term entity in contemporary metaphysics, according to which properties are themselves entities (see, e.g., Thomasson 2012, sec. 1.4).

Is Manipulability an Adequate Criterion of Reality? � 27

Unfortunately, Hacking is not very explicit about how manipulation takes us from property realism to entity realism. Based on the example he uses in this context (the polarizing electron gun PEGGY II), I take his idea to be the following: What is employed in building and operating a polarizing electron gun is not just the property of spin or the property of charge, but the systematic interplay of these properties, along with detailed knowledge of how they manifest themselves in various speci�c circumstances. With this in mind, Gross’s scenario of a future theory which may reinterpret and reassign spin, mass, and charge within a new ontology needs to be quali�ed: It will not be enough that such a new ontology includes the properties of spin, mass, and charge in some arbitrary way. Rather, it will have to respect the systematic coherence of these properties which we now associate with the concept of the electron. But this is to say that this new ontology will—at least in some sense—have to include electrons. The second argument against the su�ciency of manipulability as a criterion of reality is the following. According to Axel Gelfert (2003), the problem of the manipulability criterion is not that it yields realism about entities which may turn out to be unreal, but about entities which we already know to be unreal, namely the various kinds of quasi-particles in solid state physics. Examples include unoccupied electron states (“holes”) in semiconductors, excitons (bound electron-hole states), and magnons (coherent excitations of electron spins). All these entities can, in some sense, be manipulated and should therefore count as real by the experimental realist’s standards. Gelfert takes this to be an unacceptable consequence. Here is why: Since quasi-particles cannot exist independently of electrons, they would have to be de�ned as in some way parasitic on sets of electrons. However, no one set of electrons could be identi�ed as the material basis of a particular quasi-particle. (This is precisely the reason why quasi-particles can be ‘sprayed’ from one material sample to another without physically transforming one into the other!) Hence, the only solution would be to de�ne as the referent the total ensemble of ∼ ���� electrons. But this would be the very ensemble of particles that can support other quasi-particles (e.g. magnons in addition to excitons) as well. Hence, there would be a built-in mismatch between actual and intended referent—a mismatch that would be unavoidable in principle. (260)

Gelfert does not further explicate what he means by the “material basis” of a quasi-particle. His idea seems to be that since quasi-particles are “in some way parasitic on sets of electrons”, what we are really referring to when we speak of quasi-particles are sets of electrons. And since one and the same set of electrons can form the material basis of di�erent quasi-particles, there is a failure of reference for quasi-particle terms in the sense that terms purporting to refer to different entities actually refer to one and the same entity. I take Gelfert’s implicit

28 � Entity Realism conclusion to be that the failure of reference for quasi-particle terms implies the non-existence of quasi-particles. However, the experimental realist should resist the presupposition that we can only successfully refer to entities of which we can identify the “material basis”. Why not identify quasi-particles in the same way as electrons, namely via their causal properties? Of course, the identi�cation of particular individuals may not always be possible, but the same is true for electrons, as the long-standing debate on quantum non-individuality shows (French 2011a). In this context, I must concede that Gelfert’s critique of entity realism addresses an important point. As we will see in the next section, so-called “home truths” play an important role in Hacking’s realism, and they may lead to a mistaken description of the entities with which such a realism is concerned. For example, there is a danger of regarding it as a home truth that electrons are individual objects, which, as just remarked, might be inadequate. However, it is not entirely clear whether entity realism is really committed to the kind of home truths Gelfert attributes to it. Consider the following argument against the reality of quasi-particles: If one were to grant quasi-particles the same degree of reality as electrons, one would violate the very intuitions that lie at the heart of entity realism; namely, that there is a set of basic substantive entities that have priority over composite or derivative phenomena. A proliferation of entities would evolve into what one might call in�ationary realism. . . (Gelfert 2003, 257)

I do not see that it is “at the heart of entity realism” to think of electrons as “basic substantive entities”. Such an entity realism would indeed be problematic, as chapter 9 will show in detail. On the other hand, section 1.2 showed that a reasonable conception of scienti�c realism should accept the reality of non-fundamental entities, so quasi-particles may not be so objectionable after all. I thus agree with Brigitte Falkenburg’s (2007, 245–246) assessment of Gelfert’s critique: If it is a home truth that electrons exist as entities in their own right (i.e., in the sense of substances-on-their-own), and if this may be taken as a case against the existence of quasiparticles, all the worse for the philosopher’s home truths.

�.� From Manipulation to Explanation The experimental realist’s focus on entities as opposed to theories is sometimes taken to imply that such a realism is only committed to the existence of certain entities but not to the truth of any further statement about them, because that

From Manipulation to Explanation

� 29

would already require a commitment to theory. Such a crude entity realism invites an obvious line of criticism, which is most bitingly expressed by Alan Musgrave (1996, 20): To believe in an entity, while believing nothing further about that entity, is to believe nothing. I tell you that I believe in hobgoblins. . . . So, you reply, you think there are little people who creep into houses at night to do the housework. Oh no, say I, I do not believe that hobgoblins do that. Actually, I have no beliefs at all about what hobgoblins do or what they are like. I just believe in them.

Phrased in this way, the charge is overstated, since no entity realist has ever urged us “to believe in an entity, while believing nothing further about that entity”.� We already saw that Hacking unhesitatingly ascribes some causal properties to electrons. These property ascriptions are the “home truths” on which we rely when designing apparatus that use electrons to produce new phenomena (Hacking 1983, 265). What Hacking denies is that the conjunction of these home truths is part of a single theory. But might there not be a common core of theory, . . . which is the theory of the electron to which all the experimenters are realistically committed? I would say common lore, not common core. There are a lot of theories, models, approximations, pictures, formalisms, methods and so forth involving electrons, but there is no reason to suppose that the intersection of these is a theory at all. (264)

This raises the question what di�erentiates the “common lore” of home truths to which the experimental realist is committed from the kind of theories he wants to exclude from his realism. Many commentators have pointed out that Hacking does not have a satisfactory answer to this question and that, consequently, entity realism without theory realism is an unstable or even incoherent position (Morrison 1990; Elsamahi 1994; Psillos 1999, 256–258; Massimi 2004; Chakravartty 2007, 31– 33). I agree that Hacking’s answer is unconvincing, and in the rest of this section I will take a �rst step towards remedying this defect, thus preparing the ground for my own version of realism to be developed in part II. Hacking’s antirealism about theories is fueled partly by his distrust of inference to the best explanation (IBE) (Hacking 1983, 52–57). He considers the explana-

5 Devitt (1997, 22) may be an exception. For tactical reasons, he chooses to defend an entity realism which is merely committed to the existence of entities of certain kinds but not to the properties science ascribes to them. But when it comes to scienti�c explanation, even Devitt has to acknowledge that “phenomena are explained not by the mere existence of, say, electrons, but by electrons being the way our theory says they are” (109).

30 � Entity Realism tory power of a hypothesis “a feeble ground” (53) for belief that the hypothesis is true. This is the basis for his distinction between theoretical claims and the home truths of entity realism: The latter, unlike the former, can be warranted without relying on IBE. Some people might have had to believe in electrons because the postulation of their existence could explain a wide variety of phenomena. Luckily we no longer have to pretend to infer from explanatory success (i.e. from what makes our minds feel good). [The physicists who built PEGGY II] don’t explain phenomena with electrons. They know how to use them. (271–272)

But, as Resnik (1994) and Reiner and Pierson (1995) point out, it is unclear how successful manipulation should warrant belief in electrons (and the corresponding home truths) if not by means of some kind of IBE. At any rate, Hacking does not explain how this is supposed to work. A more informative account of how warrant accrues to claims about theoretical entities can be found in Cartwright’s (1983) version of experimental realism, to which I now turn. According to Cartwright, theory realism and entity realism both rely on IBE, but there is a crucial di�erence in the kinds of explanations that are invoked. In the case of theoretical explanations, the explanatory power resides in certain laws. In this case, Cartwright agrees with Hacking that IBE fails, because a law can be explanatory without being true. This is what she alludes to in entitling her book How the Laws of Physics Lie. But causal explanations work by postulating certain entities which are assumed to bring about the explanandum. These entities can only perform their explanatory role if they actually exist, hence IBE is valid in this case. Cartwright summarizes the relevant di�erence between the two kinds of explanation as follows: What is it about explanation that guarantees truth? I think there is no plausible answer to this question when one law explains another. But when we reason about theoretical entities the situation is di�erent. The reasoning is causal, and to accept the explanation is to admit the cause. (99)

Why is there such a di�erence in the validity of di�erent instances of IBE? Cartwright discusses several features of theoretical explanation which call into question the validity of IBE, and she claims that these features are not (or to a lesser degree) present in the case of causal explanation. The most general of these worries is that, if the explanans is related to the explanandum by nothing but a theoretical derivation, inferring the truth of the explanans commits the fallacy of a�rming the consequent: From “x explains y” and “y is true”, it does not follow that x is true. By contrast, the causal relation appealed to in a causal explanation guarantees that there is a sense in which y would not be the case unless x were the case

From Manipulation to Explanation

�

31

(Cartwright 1983, 91–93). Another problem of theoretical explanations is that in many cases, multiple mutually incompatible models can be used to explain a certain phenomenon. The same is not the case, according to Cartwright, for causal explanations (75–82). Finally, Cartwright emphasizes di�erent ways in which theoretical explanations do not proceed by strict deduction from fundamental laws but make use of approximation techniques which sometimes improve on the descriptive accuracy of the laws in question (103–127). Many authors have criticized Cartwright’s account of these features and their purported absence in causal explanations. I will turn to this discussion in chapter 4. But even if we assume for the moment that Cartwright’s distinction between causal and theoretical explanations is unproblematic, we still face the challenge of specifying what this entails in terms of the “home truths” to which the experimental realist is committed, as opposed to the theoretical claims he remains skeptical about. In other words: Which scienti�c claims can be warranted by a valid form of IBE, while others do not enjoy this kind of warrant? Where Hacking speaks of the “home truths” of experimental realism, Cartwright (1983, 8) mentions the “highly detailed causal principles and concrete phenomenological laws” to which we commit ourselves when we accept a causal explanation. But what exactly does this mean and how do these principles and laws di�er from the commitments of theory realism? Let us look at phenomenological laws �rst. The distinction between theoretical laws (e.g., Schrödinger’s equation) and phenomenological ones (e.g., Rydberg’s formula of the hydrogen spectrum)� plays an important role throughout Cartwright’s book, the basic idea being that fundamental laws are explanatory whereas phenomenological laws “describe what happens” (2). Accordingly, phenomenological laws can be con�rmed by inductive methods, while the only ground for fundamental laws is their ability to explain (88). One of Cartwright’s main tenets (and her prime argument against IBE) is that “the cost of explanatory power is descriptive adequacy” (3). As mentioned above, large parts of How the Laws of Physics Lie are devoted to showing how our explanatory practices call into question the truth of the laws we use in our explanations. In other words, the explanatory power of theoretical laws speaks against their truth. Conversely, phenomenological laws can be regarded as (approximately) true, but they do not explain.�

6 The fact that the Rydberg formula describes electromagnetic radiation of a wide range of frequencies exceeding the visible spectrum illustrates that phenomenological here has nothing to do with observable (cf. Cartwright 1983, 1–2). 7 One might think that phenomenological laws, by virtue of being laws, can (in a deductivenomological or some related sense) explain concrete experimental data. But, as Bogen and Wood-

32 � Entity Realism This shows, however, that phenomenological laws cannot be that which makes causal explanations explanatory. Another element must be responsible for this, presumably Cartwright’s “causal principles”. Unfortunately, their characterization is much more problematic than the characterization of phenomenological laws. How are we to distinguish the explanatory laws, which I argue are not to be taken literally, from the causal claims and more pedestrian statements of fact, which are? The short answer is that there is no way. (Cartwright 1983, 77) Some claims of the theory must be literally descriptive (I think the claims about the mass and charge of the electron are a good example) if the theory is to be brought to bear on the phenomena; but I suspect that there is no general independent way of characterizing which these will be. (78)

I suppose that the reason for Cartwright’s pessimism in this respect has to do with the fact that we do not have a satisfactory theory of causal explanation (161–162). Indeed, it would be most desirable to base the sought-after distinction between causal principles and theoretical laws on a theory of explanation which unambiguously tells us what makes causal claims explanatory. Now there has certainly been some progress in the research on causal explanation since Cartwright voiced her pessimism over three decades ago (e.g., Salmon 1998, Woodward 2003), but our theories of explanation are still plagued by many problems. And in the absence of a satisfactory account of causal explanation, it might seem that Cartwright’s characterization of causal principles is no less problematic than Hacking’s characterization of home truths. Nevertheless, I am more optimistic than Cartwright about our ability to circumscribe the domain of literally descriptive (and well-warranted) claims, because I do not think that a fully worked-out theory of causal explanation is necessary to do this job. For our purposes, the importance of Cartwright’s distinction between causal and theoretical explanation lies in those above-mentioned features of explanations which prevent the corresponding instances of IBE from yielding well-warranted results. So even if we are unable to fully specify what di�erentiates causal from theoretical explanations, we may still, by focusing directly on these features, be able to formulate a clear distinction between two kinds of warrant, generated by di�erent instances of IBE. I will take up this task in chapter 4, where I will claim that this distinction can sustain a defensible form of selective realism.

ward (1988) argue, data are rarely explainable in this way. Typically, if explanations of data can be given at all, they will be what Bogen and Woodward (322n17) call “singular causal explanations”.

3 NOA and the Vices of the Realism Debate The general idea of selective realism, as sketched at the beginning of the previous chapter, was to take into account the arguments against realism, while still defending realism as far as it goes. In the words of Worrall (1989), this strategy sought to bring together “the best of both worlds”. In the present chapter, I will consider an approach which could be said to focus on the worst of both worlds. Its main proponent, Arthur Fine, is (at least prima facie) not interested in combining the virtues of realism and antirealism, but rather in pointing out their respective vices, in order to argue for a complete abandonment of the very debate on scienti�c realism. As an alternative, he promotes his so-called natural ontological attitude (NOA). I will discuss two central aspects of Fine’s critique of the realism debate. The �rst is his contention that both realists and antirealists are mistaken in believing that science needs an interpretation (section 3.1). The second is Fine’s claim that both sides in the debate have an equal right to accuse the other side of relying on an unjusti�ed discrimination between observable and unobservable entities (section 3.2). In the �nal section 3.3, I will comment on the di�erence between NOA and entity realism.

�.� In Defense of Interpretation The central shortcoming of the realism debate, as Fine sees it, is that realism and antirealism,¹ despite their disagreements, essentially commit the same sins. One of them is an unjusti�ed hermeneutical attitude towards science: What binds realism and antirealism together is this. They 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. (Fine 1996, 147–148)

1 Though Fine discusses various kinds of antirealism, the passages I will focus on are almost exclusively concerned with instrumentalism. However, Fine uses this term in a broader sense than the one I introduced in section 1.1; while I associated instrumentalism with what I called “type 2a antirealism”, instrumentalism in Fine’s sense includes option 2b. In particular, Fine takes van Fraassen’s constructive empiricism to be an “attractive and strong” variant of instrumentalism (Fine 1986, 156). I will for this chapter adopt Fine’s terminology and regard van Fraassen as a paradigm instrumentalist.

34 � NOA and the Vices of the Realism Debate How to Interpret Fine’s No-interpretation View? The claim that science does not need an interpretation can be understood in several ways. I will consider them in turn, starting with what Fine (1986, 176) has to say about the interpretation of single scienti�c statements: This point is well understood in the philosophy of mathematics. For example, we all believe that there is one and only one even prime number (the number 2). That belief, however, scarcely makes number-realists (much less platonists) of us all. The point is that realism requires two distinct elements. It requires belief and it also requires a particular interpretation of that belief. Thus anti-realism, in particular instrumentalism, pursues the following strategy. If it does not withhold belief, then it o�ers instead a non-realist interpretation of that belief . . . But the reader will no doubt notice that there is an interesting third way. For one can go along with belief, but then simply not add on any special interpretation of it—neither realist nor anti-realist. That is the way of NOA.

What does it mean not to “add on any special interpretation” to a belief? In the light of Fine’s anti-hermeneutical stance sketched above, one might be tempted to read this as an advice to keep one’s beliefs free from any interpretation whatsoever. Musgrave (1989, 398) takes this position into consideration and points out the absurdity it leads to: The NOA is supposed to “accept”, say, “There is a full moon tonight” and “Electrons are negatively charged.” But does the NOA know what he has accepted? Remember, he leaves it open whether or not these statements are to be taken at face value, says nothing about how they are to be interpreted, what their ontological commitments are. The unphilosophical NOA does not just know nothing philosophical—he knows nothing at all.

Moreover, Fine’s own words do not encourage this reading. For he does “not suggest that science is hermeneutic-proof, but rather that in science, as elsewhere, hermeneutical understanding has to be gained from the inside. It should not be prefabricated to meet external, philosophical speci�cations” (Fine 1996, 148n).² So Fine would not prohibit all interpretations of scienti�c beliefs, but only all the “special interpretations”, be they realist or anti-realist. In Fine’s view, what makes an interpretation “special” (and therefore unwelcome) is the theory of truth

2 It is somewhat surprising that Fine should draw such a internal/external dichotomy, because this seems to presuppose just the kind of demarcation (between what belongs to science and what does not) that NOA explicitly opposes: “NOA places a strong emphasis on the openness of science. NOA is fundamentally anti-essentialist . . . with respect to the character of the scienti�c disciplines (which it regards as contingent, historical entities—not as things with ‘natures’ that can be separated from other human activities by ‘demarcation criteria’)” (Fine 1996, 174).

In Defense of Interpretation �

35

with which it is connected. In the case of realist interpretations, this is a correspondence theory, while antirealist interpretations are usually attached to some pragmatic or epistemic theory of truth. To Fine, all these theories are gratuitous additions to science. His NOA displays a “stubborn refusal to amplify the concept of truth by providing a theory or analysis (or even a metaphorical picture). Rather, NOA recognizes in ‘truth’ a concept already in use and agrees to abide by the standard rules of usage” (1996, 133). This no-theory conception of truth accounts for the rather puzzling fact that Fine repeatedly confesses his belief in electrons and the like, but denounces the realist’s claim that electrons really exist in the world (e.g., Fine 1996, 184). One can interpret this as attributing truth (in NOA’s minimal sense) but not correspondence-truth to the claim that there are electrons. There are two problems with this view. First, it may well be that the concept of truth “already in use” has a much more realistic character than Fine is ready to acknowledge. This point is forcefully made by Musgrave (1989, 387), who demonstrates that what Fine sees as “the standard rules of usage” for the term true (in particular what he calls “a Davidsonian-Tarskian, referential semantics”) provides realists with “as much of a correspondence theory as they need.”³ But let us suppose that Fine’s no-theory view of truth makes for an interpretation of scienti�c claims that is neither realist nor antirealist. This is where the second problem becomes virulent. By adopting this interpretative stance (thereby banning all the realist and antirealist alternatives), NOA makes a global hermeneutical commitment, of which one may wonder if it is not “prefabricated to meet external, philosophical speci�cations”. Dan McArthur (2003, 84) shares this worry: It is simply not the case that any one single epistemological stance can characterise the diverse enterprises that fall under the rubric of science. This situation is somewhat ironic. One

3 By contrast, Psillos (1999, 244) thinks that this minimal (or, as he calls it, de�ationist) sense of correspondence is not enough for realism. In particular, he sees realism as relying on “a substantive theory of reference, most typically on the Kripke-Putnam causal theory”, which is not accepted by de�ationists. But since Fine does not seem to take a stand in the debate on reference, Psillos reaches a conclusion similar to Musgrave’s: If Fine’s suggestion that we “treat truth in the usual referential way so that a sentence (or a statement) is true just in case the entities referred to stand in the referred-to relation” (Fine 1996, 130) allows a substantive (causal) theory of reference to �x the semantic values of the linguistic items, then his suggestion is fully consistent with a thick realist “correspondence” conception of truth. (Psillos 1999, 245)

36 � NOA and the Vices of the Realism Debate of NOA’s central appealing features is supposed to be its “de�ationary” stance, its refusal to adhere to any global philosophical position. However, in attempting to ban philosophical content from science, NOA fails to do justice to the diversity of science that it purports to characterise.

I hesitate to side with McArthur in attributing to NOA such a global monolithic attitude, as there are many passages that testify to NOA’s attitudinal �exibility in view of the diversity of the scienti�c enterprise. These suggest to understand Fine’s no-interpretation view in yet another way: not as directed against interpretation in general, nor against certain interpretations of single statements, but against global interpretations of science insofar as they forestall an unprejudiced approach to the single case. One of the passages that support this reading is the following: Thus NOA, as such, has no speci�c 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 commitment that emerges. Di�erent NOAers could, therefore, disagree about what exists, just as di�erent, knowledgeable scientists disagree. (Fine 1986, 176)

Understood in this way, NOA is not so much an independent alternative to the realism/instrumentalism dichotomy, but an invitation to choose between realism and instrumentalism (and possibly some other attitudes) on a case-by-case basis; for some scienti�c theories, the appropriate attitude may be realism, for others, it may be instrumentalism. I immediately admit that this is most probably not the interpretation intended by Fine, for it is sharply at odds with Fine’s fundamental criticism of realism and instrumentalism and with his intention to formulate an independent alternative to these traditional views. However, given the severe di�culties of the interpretations discussed above, it is the best I can do to make sense of Fine’s anti-hermeneutical stance. Furthermore, I consider the localism expressed in this interpretation a healthy proposal, and one that even the most determined defenders of scienti�c realism have now taken to heart, for example Psillos (1999, 161): Scienti�c realism is not an all-or-nothing doctrine, in the sense that one must either believe to an equal degree everything a scienti�c theory predicates of the world or else believe in nothing but (perhaps) observable phenomena. If scienti�c realism is to be plausible and, as most realists would urge, in agreement with actual scienti�c practice, then it must go for di�erentiated commitments to scienti�c theories, and what they entail about the world, in accordance with the evidence which supports them, as a whole and in parts.

In Defense of Interpretation �

37

In fact, the di�erentiation Psillos advocates here is even more �ne-grained than the one suggested by the above reading of NOA. It is not just that we should adopt realism with respect to some theories and instrumentalism with respect to others. Rather, we should di�erentiate our attitudes even within a given theory. In classical electrodynamics, for example, it is common to take a realistic stance towards the electric and magnetic �elds E and B, but not towards the corresponding potentials ϕ and A. It seems to me that Fine’s recommendation “to look and see as openly as one can what it is reasonable to believe in” should lead him to endorse this localism on the intratheoretical level along with the above-mentioned localism on the intertheoretical level, but nothing hinges on this supposition. What is important is that a realist who accepts Psillos’s intratheoretical localism ipso facto also accepts intertheoretical localism: If realism cannot be applied across the board within one theory, still less can it be applied across the board of all theories. The di�erent versions of selective realism mentioned at the beginning of chapter 2 also re�ect this movement away from a global realism towards a more localized treatment of the issue. Insofar as NOA is understood primarily as a criticism of the old, global versions of realism and instrumentalism, the question arises how it relates to the newer versions of these doctrines. With regard to entity realism, this question will be addressed in section 3.3.

How Interpretation May A�ect the Practice of Science In order to show what a variety of attitudes (ranging from complete agnosticism to strong belief) di�erent scientists can adopt towards a certain entity, Fine (1996, 150) cites the controversy about the existence of magnetic monopoles. I doubt that these di�erences in attitude have much to do with the scienti�c realism debate. They are di�erences about which hypotheses to accept (which is a question of physics), while the realism debate is concerned with what accepting a hypothesis amounts to (e.g., in terms of ontology). Fine, however, does not see it as a shortcoming, but as a virtue that his NOA does not have much to say on this latter issue. The reason is the following: After all, there is no reason to think that science as a whole, or inquiry more generally, has su�ered because the special issue, addressed by realism or instrumentalism, of disambiguating scienti�c acceptance has gone unanswered. To put it di�erently, the ambiguity over the character of acceptance in science that results from not raising the realism/instrumentalism question seems to be an ambiguity we can quite well live with. . . . The point is that we build models and theories (“frame hypotheses”), and act on them

38 � NOA and the Vices of the Realism Debate without necessarily settling or even addressing the interpretive questions that realism or instrumentalism raises. (Fine 1991, 93–94)

As we witnessed above, Fine’s refusal to go with any particular theory of truth led him into a slide towards realism, because the concept of truth “already in use” is arguably realist in character. Now we may wonder if the refusal to say anything about the character of acceptance will result in a similar slide. It seems to me that it will, but in the opposite direction; the concept of acceptance “already in use” will push NOA towards instrumentalism. Fine himself reminds us of what we do with our accepted models and theories even if we do not address any interpretive questions: we “act on them”, that is, we make use of them. So if “we can quite well live with” the ambiguity over the character of acceptance, this is only because we already have an understanding of the term, which happens to be an instrumentalist one. To declare any further disambiguation irrelevant (as Fine does) is not to remain neutral between instrumentalism and realism, it is to stick to instrumentalism. Neither this slide towards instrumentalism nor the slide towards realism discussed before are grave objections to NOA, if it is understood in the localist sense described above. That NOA sides with instrumentalism in some cases and with realism in others is then just what we should expect. This merely reinforces the lesson that NOA is in fact a mixture of local realisms and instrumentalisms (and possibly some other stances in yet other cases), rather than the completely independent alternative view as which Fine sometimes advertises it. But let us postpone further quarrel about philosophical labels to section 3.3 and turn to Fine’s contention that science is not a�ected by our pursuing (or not pursuing) interpretive questions. Even if this were true, would it mean that these questions should therefore not be addressed? I am not convinced that philosophy of science is only worthwhile insofar as it a�ects the practice of science.� Philosophy of science should, of course, be informed by scienti�c practice and it should be able to make sense of that practice, but it could very well meet these requirements without in turn in�uencing the practice of science. Fine’s observation that science as a whole has not su�ered from leaving the realism/instrumentalism question unanswered would then be an interesting fact, but not a reason to ban this question from philosophy of science.

4 It is even a fairly common view among philosophers of science that their discipline is in general not intended to alter scienti�c practice. As an example, consider the following response to Fine (1991) by Ernan McMullin (1991, 106–107): “If we were to adopt the maxim that questions ought not be asked about science . . . unless an answer to [them] is required for the practice of science itself, philosophy of science (so far as I can see) would have to be set aside as super�uous.”

In Defense of Interpretation

�

39

Be that as it may, I think that there is at least one way in which the realism issue does a�ect the practice of science, and I end this section by brie�y digressing into that. Fine himself acknowledges the relevance of scienti�c realism for the advancement of science in the following sense: This vision [of realism] was certainly a noble and romantic one, and deliberately so, for realism was used to defend the autonomy of science, and to attract political and economic support for it (not to mention practitioners too). It still is. (Fine 1991, 83)

On the other hand, Fine claims that “NOA’s practice would not alter the fate of science”, because NOA is, like realism, a “pro science attitude” (1991, 94).� And even if my above conjecture about NOA’s slide towards instrumentalism is right, this need not bother Fine, because instrumentalism, too, is a “pro attitude toward science” (93–94). But is it really? True, instrumentalism and NOA “tend to regard what science accepts as reasonable” (94), but, in order not to smuggle any realist connotations into the term “reasonable”, one had better replace it by “useful”. Now for many parts of science, this kind of pro attitude is enough to attract the support they need. But if we look at areas of basic research remote from any prospect of practical application, their “usefulness” gets a disturbingly circular �avor. Consider, as an example, one of the central endeavors of recent research in particle physics, the search for the Higgs boson. Even before CERN’s announcement of the discovery of a Higgs-like particle in July 2012, most physicists believed that there is such a particle. The instrumentalist may describe this situation by saying that the Higgs mechanism (which, among other things, predicts the existence of the Higgs particle) is an immensely useful tool within the standard model of particle physics, and it is therefore useful to believe in Higgs particles, since otherwise many features of the standard model could not be consistently accounted for. So far so good. But why then is it that such enormous e�ort was put into the experimental search for the Higgs particle? Of course, now that the search has met with success, we have thereby found one more task for which the Higgs hypothesis proves useful, namely explaining the data that result from this search. But I doubt that this is enough to motivate such a time-consuming and expensive activity. For all we have found out (on the instrumentalist’s view) is that the hypothesis is useful for predicting the behavior of a complicated machine (the particle detector) which we could just as well have decided not to build in the �rst place.� By

5 Cf. Fine (1996, 173): “The contrast here is with antiscience attitudes, ranging from general skepticism to the derisive contempt toward science that one sometimes encounters in the humanities.” 6 Franklin (1996, sec. 2b) o�ers the same kind of argument against van Fraassen’s constructive empiricism.

40 � NOA and the Vices of the Realism Debate contrast, the realist has a clear motivation to build such a machine, because he is con�dent that the experiments enable us to determine whether the Higgs boson is only a useful theoretical tool or a real entity in the world. Based on the localist reading of NOA mentioned above, Fine could just reply that particle physics is a discipline for which NOA (locally) shares the realist’s commitment, in accordance with the commitments of the particle physicists themselves. I am not sure that this is really the kind of view that Fine would want to endorse, given his opposition to even a modest kind of realism like entity realism (see section 3.3). But if he did, this would mean that NOA is not at all opposed to the project I am pursuing, namely to defend realism in the context of particle physics.

�.� The Principle of Fairness The purported inadequacy of the hermeneutical attitudes of realism and instrumentalism is not the only thing leading Fine to question the soundness of the realism debate itself. Another vice he detects in the debate is a certain symmetry in the arguments of the realist and the instrumentalist, creating the impression that neither of the two sides has any real prospect of winning the debate. One of the most signi�cant passages in this respect is the following: In the realm of the observable, reliability is a mark of truth (in the realist sense of external world correspondence). So, urges the realist, when we come to reliability with respect to hypotheses concerning unobservables we are entitled to mark those as true (realist style) as well. The intuition is to require epistemological parity between observables and unobservables. This is a principle of fairness, and from the realist point of view the instrumentalist violates it and is not fair to unobservables. The instrumentalist, however, may well grant the principle of fairness and, turning the table, charge the realist with violating it. For the instrumentalist sees the aim of inquiry as the same for both observables and unobservables, namely, reliability. . . . Truth, in the realist sense of correspondence with reality, discriminates epistemologically between the observable and the unobservable. . . . Thus, the instrumentalist �nds that the realist makes a mistake when, noting the rough co-variance between reliability and correspondence truth among observables, the realist projects truth (rather than reliability) as the aim and focus of inference for unobservables. For only reliability treats observables on par with unobservables from an epistemological point of view. Thus, adhering to the principle of fairness, the instrumentalist �nds the realist unfair to unobservables. (Fine 1996, 182–183)

So it seems that the debate has come to a stand-o�, the realist and the instrumentalist charging each other with unfairness: a fruitless symmetry. If this were indeed the case, we would be well advised to join NOA in turning away from this

The Principle of Fairness

� 41

unedifying battleground. My contention, however, is that the situation is much less symmetric than Fine makes it look, and that once his principle of fairness is acknowledged, the realist is in a much better position than his opponent. Since there are di�erent ways of stating what the principle of fairness (henceforth POF) exactly says, my argument against Fine will proceed in two stages. First, I will adopt a rather crude and very general reading of the POF, and show that, while the realist and the instrumentalist can both use this same principle to defend their positions, there is an asymmetry in the way in which each of them can respond to his opponent’s use of the POF. I will then move on to a more nuanced understanding of the POF, and show that the asymmetry between realism and instrumentalism with respect to the crude POF carries over to a similar asymmetry with respect to the enhanced POF. As a preliminary remark, note that the instrumentalist in the passage quoted above is not one of the “truthmongers” as described in chapter 8 of Fine (1996): Despite his pragmatic attitude towards science, this instrumentalist does not de�ne truth in terms of instrumental success or reliability, but accepts the realist’s conception of truth and unhesitatingly applies it to propositions about observables (though, unlike the realist, he does not see truth as the aim of scienti�c inquiry). Accordingly, there is (at this point at least) no ambiguity in the term true. The realist can now bring forward his argument in the following way: (1) (POF) (2)

There is warranted inference from reliability to truth in the observable realm. The observable and the unobservable realm should not be treated di�erently. There is warranted inference from reliability to truth in the unobservable realm.

It is easy to recognize this argument as an instantiation of the general realist strategy I described in section 1.1: Premise (1) is one way of expressing what fuels common sense realism, (POF) removes a possible obstacle to the legitimacy of extrapolating realism from common sense to science, thereby paving the way to conclusion (2), which is a key tenet of scienti�c realism. The instrumentalist can certainly not accept conclusion (2), so he must reject either premise (1) or (POF). For concreteness, let us look at a famous story told by van Fraassen (1980, 19–20), which suggests that his kind of instrumentalism has to accept premise (1): I hear scratching in the wall, the patter of little feet at midnight, my cheese disappears—and I infer that a mouse has come to live with me. Not merely that these apparent signs of mously presence will continue, not merely that all the observable phenomena will be as if there is a mouse; but that there really is a mouse.

42 � NOA and the Vices of the Realism Debate The context of this example is van Fraassen’s critique of inference to the best explanation (IBE). The mouse example, van Fraassen holds, needs not be seen as an instance of IBE. Since the mouse is an observable object, it is not clear that the truth, rather than the empirical adequacy of the best explanation is inferred here. But the di�erence between IBE and the more limited inference to the empirical adequacy of the best explanation (IEABE) does not matter for the present discussion, because acceptance of IEABE is enough to warrant premise (1). More recently, Ladyman et al. (1997) have argued that van Fraassen should be interpreted as rejecting abductive reasoning in toto, whether in the form of IBE or IEABE. But, as Psillos (1999, 213–215) shows, this move would let in much more skepticism about unobserved but observable entities than the constructive empiricist should be willing to accept. It seems, then, that Fine’s instrumentalist has to accept premise (1). This implies that his only chance to keep away from the realist conclusion (2) is to reject (POF). Therefore, Fine’s claim that the instrumentalist “may well grant the principle of fairness” is not correct. But this does not invalidate the second part of Fine’s argument, namely that the instrumentalist may in turn charge the realist with violating the POF. That is, the instrumentalist can use the POF for his own purpose, as follows: (3) (4) (POF) (5)

Aiming at truth discriminates between the observable and the unobservable. Aiming at reliability treats observables on par with unobservables. The observable and the unobservable realm should not be treated di�erently. Aiming at truth should be abandoned in favor of aiming at reliability.

Surely the realist is as opposed to conclusion (5) as the instrumentalist was to conclusion (2). But unlike the instrumentalist, the realist is not forced to reject the POF in order to avoid the unwelcome conclusion (5). Instead, he can (and should) reject premise (3). The above discussion of IEABE allows us to get a clear grasp of what the instrumentalist’s rationale for premise (3) is: Truth claims about observables can be warranted by IEABE, which the instrumentalist accepts as a valid form of inference. But truth claims about unobservables need something beyond IEABE, and the instrumentalist denies that there is any warrant-generating form of inference in that realm. Therefore, the instrumentalist sees the quest for truth as discriminating between observables and unobservables. However, the realist �nds the very distinction between IBE and IEABE dubious, because it places an

The Principle of Fairness

� 43

undue epistemic weight on the limits of human perception. He therefore has no reason to accept premise (3). This shows that, contrary to Fine’s view, the realist and the instrumentalist are not equally justi�ed in charging their opponent with unfairness. While the realist can indeed grant the POF even as a premise in the instrumentalist’s argument (because he can attack another premise), the instrumentalist needs to reject the realist’s use of the POF in order to escape the realist’s conclusion. In other words, the principle of fairness is not collective property of the realist and the instrumentalist (which would also make it NOA’s property), but it is squarely part of the realist’s arsenal. One might object here that my reading of the POF is inadequate. Maybe Fine’s reference to “epistemological parity between observables and unobservables” just amounts to something like this: “Epistemic practices that warrant reliability among observables also warrant reliability among unobservables”. Let me call this (POFi ). Although I do not quite see how something like (5) can be derived on the basis of (POFi ), I grant that this is a plausible reading of the POF in the context of the instrumentalist’s argument. But what about the realist’s use of the POF? According to Fine, the POF underlies the realist’s thought that “when we come to reliability with respect to hypotheses concerning unobservables we are entitled to mark those as true (realist style) as well”. But surely (POFi ) is irrelevant for this entitlement, because it is explicitly restricted to reliability. So the realist seems to need some kind of (POFr ) which di�ers from (POFi ), presumably by replacing “reliability” with “truth”. This would render Fine’s talk of the principle of fairness rather misleading. Now this ambiguity with respect to “the” POF may not a�ect Fine’s main argument. Even if there are actually two POFs instead of one, the symmetry which Fine postulates between realism and instrumentalism may still obtain in the sense that the realist can accuse the instrumentalist of violating (POFr ), while the instrumentalist can accuse the realist of violating (POFi ). However, this is not the case. Even if we assume that the instrumentalist argument for (5) still goes through if (POF) is replaced by (POFi ), the argument will still need to include a premise similar to (3), which the realist can reject. Therefore, the realist can grant the instrumentalist his version of the POF. The converse is not true, because regardless of whether (POF) or (POFr ) is used in the realist argument for (2), rejecting this premise is still the only option for the instrumentalist if he wants to avoid the conclusion (2). In sum, there is still an asymmetry between the realist and the instrumentalist in that the former can accept (POFi ) while the latter cannot accept (POFr ).

44 � NOA and the Vices of the Realism Debate

�.� NOA, Entity Realism, and the Homely Line Section 3.1 showed that the most plausible reading of NOA is as a localist attitude which advocates realism for some domains and instrumentalism for others. But Fine would probably want to resist the characterization of NOA as just another form of selective realism. Rather, he would see things the other way round, claiming that realism is coming closer and closer to NOA: In fact, there has been a rather dramatic shift in focus. Although there are many who still �y a realist �ag, the doctrines they defend are seldom those of grand old realism. Rather, the dominant trend today is to abandon the old project of providing an argument for realism in general and to pick and choose where to place the �ag. . . . Just as NOA urged, realism has gone piecemeal, and that is an advance. (Fine 1996, 181)

And, more speci�cally addressing the kind of selective realism I discussed in chapter 2, Fine writes: It may well be that “entity realism” is just a special version of NOA. That depends on how the truth of existence claims is to be understood (or not). I suspect that Hacking’s (1983) version is like NOA, and that the realist label is just his way of trying to redirect philosophical discussion by giving the new direction an old and honoured name. (Fine 1986, 176)

This resonates with Suárez’s proposal (mentioned in section 2.1) to transform entity realism into a mere “experimental attitude”, which would then be part of NOA. But it seems to me that there is more realism in Hacking’s position than Fine and Suárez are ready to acknowledge. Fine is certainly right in not attributing to Hacking a correspondence theory of truth. In that respect, Hacking’s view “is like NOA”. But if it was already di�cult to tell apart Fine’s “referential semantics” from what a realist needs by way of correspondence (see section 3.1), this is even more so with Hacking. For, unlike Fine, Hacking (1983, 104–108) does take a stand on reference, and though he may not subscribe to a full-blown causal theory of reference of the type that Psillos deems necessary for realism (see footnote 3 above), he strongly emphasizes the importance of causal interaction for the �xing of reference, primarily with respect to everyday objects (such as cats and cherries), but also with respect to scienti�c entities (such as muons and mesons). This shows that, insofar as there is a gap between realism and NOA, Hacking is closer to the realism side of the gap than to the NOA side. The emphasis on causal interaction is linked to another feature that di�erentiates entity realism from NOA: Entity realism tries to establish a criterion for when IBE is warranted and when it is not, by exploiting the di�erence between causal and non-causal reasoning. As Fine (1991, 84) puts it: “what grounds entity

NOA, Entity Realism, and the Homely Line

�

45

realism is a causalism that picks out causes from among the bag of theoretical entities, and holds them up as special”. He then goes on to denounce such picking, because “like any other, a causal story can be reliable without being true, hence without the cause being real” (86). Unfortunately, Fine does not give an argument for this claim. (There are other authors who do, and I will discuss them in section 4.4.) Instead, he presupposes it as part of an instrumentalist argument against entity realism, similar to the one we already encountered in section 3.2. Just as he did there, Fine promises to present an instrumentalism which “treats all entities (observable or not) perfectly on par” in that “according to instrumentalism what we want from our theories, posits, ideas etc. in all the various contexts of inquiry is instrumental reliability” (ibid.). But this by itself would not allow the generation of any knowledge about the external world: I can never know whether there is a notebook on my desk, because the best any inquiry can give me is a reliable hypothesis that there is a notebook on my desk. Fine clearly sees the implausibility of this and is quick to amend the standpoint of his instrumentalist: Of course if the cause happens to be observable, then the reliability of the story leads me to expect to observe it (other things being equal). If I make the observation, I then have independent grounds for thinking the cause to be real. If I do not make the observation or if the cause is not observable, then my commitment is just to the reliability of the causal story, and not to the reality of the cause. (Fine 1991, 86)

But with this addition, it is no longer the case that all entities (observable or not) are treated “perfectly on par”. Just as we have seen in section 3.2, instrumentalism and the principle of fairness do not go together very well. Now there may be some philosophers who do not mind (POF) being swept overboard, but Fine is not among them. Therefore, after having sent instrumentalism into a battle against entity realism, he suddenly thinks “we might pause to wonder whether it is really important to declare our allegiance in this matter either to realism or to instrumentalism, either in general or in particular” (93). Well and good, but in the absence of instrumentalism, Fine lacks any argument against entity realism. This by itself is no reason for preferring entity realism to NOA. But I will now argue that entity realism is in fact superior to NOA, because the latter can only establish its own position by tacitly relying on the argumentative work of the former. While the two of them agree that the global explanationist defense of scienti�c realism fails, they di�er in their reaction to this diagnosis: Entity realism, seeing the need to defend at least some forms of abductive reasoning, seeks to de�ne the limits within which such reasoning is still warranted. NOA, distrustful of demarcations, completely retreats from the realist game and contents itself with the following “homely line”:

46 � NOA and the Vices of the Realism Debate I certainly trust the evidence of my senses, on the whole, with regard to the existence and features of everyday objects. And I have similar con�dence in the system of “check, doublecheck, check, triple-check” of scienti�c investigation, as well as the other safeguards built into the institutions of science. . . . When the homely line asks us, then, to accept the scienti�c results “in the same way” in which we accept the evidence of our senses, I take it that we are to accept them both as true. (Fine 1996, 126–127)

It is of course quite probable that if we were to justify our “con�dence in the system of . . . scienti�c investigation”, we would (at least partly) have to rely on the abductive strategies that NOA wants to stay away from. But Fine does not mind that, because the good thing about the homely line, as he sees it, is that there is no need to argue for it, since it is not disputed by anyone: Let me use the term “antirealist” to refer to any of the many di�erent speci�c enemies of realism: the idealist, the instrumentalist, the phenomenalist, the empiricist (constructive or not), the conventionalist, the constructivist, the pragmatist, and so forth. Then, it seems to me that both the realist and the antirealist must toe what I have been calling “the homely line.” That is, they must both accept the certi�ed results of science as on par with more homely and familiarly supported claims. (Fine 1996, 128)

Already my discussion in section 1.1 showed that this is not the way I see the realism debate. If we start with common sense realism (as antirealists like van Fraassen do), then the whole point of antirealism is to deny that scienti�c results should be accepted “on par with more homely and familiarly supported claims.” It even seems to me that this has been the core of antirealism ever since Osiander’s recommendation to regard the Copernican system merely as a tool for calculations (famously quoted in Duhem 1954, 42). Besides, Fine (1996, 128n) himself has to admit that such thinkers as Niels Bohr and Bas van Fraassen do not (at least not unambiguously) accept the homely line. But if the homely line is disputed, then it needs a defense. Simply declaring the realism debate �nished by saying that no defense is asked for, as NOA does, is not an adequate response. Entity realism’s advantage over NOA is that it at least tries to provide such a defense. At the same time, chapter 2 showed that this defense still has some weaknesses. In the following chapters, I will attempt to overcome them.

�

Part II: Causal Realism Test everything. Hold on to the good. 1 Thessalonians 5:21

4 Causal vs. Theoretical Warrant As discussed in section 2.3, Nancy Cartwright’s (1983) version of entity realism rests on the distinction between theoretical and causal explanation and on the claim that IBE works reliably only in the latter, not in the former case. In other words, entity realism claims that inference to the most likely cause (IMLC) is a valid inference scheme, while inference to the best theoretical explanation (IBTE) is not. While I agree that the distinction between causal and theoretical explanation is of crucial importance for realism, I think it needs to be re�ned and modi�ed in several respects. This will be done in the �rst three sections of this chapter. The concluding section 4.4 will then show that the causal realism which emerges from these re�nements and modi�cations does no longer fall prey to the objections that were raised against Cartwright’s entity realism. The central element of causal realism is the distinction between causal and theoretical warrant, introduced by Suárez (2008). Like Cartwright, Suárez holds that causal explanation provides stronger grounds for existential commitments than theoretical explanation. But instead of drawing a sharp contrast between IMLC as a valid inference scheme on one side and IBTE as an invalid one on the other side, Suárez construes the di�erence between the two types of inference as a di�erence between two types of warrant that are generated by IMLC and IBTE, respectively. He thereby weakens Cartwright’s (1983, 93) claim that accepting a causal explanation provides “conclusive reason” to believe in the reality of the cause: “Conclusive” is thus to be understood as a relative term: Causal warrant is conclusive in comparison with the warrant provided by an IBTE. No existential commitment derived from an IMLC can be defeated by any amount of theoretical warrant to the contrary. . . . The only defeater of causal warrant in favour of the existence of x is causal warrant of the same strength against x. (Suárez 2008, 145)

But even with this relativization in place, the causal realist faces the challenge of explicating what it is that di�erentiates causal from theoretical warrant. As I see it, Suárez’s reply to that challenge is on the right track, but incomplete. In the following presentation of the distinction between causal and theoretical warrant, I will therefore be quite brief on those aspects which I think Suárez already got right, while elaborating on what needs to be added to his characterization in order for it to be completely convincing. In a nutshell, I characterize theoretical and causal warrant as follows: 1. Every instance of IBE generates theoretical warrant.

50 � Causal vs. Theoretical Warrant 2.

An instance of IBE generates causal warrant if and only if the corresponding explanation additionally ful�lls the criteria of non-redundancy, material inference and empirical adequacy.

I now turn to a discussion of these three criteria.

�.� Criterion 1: Non-redundancy If we are to infer the truth of a hypothesis on the basis of its ability to explain a given phenomenon, we need to make sure that no other hypothesis provides an equally satisfactory explanation of the phenomenon. Concerning the way in which such redundancy is removed, Suárez (2008, 155) identi�es an important di�erence between causal and theoretical explanations: The nonredundancy requirement is met to a much larger degree by IMLC than by IBTE . Scientists establish which putative cause is nonredundant through controlled intervention and manipulation in laboratory conditions—and only then have they got reason to believe that the putative cause is genuinely responsible for the phenomenon. By contrast, we have much poorer and controversial methods to establish which among all possible empirically adequate theoretical explanations of a phenomenon is the most probable one. . . . Is the most probable theoretical explanation the simplest one, the most ontologically parsimonious, the most familiar, the one that preserves the greatest amount of structure from previous theories, the one that explains a greater number of independent phenomena, etc, etc, etc. As a consequence there is much more redundancy, in the form of underdetermination, in the case of theoretical explanation.¹

If this is correct, then it is reasonable to spell out the criterion of non-redundancy in the following way: A hypothesis is non-redundant exactly if there is no other serious hypothesis that agrees with the experimental results. (I will specify below what I mean by “serious” here.) Accordingly, I will not count an explanation as non-redundant if it is merely marked out by theoretical virtues of the type mentioned in the penultimate sentence of the above quotation.²

1 For a similar defense of IMLC based on non-redundancy, see Clarke (2001, sec. 5). 2 This might be a slightly stronger understanding of non-redundancy than what Suárez has in mind. He seems to trace the redundancy in the case of theoretical explanation to the fact that there are so many di�erent (and possibly con�icting) theoretical virtues. Thus, it seems that he would count an explanation as non-redundant if it is favored by all (or most) theoretical virtues, even if there is an empirically adequate (but unambiguously less virtuous) competitor. By contrast, such an explanation would count as redundant on my understanding of the criterion. In section

Criterion 1: Non-redundancy

� 51

However, some objections can be raised against the claim that there is much more redundancy in theoretical than in causal explanations. Let us �rst look at Margaret Morrison’s (1994, 127) assertion that “causal explanation exhibits the same kind of redundancy present in theoretical explanation”. Morrison’s paper is directed against Cartwright (1983, 78–81), who uses an example from laser theory to illustrate the contrast between redundant theoretical and non-redundant causal accounts. Discussing the very same case, Morrison concludes that there is actually no such contrast. It seems to me, however, that what is established by Morrison’s analysis is not so much a redundancy of causal accounts but a multiplicity. And multiplicity of causal accounts by itself is no threat to the validity of IMLC. Instead, it is what we should expect, especially if we insist (as Morrison rightly does) that causes need to be speci�ed more precisely than in Cartwright’s generic causal story. It should not surprise us that a phenomenon can in general not be attributed to a single cause, but usually results from a combination of several causal factors or partial causes. It is in this sense that we may tell more than one causal story, but there is no redundancy here, because all these stories are needed to get the causal account right. Of course, we need to specify in each case precisely which causal factors are relevant (and to what degree), but this is often (though not always; see below) possible by the same methods we have already encountered in the above quotation: controlled intervention and manipulation in laboratory conditions.³ But now Morrison (1994, 136) can argue that this need for speci�cation again threatens the distinction between causal and theoretical explanations: Just as the mathematical treatment is determined by the kind of phenomena one is dealing with, causal stories also involve speci�c contextual features rather than a generic story applicable to all cases. A generic story could not qualify as a phenomenological law since it would require a ceteris paribus clause to guarantee its truth. Instead, the phenomenological laws are the ones that govern each individual case, but if we acknowledge their uniqueness as a basis for their “truth” then we must also acknowledge the uniqueness of speci�c theoretical descriptions.

It is debatable in how far the phenomena really determine the theoretical treatment in the same way as they determine which causal stories have to be taken into account, but I grant that uniqueness of theoretical description is sometimes achievable. The lesson to be drawn from this is that non-redundancy is a nec-

5.2, I will illustrate and justify this stronger reading of the criterion in the context of the neutrino hypothesis. 3 For a well-developed account of causation along these lines, see Woodward (2003).

52 � Causal vs. Theoretical Warrant essary but not a su�cient condition for causal warrant. Non-redundancy of an explanation by itself does not yet guarantee that the explanatory hypothesis is causally warranted, other criteria need to be taken into account as well. The di�erence between theoretical and causal warrant in terms of redundancy can also be questioned on another level, in connection with the general discussion of the underdetermination of theories by evidence (UTE).� One might think that there is always (that is, also in the case of causal explanations) a multitude of mutually incompatible possible explanations between which experimental evidence of the type Suárez speaks of is unable to decide. Consider, for example, a hypothesis H which constitutes a genuine (causal or theoretical) explanation for some phenomenon. Now it is always possible to construct an alternative hypothesis H ′ asserting that the world behaves exactly as H says if the universe is observed, but if no one is looking, the world behaves in ways which are incompatible with H.� Obviously, we should not let cases like this cast doubt on our warrant for belief in H, even though there is (by construction) no empirical way to determine whether H or H ′ is true. The reason is simply that this kind of UTE does not just threaten our warrant for scienti�c hypotheses but for virtually all our knowledge. It is, of course, possible to deny that any claim can be warranted, but to do so is to leave the debate on scienti�c realism and to go for general skepticism. Thus, there is a sense in which also the causal realist relies on the evaluation of theoretical virtues: If a hypothesis is so ad hoc that taking it into account would result in general skepticism, then it is excluded even if there is no experimental evidence against it. This explicates what I mean by “serious” in the above statement of the criterion of non-redundancy. A more interesting question about the status of the theoretical virtues arises in cases of non-trivial underdetermination (or redundancy), that is, cases where

4 UTE is not quite the same phenomenon as redundancy of explanations, but the di�erence is irrelevant here. Nevertheless, it may be helpful to spell out how the two concepts di�er. Apart from the fact that UTE is not restricted to explanatory contexts, the main di�erence between UTE and redundancy of explanations is, I think, this: In cases of UTE, the theories in question are generally required to be empirically adequate, whereas (as we shall see in section 4.3) no such requirement is made in the case of redundant (theoretical) explanations. This accounts for the rather puzzling fact that no one (to my knowledge) has so far contested Cartwright’s (1983, 79) claim that “this kind of redundancy of treatment . . . is common throughout physics”, while it is generally agreed that cases of genuine (non-trivial) UTE are very rare (Stanford 2001, S5–S7). 5 The example is modeled on a similar example from Kukla (1993). Stanford (2001, S2–S7) gives a concise review of the di�erent strategies pursued in the literature to construct empirically equivalent rivals of existing theories. Stanford’s critique of all these attempts is also the source of my argument against taking them seriously in the context of scienti�c realism.

Criterion 1: Non-redundancy

�

53

mutually incompatible, scienti�cally respectable hypotheses account for the same phenomena. Here the causal realist denies that the evaluation of theoretical virtues can e�ectively eliminate redundancy. This attitude resembles what Psillos calls the “ignorance response” to UTE, which consists in admitting that, if more than one theory is compatible with some evidence, we simply do not know (and may never be in a position to know) which of the competing theories is true. Psillos (1999, 169) thinks that this concedes too much ground to the antirealist: But admitting that the “ignorance response” is a viable realist answer to UTE is tantamount to conceding the main point that realists need to defend: that there is space for rational belief in one of the two empirically congruent theories. Instead of addressing the real challenge that UTE poses to realism—that there is no possibility of warranted belief in one of the two theories—the “ignorance response” simply sidesteps it.

This criticism has some force to it, so it would be unwise for the causal realist to completely reject the epistemic value of the theoretical virtues. On the other hand, Suárez is right that comparing hypotheses by experimental tests is much more reliable than evaluating their theoretical virtues. Requiring non-redundancy (in the sense de�ned here) as a necessary criterion for causal warrant takes into account both of these acknowledgements. The theoretical virtues have their role to play in the generation of theoretical warrant. However, they are incapable of generating causal warrant, because they cannot eliminate redundancy. For this, experimental testing is needed. Despite my criticism of Morrison’s example above, I do not claim that the elimination of redundancy by means of experimental testing is possible in all cases of causal explanation. This shows that the distinction between causal and theoretical warrant, although inspired by Cartwright’s entity realism, di�ers from her distinction between causal and theoretical explanations. The mere fact that an explanation cites a causal story is not su�cient to generate causal warrant. If there is a di�erent causal story, incompatible with the �rst one, which also explains the phenomena in question, then, due to failure of non-redundancy, none of the two stories is causally warranted. We then have to turn to an evaluation of theoretical virtues in order to see whether there is at least theoretical warrant for one of the stories. It might thus seem that, in such a case, causal realism simply collapses into a standard form of scienti�c realism as defended by Psillos, for example. However, I will argue in chapter 7 that this impression is mistaken, by discussing what is (to my knowledge) the clearest case of redundancy of causal explanations and by showing that the distinction between causal and theoretical warrant is important even in such a context.

54 � Causal vs. Theoretical Warrant

�.� Criterion 2: Material Inference An important intuition concerning the di�erence between theoretical and causal explanations is that the central explanatory role in the former case is played by the laws from which the explanandum is derived,� while the emphasis in the latter case is rather on the objects, events and processes which the explanans describes as causally responsible for the phenomena in question. Correspondingly, the application of IBE is, in the case of a theoretical explanation, expected to establish the truth of some laws, while in the case of a causal explanation it is taken to argue for the reality of some causally e�cacious entities. Suárez (2008, 157–159) describes this di�erence by saying that IBTE is a formal inference, whereas IMLC can be carried out as a material inference. Then, comparing two examples of inference, one in the formal, the other in the material mode, he concludes that “the latter type of inference is more robust—it contains fewer steps where it may go wrong” (158). While I agree with Suárez’s conclusion, I �nd his argument in support of it far from convincing. To see how it could be improved, let us look at Cartwright’s (1983, 76) argument for the di�erence in robustness between IMLC and IBTE: Causes make their e�ects happen. We begin with a phenomenon which, relative to our other general beliefs, we think would not occur unless something peculiar brought it about. . . . The peculiar features of the e�ect depend on the particular nature of the cause, so that—in so far as we think we have got it right—we are entitled to infer the character of the cause from the character of the e�ect. But equations do not bring about the phenomenological laws we derive from them (even if the phenomenological laws are themselves equations). Nor are they used in physics as if they did.

According to Cartwright, the di�erence between causal and theoretical explanations lies in the fact that there is a relation of “making happen” or “bringing about” only in the former, not in the latter case. In order to grasp how this relates to the di�erence between material and formal inference, we need to understand more precisely what “bringing about” means. As a �rst approximation, one might think that if an event of type x brings about an event of type y, then the occur-

6 The idea that the explanandum is logically derived from the explanans obviously stems from the DN-model of explanation, and it might seem unfair to saddle the advocate of IBTE with this outmoded account of explanation. However, the most appropriate candidate for a more modern approach to theoretical explanation is the uni�cation model developed by Michael Friedman and Philip Kitcher, and this approach “retains the Hempelian idea that to explain a phenomenon is to produce an argument whose conclusion describes the phenomenon” (Kitcher 1989, 431).

Criterion 2: Material Inference

� 55

rence of x is a necessary condition for the occurrence of y, so that, whenever we observe y, we can be sure that x occurred as well. The inadequacy of this proposal becomes obvious once we consider how causal reasoning works in actual science. If, for example, particle physicists want to test whether some process x really occurs, they typically try to detect the products of x. More precisely, they calculate what kind of signal the process x should produce in their detector, and then they look for this signal (let us denote it by y) in the experimental data. Now typically, the mere occurrence of y by no means establishes the reality of x, because there are usually some alternative ways in which a signal of type y could have been produced. This is what physicists call “background”. A case for x will only be made if there is a part in the counting rate for y that cannot be attributed to background. It will then in general be false to say that whenever y occurs, x has occurred as well. But it will be true that at least in some cases (though we may not know in which ones), y would not have occurred if x had not occurred. The truth of this counterfactual statement for at least some tokens of the event type y is an essential part of what it means for x to bring about y (regardless of the di�culties that a general account of causation in terms of counterfactuals may face). And the truth of this counterfactual, in conjunction with a su�cient number of y-occurrences, establishes realism with respect to x. Now it might seem that there is a similar counterfactual dependence between the explanandum and the explanans in the case of a theoretical explanation, because such explanations contain laws, and an essential aspect of lawhood is the ability to support counterfactuals. But these counterfactuals are not of the right sort to support realism about laws in the same way as the above counterfactuals support realism about causes. Here is why: Laws support counterfactual claims concerning their instances. For example, Boyle’s law supports claims like “if the volume of this gas was reduced at constant temperature, its pressure would increase”. But what IBTE aspires to establish is not the truth of a singular statement but of the law itself. In order to achieve this, a claim of the following form would be needed: “If law L did not hold, phenomenon y would not occur.”� This is actually not just a counterfactual, but a counterlegal statement, and the mere fact that L explains y implies nothing about the truth of such a statement. What it implies is that L is part of a su�cient condition for the occurrence of y, in the sense that L, conjoined with some initial conditions, allows us to derive a statement describing y. But using L in a theoretical explanation involves no speculation about what

7 By speaking of y as an occurrent phenomenon, I am assuming that the explanandum is a singular fact, but the argument is just the same if y is itself a (phenomenological) law, as Cartwright assumes in the above quotation.

56 � Causal vs. Theoretical Warrant would happen if L did not hold. By contrast, as we have seen above, citing an entity x in a causal explanation of y essentially involves a claim about what would (or would not) have happened had x not occurred. In other words: the counterfactual statements that give rise to causal realism are an integral part of causal explanations. Therefore, the connection between scienti�c explanation and realism is much tighter in the case of causal explanations than in the case of theoretical explanations. It is this di�erence that the criterion of material inference aims to capture by stating that there is causal warrant only if the result of an IBE is a concrete matter of fact, as opposed to a law.

A Dilemma for Causal Realism: What Counts as a Concrete Matter of Fact? The soundness of the characterization just given stands and falls with how one draws the line between concrete matters of fact and laws. Does, for example, the statement that electrons are negatively charged express a concrete matter of fact (because electrons are concrete entities) or a law (because it is a general statement)? If the latter, then it seems that hardly any statement of scienti�c interest can be arrived at by material inference, so that a realism based on this criterion faces the threat of vacuity. (As we saw in section 2.3, this was one of the main objections against entity realism.) On the other hand, if the negative charge of electrons counts as a concrete matter of fact, why should the fact that electrons are accurately described by quantum electrodynamics (QED) not also count as such a fact? But then it seems that the criterion of material inference would be trivially satis�ed by even the most theoretical explanations, if only they contain references to at least some concrete entities. (This is the kind of criticism that Massimi (2004) advances against entity realism.) To solve this dilemma, we need to take a closer look at the causal claims which underlie IMLC. And this �rstly requires some clari�cation on the relata of the causal relation. So far, I have been very permissive in this respect, allowing objects, events, facts or phenomena to �gure as causes or e�ects. My reason for doing so is that I do not think that any of these is the right way to speak about causation. They are all useful in certain contexts (and I will continue to use them whenever there is no danger of confusion), but if we want to be precise, we have to speak of causation in terms of causal properties (Chakravartty 2007, 107–111). This is not the place to enter into a detailed metaphysical discussion of causal relata, but the observation just made helps us to see more clearly what needs to be done in order to �esh out the criterion of material inference: What is to count as material inference needs to be de�ned in terms of the kinds of properties that

Criterion 2: Material Inference

� 57

can be ascribed to entities as a result of such inference (let us call them material properties). Which properties belong to that class? An obvious proposal would be to identify them with what Chakravartty (2007, 41) calls causal properties, that is, properties which “confer dispositions for relations, and thus dispositions for behavior on the particulars that have them”. But this is not precise enough, because the above-mentioned property of “being accurately described by QED” also confers a disposition for behavior on the electron (at least in some sense).� The necessary re�nement of this proposal can be found, I think, in James Woodward’s manipulability account of causation. Woodward (2003, 112) holds that “for something to be a cause we must be able to say what it would be like to change or manipulate it”.� The rationale for this requirement is that we usually test causal claims by intervening on the putative cause and observing whether the putative e�ect changes accordingly. While it is, on Woodward’s account, not essential for causal claims that such an intervention is possible in practice (see Woodward 2003, 86– 91, 127–133), it is essential that there is a well-de�ned notion of what it would mean to intervene on the putative cause. This requirement furnishes us with the necessary distinction between material and formal properties. Let us apply it to the two examples constituting the above dilemma. It is perfectly clear what it means to modify the property of negative charge. (In some cases, this is even practically possible, by performing an experiment with positrons instead of electrons, but we may also envisage hypothetical experiments in which the charge of the electron is set to a value other than ±e.) By contrast, it is not clear what it would mean to intervene on the property of being accurately described by QED. It could mean to modify the elementary processes that QED describes, to change the way in which they are superposed, to introduce an entirely new interaction, etc. So we arrive at the following de�nition of material inference: An inference in the material mode

8 On the other hand, it would be too restrictive to identify the material properties with what Chakravartty (2007, 47) calls detection properties, that is, “causal properties one has managed to detect”. The criterion of material inference should only be concerned with what can be established, not with what has been established. Otherwise, the criterion would by itself be su�cient for causal warrant, thus rendering the other criteria super�uous. So what we need here is not a speci�cation of detection properties, but of detectable properties. 9 With regard to the question of what the relata of the causal relation are, Woodward introduces yet another candidate, namely changes in the values of variables. He also mentions some cases in which thinking of causation in terms of properties (as Chakravartty does) would lead us astray (Woodward 2003, 112–114). However, the con�ict between the two approaches can be resolved by excluding the properties appearing in these examples from the domain of causal properties.

58 � Causal vs. Theoretical Warrant is one that results in ascribing to a concrete entity a property for which there is a well-de�ned notion of what it means to modify it.¹�

Causal Realism without Causal Fundamentalism To conclude the discussion of material inference, I must address a more fundamental objection to the legitimacy of this criterion. Many will view the causal notions which appeared throughout the discussion as an element of dubious metaphysics. They deny that there are causes in nature and that advanced science is concerned with causal claims. One way to answer this objection would be to enter into a discussion of the fundamental structure of the world and to argue that causation has a role to play there. There are well-developed arguments for this, both from pure metaphysics (e.g., Bird 2007) and from the re�ection on contemporary physics (Esfeld 2010); furthermore, Chakravartty (2007) has recently integrated a metaphysics of causal properties into a defense of scienti�c realism. While I am not opposed to this line of reasoning, I suspect that it claims more than what the realist needs. What these authors defend is a kind of causal fundamentalism (the view that causality is a fundamental trait of nature), while causal realism, as I understand it, can be argued for independently of how one thinks about these fundamental issues. For this move, I take inspiration from one of today’s �ercest critics of causal notions, John Norton. In his essay Causation as folk science (2007), Norton denounces causal fundamentalism, but at the same time admits that causes are in some sense real. To illustrate this, he draws an analogy between causes and gravitational forces: Although gravitational forces do not appear in our most fundamental theory of gravitation (general relativity), they have a derivative reality in the sense that the classical Newtonian theory incorporating them can be recovered as a limiting case of general relativity. Just like gravitational forces, causes can be considered real, even though they are not fundamental: Causes, caloric¹¹ and gravitational forces have a derivative reality. They are not �ctions insofar as they are not freely invented by us. Our deeper sciences must have quite particular

10 While this excludes formal and higher order properties, it is not part of causal realism to insist that only intrinsic properties are allowed. So there is no contradiction between causal realism and structural realism, as long as the latter conceives of structures as (�rst-order) relations between particulars (Chakravartty 2007, sec. 2.3). 11 Besides the analogy with gravitational forces, Norton also compares causes to the caloric, the alleged heat substance of pre-modern theories of heat. But this analogy is ill chosen, because the caloric theory of heat is not just superseded (like Newtonian mechanics) but thoroughly discredited. Accordingly, no one would nowadays seriously claim that caloric is real, while there are

Criterion 3: Empirical Adequacy

�

59

properties so that these entities are generated in the reduction relation. . . . But then they cannot claim the same reality as the fundamental ontology. (Norton 2007, 31)

Granted, more would have to be said about what this “derivative reality” amounts to ontologically. I do not have a story to tell how causes can be “generated in the reduction relation” on the basis of a non-causal fundamental ontology. But neither do I think that such a story is necessary for the purpose of defending the use of causal concepts in an argument for scienti�c realism. The reason for this is that the very debate on scienti�c realism is based on assuming the reality of non-fundamental entities for which we do not (in general) have such a reduction story, namely the objects of common sense (see chapter 1). By treating causes as real without pronouncing on their fundamentality, causal realism gains a basis that can be accepted by a wider audience than just the supporters of causal fundamentalism.

�.� Criterion 3: Empirical Adequacy According to van Fraassen (1980, 12), “a theory is empirically adequate exactly if what it says about the observable things and events in this world, is true—exactly if it ‘saves the phenomena’.” Being thus de�ned as a property of theories, empirical adequacy does not seem to have much to do with causal warrant. I nevertheless claim it to be a criterion for causal warrant, so I must �rst explain what it means for a causal explanation to be empirically adequate, and then argue that this condition is typically ful�lled for causal, but not for theoretical explanations. What an explanation says about observables is not given by the explanatory hypotheses alone but by what can be deduced from them (with the help of some auxiliary hypotheses). Therefore, to speak of the empirical adequacy of an explanation requires that it makes sense to speak of the deductive consequences of the explanans. Now if (as I suggested at the beginning of the last section) the central explanatory role in a causal explanation is played by concrete particulars (objects, events, processes), then it may sound like a category mistake to say that we deduce anything from such an explanans. However, the subsequent discussion made it clear that causal explanations are associated with certain descrip-

good arguments to hold that gravitational forces are real (Wilson 2007). Norton’s mistake is that he focuses on just one aspect of the caloric theory, namely the claim that heat is a conserved �uid. The actual theory made various other claims about the caloric, and these render a recovery of the caloric theory as a limiting case of today’s kinetic theory impossible (Chang 2003).

60 � Causal vs. Theoretical Warrant tions (statements of what I have called concrete matters of fact), and these do, of course, have logical implications. Some of these implications are statements about observable things and events, and if these statements are true, the criterion of empirical adequacy is ful�lled. But is empirical adequacy not something we should require from any explanation, theoretical or causal? After all, even an antirealist like van Fraassen, who denies that explanation is connected with truth, agrees that it is connected with empirical adequacy. Nonetheless, Cartwright (1983, 159–160) argues against this connection in the case of theoretical explanations: [Van Fraassen and realists like Sellars] have in common a surprising respect for theory. Both expect that theories will get the facts right about the observable phenomena around us. . . . This is not how I see good theories working. The observational consequences may be a rough match to what we suppose to be true, but they are generally not the best we can do. If we aim for descriptive adequacy, and do not care about the tidy organization of phenomena, we can write better phenomenological laws than those a theory can produce.

To support her claim, Cartwright (1983, essay 6) discusses various theoretical explanations in which the explanandum does not follow from the explanans by strict deduction, but only with the help of approximation procedures that improve on the descriptive accuracy of the fundamental laws and are dictated solely by the facts to be derived. In cases like these, it is hard to see how the explanatory relation should provide warrant for the truth of the explanans. Realists may hope to overcome this predicament by building upon Larry Laudan’s and Jarrett Leplin’s (1991, 461–465) argument that a theory can be supported by empirical evidence that is not part of its deductive consequences. For the case of IBTE, this would mean that a theory’s explaining of some observed phenomenon would provide warrant for its truth, although the phenomenon is not strictly deduced from it. But I contend that even if this argument is successful, the warrant it provides will clearly be theoretical, not causal. For the “indirect empirical support” that a hypothesis H� can, according to Laudan and Leplin (1991, 464), gain from some empirical evidence e that is not directly entailed by it, depends on H� being derivable from a more general theory T, which also entails another hypothesis H� , of which e is a consequence.¹² So whatever warrant there is for H� derives from warrant for T, and this is only theoretical warrant, because the inference from e to T (via H � ) will in general not satisfy the criterion of material

12 Laudan and Leplin also consider other modes of nonconsequential empirical support, such as analogical reasoning, but these seem even more vulnerable to the charge that what is transmitted here is just theoretical and not causal warrant.

Criterion 3: Empirical Adequacy

� 61

inference. Neither is T’s ability to unify H� and H� (or any other theoretical virtue T may possess) su�cient to ensure non-redundancy. In fact, as seen in the above quotation, Cartwright even argues that a theory’s ability to achieve a “tidy organization of phenomena” works counter to its empirical adequacy. This clari�es why empirical adequacy is more tightly connected to causal than to theoretical explanations. In theoretical explanation, uni�cation of diverse phenomena is essential (see note 6 above); without it, a hypothesis would not count as explanatory. By contrast, though uni�cation is certainly desirable in causal explanations, a causal hypothesis can to some extent be explanatory without it, by relying on concrete matters of fact and fairly speci�c phenomenological laws. Having a more limited range of application, statements about such facts and laws have a higher chance of being empirically adequate. However, these considerations point to an important ambiguity in the notion of empirical adequacy, which needs to be removed if this criterion is to be coherently applied. Having explicated empirical adequacy as a theory’s ability to “save the phenomena”, van Fraassen (1980, 12) emphasizes that “this refers to all the phenomena: these are not exhausted by those actually observed, nor even by those observed at some time, whether past, present, or future.” A strong reading of “all” in this quote would imply that only a theory of everything can be considered as empirically adequate. This would render the notion of empirical adequacy quite useless for our purposes (and, I suppose, for the debate in general too). “All phenomena” must therefore be understood relative to some domain of application. But if we interpret it too liberally, the notion will be equally deprived of its usefulness, for then, empirical adequacy is too easy to get. I illustrate this with a well-known example, which will be discussed in more detail in the next section. In many experiments, electrons behave like tiny billiard balls, moving on wellde�ned trajectories, subject to local interactions. Still, we would not say that this classical picture of the electron is empirically adequate, because there are other experiments (e.g., the double slit), producing phenomena which are incompatible with the classical theory. What is needed, therefore, is a speci�cation of a domain neither too vast nor too narrow, within which all the phenomena need to be saved. Furthermore, in order to be suitable for realism, such speci�cations should be objective in the sense of not depending on the kind of experiment we choose to perform. The most common way to accomplish this is by considering limiting cases with respect to some physical parameter. The example of gravitational forces, familiar from section 4.2, illustrates this well. Newton’s theory of gravitation does not save all the phenomena, nor even all gravitational phenomena (the perihelion precession of Mercury being a famous counterexample). But its domain of application can be objectively circumscribed by requiring that the

62 � Causal vs. Theoretical Warrant considered distances are large compared to the Schwarzschild radii of the considered objects.

�.� Causal Realism’s Advantages over Entity Realism Many of the arguments in the previous sections made use of Cartwright’s distinction between causal and theoretical explanation. Given this dependency of my causal realism on (a central element of) Cartwright’s entity realism, I must now show that the points of criticism that were raised against entity realism are no threat to causal realism. Section 4.2 already contains causal realism’s response to the charge of incoherence described in section 2.3: It is not the case that causal realism only warrants a commitment to certain entities without being committed to the truth of any statements about them. The distinction between formal and material properties clearly shows how causal realism can be committed to certain scienti�c statements without collapsing into ordinary theory realism: Statements concerning formal properties are based on IBTE, so the causal realist is not (or at most tentatively) committed to them. By contrast, statements concerning material properties can be warranted by IMLC, therefore belief in these statements (to which the causal realist is strongly committed) does not imply belief in the rest of the theory, for which there is only a weaker kind of warrant. Two other charges against entity realism remain to be discussed. The �rst is that such a realism is based on a mere convention, the second is that it is contradicted by examples from actual science.

Conventionality Robert Pierson and Richard Reiner (2008) argue that the alleged di�erence between IBTE and IMLC is largely conventional: Due to the work of antirealists like Duhem and van Fraassen, we have grown accustomed to the thought that theories can be explanatory without being true, whereas we may still think that no causal story can be explanatory without citing the actual causes. Another way of expressing this di�erence is to say that “causally explains” is usually taken to be a success term, while “[theoretically] explains” is not (Pierson and Reiner 2008, 272). Pierson and Reiner question this convention as arbitrary, but this is only the �rst part of their criticism. They then go on to claim that even if we accept the convention, IMLC will be unable to transmit warrant from its premises (“P causally

Causal Realism’s Advantages over Entity Realism � 63

explains Q” and “Q is true”) to its conclusion (“P is true”), because if “causally explains” is interpreted as a success term, then the �rst premise is only warranted insofar as there already is warrant for the conclusion. If P is not describing an actual cause, there is no causal explanation in the �rst place (278, 281). However, the criteria discussed in the previous sections show that the di�erence in validity between IMLC and IBTE rests on more than mere convention and that evidence for “P causally explains Q” can be gathered independently of evidence for the truth of P.¹³ This independence is underlined by the fact that each of the three criteria is, by itself, insu�cient to guarantee causal warrant. Only when they are jointly ful�lled is there causal warrant for the truth of P. In this context, another advantage of causal realism over entity realism deserves to be mentioned. Within Cartwright’s entity realism, the criterion of nonredundancy is a necessary condition for an explanation to qualify as causal, which has the following consequence: If scientists, after a process of careful experimental testing, �nally succeed in eliminating redundancy of explanations for a certain phenomenon, then the one explanation that emerges as the only survivor, having hitherto been a theoretical explanation, will now suddenly become causal. This contradicts the intuition that being causal should be an intrinsic and timeindependent feature of an explanation. Cartwright (1983, 81) seems to recognize this problem and consequently holds that the causal/theoretical distinction does not depend on whether there actually is a redundancy of explanations, but on whether or not we tolerate redundancy: Di�erent [theoretical] approaches are useful for di�erent purposes; they complement rather than compete. . . . We do not have the same pragmatic tolerance of causal alternatives. We do not use �rst one causal story in explanation, then another, depending on the ease of calculation, or whatever.

Unfortunately, this way of interpreting the requirement of non-redundancy does little to defend entity realism against the conventionality objection, because it makes the causal/theoretical distinction depend on our attitude towards redundancy, which is just as much a matter of convention as our habit of taking “causally explains” (but not “theoretically explains”) as a success term. But for the causal realist who has assimilated Suárez’s re�nement of the causal/theoretical distinction, there is no need to move towards such a subjective interpretation of non-redundancy. If the distinction is no longer between causal and theoretical explanations but between causal and theoretical warrant, then

13 For more details on this point, see Suárez (2008, 153–155). Suárez does not directly reply to Pierson and Reiner, but to an equivalent argument by Hitchcock (1992, 160–166).

64 � Causal vs. Theoretical Warrant the above scenario of something theoretical that suddenly becomes causal poses no threat at all. Nobody expects the character of warrant to be intrinsic and timeindependent. In the course of inquiry, it repeatedly happens that a theoretically warranted hypothesis at some point acquires causal warrant, and the elimination of redundancy is the obvious way to achieve this transition.

Counterexamples to the Validity of IMLC Christopher Hitchcock (1992, sec. IV) argues against the validity of IMLC by giving two counterexamples to the claim that acceptance of a causal explanation in science necessarily involves belief in its truth. This objection is obviously related to the previous one in that it also targets the distinction between IMLC and IBTE. More precisely, it blocks a possible reply to the conventionality objection, namely the causal realist’s claim (in the spirit of Cartwright 1983, essay 4) that the difference between IMLC and IBTE is not just a matter of arbitrary convention, but �rmly rooted in actual scienti�c practice. Not so, says Hitchcock, there are examples showing that scienti�c practice is actually opposed to treating IMLC di�erently from IBTE. What are these examples? They both come from quantum mechanics, and since they are closely related to each other, it su�ces to look at one of them. Consider the following well-known phenomenon: In a double-slit experiment with quantum objects (e. g., electrons), an interference pattern appears on the screen behind the two slits. But as soon as we perform a measurement to �nd out through which of the two slits each electron passed, the interference pattern disappears. This is in some textbooks explained by the following line of thought: Suppose that we try to determine which slit a given electron passes through by bouncing a photon o� of it as it passes through. The wavelength of the photon must be small enough that the position measurement on the electron after it has passed through the �rst screen is accurate relative to the distance between the two slits. The collision of the photon with the electron will impart some momentum to the electron in the vertical direction, with the amount of momentum transmitted being proportional to the frequency of the photon. If the electron receives a su�cient kick, its vertical displacement upon reaching the screen will be of the same order of magnitude as the distance between the crests of the interference pattern, so the interference pattern will be destroyed. (Hitchcock 1992, 170–171)

Hitchcock now claims that this explanation is accepted in the sense that it appears in many physics textbooks, but not believed to be true, because it contradicts some �rmly established results from quantum mechanics, particularly the non-existence of simultaneously well-de�ned position and momentum values for electrons and photons and the resulting absence of classical particle trajectories.

Causal Realism’s Advantages over Entity Realism � 65

It thus seems that, as Psillos (2008, 169) puts it, Hitchcock raises “the spectre of van Fraassen”: While van Fraassen (1980) showed us that acceptance of theoretical explanations does not commit us to belief in their truth, Hitchcock now does the same for (at least some cases of) causal explanations. But there is in fact no such alignment between van Fraassen and Hitchcock, and the illusion of their achieving the same thing turns on an equivocation of the term “accept”. Hitchcock (1992, 174) understands the term in a very broad sense: All of these causal stories are accepted in the sense that members of the scienti�c community deem them appropriate for the role they play; even the just-so stories can be accepted as serving important heuristic roles.

Van Fraassen’s notion of acceptance is much narrower. According to him, empirical adequacy is a precondition for considering a theory acceptable (1980, 95). Now, is the explanans in Hitchcock’s example empirically adequate? It surely “saves the phenomena” for this particular experiment, but we do not have to look far for phenomena which are incompatible with this classical picture; the interference pattern in the original double-slit experiment (without detection through which slit each electron passes) cannot be explained by classical electron trajectories and local interactions.¹� Therefore, Hitchcock’s story can only serve as a counterexample to the validity of IMLC if we do not require causal explanations to be empirically adequate. But, as discussed in section 4.3, empirical adequacy is a necessary condition for causal warrant, so Hitchcock’s argument has no force against causal realism.¹�

14 Notice that it is the commitment to electron trajectories and local interactions which renders the classical picture empirically inadequate. Orthodox interpretations of quantum mechanics claim that it is the commitment to continuous particle trajectories that causes the trouble, but we will see in section 8.1 that there is an empirically adequate account of the double slit which is in fact committed to particle trajectories, based on Bohmian mechanics. What this theory abandons is the idea of purely local interactions. Furthermore, chapter 7 will show that, as soon as the results of Bell-type experiments are taken into account, no theory holding on to a certain locality condition can be empirically adequate. This suggests that it is not so much the commitment to particle trajectories as the commitment to local interactions that is problematic in the classical picture. 15 Suárez’s reply to Hitchcock’s objection is similar to mine, but since he does not formulate it in terms of empirical adequacy, it is not very convincing. Suárez (2008, 152) denies (as I do) that Hitchcock’s example really is an acceptable causal explanation, but it seems that his only reason for doing so is that this causal story is not believed to be true:

66 � Causal vs. Theoretical Warrant

The defender of causal explanation takes causal explanation to be a success term. So if we don’t believe in the “causes” appealed to in the story then the explanation the story o�ers—if any—cannot be said to be causal. But the point of Hitchcock’s argument was precisely to deny that causal explanation needs to be taken as a success term. So simply insisting on this linguistic practice will not do. In another passage, Suárez (2008, 159) seems to imply that the criterion of material inference helps to exclude Hitchcock’s counterexamples, but this will not work either, because these causal stories do not seem to depend on any formal inference. I conclude from this that the most e�ective reply to Hitchcock consists in requiring empirical adequacy as a criterion for causal warrant.

5 Causal Warrant for the Neutrino: A Case Study The concepts prepared in the previous chapter are now applied to a concrete case from the history of particle physics, the discovery of the neutrino.¹ My aim in this chapter is to show that the concept of causal warrant, which was up to now introduced in a rather abstract way, can do some real epistemological work when applied to a concrete scienti�c problem. More precisely, I will argue for the following thesis: Any philosophy (realist or antirealist) failing to appreciate the distinction between theoretical and causal warrant will be unable to grasp the signi�cance of what physicists call “the detection of the neutrino”. Therefore, insofar as we think that our philosophy of science should be able to make sense of the actual practice of science, this case study amounts to a strong argument for causal realism.

�.� Bohr and Pauli on Beta Decay The neutrino hypothesis has its origin in a problem which appeared around 1914 and became ever more pressing towards the end of the 1920’s: the continuous energy spectrum of electrons in nuclear beta decay. For a long time, physicists had tried to attribute this phenomenon to secondary processes in the atomic shell, assuming that the primary electron spectrum is discrete, in analogy to alpha decay. But this option had to be abandoned when calorimetric measurements showed that the emitted electrons did not really lose any energy in the putative secondary processes (Ellis and Wooster 1927; Meitner and Orthmann 1930). The inevitable conclusion that the continuous spectrum was actually a feature of the decay itself con�icted squarely with the well-founded assumption that the nucleus has discrete energy levels before and after the decay. In 1930, Wolfgang Pauli proposed what he called “a desperate way out” of this problem (Pauli 1985, 39; translated in Pais 1986, 315). He suggested that a hitherto unknown, neutral spin-1/2 particle of small mass might carry away part of the decay energy without itself being detected. Pauli called this hypothetical particle “neutron”, but this name was soon to be replaced by “neutrino”, a term coined by Enrico Fermi.² Beside the neutrino hypothesis, there were other proposals to

1 Historically oriented accounts of this case are given by Pauli (1984, English translation in: Winter 2000, 1–22), Sutton (1992) and Franklin (2004). The early phase (1930–1934) is treated in more detail by Pais (1986, sec. 14(d)) and Jensen (2000, chap. 6). 2 Apart from this verbal interrelationship, the neutrino hypothesis is also connected to what we now call “neutron” in a more interesting way: Prior to the discovery of the neutron in 1932, atomic

68 � Causal Warrant for the Neutrino: A Case Study account for the continuous beta spectrum, the most prominent being Niels Bohr’s idea that energy conservation might fail at the nuclear level.³ Comparing these two proposals in the light of our discussion in chapter 4, the most obvious di�erence between them is that Pauli’s hypothesis meets criterion 2 (material inference) while Bohr’s explanation is clearly formal in character; it does not tell us what causes the continuity of the spectrum. This may be one of the reasons why Pauli, in a letter to Bohr in 1929, denied Bohr’s proposal even the status of an explanation: “So we really do not know what is going on here! You do not know it either, you can only give reasons why we do not understand anything.”� On the other hand, its lack of speci�city saved Bohr’s hypothesis from any con�ict with criterion 3 (empirical adequacy). The neutrino hypothesis, by contrast, had serious problems in this respect, as Pauli himself acknowledged in a letter to Oskar Klein in 1930: “So if the neutrons really existed, it would hardly be understandable that they have not yet been observed. Therefore, I also do not myself entirely believe in the neutrons.”� This problem gradually disappeared in the years to come, when it became clear how weak the neutrino’s interaction with matter actually was. With Bohr’s and Pauli’s hypotheses both (more or less) conforming to experimental data, there was obviously no causal warrant for the neutrino at that time, since criterion 1 (non-redundancy) was not satis�ed. This assessment is not signi�cantly altered even if one takes into account that the neutrino does not only ensure energy conservation in beta decay, but also conservation of momentum and angular momentum;� if one is willing to admit energy non-conservation, it is

nuclei were believed to consist of protons and electrons. This implied that nitrogen or lithium-6 nuclei consisted of an odd number of fermions, which contradicted experimental results assigning integer spin to these nuclei. Pauli now thought that the neutrino hypothesis would also solve this problem, because it would make the nuclear spin depend not only on the number of protons and electrons, but also on the number of neutrinos (which Pauli (1985, 39) believed to “exist in the nucleus”). But this explanatory role of the neutrino became obsolete when it was discovered that nuclei do not actually consist of protons and electrons, but of protons and neutrons. 3 For a survey of other proposals, see Jensen (2000, sec. 6.3). 4 “Wir wissen also wirklich nicht, was da los ist! Du weißt es auch nicht, kannst nur Gründe dafür angeben, warum wir nichts verstehen” (Pauli 1979, 513). 5 “Wenn die Neutronen also wirklich existieren würden, wäre es wohl kaum verständlich, daß man sie noch nicht beobachtet hat. Deshalb glaube ich auch selber nicht so ganz an die Neutronen” (Pauli 1985, 45). Recall that the term “neutrino” had not yet appeared in 1930. 6 That the neutrino is relevant not only for energy but also for momentum conservation became clear in the recoil measurements carried out from 1936 onwards. (For a survey of the most important experiments, see Crane 1948.) Its importance for angular momentum balance stems from the following fact: If nothing but an electron were emitted in beta decay, the nuclear spin would have

The Impact of Fermi’s Theory and the Need for Direct Detection �

69

a small step to also allow for non-conservation of the other two quantities. However, it is an instructive thought experiment to ask with how much causal warrant we should credit the neutrino hypothesis on the basis of these observations, even if we assume (counterfactually) that criterion 1 had been satis�ed by excluding Bohr’s hypothesis (and all other alternatives). Surely we would be warranted in inferring the reality of certain (causal) properties: the ability to carry away energy, momentum and angular momentum. But would we thereby be equally warranted to say that there is some object (the neutrino) which has these properties? The following analogy might illustrate this worry: Classical electrodynamics provided compelling evidence for the existence of electromagnetic waves, and it was only natural to assume that these were waves of something. But as it turned out, the underlying something (the ether) does not exist. May not the same apply to the neutrino? At this stage of enquiry, I see no reliable basis to refute this argument. One may claim that the simultaneous manifestation of three di�erent properties (or at least two, given that the carrying away of energy and momentum might be viewed as the same property) provides some reason to believe in an underlying object having those properties, but this claim is not very convincing, since this simultaneous manifestation of properties has so far only been observed in the context of one kind of process, nuclear beta decay. Hence the need (to be discussed in detail in the next two sections) to detect the neutrino at a location other than that at which it was created.

�.� The Impact of Fermi’s Theory and the Need for Direct Detection In his Versuch einer Theorie der β-Strahlen, Enrico Fermi (1934) described the emission of an electron and a neutrino from a nucleus in close analogy with the emission of photons from an atom. He thereby replaced the traditional conception of beta decay as an emission of electrons which had previously been part of the nucleus (cf. footnote 2 above) by the quantum �eld theoretical conception of creation and annihilation of particles. Furthermore, he integrated Pauli’s rather qualitative neutrino hypothesis into a quantitative theoretical framework.� The neutrino

to change by 1/2, that is, change from a half-integral to an integral value (or vice versa), contrary to what is observed in experiment. 7 Interestingly, the competing hypothesis of energy non-conservation had—some months earlier—also been integrated into a theory of beta decay (Beck and Sitte 1933). However, this approach was soon eclipsed by the successes of Fermi’s account (Jensen 2000, sec. 6.6).

70 � Causal Warrant for the Neutrino: A Case Study thereby became what is properly called a theoretical entity, and the impressive corroboration of Fermi’s theory in the years that followed was by most physicists taken as con�rming the existence of the neutrino. And even those who felt that a more direct proof was needed had to recognize that “the theory was so attractive in its explanation of beta decay that belief in the neutrino as a ‘real’ entity was general” (Reines 1996, 317). The question now is on what kind of warrant this general belief in the existence of the neutrino was based. Had the neutrino hypothesis somehow acquired causal warrant by being part of Fermi’s theory? I do not think so, for two reasons. First, the problem discussed at the end of the last section is obviously not a�ected by integrating the posited entity into a successful theory. Remember that the ether hypothesis was part of an immensely successful theory (classical electrodynamics), but it nevertheless turned out to be untenable. Second, though it may seem as if the unmatched success of Fermi’s theory helped the neutrino hypothesis to ful�ll criterion 1, a closer inspection shows that redundancy of explanations is still a problem. While it is true that, thanks to Fermi’s theory, the neutrino hypothesis proved to be incomparably more fruitful than Bohr’s hypothesis (which, by the way, was abandoned by Bohr himself in 1936), this did not actually exclude energy non-conservation as a possible explanation of the continuous beta spectrum. Nor did it render Bohr’s hypothesis so implausible that adopting it would amount to general skepticism, as is the case with hypothesis H ′ discussed in section 4.1. Instead, the epistemic situation at that time is accurately described by the following quote from 1948: While the [neutrino] hypothesis has had great usefulness, it should be kept in the back of one’s mind that it has not cleared up the basic mystery, and that such will continue to be the case until the neutrino is somehow caught at a distance from the emitting nucleus. Some physicists prefer to say simply that energy and momentum are apparently not conserved, giving full recognition, of course, to the energy and momentum relations that have been established experimentally, and to the success of the beta-ray theory which has been built upon the neutrino hypothesis. Perhaps all one can say is that this is a matter of taste. (Crane 1948, 278)

Another way of making this point is to say that the evidence that con�rms the neutrino hypothesis via Fermi’s theory does so only indirectly (in the sense of Laudan and Leplin, as discussed in section 4.3), and as we have seen, this a�ords only theoretical, not causal warrant. The dissatisfaction of some physicists with the credentials of the neutrino’s existence can thus be seen as resting on the distinction between theoretical and causal warrant. As long as a hypothetical posit has only theoretical warrant in its favor, skepticism about its actual existence is well founded, despite the success

The Impact of Fermi’s Theory and the Need for Direct Detection �

71

of the theory incorporating it. Accordingly, it is reasonable to stress the importance of the neutrino’s direct experimental detection. Conversely, any philosophy failing to appreciate the distinction between theoretical and causal warrant will have to downplay this importance. The constructive empiricist, for example, will have to insist that a (so-called) neutrino detection can at most point to the empirical adequacy of the neutrino hypothesis, never to its truth. But insofar as Fermi’s theory already provides overwhelming evidence for the empirical adequacy of the neutrino hypothesis, the epistemic impact of a neutrino detection would be quite insigni�cant. Similarly, an antirealist in the spirit of Stanford (2006) will deny that an experimental detection of the neutrino could justify belief in its existence, even if such an experiment managed to show that no other explanation (such as Bohr’s hypothesis of energy non-conservation) accounted for the relevant phenomena. This is due to the problem of unconceived alternatives (already encountered in section 1.2 and to be discussed in detail in chapter 6): Elimination of all but one hypothesis is not su�cient to justify belief in this last contender, since there are always other possible hypotheses, which are not excluded, because no one has even thought of them yet. It is not too surprising that antirealist positions tend to understate the importance of what physicists call a detection of some unobservable entity. But the same tendency is present in IBTE-based (or non-causal) versions of scienti�c realism, such as Psillos’s. We have seen that, gauging by theoretical virtues, the neutrino hypothesis provided by far the best explanation of the observed phenomena. True, Bohr’s alternative assumption remained a possible option, but since it contradicted Fermi’s theory and did not allow for any explanations or predictions over and above the ones for which it had been introduced, there was no sense in which it could claim to be an equally good explanation of the phenomena as the neutrino hypothesis. In other words, the neutrino hypothesis had as much theoretical warrant as anyone could wish for. But then, if no distinction between the strength of theoretical and causal warrant is drawn, the overwhelming (theoretical) warrant in favor of the neutrino hypothesis already put the existence of the neutrino beyond reasonable doubt, rendering its direct experimental detection super�uous. Due to this consequence, non-causal realism is at odds with the convictions of a signi�cant part of the scienti�c community, including the Nobel Committee, who awarded the 1995 Physics Price to Frederick Reines “for the detection of the neutrino”. This by itself is, of course, not a telling argument against non-causal realism; physicists may be mistaken about what is scienti�cally important and what

72 � Causal Warrant for the Neutrino: A Case Study is not.� But if that were the case, then we would have to advocate major changes in the present and future priorities of particle physics, and this is in general not what scienti�c realists do. The realist might try to bring his position into line with scienti�c practice by saying that the neutrino’s detection is an important test, even though there already is high con�dence in the correctness of the neutrino hypothesis, due to the success of Fermi’s theory. Surprises, he might say, are always possible. But this amounts to conceding that the best explanation may not be good enough after all, and it invites the antirealist’s question as to why this speci�c test (in contrast to all the previous tests) is supposed to take us from a state in which we are still prepared for surprises to a state of �rm belief in the truth of the neutrino hypothesis. I will argue in the next section that the concept of causal warrant enables the causal realist to give a convincing answer to this question. But let me �rst dismiss a possible attempt to answer the question without relying on causal warrant. The non-causal realist might say that the signi�cance of the neutrino’s detection lies in the fact that it provides Fermi’s theory with novel predictive success, while the theory could previously only boast explanatory success. This is a natural suggestion, given the centrality of novel predictions in contemporary arguments for scienti�c realism (see, e.g., Psillos 1999, 100–102; Alai 2014; see also section 6.1 below). However, Fermi’s theory had made successful novel predictions long before the neutrino was detected. For example, it predicted the possibility of electron capture, a process in which a proton in a nucleus absorbs one of the orbital electrons, forming a neutron and a neutrino. This process was subsequently detected by Luis Alvarez (1937). Therefore, we will have to look for something else than just predictive success, if we want to grasp the signi�cance of the neutrino detection experiments.

�.� The Detection of the Neutrino by Reines and Cowan In the mid-1950’s Frederick Reines and Clyde Cowan performed a series of experiments by which they succeeded in directly detecting the neutrino. It may, of course, not be obvious what “directly” means in this context; neutrinos are neither visible themselves nor do they produce visible tracks in particle detectors. All that can be detected are the products of so-called inverse beta decay, a process in

8 Post (1975, 25) seems to take this view concerning the detection of the neutrino.

The Detection of the Neutrino by Reines and Cowan

� 73

which a proton absorbs a neutrino� and turns into a neutron, emitting a positron: ν¯ + p → β+ + n. The cross section for this reaction is exceedingly small (around 10−�� cm� ), so a high neutrino �ux is needed for the detection to be successful. Since neutrinos are generated in beta decays, and these occur most frequently in the products of nuclear �ssion, it is natural to look for inverse beta decay in the neighborhood of a nuclear reactor, and that is where Reines and Cowan placed their detector.

Fig. 5.1. Schematic diagram of inverse beta decay detection. (Reprinted �gure with permission from Reines et al. 1960, 159. © 1960 by the American Physical Society.)

The detector consisted of a container �lled with water and cadmium chloride, which was sandwiched between two liquid scintillation detectors, surrounded by photomultipliers (see �gure 5.1). If a neutrino, coming from the reactor, initiates an inverse beta decay in the water tank, the emitted positron is almost immediately annihilated, and the two gamma rays originating from this event produce

9 More precisely, an antineutrino, but it is customary to neglect the di�erence between neutrino and antineutrino in this context.

74 � Causal Warrant for the Neutrino: A Case Study a prompt signal in the liquid scintillators. The produced neutron loses part of its energy via multiple collisions with protons in H� O and is then captured by a cadmium nucleus. This in turn generates gamma rays, which are registered as a delayed signal in the scintillation detectors. In sum, the signature for a neutrino event is given by two gamma signals, temporally separated by a few microseconds (see �gure 5.2).

Fig. 5.2. Candidate for a neutrino event. The two small peaks on the left stem from the gamma rays produced by positron annihilation (recorded simultaneously in the upper and the lower liquid scintillator). The large peaks on the right indicate the neutron capture event. (Reprinted �gure with permission from Reines et al. 1960, 162. © 1960 by the American Physical Society.)

In the main measurement series, around 4.5 events with this signature were recorded per hour.¹� By observing how this counting rate responded to the manipulation of various parameters (e.g.: switching o� the nuclear reactor, changing the time frame for the gamma coincidence, or adding more shielding material), Reines and Cowan determined which part of this counting rate was due to background e�ects. The remaining signal rate was �.�±�.� neutrino events per hour.¹¹

10 Cf. table II in Reines et al. (1960, 164). The detector used in this experiment consisted of two overlapping parts, called the “top triad” and the “bottom triad”. The value of 4.5 events per hour results from adding the rate measured in the top triad (919 counts in 379.1 hours) to the event rate in the bottom triad (815 counts in 383.5 hours). 11 More precisely: �.�� ± �.�� events per hour in the top triad, �.�� ± �.�� events per hour in the bottom triad (Reines et al. 1960, 164).

The Detection of the Neutrino by Reines and Cowan � 75

Experimentally Eliminating Redundancy So why was this a more direct proof of the neutrino’s existence than earlier experiments? What exactly did Reines and Cowan achieve that was not achieved before? Given our previous discussion, it is natural to turn to criterion 1 and suspect that Reines and Cowan managed to eliminate redundancy by somehow excluding Bohr’s hypothesis of energy non-conservation. And indeed, this hypothesis plays an important role in the assessment of their experiment, but it is a more delicate role than simply that of providing an alternative to the neutrino hypothesis. To see this, let us look at an extract from an article by Bruno Pontecorvo ([1946] 2000). He �rst summarizes the explanatory situation concerning beta-decay: We see that the fundamental facts can be reconciled only with one of the following alternative assumptions: (1) The law of the conservation of the energy does not hold in a single β process. (2) The law of the conservation of the energy is valid, but a new hypothetical particle, undetectable in any calorimetric measurement—the neutrino—is emitted together with a β-particle in a β transition in such a way that the energy available in such a transition is shared between the electron and the neutrino. (Pontecorvo [1946] 2000, 23)

Then, after discussing a number of neutrino experiments, in particular some that included measurements of the recoil energy of the nucleus, Pontecorvo comments: It should be noticed that experiments of this type, while of fundamental signi�cance in the understanding of the β process, cannot bring decisive direct evidence on the basic assumption of the existence of the neutrino. This statement can be understood if we keep in mind that recoil experiments are interpreted on the basis of the laws of the conservation of the energy and momentum in individual β processes, i.e., on the basis of alternative (2), which, in e�ect, corresponds essentially to the assumption of the existence of the neutrino. Direct proof of the existence of the neutrino must, consequently, be based on experiments, the interpretation of which does not require the law of conservation of energy, i.e., on experiments in which some characteristic process produced by free neutrinos (a process produced by neutrinos after they have been emitted in a β disintegration) is observed. (24)

This is precisely what was done in Reines’s and Cowan’s experiment. What marks this experiment out as a direct detection is therefore not so much that it eliminates Bohr’s hypothesis (1) as an alternative to the neutrino hypothesis (2), but that it provides evidence for (2) that does not itself presuppose the negation of (1).¹²

12 This is a slightly di�erent formulation of non-redundancy than the one given in section 4.1, and it has the advantage of describing more accurately what Reines and Cowan actually achieved.

76 � Causal Warrant for the Neutrino: A Case Study The Importance of a Causal Link But does the interpretation of the Reines-Cowan experiment really no longer depend on the validity of energy conservation? A diehard critic of (2) will deny this, claiming that the experiment can only imply the existence of the neutrino under the assumption that energy is conserved in inverse beta decay. Without this assumption, we should expect inverse beta decay to occur spontaneously, not needing a neutrino that produces it.¹³ Reines and Cowan would then simply have observed the process p → β+ + n, mistaking it for a neutrino-induced e�ect. To counter this critique, it is important to remember one key aspect of the experiment, namely that the neutrinos which were supposed to generate inverse beta decays in the detector originated from a nuclear reactor. Now the interesting thing about a nuclear reactor is that it enables the experimenter to see if the event rate for inverse beta-decay changes when the reactor is turned on or o�. And indeed, the signal rate was signi�cantly higher when the reactor was operating than when it was shut down. It is here that the evidence for the neutrino becomes compelling, for the critic can only maintain his position by choosing one of the following options: 1. Deny that the correlation between the reactor status and the event rate suggests a causal link between the two. 2. Insist that the causal link between the reactor and the detector is mediated by something other than neutrinos. This may either be (a) some familiar objects, such as gamma rays or neutrons, or (b) some hitherto unknown entities. Option 1 is a variant of the well-known phrase that “correlation does not imply causation”, but to invoke it here is to overlook the fact that Reines and Cowan did not just observe a correlation between reactor status and event rate, but rather found that turning on the reactor is an e�ective strategy to increase the rate of inverse beta decay. Or, to use the terminology of Woodward (2003, 45, 98), turning on the reactor is an intervention on the reactor status which changes the value of the event rate. If one were to deny that this establishes a causal link, one would also have to be skeptical about causation in everyday situations, for example that there is a causal link between regular toothbrushing and increased oral health. However, the two formulations are epistemically equivalent: independent evidence for the neutrino hypothesis amounts to evidence against the competing hypothesis. 13 This process would involve a creation of energy in just the same way as ordinary beta decay would—according to hypothesis (1)—involve a destruction of energy. Bohr had originally thought that this might explain the production of stellar heat (Jensen 2000, 149).

The Detection of the Neutrino by Reines and Cowan � 77

Such a level of skepticism would take us beyond the debate on scienti�c realism in much the same way as hypothesis H ′ discussed in section 4.1. Option 1 is therefore not to be taken seriously in the present context. Option 2a, by contrast, has to be taken very seriously, and Reines and Cowan indeed took it so. They performed various tests to exclude any in�uence of known reactor products on the detector (see Reines et al. 1960). It is safe to say that they succeeded in excluding option 2a. This leaves us with option 2b. Here we should �rst note that the critic can not just postulate a new entity (let us say, a meutrino) which has all the properties of the neutrino, but is not a neutrino. Given that there is nothing that makes a neutrino a neutrino apart from its properties, such a strategy would merely amount to renaming the neutrino.¹� The critic thus has to make clear how the meutrino di�ers from the neutrino. But however he chooses to do that, the meutrino will have to share some properties with the neutrino, in order to explain the correlation between reactor status and event rate. In particular, the meutrino will have to be produced in the reactor and possess the ability to travel to the detector and produce an inverse beta decay there. So even if we were to admit that Reines and Cowan might have detected meutrinos instead of neutrinos, it would still not have been the case that inverse beta decay occurred spontaneously, violating energy conservation. In sum, we see that the evidence in favor of the neutrino hypothesis provided by the Reines-Cowan experiment does not depend on the validity of conservation laws at the nuclear level. Furthermore, it is clear that Bohr’s hypothesis now fails with respect to criterion 3, because it cannot explain the correlation between the status of the nuclear reactor and the rate of inverse beta decays in the detector. This implies that the neutrino hypothesis now ful�lls criterion 1 and therefore, given that criterion 2 is still unproblematic, the hypothesis is now causally warranted. Finally, notice that the neutrino’s properties no longer just manifest themselves in the context of its creation (in beta decay), but also in the context of a second process that would not have occurred without the neutrino’s presence (inverse beta decay). This dispels the worry (from section 5.1) that causal warrant may just extend to the properties of the neutrino, but not to the underlying object (the neutrino itself). It does not, of course, resolve the metaphysical question what objects are; they might be nothing more than bundles of properties. But some bond between the di�erent neutrino-like properties is necessary in order to explain the

14 A similar argument (regarding the case of electrons instead of neutrinos) is given by Clarke (2001, 720).

78 � Causal Warrant for the Neutrino: A Case Study causal link that Reines and Cowan found. Something must have traveled from the reactor to the detector, and this is what we call neutrino. The main goal of this chapter has been to illustrate and substantiate the distinction between causal and theoretical warrant. But of course, an argument for the strength of causal warrant relative to theoretical warrant is not yet an argument for the claim that causal warrant is strong enough to withstand the antirealist challenges. Therefore, the next chapter will apply the notion of causal warrant in a response to what I take to be the strongest contemporary argument against realism.

6 The Problem of Unconceived Alternatives Kyle Stanford (2006) argues against scienti�c realism by claiming that much of our scienti�c knowledge is subject to what he calls the problem of unconceived alternatives (PUA). The importance of this problem lies in its detrimental e�ect on eliminative inferences, that is, inferences which proceed by formulating di�erent hypotheses, testing them and, if only one of them survives all the tests, concluding that it must be true. As Stanford notes, such a procedure is only reliable “when we can be reasonably sure that we have considered all of the most likely, plausible, or reasonable alternatives before we proceed to eliminate all but one of them” (2006, 29). Therefore: (PUA) Many eliminative inferences in science are unreliable, because there are plausible and su�ciently distinct alternative hypotheses that are not taken into consideration. This formulation of the PUA is intended to show that Stanford is not attempting to undermine eliminative inference in general, but only some instances of it. The interesting question then is, of course, which instances these are. Stanford’s (2006, 33) answer is that the PUA threatens “our e�orts to theorize about the most fundamental aspects of the constitution and dynamics of the various domains of nature”. In order to argue for this claim, he identi�es in the historical record of fundamental theorizing a pattern of recurrent, transient underdetermination (RTU) and then performs what he calls a new induction (NI)¹ so as to generalize from past theorizing to all (including present) theorizing. In more detail, the argument goes as follows: (RTU) Past theorists failed to consider relevant and radically distinct alternatives to their own theories, namely the theories that came to be accepted in the course of later inquiry. These theories were plausible, since they were at least as well con�rmed by the available evidence as the previous theories. This shows that past theorists were subject to the PUA in their fundamental theorizing. (NI) Since present theorists are not relevantly di�erent from past theorists, it follows by induction that they, too, are subject to the PUA in their fundamental theorizing.

1 By calling his argument a “new induction”, Stanford emphasizes the contrast (but also the connection) between his argument and the pessimistic (meta-)induction usually attributed to Laudan (1981; see my discussion in section 1.1). For a detailed analysis of how the NI di�ers from the “old” pessimistic induction, see Magnus (2010).

80 � The Problem of Unconceived Alternatives The immediate consequence of (NI) is that realism about the claims of present fundamental theories should be rejected insofar as these claims result from eliminative inferences. Stanford (2006, chap. 8) thus advocates an epistemic instrumentalism with respect to these claims. Equipped with this rough sketch of Stanford’s argument, I will now give an overview of the most important objections raised by his critics, grouping them according to which link of Stanford’s argumentative chain they are attacking. I cannot discuss these objections in detail, but I will highlight some of their shortcomings, in order to show that more work is required to refute Stanford’s argument. Section 6.2 then attempts to do the necessary work by �eshing out a response to Stanford’s challenge, based on causal realism. Finally, section 6.3 applies these considerations to a concrete case from the history of physics, on which Stanford has commented in some detail: Jean Perrin’s work on Brownian motion and the atomic hypothesis.

�.� Previous Attempts to Undermine the PUA Realist criticisms of Stanford’s reasoning fall into three camps. The �rst denies that the historical cases of unconceived alternatives mentioned in (RTU) are plausible in the relevant sense, the second denies that these alternatives are su�ciently distinct from the theories available at the time, and the third seeks to block the induction described in (NI) by pointing out di�erences between the past and the present development of science. I will discuss these three camps in turn, arguing that none of them has so far produced a completely convincing response to the PUA.

The Plausibility of Unconceived Alternatives Unconceived alternatives can only generate a problem for scienti�c realism if they have some plausibility. Stanford (2006, sec. 1.2) is therefore very careful to distance the PUA from what he calls Cartesian fantasies, that is, far-fetched skeptical scenarios which, if taken seriously, would call into question any knowledge claim whatsoever. One line of criticism against Stanford is that the notion of plausibility employed in (RTU) threatens to collapse this distinction between serious alternatives and skeptical fantasies. To see how, notice �rst that it is not part of this notion that the alternative theories would have seemed plausible to the scientists who had failed to conceive of them: Although special relativity seems plausible to us, it is unlikely that 18th century physicists would have found it plausible,

Previous Attempts to Undermine the PUA �

81

had they conceived of it (Magnus 2006; 2010, sec. 4).² Stanford (2009a, 380–381) admits this, but he insists on calling such theories “plausible” simply because they actually came to be accepted by later scienti�c communities. But then, the critic continues, we should also regard some skeptical scenarios as “plausible” in this sense. For example, we presently �nd the possibility that all our sensory impressions are generated by a giant computer simulation wildly implausible, in the same way as Newtonian physicists would have found relativity theory implausible. But if someone in the future discovered that the simulation scenario was actually true, then we would have to conclude that it had been a serious possibility all along, hence it should count as “plausible” in Stanford’s sense (Magnus 2010, 810). If this were so, then Stanford’s cherished distinction between Cartesian fantasies and scienti�cally serious possibilities would collapse, and the realist could conclude that we need not worry about the PUA any more than about Cartesian skepticism. The problem with this objection is that it ignores the role of RTU in delimiting the scope of the PUA. The history of science shows that there were unconceived alternatives to past scienti�c theories and that they are plausible in the abovementioned sense of being accepted by a later scienti�c community. By contrast, the history of pondering on skeptical scenarios of the computer simulation type does not show any such development and therefore does not give us any reason to take these possibilities seriously. Therefore, the PUA is a distinctive challenge which should worry the scienti�c realist, quite apart from general skeptical considerations.

The Distinctness of Unconceived Alternatives The second way to attack the premise (RTU) is to deny that the alternative theories which past scientists failed to consider are in fact “radically distinct” from the theories they did consider. If there is su�cient continuity across scienti�c theories over time, then the existence of unconceived alternatives is no longer a threat to realism, because these alternatives, although unconceived, can be expected to be relevantly similar to the theories at hand. The PUA is then no longer a reason to doubt that these theories are at least approximately or partially true. Since realists have employed similar strategies to counter the conventional pessimistic (meta-)induction, Stanford devotes two chapters (2006, chaps. 6 and 7) to criticiz-

2 Indeed, it has been argued that special relativity was not just implausible, but even inconceivable for these physicists (French 2011b, sec. 2).

82 � The Problem of Unconceived Alternatives ing such attempts.³ His central claim is that the appeal to continuity is vacuous, because the realist can only retrospectively identify those parts or aspects of earlier theories that were retained in later theories, and consequently, this strategy “allows us to trust only some of what current theories tell us about the natural world . . . while leaving us completely unable to be con�dent in our ability to discern just which parts of our theories actually constitute this privileged class of theoretical claims” (2006, 153). In response, realists have insisted on our ability to identify the trustworthy parts not only of past theories, but also of our current theories. Our strategies for performing this di�cult task may not yet be perfect, but there is no principled reason to think that it cannot be done (Psillos 2009, sec. 4.2). Furthermore, continuity between successive theories does not necessarily require that some parts of the earlier theory are explicitly retained in the later theory, it is su�cient that the earlier theory survives as a limiting case of the later theory (En�eld 2008). But in either of these cases, the claim that it is in principle possible to identify some aspects of current theories which are likely to survive future theory change does not go to the heart of Stanford’s challenge. A convincing response to his critique of previous realist strategies must do more than simply assert that one can do better; it must show explicitly how one can do better. Several ways to do so have been proposed in the recent literature. One group of strategies seeks to identify those parts of theories which are in some sense responsible for their empirical successes, and claims that we can expect these to be retained in later theories. The focus here can be either on successful novel predictions (Saatsi 2009) or on the comparative success of a theory relative to its predecessor (Harker 2013). One problem for this approach is that the �xing of criteria for what is to count as genuine success leads to a dilemma, as Stanford (2009a, 384) points out: On the one hand, if the criteria are too strict, then much of contemporary science will be excluded from realist treatment. For example, exclusive concern with novel predictive success at the expense of explanatory success will result in antirealism about large parts of some scienti�c �elds, such as geology or evolutionary biology. On the other hand, even a very restricted realism seems to be confronted with some counterexamples, such as Poisson’s spectacularly success-

3 The fact that these realist strategies are equally e�ective against the (old) pessimistic induction as against the NI (if they are e�ective) has prompted Chakravartty (2008, 153) to call the NI “a novel red herring”. This verdict is justi�ed as long as one is only concerned with this particular type of realist strategy. However, as Magnus (2010, sec. 5) points out, there are other realist strategies which are e�ective against the pessimistic induction, but not against the NI.

Previous Attempts to Undermine the PUA �

83

ful novel prediction of a bright spot at the centre of the shadow of a circular disc, which rested on an assumption later abandoned, namely the ether hypothesis. Another problem for this type of strategy is that it is often hard to tell which theoretical assumptions are really responsible for a particular empirical success. For illustration, consider the example of Rutherford’s atomic model, as discussed by Harker (2013). Rutherford’s model di�ered from Thomson’s previous model in two respects, by postulating (i) a positively charged nucleus and (ii) a uniformly distributed negative charge surrounding it. Harker (2013, 80) admits that Rutherford used both of these assumptions “for purposes of explaining the scattering results”, which constituted the main empirical success of his model compared to Thomson’s. But surely we would not want to be realists about assumption (ii), so we need to claim that only assumption (i) was responsible for the model’s success. The challenge then is to distinguish in a principled way between those theoretical postulates which were responsible for success and those which were not. Unfortunately, Harker’s observation that “by Rutherford’s own admission, [assumption (ii)] was an idealization and in no sense con�rmed by the data” (91) does not go beyond Psillos’s (1999, chap. 6) earlier attempts to base such a distinction on the scientists’ own judgments, which Stanford (2006, sec. 7.3) has shown to be unreliable (see also Chang 2003). A �nal proposal on how to identify trustworthy theory parts uses the notion of minimally interpreted mathematical parts of theories (Votsis 2011). The guiding idea is that mathematical structure is often preserved in theory change, and an appropriate interpretation of that structure might yield a realism capable of withstanding the PUA. However, much hinges on what exactly one means by “minimal interpretation” here. Votsis (2011, 1229) characterizes it as “the bare minimum that needs to be assumed in order for there to be an appropriate inferential relation” from the equations to the empirical predictions. Without further quali�cation, this seems compatible with a strict instrumentalism, according to which theories are nothing more than useful tools for deriving empirical predictions. Hence, it is not immediately clear how the appeal to minimally interpreted mathematical parts of theories is supposed to support any kind of realism. The notion of a “minimal interpretation” was introduced by Anjan Chakravartty, who seems to acknowledge the problem I have just outlined, when he writes: Antirealists will have di�erent views here, regarding what constitutes an appropriate minimal interpretation, and consequently what is warranted. But my present concern is with realism. (Chakravartty 2007, 52)

84 � The Problem of Unconceived Alternatives So it seems that what the realist needs is not a minimal interpretation tout court, but a minimal realist interpretation. The realism of an interpretation consists, according to Chakravartty, in its commitment to certain detection properties, that is, properties which “are connected via causal processes to our instruments and other means of detection” (48). Such causal considerations are also at the heart of Chakravartty’s (2008) response to Stanford. I take this to be the most promising realist strategy, and it will be discussed in detail in section 6.2. But as we will see, the support for realism comes from this causal component, not from the minimal interpretation of mathematical structure.

The Induction from Past to Present Having discussed several arguments against the premise (RTU) , let us now consider the inductive step performed within (NI). The validity of such an inference from the history of science to our present situation depends on the assumption that the present is relevantly similar to the past. This assumption can be challenged in various ways. First, if the evidence supporting our present best theories is qualitatively better than the evidence that supported our best theories in the past, then an induction from past to present theories might be problematic. In other words, while an induction of the (NI) type may count as some kind of second-order evidence against the truth of present theories, such evidence needs to be balanced against the �rst-order evidence we have in support of these theories (Psillos 2009, sec. 4.4), and this evidence is, in some cases at least, massively more impressive than the evidence for the theories on which Stanford bases his NI (Votsis 2007). However, as a general strategy against the NI, this idea faces the dilemma familiar from the previous subsection: If the standard of �rst-order evidence required for realism is set too high, then much of present science will fail to meet that standard; if it is too low, then many theories now considered false will meet the standard, and the NI will not be blocked. Furthermore, pointing to di�erences between past and present theories is not a su�cient response to the NI, because the NI is, as Stanford (2006, sec. 2.3) emphasizes, an induction over theorists rather than theories. The second way of criticizing the NI takes this fact into account and argues that present scientists di�er relevantly from past scientists, due to improvements in scienti�c methodology (Roush 2010; Devitt 2011). It seems clear that there is indeed such a di�erence between the past and the present. What is less clear is whether the di�erence in scienti�c methodology bears on how well scientists are equipped to deal with the

Causal Knowledge as a Criterion for the Realist � 85

PUA.� If improvements in methodology just lead to more empirical success, but not to a better grasp of the space of alternative theories, then present theorists do not di�er relevantly from past theorists, as far as the PUA is concerned. Finally, one might argue that Stanford’s induction from past to present can be blocked even if it is agreed that present theorists are, individually, not relevantly di�erent from past theorists. This is because the scienti�c community is simply much larger now than it was in the past (Fahrbach 2011, see Wray 2013 for criticism), and because later scientists have had more time for their increasingly robust scienti�c traditions to uncover alternatives (Ruhmkor� 2011). But notice that these developments have been accompanied by other developments that might well have the opposite e�ect: It is at least plausible that the contemporary mechanisms of research funding lead to a decrease in the investigation of fundamentally distinct alternatives to dominant theories (Stanford 2006, 132). Therefore, although there is an undeniable di�erence between the scienti�c community in the present and the past, it is not self-evident that this di�erence undermines the NI.

�.� Causal Knowledge as a Criterion for the Realist As the discussion in the previous section has shown, the realist’s appeal to continuity across theory change as an argument against the PUA needs to be made precise in terms of a criterion that permits the identi�cation of parts of theories which are likely to be retained in the future. Furthermore, this criterion must be applicable prospectively, that is, it must not depend on knowing which parts were in fact retained in later theories. Chakravartty (2008, 155) proposes detailed causal knowledge as such a criterion, arguing that “if one has a detailed enough causal knowledge of something, knowledge that allows one to manipulate it in highly systematic ways, then there is no better warrant for knowledge”. This suggestion alone would not go a long way towards rebutting (NI), since Stanford has already dealt with realist attempts to capitalize on causal considerations. But as I will now try to show, Chakravartty’s proposal has the resources to withstand Stanford’s

4 While Devitt (2011) does not address this issue, because he thinks that the onus is on the antirealist to show that the past and the present are relevantly similar (290n11), Roush (2010, 55) explicitly connects improvements in methodology with the PUA. In support of the view that there is such a connection, she refers to her earlier treatment of Perrin’s work on the atomic hypothesis (Roush 2005, 218–223). I will discuss this example in section 6.3, where we will see that her approach is, by itself, not su�cient to establish the immunity of the atomic hypothesis with respect to the PUA.

86 � The Problem of Unconceived Alternatives criticisms. However, his account is rather sketchy, so it is necessary to look at the relevant passages of Stanford (2006) in some detail, in order to see why Stanford’s criticism of earlier causal strategies invoked by realists does not apply to Chakravartty’s strategy.

How Chakravartty’s Proposal Di�ers from Earlier Causal Strategies Stanford (2006, sec. 6.3) discusses several realist attempts to argue for successful reference of theoretical terms despite changing theoretical descriptions associated with these terms. Of interest for our purpose is his critique of Psillos’s causaldescriptivist account of reference, which argues for referential continuity across theory change by allowing for changes in theoretical descriptions as long as they leave the “core causal description” of a given entity intact. For example, Psillos (1999, 286) claims that there is referential stability between nineteenth and twentieth century electrodynamics in that “the term ‘luminiferous ether’ may be seen as referring to the electromagnetic �eld”, because “the core causal description associated with the term ‘electromagnetic �eld’ takes up the core causal description associated with the term ‘ether’ ”. Stanford’s (2006, 151) reply is as follows: Of course, this account of the matter invites the realist to choose the core causal descriptions she associates with the central terms of past theories rather carefully, with one eye on current theories’ claims about nature, so there is more than a whi� of ad hoc-ery about the proposal.

And further: This case for the referential status of central terms in successful past theories simply invites from the historical record a renewed form of the pessimistic induction itself, this time concerning our ability to distinguish (at the time a theory is a going concern) which of our beliefs about an entity are actually part of its core causal description. (152)

What Stanford has in mind here is the fact that past scientists have repeatedly included claims we would now consider utterly false in the core causal descriptions of their theoretical entities. As an example, he mentions Maxwell’s claim that the ether must be a material substance, because the energy transmitted by electromagnetic waves “cannot be contained in any vessel except the inmost substance of material things” (Maxwell [1873] 1955, 493; quoted in Stanford 2006, 152). Generalizing from such examples in the spirit of the pessimistic induction, Stanford concludes that “we cannot rely on our own judgments about which of the descriptions we associate with our own terms are genuinely part of their own core causal descriptions” (153).

Causal Knowledge as a Criterion for the Realist � 87

Returning now to Chakravartty’s proposal, it is tempting to identify what he calls “detailed causal knowledge of something” with Psillos’s “core causal description” (of something) and, consequently, to suspect that Stanford’s critique is equally damaging to both of these proposals. But this impression is misleading, as can be seen by attending to another important aspect of Chakravartty’s position: his insistence that scienti�c realism is “�rst and foremost a realism about well-con�rmed properties” (2008, 155). On this view, continuity between successive theories is ensured not by their referring to the same entities, but the same properties. The di�erence between the two accounts is quite subtle, because Psillos (1999, 293), too, emphasizes continuity at the level of properties. However, he identi�es them as “properties by virtue of which [the posited entities] play their ascribed causal role”. This invites Stanford’s complaint that we may only be able to single out which properties these are with the bene�t of hindsight. Taking again the above example, we nowadays know that the electromagnetic �eld can play its causal role without being a material substance, but as we have seen, Maxwell explicitly denied that this could be true of the ether, and Psillos does not give us a criterion on the basis of which Maxwell and his contemporaries could have thought otherwise.� By contrast, Chakravartty’s identi�cation of the properties to which realists should be committed makes no reference to the entities supposed to possess these properties. Realism, Chakravartty (2007, 47) argues, is primarily concerned with “causal properties one has managed to detect” (detection properties). On the other hand, realists should remain agnostic with regard to “any other putative properties attributed to particulars by theories”, which he calls auxiliary properties. The question of how to demarcate detection properties from auxiliary properties will be discussed below. For the moment, I just note that one does not need to rely on retrospective judgement to perform that task. For example, I will argue that even at the time of Maxwell, physicists could have classi�ed the amplitudes or frequencies of (what they took to be) ether waves as detection properties, while recognizing the ether’s materiality as an auxiliary property. If this is so, then Stanford’s argument against our ability to reliably �x the core causal descriptions of our theoretical entities does not translate into an argument against our ability to tell detection properties from auxiliary properties. This re�nement of realism is also e�ective against a second objection raised by Stanford to the realist’s use of causal reasoning. Having criticized Philip

5 When Psillos (1999, 139) discusses Maxwell’s “di�erentiated attitude towards the several parts of a theory in view of what evidence supports them”, he may have in mind something similar to the criterion I will discuss in the following. But his failure to state it explicitly (or to provide any other criterion) invites Stanford’s charge of rationalization post hoc.

88 � The Problem of Unconceived Alternatives Kitcher’s (1993, sec. 5.4; 2001, sec. 5) distinction between “working posits” and “presuppositional posits”, Stanford (2006, sec. 7.2) considers a possible improvement of this distinction on behalf of the realist, to the e�ect that realists might want to commit themselves to the existence of only those posits to which theories ascribe direct causal roles, treating other posits as merely presuppositional (or “idle”). He takes a dim view of this idea: Although perfectly natural, this suggestion seems to run afoul of any number of discarded theoretical posits that were ascribed direct causal roles in the production of phenomena by the successful explanatory practices of their respective theories. To such familiar examples as phlogiston and caloric �uid we’ve seen that we might fairly add Darwin’s gemmules, Galton’s stirp, and Weismann’s biophors. (Stanford 2006, 172)

A realist who has taken Chakravartty’s lesson to heart need not be embarrassed by Stanford’s list of theoretical entities which were once believed to play a causal role but were subsequently abandoned. As we have seen above, there may well be discontinuity on the level of entities, but continuity on the level of properties. Thus, a sophisticated realist does not need to (implausibly) deny that, for example, past theories indeed ascribed a direct causal role to caloric. And he can even admit that Hasok Chang (2003) has refuted Psillos’s (1999, 130) claim that “caloric” was not a central term (see also Stanford 2006, 175–179). But he can maintain that the properties which were ascribed to caloric and which we now regard as mistaken were just auxiliary properties, while a case can be made for the retention of some detection properties which appeared in the caloric theory, such as the speci�c heat of air (c p and c v ). Chakravartty’s focus on properties instead of entities might make him vulnerable to another one of Stanford’s arguments, a variant of which we have already encountered in section 6.1. This new strategy, the objection goes, secures nothing more than a pyrrhic victory for the realist, since a realism which forfeits its commitment to theoretical entities, settling for the mere commitment to (some) properties, does no longer deserve to be called realism. However, the fact that properties are the primary focus of the realist’s response to (NI) does not imply that his commitment is limited to properties alone. Once the reality of certain properties is established, the realist can argue for the existence of entities, based on the fact that properties often cohere to form interesting units (Chakravartty 2007, 63– 66). We have already seen an example of this at the end of chapter 5: Belief in the neutrino is grounded on the coherent manifestation of neutrino-like properties in di�erent contexts. Another way of performing the step from properties to entities will be discussed in section 6.3.

Causal Knowledge as a Criterion for the Realist � 89

Causal Realism and the Detection/Auxiliary Distinction The analysis so far has shown that Stanford’s arguments against the realist’s appeal to causal knowledge do not, by themselves, threaten Chakravartty’s proposal. On the other hand, it has also shown that the viability of this new causal strategy depends crucially on our ability to tell detection properties from auxiliary properties “at the time a theory is a going concern”, as Stanford puts it. This necessitates a closer look at Chakravartty’s (2007, 47–54) account of the demarcation between detection and auxiliary properties. Here is the gist of his suggestion: Detection properties are connected via causal processes to our instruments and other means of detection. One generally describes these processes in terms of mathematical equations that are or can be interpreted as describing the relations of properties. As I will attempt to show, one can thus identify detection properties as those that are required to give a minimal interpretation of these sorts of equations. (48)

But, as discussed in section 6.1, the problem with the notion of “minimal interpretation” is that the corresponding commitment depends crucially on which reading of “minimal” is applied. Without further quali�cation, Chakravartty’s (2007, 48) advice to treat as auxiliary anything that “goes beyond what is minimally required to do the work of science: to make predictions, retrodictions, and so on” just results in instrumentalism. To be of any use for the realist, “minimal interpretation” thus needs to be read as “minimal realist interpretation”. Chakravartty makes this reading explicit in his discussion of Fresnel’s equations: “The existence of certain properties is minimally required to give a realist interpretation of these equations” (49, emphasis added). Unfortunately, Chakravartty does not o�er an explication of the appropriate (realist) understanding of minimality that does not itself depend on having already identi�ed some properties as indispensable. This makes his claim that “one can . . . identify detection properties as those that are required to give a minimal interpretation” somewhat misleading: It is not the case that we can approach the relevant equations equipped with a well-understood notion of “minimal interpretation” in order to learn which of the properties described by them are detection properties. Rather, it seems to be the other way round: Only once we know which the detection properties are, are we in a position to give an appropriate minimal (realist) interpretation of the equations in question. This is not to say that Chakravartty’s characterization of detection properties is unilluminating, but its success does not depend on identifying them by means of a minimal interpretation. In fact, Chakravartty seems to elucidate both concepts (detection property and minimal interpretation) with the help of a prior notion of causal connection or causal contact. Not only does he (as seen in the above quotation), characterize detection properties as the ones which are causally connected

90 � The Problem of Unconceived Alternatives to our instruments, but he also summarizes his “recipe of the minimal interpretation” as a commitment “only to structures with which one has forged some significant causal contact” (54). Of course, this only pushes the problem one step back. In order to demarcate detection properties from auxiliary properties, we now have to know what counts as signi�cant causal contact. It is at this point that Chakravartty’s account can pro�t from the considerations of causal realism, as I have developed them in chapter 4. More precisely, I suggest that my account of causal warrant adequately captures and explicates the intuitions behind Chakravartty’s detection/auxiliary distinction. By providing such an explication, causal realism turns Chakravartty’s (2008) sketch into a robust strategy against Stanford’s NI. In footnote 8 of section 4.2, I already hinted at the basic idea of how causal warrant connects to the concept of a detection property: A detection property does not only ful�ll the criterion of material inference, but all three criteria associated with causal warrant. Put more simply, a detection property is one for which we have causal warrant. In that same footnote, I used the term “detectable properties” as a synonym for “material properties”. This synonymy is justi�ed by the fact that the criterion of material inference, as discussed in section 4.2, explicates with what kind of properties we can in principle establish some causal contact. Whether the causal contact is in fact established (in other words: whether the detectable property becomes a detection property) then depends on the other two criteria, nonredundancy and empirical adequacy. This matches very well with Chakravartty’s account: Empirical adequacy captures his idea that detection properties allow us to make predictions, retrodictions and so on, while non-redundancy expresses the belief that these properties are indispensable to performing these tasks.� However, Stanford’s challenge creates a problem for this characterization, because it threatens the applicability of the criterion of non-redundancy. To take the PUA seriously is to acknowledge that even if a certain hypothesis seems to be the only one which can account for some phenomena, we may not be justi�ed in asserting that this is really so. As a consequence, if non-redundancy were understood in the strong sense of there being no other empirically adequate hypothesis at all (whether known or unknown), then we could never claim any causal warrant in cases in which the PUA is a matter of concern, because we could never

6 The foregoing considerations concerning the “minimal interpretation” may have raised the worry that the criterion of non-redundancy can never actually be ful�lled, because there is always the alternative to choose an instrumentalist interpretation of equations instead of a realist one, and both options allow us to make the same predictions, retrodictions etc. However, this is a kind of redundancy which, as I have argued in section 4.1, is not a proper subject for the debate on scienti�c realism, because it is just as damaging to common sense realism as to scienti�c realism (see also Stanford’s (2001, S3–S4) comments on the “Craigian reduction” of a theory).

Causal Knowledge as a Criterion for the Realist � 91

justi�ably claim that the requirement of non-redundancy has been met. Therefore, if causal realism is to do any work in solving the PUA, non-redundancy has to be understood in the weaker sense of there being no other known hypothesis that accounts for the phenomena. Distinguishing these two senses of nonredundancy allows us to reformulate Chakravartty’s claim that our knowledge of detection properties is immune to the PUA in the following way: Hypotheses that are causally warranted, which is to say that they conform to the criteria of nonredundancy (in the weak sense), material inference and empirical adequacy, are likely to be non-redundant in the strong sense. To see how this explication improves on Chakravartty’s proposal, let us return to the example of ether waves introduced above. The challenge is to show how one can recognize the amplitude of such a wave as a detection property and the substantiality of the corresponding medium as an auxiliary property, without yet knowing about the former’s retention and the latter’s rejection in the subsequent theory. Working with Chakravartty’s de�nition, it is not easy to see how this could be done, because there is a sense in which the ether’s substantiality is just as much “connected via causal processes to our instruments” as the amplitudes of ether waves. More precisely, if one presupposes Maxwell’s belief (quoted above) that no energy can be transferred without a substantial medium, then we actually detect the ether’s substantiality whenever we detect the energy of an electromagnetic wave. The reason why the ether’s substantiality should (unlike the amplitude of an ether wave) still not count as a detection property becomes clear once we turn to the concepts of causal realism, in particular the criterion of material inference. The criterion is certainly ful�lled in the case of a light wave’s amplitude, because there is a well-de�ned notion of what it means to modify that property. There are well-known experimental procedures for this, and even in cases where experimental intervention is not possible in practice, it is perfectly clear what it would mean to intervene on the amplitude of an electromagnetic wave. By contrast, no such procedures or well-de�ned notions exist in the case of the ether’s substantiality. Even if one accepts Maxwell’s presupposition connecting substantiality to the capacity for energy transfer, it is not clear what it would mean to intervene on that property, and hence, the corresponding hypothesis does not meet the requirement of material inference. If, on the other hand, one insists that there is a well-de�ned notion of modifying substantiality independently of the capacity for energy transfer (after all, we now know that energy transfer can occur with or without a substantial medium), then the possibility of an electromagnetic �eld without a substantial ether is no longer an unconceived alternative but a known one, and hence, non-redundancy fails even in the weak sense mentioned above. In either case, it follows that there is no causal warrant for the ether’s substantiality, which therefore fails to be a detection property.

92 � The Problem of Unconceived Alternatives This shows that, by drawing on the conceptual resources of causal realism, the distinction between detection properties and auxiliary properties can be made su�ciently precise to underlie a realism with the prospect of withstanding Stanford’s argument from unconceived alternatives. The extent to which this prospect is realized will depend on how well this account �ts the historical record of scienti�c reasoning, hence the need for detailed case studies as the one presented in the next section. What has been shown so far is that none of the arguments by which Stanford attacked earlier versions of realism are e�ective against the causal strategy described in this section, such that there is (until further notice) no reason to worry about unconceived alternatives to causally warranted claims. Furthermore, it seems unlikely that Stanford’s arguments can be modi�ed so as to demonstrate causal realism’s vulnerability to the PUA. The reason for this is that, as discussed in chapter 4, causal warrant is closely associated with causal explanations of experimental phenomena, rather than with fundamental theorizing. Now admittedly, it would not be correct to say that unconceived alternatives are only a problem for fundamental theorizing, not for explaining experimental results. Indeed, experimenters regularly grapple with questions such as: Have we taken into account all external in�uences on the apparatus? Did nothing go wrong in the calibration process? Is the signal more than just an artefact of our sophisticated data analysis? In dealing with these questions, it happens that relevant alternative explanations of the phenomena are overlooked. But the historical record of these unconceived alternatives does not possess the right kind of structure to support Stanford’s NI. Unconceived alternatives appear here and there in the history of scienti�c experimentation, but there is nothing like the systematic pattern described in (RTU), and accordingly, there is no analogue to (NI) which could call into question the scientists’ ability to reliably identify the causes of experimental phenomena.

�.� Causal Realism, Unconceived Alternatives, and the Atomic Hypothesis The previous section ended with a claim about the history of experimental science, and such a claim can of course be disputed on historical grounds. In fact, Stanford (2009b) can be interpreted as doing just that. Discussing Jean Perrin’s experimental work on Brownian motion and the associated case for the existence of atoms and molecules, Stanford argues that Perrin’s experimental genius did not prevent him from being subject to the PUA. The present section will address this objection. Analyzing the case of Perrin will also allow me to connect the rather

Causal Realism, Unconceived Alternatives, and the Atomic Hypothesis �

93

abstract considerations of the previous section with actual scienti�c practice. In particular, this historical example will illustrate three central elements of the causal realist’s response to the PUA: the distinction between detection and auxiliary properties, the di�erence between fundamental theorizing and the causal explanation of phenomena, and the inferential step from detection properties to the reality of unobservable entities.

Perrin and the Philosophers: Some Initial Observations Brownian motion is an irregular movement of microscopic particles suspended in a �uid, discovered by the botanist Robert Brown in 1827. Perrin was by no means the �rst to interpret this phenomenon as an e�ect of the constant movement of the molecules which constitute the �uid, but there is a wide consensus that the unprecedented accuracy of his experiments on Brownian motion, carried out between 1908 and 1911, played a crucial role in establishing that atoms and molecules actually exist.� However, there is considerable (and ongoing) disagreement about why Perrin’s work played such a crucial role. The proposed answer which concerns us here is the one given by Sherrilyn Roush (2005), but before discussing this approach, some remarks about other philosophical treatments of this historical episode will help to set the stage. The philosophical literature on Perrin’s work is so comprehensive that I will not try to give a complete summary here. I merely wish to discuss the part of the debate which is most closely related to causal realism, namely the arguments surrounding Cartwright’s (1983, 82–85) and Wesley Salmon’s (1984, 213–227) reconstruction of Perrin’s reasoning in terms of a causal inference. The two accounts are not identical (Cartwright sees Perrin’s argument as an instance of a more general inference scheme, namely inference to the most likely cause, while Salmon focuses on the common cause structure of Perrin’s argument), but they are closely enough related to be treated together in the present context. The central idea in both treatments is that the atomic hypothesis causally explains a wide variety of very di�erent physical phenomena and that it would be an incredible coincidence if it managed to do so without being true. Even though this approach has been widely criticized, it still seems essentially

7 It is not immediately clear whether Stanford (2009b) departs from this consensus or merely criticizes one particular way of characterizing Perrin’s achievement. I will return to this question at the end of this section. For explicit criticism of the consensus, see van Fraassen (2009), but also the response by Chalmers (2011).

94 � The Problem of Unconceived Alternatives correct to me. I will therefore brie�y address the di�erent lines of criticism that have been directed against it. Deborah Mayo (1986) argues that Cartwright misses an important element when she takes Perrin to infer the reality of molecules from the remarkable agreement in the estimates for Avogadro’s number based on entirely di�erent physical phenomena. Before making such an “argument from coincidence”, Perrin had to check what Mayo (1986, 49) calls “the experiment’s internal validity”. For the study of Brownian motion, this meant to demonstrate that the movement of a Brownian particle was indeed a random process, not depending on the movement of neighboring particles. This is the same element of Perrin’s argumentative chain that Roush (2005) focuses on, and my discussion of Roush’s position will show that causal realism can readily accommodate this element. Indeed, it seems that Mayo herself sees her approach not so much as contradicting, but as re�ning Cartwright’s analysis (see also Mayo 1996, 216–217). A second line of criticism is that the causal inference which Cartwright and Salmon take to be the essence of Perrin’s reasoning can only assure that the phenomena in question have a (common) cause, but not what that cause is (Achinstein 2001, 250;� Psillos 2011, 358n14). This critique is reminiscent of a well-known objection against entity realism (see section 2.3). The point I made in section 4.2, namely that causal realism is not only committed to certain entities, but also to some of their properties, will be further illustrated below in response to Roush’s and Stanford’s account of Perrin’s achievement. This will show that causal realism provides a substantial (though not exhaustive) characterization of what the cause of Brownian motion is. Finally, it has been argued that Cartwright and Salmon cannot explain why Perrin’s work was such a decisive step in convincing the scienti�c community of the reality of atoms, since the results they take to be central seem to have been available well before Perrin took on the issue (Maddy 2007, 404; Psillos 2011, 358n14). As mentioned above, I accept that the argument from coincidence, which Cartwright and Salmon emphasize, needs to be complemented by Mayo’s insight about Perrin’s demonstration of the randomness of Brownian motion. Therefore, I agree with this critique to the extent that the Cartwright–Salmon account is not quite su�cient to fully appreciate the importance of Perrin’s work. On the other hand, the alternative accounts put forth by Maddy and Psillos strike me as even less convincing explanations of Perrin’s historical importance. Maddy (2007, 406) observes that “before Perrin’s successes, the case for the existence of atoms had

8 Interestingly, this criticism does not appear in Achinstein’s (2002) paper dedicated to Salmon’s argument.

Causal Realism, Unconceived Alternatives, and the Atomic Hypothesis �

95

hinged on aggregate behavior” and claims that “Perrin’s accomplishment was to establish a link to [the] behavior of individual molecules”. But this last claim is not true in any but a very loose sense, given that a Brownian particle undergoes some ���� collisions with molecules of the surrounding �uid per second (Chandrasekhar 1943, 23). Psillos (2011, sec. 3.3) reconstructs Perrin’s reasoning by means of a probabilistic argument, but it remains unclear why Perrin’s experiments should have been crucial for establishing any of the argument’s three premises. The �rst premise states that the likelihood ratio P(n = N | −AH)/P(n = N | AH) (where AH stands for the atomic hypothesis and n = N for the claim that the number of molecules in a mole is equal to Avogadro’s number N) is very small. This follows from Perrin’s theoretical model and does not depend on any of his experimental results. By contrast, the second premise does report an experimental result, namely that n = N is the case. But if, as Psillos claims, “most of the ways to calculate Avogadro’s number were known (and they were in agreement) before Perrin brought them together in his books” (358n14), then establishing this premise was not a new achievement. The same is true for the third premise, stating that P(AH) is “not very low”. According to Psillos, Perrin evaluated this prior probability “by eliminating several alternative potential explanations of Brownian movement” (357), but clearly, others had done this before him (Nye 1972, 21– 28).

Roush and Stanford on Perrin Building upon Mayo’s ideas, Roush takes the decisive element of Perrin’s research to be his demonstration that Brownian motion is completely random, exhibiting “no systematic e�ects, no dependencies or correlations between the motions of one particle and another or tendencies in the motion of a single particle”. Roush goes on to claim that, by demonstrating the complete randomness of Brownian motion, Perrin con�rmed what she calls “the modest atomic hypothesis”, that is, the hypothesis “that there are atoms and molecules, understood merely as spatially discrete sub-microscopic entities moving independently of each other, i.e., at random” (Roush 2005, 219). This is the claim that Stanford attacks in his (2009b) paper. The disagreement between Roush and Stanford is illuminating, because, from the perspective of causal realism, both of them are partly right, but also partly wrong about this case. Let us see why. Unsurprisingly, Stanford takes issue with Roush’s (2005, 219) claim that “there do not seem to be any hypotheses that could

96 � The Problem of Unconceived Alternatives explain a random walk in the Brownian particles that are not included within this [modest] atomic hypothesis”. He cites two hypotheses as counterexamples, taken from Fine (1991, 91): Instead of being caused by molecular collisions, Brownian motion could be due to electrostatic forces among the particles themselves (in conjunction with exchange forces with the medium) or it could be uncaused, a manifestation of some fundamental randomness in nature. I do not think that the �rst example really supports Stanford’s argument, because any causal in�uence of electrostatic forces among the Brownian particles on their movement would result in some correlation between the movements of nearby particles, thus contradicting Perrin’s conclusion that Brownian motion is fully random.� Since Stanford does not dispute that Perrin indeed established this conclusion, Fine’s �rst example is of no use to him. By contrast, Fine’s second example does support Stanford’s argument by accounting for complete randomness in Brownian motion without relying on the atomic hypothesis. This shows that Roush is wrong when she equates Perrin’s demonstration of randomness with a con�rmation of the atomic hypothesis.¹� Using the terminology introduced above, we can summarize the argumentative situation up to this point by saying that the randomness of Brownian motion

9 Stanford (2009b, 260) anticipates this kind of argument and admits that “the bare appeal to ‘electrostatic forces’ does not immediately or straightforwardly entail random motion of the Brownian particles”. He nevertheless deems such an entailment possible, depending on “the characteristics of the electrostatic forces and their interaction with the exchange medium”. The idea seems to be that the electrostatic forces interact with the medium in such a way that they cancel out any correlation between the movements of nearby particles. However, the plausibility of such a model is undermined by Svedberg’s experiments of 1907, which showed that the electrostatic properties of the medium had no in�uence on the Brownian movement of the particles (Nye 1972, 124–125). 10 To be fair, Roush does not completely equate the two claims. She considers the possibility that something other than the (modest) atomic hypothesis could explain a random distribution of the motion of Brownian particles, but dismisses this possibility on the basis of the following two-stage argument: First, she argues that even if we (generously) assign a prior probability of 0.5 to the totality of these unknown alternatives, we still end up with a posterior probability for the atomic hypothesis of ≥ �.�, “better than more likely than not”. But surely a realist wants to claim signi�cantly more than that. So Roush adds that the atomic hypothesis also explains several other phenomena, including those Perrin used to determine Avogadro’s number. These pieces of evidence, Roush concludes, “increase the probability that there are atoms above the minimum probability I have argued for here” (Roush 2005, 221). Hence, Roush can only infer the reality of atoms from the randomness of Brownian motion by relying on the traditional argument from coincidence. I think this is the right way to go (see below), but it undermines her claim that measurement of Avogadro’s number is not needed to con�rm the modest atomic hypothesis (218–219).

Causal Realism, Unconceived Alternatives, and the Atomic Hypothesis �

97

became, thanks to Perrin, a detection property. And while I agree with Stanford that this does not yet establish the reality of atoms, I also agree with Roush that it was a signi�cant achievement. Drawing on the resources of causal realism, it is even possible to explain precisely how Perrin managed to turn randomness from an auxiliary into a detection property. Among the three necessary criteria for such a transition (see chapter 4), non-redundancy is the least relevant to this case, for two reasons: First, the logical space of possibilities is rather simple here, “since the motion is either random or it is not” (Roush 2005, 219; quoted approvingly in Stanford 2009b, 257). Second, and more importantly, considerations of nonredundancy cannot even start before there is a clear notion of what it would mean for the system to have one or the other of the alternative properties. In other words, it needs to be established that the randomness hypothesis meets the criterion of material inference. Perrin did this by deriving precise quantitative expectations about the behavior of Brownian particles from Einstein’s (1905) theory of Brownian motion, which depended crucially on the supposition that the motion is completely irregular (Mayo 1996, sec. 7.4). The comparison of these predictions with the experimental results then showed that the randomness hypothesis also satis�ed the criterion of empirical adequacy, whereas its negation was incompatible with experimental data. Taken together, these �ndings endow the hypothesis with causal warrant, which amounts to saying that randomness is a detection property.

From Brownian Motion to the Reality of Atoms Hence, the causal realist and Roush agree that detecting the randomness of Brownian motion was an important step in Perrin’s argument. Furthermore, due to the simplicity of the logical space of possibilities (either random or not), not even Stanford views this part of Perrin’s claim as threatened by the possibility of unconceived alternatives. But how do we get from here to a robust (PUA-proof) realism about atoms? As a �rst step, let us take a more detailed look at Fine’s second hypothesis mentioned above, which I shall call the no-cause hypothesis. Stanford (2009b, 261) admits that it is not easy to see how the appeal to fundamental, uncaused randomness could plausibly recapture Perrin’s explanation for the vertical distribution of Brownian particles at equilibrium. Perrin (1910, 530) described this distribution quantitatively by means of the following equation: � n W log � = ϕ(∆ − δ)gh, � n where n� and n denote the concentration of Brownian particles at two levels separated by a height h, W is the mean (translational) kinetic energy of the particles, ϕ their volume, ∆ their density and δ the density of the �uid in which they are

98 � The Problem of Unconceived Alternatives suspended. Now there is a sense in which Stanford here concedes more to the realist than he needs to. As Alan Chalmers (2011, 721) emphasizes, Perrin was able to derive the above formula assuming only random motion of the Brownian particles, Newtonian mechanics and some elementary statistics, but nothing about molecules or the kinetic theory. In this sense, the no-cause hypothesis can explain the (experimentally observed) exponential decrease of particle concentration as a function of height. What it cannot explain is the rate of this decrease, since it does not tell us anything about the particle energy W. Nor can W be measured, because the velocity of the Brownian particles changes much too quickly to permit a reliable measurement (Perrin 1910, 528–529). In the absence of a constraint on W, there is no constraint on the value of n� /n either, so the no-cause hypothesis would also be compatible with a uniform vertical distribution of particles (n� /n → �) or with all the particles sinking to the bottom of the vessel (n� /n → ∞).¹¹ Assessing the shortcomings of the no-cause hypothesis in this way allows us to spell out precisely what we need to assume about the cause of Brownian motion in order to arrive at a satisfactory explanation of the experimental results. Following the analysis by Chalmers (2011, sec. 5), I suggest that not the complete kinetic theory is needed, but only two of its central elements. The �rst one is a special case of the equipartition theorem, namely the assumption that, in equilibrium, the molecules constituting the liquid in which the Brownian particles are suspended have the same mean (translational) kinetic energy w as the Brownian particles themselves: w = W. This does not yet constrain the value of W, until the energy of a molecule is linked to macroscopically accessible magnitudes. The second assumption does this by postulating a speci�c value for Avogadro’s number N, which connects w to the temperature T and the universal gas constant R via the equation w = �RT/�N (Perrin 1910, 517). If N = � · ���� mol−� is assumed, the calculated value for n� /n agrees with what is experimentally measured. Conversely, measurements of n� /n (and ϕ, ∆ etc.) can now be used to determine the value of N with high accuracy. Of course, invoking the atomic hypothesis in the form of the two assumptions described here is not the only possible way to explain the vertical distribution of Brownian particles. Experimentation is a tricky business, and there is always the possibility that what the experimenter observes is not an e�ect of the natural processes he intended to study, but simply an artefact of the experimental setup. This is where the argument from coincidence, as emphasized by Cartwright and

11 This argument occupies a central position in Perrin (1910, 554), as Chalmers (2011, 726) points out.

Causal Realism, Unconceived Alternatives, and the Atomic Hypothesis �

99

Salmon (but also by Chalmers 2011, 724–726), becomes crucial. If studying the vertical distribution of Brownian particles were the only way to establish a link between Avogadro’s number and observable phenomena, there would be room for skepticism about the atomic hypothesis. But this is not the case. In fact, there are several di�erent phenomena from which we can calculate N. It would be an incredible coincidence if each of these observations were an artefact, and yet all agreed about the value of N (Cartwright 1983, 84). The argument from coincidence thus allows us to rule out the artefact hypothesis as a reasonable explanation of the observed phenomena, just as the no-cause hypothesis was ruled out by the previously discussed argument. But this does not yet address Stanford’s worry about hitherto unconceived hypotheses. In order to do this, it remains to be shown that the atomic hypothesis belongs to a class of claims which is not threatened by the PUA, because it does not provide the basis for an induction of the form (NI). According to causal realism, this requires showing that the atomic hypothesis enjoys causal warrant. The foregoing discussion has already shown that it ful�lls the criteria of empirical adequacy and (weak) non-redundancy, since it accurately predicts what is experimentally observed, while the known alternative hypotheses are unable to do so. The assessment with respect to material inference is less straightforward, because it is not immediately clear whether there is a su�ciently well-de�ned notion of what it means to modify the properties to which the atomic hypothesis refers. In particular, while it is easy to see what it means to modify the numerical value of N, one needs to specify what this means in terms of physical properties; a possible world in which the number of molecules in one mole of substance is N ′ instead of N may be seen as a world which has exactly the same physics as our world, but in which people simply use another de�nition of “one mole”. In order to relate a change in N to a real change of physical properties, one needs to keep the de�nition of the mole �xed while modifying the value of N. This change then corresponds to a change in physical properties, namely the masses of all atoms, such that the number of atoms in 12 grams of pure carbon-12 is no longer �·���� , but something else.¹² Of course, there is no way in which we could actually bring about such a change, but practicability is not part of what the criterion of material inference requires. It is su�cient that

12 There is an additional subtlety here, because the de�nition of the mole, in virtue of the phrase “12 grams”, implicitly refers to atomic masses, if “gram” is de�ned in terms of the mass of a material body (as it still is according to the International System of Units). “Keeping the de�nition of the mole �xed” therefore requires that mass be de�ned in an independent way, e.g., in terms of Planck’s constant.

100 � The Problem of Unconceived Alternatives the relevant change is well-de�ned, and this is the case here. In sum, the atomic hypothesis meets all the criteria for causal warrant. The fact that the experiments on Brownian motion do not merely con�rm some predictions of the kinetic theory in general, but lend causal warrant to some speci�c assumptions, is a crucial achievement of Perrin’s work (for a similar assessment, though not formulated in terms of causal warrant, see Chalmers 2011, 723). More speci�cally, it provides us with a concrete historical example of the contrast between fundamental theorizing and explaining experimental phenomena, which I emphasized at the end of section 6.2; the kinetic theory as a whole is a product of the former activity, but when we manage to direct the successes of its predictions to some speci�c assumptions, we engage in the latter. Since, by Stanford’s own admission, his NI applies only to fundamental theorizing, not to explaining experimental phenomena, there is no reason to expect any unconceived alternatives to the atomic hypothesis when it comes to explaining the results of Perrin’s experiments. And since the known alternatives (notably the no-cause and the artefact hypothesis) were found wanting, the case for the reality of atoms and molecules is complete.

What We Know about Atoms But even if the truth of the atomic hypothesis is now secured, the above attempt to dissociate it from (parts of) the kinetic theory might raise the worry that the hypothesis thereby becomes uninformative. Stanford expresses this worry concerning Roush’s modest atomic hypothesis, when she dissociates it from the idea of exact localization, “to accommodate the possibility, later discovered, that atoms exhibit the quantum mechanical property of not being fully localized” (Roush 2005, 219). The subtleties of quantum non-locality do not matter here, what matters is just that an idea which at one time was associated with the atomic hypothesis (namely exact localization) later came to be separated from it, so we might think that the two ideas should always have been kept separate. Stanford (2009b, 262) comments: Such retrospective retelling of the story threatens to treat the modest atomic hypothesis simply as a placeholder or a bare name for whatever further inquiry ultimately decides about the causes of the phenomena that occasioned its introduction.

But we have already seen that this criticism has no force against my account of the atomic hypothesis. Firstly, it should be clear by now that the distinction between what does and does not belong to a hypothesis, on the basis of what is and is not causally warranted, can be drawn without the bene�t of hindsight, so there

Causal Realism, Unconceived Alternatives, and the Atomic Hypothesis

� 101

is nothing problematic about “retrospective retelling of the story”. Secondly, the atomic hypothesis, as construed above, is much more than a placeholder; it makes substantial claims about molecules, most notably, how many of them there are in a certain amount of substance.¹³ Finally, since N connects macroscopic with molecular magnitudes, the atomic hypothesis directly yields further information about molecules, for example concerning their masses or their approximate dimensions. This is not to say that the atomic hypothesis gives us exhaustive knowledge about atoms and molecules, but it de�nitely gives us more than a bare name. It is now time to �nally address a question which I have already hinted at in footnote 7 above: Does Stanford really claim that we have insu�cient evidence for the reality of atoms, or does he just argue that Roush’s account is insu�cient to support such a realism? His (2009b) paper can be interpreted as defending only the second claim, with which I obviously agree. However, the following passage (taken from a di�erent paper) indicates that he also embraces the �rst claim: Are atoms and amoebae really on epistemological equal footing? Although we can point to a glowing blue dot in a suitably prepared photograph and say “see, that’s an atom,” virtually all of what we think we know about atoms comes from the role they play in a highly elaborate fundamental theory we have adopted because its empirical accomplishments are so much more impressive than those of any competing account we know of concerning the �ne structure of matter. But quite a lot of what we know about amoebae (how fast they move, what they eat, how often they reproduce, etc.) does not come to us in this way, but in a variety of other ways by means of which we routinely gather knowledge about the world around us (even if this knowledge is also ultimately “theoretical” in character). (Stanford 2009a, 389)

This suggests that, although Stanford advocates realism about amoebae (which distinguishes him from a constructive empiricist), he remains skeptical about atoms. I have argued in this section that taking into account Perrin’s complete argument (rather than just the part Roush focuses on) undermines such a skeptical position. But even if we suppose that this argument fails and that Perrin did not succeed in rendering the atomic hypothesis immune to the threat of unconceived alternatives, I doubt that it is, a century after Perrin, still possible to coherently oppose realism about atoms on the basis of the PUA. The reason for this is that we today possess evidence for atoms which is much more obviously immune to the PUA than the kind of evidence Perrin could produce. Consider, for example, the famous experiment in which Donald Eigler and Erhard Schweizer used 35 xenon atoms to write “IBM” on a nickel surface (�gure 6.1). The point is not that

13 This commitment to countable entities is yet another illustration of the claim that causal realism is not just a realism about certain properties, but also extends to entities.

102 � The Problem of Unconceived Alternatives

Fig. 6.1. Scanning tunneling microscope image of xenon atoms positioned on a nickel surface. (Eigler and Schweizer 1990, 525; reprinted by permission from Macmillan Publishers Ltd: Nature, © 1990)

we “see” these atoms “in a suitably prepared photograph”, but that our ability to produce such pictures testi�es to our knowledge about atoms (how they behave when placed on a metallic surface, how they can be moved from one place on the surface to another, etc.).¹� And this knowledge does not come from a fundamental theory, but from what is experienced in the laboratory. In fact, Eigler and Schweizer (1990, 525) admit that they have only an incomplete theoretical understanding of the interaction between the xenon atoms and the nickel surface or between the xenon atoms and the tip of their scanning tunneling microscope, but this did not stop them from knowing how to position the atoms. To put it another way, the claim that Eigler and Schweizer really manipulated single atoms and that these really have the properties which make such manipulation possible does not depend on any eliminative inference. Rather, these claims are of the same kind as the ones Stanford cites with regard to amoebae, for which the PUA has no relevancy whatsoever.

14 This is obviously an instance of Hacking’s (1983) argument from manipulability. The shortcomings of Hacking’s general position, which I discussed in chapter 2, do not prevent the argument from working perfectly well in the case discussed here.

�

Part III: The Quantum Challenge The problem was so di�cult that it was hard even to get a wrong idea about it. Abraham Pais, Inward Bound

7 Causal Realism in the Context of Bell-Type Experiments Having defended causal realism against general philosophical objections, it is now time to confront the speci�c challenges of quantum theory. In the present chapter, I will focus on the phenomena surrounding the so-called EPR correlations¹ between distant measurements on pairs of quantum particles. Such correlations are experimentally well con�rmed, and the main task of this chapter will be to investigate how a causal explanation of these correlations can be given. This endeavor faces two major di�culties, namely that any such explanation must include some non-local elements (section 7.1) and that the experimental results leave many aspects of the possible explanations undetermined (section 7.2). The second of these di�culties is especially troubling for causal realism, because it results in a failure of the criterion of non-redundancy. I will formulate a response to this problem and show by means of two examples how recent experiments can help us deal with this underdetermination (section 7.3).

�.� Bell-Experiments: Causal Warrant for Superluminal Causation The history of Bell’s inequalities is a fascinating example of how scienti�c investigation can transform a philosophical question. What started as a purely metaphysical inquiry in Einstein et al. (1935) became an (at least in principle) empirically testable prediction through the work of John S. Bell (1964) and an inspiration for a host of increasingly sophisticated experiments in the last four decades.² Out of the variety of di�erent Bell-type inequalities, I will here brie�y introduce the one that is probably most widely used, namely the so-called CHSH inequality (Clauser, Horne, Shimony and Holt 1969). The context is David Bohm’s version of the EPR thought experiment, in which a source emits two particles in opposite directions, where on each of them a spin measurement is performed. Each of the two measuring devices can be set up in a particular way, characterized by the setting parameters a (for the left hand apparatus) and b (for the right hand

1 The name derives from the fact that these correlations were �rst described by Einstein, Podolsky and Rosen (1935). I will also use the term “Bell correlations” to refer to the same phenomenon. 2 For a concise overview of this development, see Aspect’s introduction to Bell (2004). A more detailed historical account is given by Whitaker (2012).

106 � Causal Realism in the Context of Bell-Type Experiments apparatus), respectively. The variables A and B denote the outcomes of the two measurements, each of which is assumed to yield one of the two possible results +� or −�. We allow the measurement results to depend on a variable λ (traditionally called the “hidden variable”), which is distributed according to a normalized probability distribution ρ(λ) and is independent of a and b. The decisive assumption now is that neither the setting nor the outcome in the left hand apparatus in�uences what happens in the right hand apparatus, and vice versa. Formally, this reads: A = A(a, λ); B = B(b, λ). (7.1) I now de�ne the correlation function � . E(a, b) = A(a, λ)B(b, λ)ρ(λ)dλ .

These assumptions and de�nitions su�ce to derive the CHSH inequality, which combines the correlation functions for two possible settings a, a ′ and b, b′ for each measuring device, as follows: . S = |E(a, b) − E(a, b′ )| + E(a′ , b′ ) + E(a′ , b) ≤ � .

(7.2)

Quantum mechanics violates this inequality: For suitable choices of a, a′ , b and √ b′ , the parameter S can take values up to � �. Since S can be measured experimentally, the Bell/CHSH inequality provides a possibility to test quantum mechanics against the class of theories satisfying the assumptions which went into the inequality’s derivation. Many such Bell tests (which I will also call EPR experiments) have been performed, con�rming the quantum mechanical predictions with high accuracy. But while it is widely agreed that Bell experiments tell us something important about the fundamental structure of reality, it is far less clear what exactly they tell us. I will not review this debate (see, e.g., Berkovitz 2008), but I will present an argument for one particular claim which I take to follow from the violation of Bell’s inequalities, namely that there are superluminal causal in�uences in nature. The argument was developed by Tim Maudlin (2011, chap. 5; �rst published in 1994), and it is remarkably general, in two senses: First, it claims not to depend on any speci�c account of causation, but to involve only the most uncontroversial application of that notion. This relieves Maudlin of the duty to enter into the complex and ongoing debate between the di�erent philosophical accounts of causation. Second, it claims to hold independently of which particular solution to the quantum measurement problem one happens to prefer.

Bell-Experiments: Causal Warrant for Superluminal Causation

�

107

Maudlin’s Argument for Superluminal Causal Influences Maudlin starts by specifying a su�cient condition for a causal implication between two events: “The local physical events A and B are causally implicated with one another if B would not have occurred had A not (or vice versa)” (Maudlin 2011, 117; departing from Maudlin’s notation, I designate events by italic letters for the sake of clarity).³ Obviously, the fact that A and B are causally implicated with one another in this sense does not yet imply that either A caused B or vice versa. If two television sets are tuned to the same program, it is correct to say that a certain picture would not have appeared on the �rst screen, had it not appeared on the second one. But it is not the case that the appearance of the picture on one of the screens caused the appearance on the other. Instead, there is a common cause for the two events, namely the signal sent out by the broadcasting company. In the context of EPR experiments, it is very natural to think that the observed correlations are due to a common cause, since these experiments typically involve particles coming from a single source,� detected at di�erent locations. Since the particles do not travel faster than the speed of light, the event of their emission at the source lies in the backward light cones of both the detection events. Therefore, if the emission could serve as a common cause explanation for the correlations, there would be no need for superluminal causation. So in order to argue for superluminal causation, we do not only need to show that two space-like separated events A and B are causally implicated with one another, but also that the implication cannot be traced to an event situated in the backward light cones of A and B. This is captured by Maudlin’s (2011, 118) su�cient condition for superluminal in�uences: (SI) Given a pair of space-like separated events A and B, if A would not have occurred had B not occurred even though everything in A’s past light cone was the same then there must be superluminal in�uences. It is obvious from the context that by “in�uences” Maudlin hear means “causal in�uences”. Notice that the claim is not that there is a direct causal in�uence from either A to B or B to A. The causal connection between A and B may be due to

3 Maudlin’s condition bears some resemblance to the counterfactual claim discussed in section 4.2, but there is an important di�erence in the focus of inquiry: In section 4.2, we started from an observable event y and asked about the reality of its unobservable cause x. Here, A and B are both observable events (typically the outcomes of measurements) and the question is whether there is a causal link between them. 4 As we will see in chapter 8, this is not a necessary condition. In an entanglement-swapping experiment, particles coming from di�erent sources can violate Bell’s inequalities.

108 � Causal Realism in the Context of Bell-Type Experiments a common cause C, but the condition (SI) states that C must (at least partly) lie outside A’s backward light cone. But this is to say that there is superluminal causation between C and A. So whether we opt for direct causation between A and B or for some common cause, in either case there is superluminal causation. How do we evaluate a counterfactual statement like the one in (SI), in order to decide whether (SI) is actually ful�lled in the context of EPR-experiments? Maudlin’s (2011, 120) answer is that “if we have gotten the laws of nature right, then we can know about at least some unrealized possibilities. Given a set of laws we may be able to evaluate counterfactuals, and thereby to discern some causal connections”. At this point, a contradiction with causal realism might seem to arise, since, as argued in section 4.2, causal realism maintains that our knowledge of the laws of nature is signi�cantly less secure than our knowledge of causes. But if the evaluation of a causal claim like (SI) depends on a knowledge of certain laws of nature, then it seems that, contrary to what the causal realist believes, laws are epistemically prior to causes. However, this seeming contradiction can be resolved by remembering the distinction between fundamental and phenomenological laws (section 2.3). It is only the former that arouse the causal realist’s suspicions, because their acceptance depends crucially on their explanatory virtues, which, as discussed in section 4.2, are insu�cient to establish their truth. By contrast, phenomenological laws derive their support from the simple fact that they accurately describe what is observed in experiments. The causal realist can endorse laws of this type wholeheartedly, and nothing more is required here. Consider, for example, the �rst part of (SI), the claim that “A would not have occurred had B not occurred”. No deep theory is needed to justify this claim. Once we accept that there is a systematic correlation between the measurement outcomes in the left and the right wings of an EPR experiment (“systematic” in the sense that it can be expressed by a phenomenological law), we may infer that in at least some cases, the left outcome would have been di�erent, had the right outcome been di�erent.� A somewhat more detailed treatment is needed to assess the second part of (SI), namely the claim that even if we held �xed everything in the past light cone

5 Maudlin formulates his argument in terms of perfect correlations, and in this case it is always true that the left outcome would have been di�erent had the right outcome been di�erent. This is of course highly idealized. However, Maudlin’s argument goes through even with imperfect correlations, as long as they are assured to be non-accidental (and only the most radical skeptic will doubt that this latter fact can be established experimentally). The reason to assume perfect correlations is just that the argument can then be formulated in simpler terms. I will follow that practice below.

Bell-Experiments: Causal Warrant for Superluminal Causation �

109

of A (or B), the correlation between A and B would remain intact. But even here, the evaluation of the counterfactual does not depend on any speci�c theory. It only has to take into account that the measurement process which gives rise to the events A and B can be either deterministic or (irreducibly) stochastic. The two cases require two di�erent treatments, but the result will be the same. In the deterministic case, the assumption that a common cause located in the intersection of the backward light cones of A and B is responsible for the correlation implies a Bell-type inequality. The experimentally well-con�rmed violations of such inequalities rule out any local-deterministic common cause model for the EPR correlations. In other words, assuming a deterministic measurement process, the correlation between the events A and B cannot be attributed to any causal factor located in the intersection of their past light cones. Thus (SI) is essentially� satis�ed in this case. That the same is true for indeterministic models can most easily be seen in the case of a perfect correlation between A and B. If the measurement process that leads to A is truly stochastic, it could have come out di�erently even if its complete backward light cone remained unchanged. But had A come out di�erently, so would B (and vice versa), as required by (SI). Since (SI) holds for deterministic as well as indeterministic models, it follows that the existence of superluminal causation is established independently of any speci�c approach to the measurement problem.

Causation, Manipulability, and Signaling One reaction to the above verdict is to ask whether Maudlin has rigged the game by helping himself to too weak a notion of causation. If there is a causal relation between A and B, should it not at least in principle be possible to bring about a variation in B by manipulating A? And if so, should it not be possible to use that manipulation to send a signal from A to B, thereby violating some no-signaling theorem (e.g., Ghirardi et al. 1980)? Maudlin (2011, 135-141) discusses this question in a section entitled “But is it causation?”. There he argues that the exploitability or controllability of the causal relation should not be part of the concept of causation and that no-signaling should therefore not be taken to imply no-causing: “In general if one adds control of one variable to a counterfactual-supporting con-

6 There is a small argumentative gap here, because (SI) requires holding �xed the entire past light cone of A and not just the part that overlaps with the past light cone of B. See Maudlin (2011, 122) for an argument closing this gap.

110 � Causal Realism in the Context of Bell-Type Experiments nection one gets signaling, but the addition is strictly irrelevant to the existence of the causal connection” (137). No matter whether or not one agrees with this characterization, one might ask at this point if the issue is relevant at all. It certainly is interesting to learn about these non-local dependencies, but does it make any di�erence whether we call them causal or not? Well, from the perspective of causal realism, it does make a di�erence which structures are causal and which are not, because this a�ects the decision on how far the realist commitment should extend. Furthermore, we have seen in earlier chapters that practical manipulability has played an important role in some of the arguments for causal realism. It thus seems worthwhile to consider in more detail the following two questions: First, why should the causal realist want to insist on tying causation to manipulability in the context of EPR experiments? And second, is there any chance of applying a manipulability account of causation to the EPR scenario? As we saw above, the argument for superluminal causation (in Maudlin’s sense) is independent of any speci�c choice of theoretical model and it is backed by strong experimental evidence. This is precisely the kind of argument that the causal realist values, as it a�ords causal warrant for its conclusion. But what exactly does this commit him to? To the reality of superluminal in�uences, of course, but of what kind of in�uences and between which relata? Maudlin’s argument can only be as general as it is by refusing to answer these questions. For illustration, let us look at two possible ways to account for EPR correlations.� The most obvious option is to postulate a direct superluminal in�uence from A to B (or vice versa). Apart from being faster than light, such in�uences would be unusual in yet other ways: their strength does not seem to diminish when the distance between A and B is increased� and the in�uence is discriminating in that it only a�ects particles that have previously interacted with each other (Maudlin 2011, 21–22). The second option tries to avoid such action at a distance by denying that there are two entities, one in each wing of the experiment, in�uencing each other

7 I will in the following refer to these two ways as “two models”, although each one of them actually represents a whole class of models. I am not here concerned with the di�erences within these classes, but only with the di�erences between them. Neither am I concerned with other, more exotic causal models for EPR, involving, e.g., backwards causation. For a more detailed study of causal models for the EPR correlations, see Suárez (2007). 8 One might say that this is actually not so unusual; the strong nuclear force, as described by quantum chromodynamics, even increases with increasing distance between the interacting particles. But such behavior is restricted to subatomic distances. By contrast, EPR correlations have been shown to extend over distances of several kilometers.

Bell-Experiments: Causal Warrant for Superluminal Causation

�

111

across a space-like interval. Rather, there is one single quantum object (misleadingly called “two-particle system”) that brings about events A and B. If we insist on describing such a system as being somehow composed of two parts, we must acknowledge that the state of this system is non-separable, which is to say that it is not determined by some intrinsic states of its parts. But of course, this model still includes a superluminal in�uence: Event A, for example, is caused by the whole, non-separable quantum system which spans both wings of the experiment, so A is in�uenced by something which is not con�ned to its past light cone. I postpone to section 7.3 the discussion on which of these models is preferable (or less objectionable). The point here is that they are both compatible with what experiments tell us about EPR-like arrangements. The situation is similar to one that appears frequently in empirical research based on statistics: We observe a correlation between two variables, but we do not know whether this correlation is due to a direct causal in�uence from one variable to the other or to a common cause of the two variables. This analogy allows us to see why causal realism places such emphasis on the practical manipulability of alleged causes. For this is precisely what enables us in many cases to discern the causal structure that holds between the variables. Consider once again the above example of the two TV sets. The fact that we can manipulate the image on one of the screens, but that we cannot thereby in�uence what appears on the other screen, strongly favors the common cause hypothesis over the hypothesis that one of the images causes the other. The controllability of a causal factor therefore has an important epistemic signi�cance over and above the obvious fact that a controllable causal factor may open the way for technical applications. In the absence of controllability, we might have strong empirical evidence for the existence of some causal relations (as Maudlin’s argument shows), but we do not have this type of evidence for claims about the precise causal structure of the situation. This brings us to the question of whether we can �nd such manipulable causal factors in an EPR-type situation. So far, I only mentioned that the no-signaling theorems of quantum mechanics seem to preclude the possibility of changing B by manipulating A. For the moment, I do not discuss the question how conclusive these theorems are (see section 7.3 for some remarks on this issue). What is important here is that the impossibility of signaling from A to B can arise in two fundamentally di�erent ways: either because there is no direct causal link between A and B or because it is impossible to manipulate A. The TV example is an illustration of the �rst case; we can manipulate the image on the �rst screen, but this does not in�uence the second screen. By contrast, the EPR scenario is usually taken to be an instance of the second case; here the measurement outcome A is beyond our control, so even if it directly in�uenced B, we could not use this in�uence to

112 � Causal Realism in the Context of Bell-Type Experiments send a signal. If this is true, then a manipulability account of causation is not applicable to the EPR case, because there is no intervention variable (in the sense of Woodward 2003, 98) for A with respect to B, that is, we cannot control A in the right way to test its putative causal in�uence on B (Hausman and Woodward 1999, 565–567). Suárez (2013) claims that this conclusion is premature, because the availability of interventions depends on the details of the causal hypothesis under test and on the interpretation of quantum mechanics that is adopted. More speci�cally, he argues that the event S a of setting up the spin measurement device in the left wing of the experiment may be seen as an intervention on A with respect to B.� The argument has four parts, corresponding to the four conditions on intervention variables, mentioned in Woodward (2003, 98). I will focus on the �rst condition, and claim that, although there is a reading on which S a satis�es it, this does not make S a a suitable intervention variable for testing the causal structure underlying the correlations between A and B. I further claim that this assessment does not depend on any speci�c interpretation of quantum mechanics, because it only makes use of uncontroversial facts about the probabilities of measurement outcomes. Woodward’s �rst condition for something to be an intervention on X is simply that it causes X. So our question is whether S a is a cause of A. Suárez (2013, 209) comments on this as follows: The answer to the question is surprisingly elusive. If by “being a cause” we mean “actively changing the value” of [A], or its probability, then the answer must be negative, for that probability is never anything other than �/� (if it is de�ned at all). But if we just mean “determining the value” of [A], or its probability, then certainly the setting event determines that the probability of a particular outcome of the spin measurement is �/�.

It is not immediately clear what Suárez means by “determine” here. Let us, for example, denote the probability for a “spin up” outcome in the left wing by P(↑A ). Now if the setting event determines P(↑A ), should that not imply that P(↑A ) somehow depends on the setting? Yet (as Suárez acknowledges) P(↑A ) is always �/�, regardless of how the measurement device is set up. But if no dependence of P(↑A ) on S a is required for claiming that S a determines P(↑A ), then any variable (e.g., the surface air temperature at the north pole) can be said to “determine” P(↑A ). Since

9 In order to maintain consistency with the rest of the chapter, I have interchanged the roles of A and B vis-à-vis Suárez’s formulation. Due to the symmetry of the situation, this does not change the argument. Furthermore, I use capital instead of lower case letters to denote the measurement outcomes.

Bell-Experiments: Causal Warrant for Superluminal Causation

�

113

this would amount to an unacceptable trivialization of the term “determine”, there must be some other way to make sense of Suárez’s claim, and I suspect that it is by attending to his parenthetical remark “if it is de�ned at all”. This seems to suggest that the setting event can determine whether P(↑A ) has a value or not, although it does not determine what that value is. Presumably, if the experimenter in the left wing chooses not to make a spin measurement at all, then it does not make sense to ask what the probability for a particular outcome will be. In that sense, P(↑A ) does indeed depend on S a , and consequently, S a can be said to be a cause of A. If we further agree with Suárez that S a also satis�es the other conditions on intervention variables, then the setting of the apparatus can indeed be regarded as an intervention in Woodward’s sense. Furthermore, the availability of this kind of intervention su�ces to counter Hausman’s and Woodward’s (1999, 566) claim that “the measurement result on one wing is not really a distinct event from the result on the other wing”: The two events are distinct, since it is possible to intervene in such a way that there is a measurement result on one wing, but not on the other. Unfortunately, this does not mean that the manipulability account of causation can pro�tably be applied to the EPR case. For the purpose of deciding whether the observed correlations are due to a common cause of A and B or a direct causal in�uence between A and B, the intervention described here is useless, because it concerns nothing but the fact whether or not a measurement is made. Performing such an intervention in a Bell-type experiment would result in a series of measurements within which A would for some runs of the experiment not have a value because no measurement was performed in the left wing. The problem with these runs is that they tell us nothing about EPR correlations; it is only by performing measurements in both wings that we can learn something about the correlations between A and B. An analyst interested in EPR correlations would therefore simply have to disregard all the data from runs in which only one measurement was taken. But this is to say that the intervention variable S a , understood as the choice to perform or not to perform a measurement, is epistemically inert. Setting up the measurement apparatus is therefore not an intervention in the relevant sense. Since there does not seem to be any other realistic proposal on what could count as such an intervention, I conclude that we cannot rely on manipulability to analyze the causal structure of EPR experiments.

114 � Causal Realism in the Context of Bell-Type Experiments

�.� Causal Realism and Underdetermination in Quantum Mechanics The result of the last section seems to have a rather devastating implication for causal realism in the context of EPR experiments. If causal realism can only get o� the ground when manipulation of the relevant factors is possible and if quantum mechanics prohibits such manipulation in the EPR cases, then it seems that causal realism is simply irrelevant in this context. In the rest of this chapter, I will argue that this view is too pessimistic, and that causal realism has important things to say about EPR, even though it cannot solve all the problems by itself. As a preliminary step in this argument, let me �rst spell out the problem more precisely and investigate its connection with the more general theme of underdetermination in quantum mechanics.

Underdetermination Between Direct Cause and Common Cause Models Using the concepts of section 4.1, we can formulate our problem as follows: There is no causal warrant for a concrete causal account of the EPR correlations, because there is a redundancy of explanatory models. In other words, the choice between the two models introduced in the previous section is underdetermined by the empirical data. A �rst response to this situation might be to question whether there really is a substantial di�erence between the two models. After all, is postulating a superluminal in�uence from event A to event B not just another way of saying that whatever brought about A did also bring about B? To answer this question, let us try to see more precisely what it is that di�erentiates a direct cause (DC) model from a common cause (CC) model. A minimal requirement for a DC model is that there should be some kind of asymmetry between the two events. Otherwise, it would not make sense to say that one event caused the other, and the DC model would collapse. What could serve as a basis for such an asymmetry? As discussed above, manipulability in the sense of an ability to modify B by modifying A (but not vice versa) cannot do the job here. Nor can the supposed causal asymmetry be grounded on a temporal asymmetry in the sense that the earlier event in�uences the later one, but not vice versa: Since A and B are space-like separated, there is, according to special relativity, no objective time order between them. For instance, there is a class of reference frames in which A and B are simultaneous.

Causal Realism and Underdetermination in Quantum Mechanics

�

115

In these frames, the causal in�uence cannot be properly said to “travel” from A to B, as van Fraassen (1991, 351) points out:¹� To speak of instantaneous travel from X to Y is a mixed or incoherent metaphor, for the entity in question is implied to be simultaneously at X and at Y—in which case there is no need for travel, as it is at its destination already. . . . one should say instead that the entity has two (or more) coexisting parts, that it is spatially extended.

This may not be a fatal objection; one might simply accept that causal in�uences do not “travel” in an ordinary sense. But it seems to me that this move would undermine the main motivation to take DC models seriously in the �rst place, which consists precisely in the intuition that causal in�uences should somehow propagate continuously through space. Remember the main challenge for CC explanations of EPR correlations: The common cause must be non-local, that is, its producing an event in the left wing must somehow take into account what happens in the right wing (and vice versa). By contrast, the DC model promises to do without such mysteriously extended causes; within a DC model, all carriers of causal in�uences can be arbitrarily well localized at all times, it is just that some of them travel faster than light. But as soon as we allow “instantaneous travel”, the DC model loses this explanatory advantage over the CC model, and so there seems to be little reason to postulate an empirically inaccessible causal asymmetry between the two measurement events. The obvious way around this problem is to forgo Lorentz invariance by postulating a preferred inertial reference frame.¹¹ This introduces an objective timeordering even for space-like separated events, so that the causal asymmetry necessary for the DC model can then be grounded on a temporal asymmetry.¹² Of course, such a commitment to a preferred (but presumably undetectable) foliation

10 In yet other reference frames, B precedes A, so the causal in�uence travels backwards in time. This is potentially even more problematic than “instantaneous travel” between two simultaneous events, but for the present context, it is su�cient to discuss the latter case. 11 More precisely, one does not need to single out one reference frame, but only a foliation of space-time. Such a foliation corresponds to a class of (global) reference frames, which all agree about spatial and temporal distances between events. Since the di�erence between a choice of foliation and a choice of reference frame will play no role in the following, I will use the two notions interchangeably. 12 The strategy of grounding the causal asymmetry on a temporal asymmetry can also be pursued without postulating a preferred foliation. But then, the answer to the question which of two causally connected space-like separated events is the cause and which is the e�ect becomes frame-dependent, in accordance with the so-called reinterpretation principle (Recami 1978). However, in order to account for EPR correlations, the DC model needs an objective identi�cation of cause and e�ect. This is why tachyons, even if their existence is not prohibited by special rel-

116 � Causal Realism in the Context of Bell-Type Experiments of space-time is not a very attractive feature of the DC model. But then, one should remember that the mysterious non-separability included in the CC model is a similarly unattractive feature. This is precisely why Chang and Cartwright (1993, 182) introduce a DC model to describe the EPR correlations, dismissing the alternative view which attributes the correlations to “a kind of wholism” that pertains to the two-particle system: After all, what is this kind of wholism but an acknowledgment that di�erent parts of a system do in�uence each other in some ways that cannot be reduced to well-understood methods by which independent systems interact with each other? If we had wholism without superluminal information transfer, we would have to have some connections whose nature is equally mysterious.¹³

I have arrived at the postulate of a preferred reference frame in the process of �nding a basis for the asymmetry which is necessary to distinguish the DC model from the CC model. But now it should be noticed that this asymmetry is not su�cient to draw the sought-after distinction: As soon as we postulate a preferred foliation, there is an asymmetry between the two events A and B (the earlier one in�uencing the later one but not vice versa), regardless of the kind of causal model we assume. In this case, the di�erence between the two models has to do with the non-local holism discussed in the previous two paragraphs and comes out clearly when we consider what happens if the temporal distance between A and B approaches zero. The holism of the CC model implies that the common cause can, in principle, produce its two e�ects at di�erent places simultaneously. By contrast, the DC model is separable in the sense that the causal in�uence travels from one event to the other, which takes a �nite (though possibly very small) amount of time. This di�erence will be important for the experimental tests to be discussed in section 7.3.

ativity, are of no use for explaining EPR correlations in a Lorentz invariant way (Maudlin 2011, 70–72). 13 Chang and Cartwright do not address the question whether this criticism equally applies to the CC model they introduce earlier in their paper. It seems to me that it does, although their model does not invoke the holistic nature of the two-particle system. In fact, once the two particles have left the source, the two-particle system plays no role in Chang’s and Cartwright’s model, because they take the emission event by itself as a (partial) common cause of the measurement outcomes, without there being any continuous connection between cause and e�ect (177–181). This seems at least as mysterious as the holistic connections they criticize. For a further investigation of common cause models for EPR, see San Pedro (2012) and Egg and Esfeld (2014).

Causal Realism and Underdetermination in Quantum Mechanics

�

117

More Underdetermination: The Quantum Measurement Problem Now that we have acquired a su�cient understanding of the di�erence between DC and CC models as causal accounts of EPR correlations, let us ask how this case of underdetermination relates to a more well-known case of underdetermination, namely the one between di�erent solutions to the quantum measurement problem. Very roughly, this problem consists in accounting for the fact that measurements have de�nite outcomes, while quantum mechanics (in general) describes physical systems as being in superpositions of di�erent states. In physics textbooks, the problem is solved by introducing the so-called collapse postulate, according to which, whenever a measurement is performed, superpositions disappear and the system ends up in a de�nite state. But as long as there is no clear characterization of what counts as “measurement”, this solution is utterly unsatisfying (Bell 1990a). One can of course dissolve the problem by taking an instrumentalist stance towards quantum mechanics; superpositions of states do then not occur in the real world, but only in the theoretical formalism, which is interpreted as nothing but a useful tool to derive empirical predictions. Alternatively, one can take the opposite direction and claim (in the spirit of Everett 1957) that the world contains only superposed states. The challenge then is to explain why it nevertheless seems to us as if our experiments had de�nite outcomes. I will not further consider the instrumentalist and the Everettian response to the measurement problem, because they do not �t well with realism; while the former is in con�ict with scienti�c realism, the latter contradicts common sense realism. For my discussion of quantum mechanical underdetermination, it will su�ce to look at the following two (classes of) responses to the measurement problem, which I will call realistic: The �rst one holds that quantum mechanics is an incomplete description of physical reality and that physical systems always have de�nite states. According to this view, when quantum mechanics describes a system as being in a superposition, it simply expresses our ignorance about the actual state of the system. The best known proposal in this class of solutions to the measurement problem is Bohmian mechanics (Bohm 1952). The second class of realistic solutions takes the above-mentioned collapse postulate seriously, but does not tie it to the dubious notion of measurement. Instead, collapses (or state reductions) are supposed to occur spontaneously from time to time. Since the probability for such a collapse increases with the number of particles involved, macroscopic objects

118 � Causal Realism in the Context of Bell-Type Experiments are (almost) always in de�nite states. The classic example for such an account is the GRW theory, named after its developers Ghirardi, Rimini, and Weber (1986).¹� Both realistic solutions to the measurement problem described here are able to reproduce the empirical predictions of standard quantum mechanics, but they make very di�erent claims about the physical reality which underlies these predictions. We are therefore confronted with another case of underdetermination, and the question is how this relates to the underdetermination between the two causal models described above. The most striking di�erence between the two cases is probably that the two causal models I described are exceedingly simple, naïve and coarse-grained, while research on the measurement problem has led to a number of highly elaborate, rigorous and di�erentiated theories (or interpretations). Given this contrast, one might question the relevance of the �rst case of underdetermination. However, I do not think that the DC/CC issue loses its relevance in light of the impressive theoretical progress concerning the measurement problem, because even if we have settled for a particular solution to the measurement problem, we may still be ignorant about the causal structure underlying the EPR correlations. Let me illustrate this in the context of the two examples for realistic solutions to the measurement problem introduced above, namely Bohmian mechanics and the GRW theory. Notice �rst that Bohm’s theory is committed to a preferred foliation of spacetime, so the Bohmian already accepts what I called above an “unattractive feature” of the DC model. In its original formulation (Bohm 1952), the theory is also committed to a “quantum mechanical potential” giving rise to non-classical forces. This �ts well with the DC picture of direct causal in�uences between distant particles. Bohm and Hiley (1993, 293) even envisage the possibility “that the long range connections of distant systems are not truly nonlocal, as is implied by the quantum theory, but that they are actually carried in the preferred frame at a speed that is �nite, but very much greater than that of light”. Such a proposal is very much in line with the DC model. On the other hand, the equations of Bohmian mechanics are such that the motion of a particle at a given time depends on the positions of the other particles at this same time, which is in tension with the requirement that causal in�uences should not “travel instantaneously” in the DC model. Furthermore, more recent formulations of the theory no longer include the quantum potential or any kind of forces (Goldstein 2001, sec. 5), and they regard the wave function simply as a part of the law according to which the

14 For a detailed introduction to the measurement problem and its proposed solutions, see Wallace (2008). Discussions of the measurement problem in the wider context of underdetermination and realism are given by Cordero (2001) and Lyre (2010).

Causal Realism and Underdetermination in Quantum Mechanics

�

119

con�guration of particles evolves (Dürr et al. 1997). In this form, Bohmian mechanics does not seem hospitable to the idea of a causal in�uence traveling from one particle to the other at a �nite speed. Rather, the suggested picture for the EPR case seems to conform to the CC model, featuring a non-separable two-particle system which, together with the two measuring devices, produces the correlated results. In contrast to Bohmian mechanics, which unambiguously is a theory about particles, it is not immediately clear about what kind of objects the GRW theory speaks (Allori et al. 2008; see also Maudlin 2011, 229-238). In one version (called GRWf ), the theory is merely committed to the existence of some scattered events in space-time (the �ashes), corresponding to the spontaneous collapses of the wave function. It seems rather di�cult to reconcile any intuitive causal model with this version of the theory; since there is nothing in between the �ashes, the causes would have to act across spatiotemporal gaps, which is somewhat mysterious (see footnote 13 above; for a proposal to supplant the �ash ontology with an ontology of dispositions, such that the �ashes become the manifestations of the dispositions to collapse, see Dorato and Esfeld 2010). According to the other version of the GRW theory (GRWm), the quantum mechanical wave function describes a matter density in space. Like Bohmian mechanics, this theory is committed to a preferred foliation of space-time. It is then usually assumed that, whenever the wave function collapses, the matter density immediately reduces to an almost pointlike object in space. Again, this includes the kind of holism associated with the CC model, because the matter density disappears simultaneously in (almost) all regions of space, regardless of its spatial extension prior to the collapse. In order to avoid such a holism, one could postulate a non-instantaneous collapse dynamics, according to which the collapse starts at a speci�c point (which will usually be located in the region where the matter density interacts with a macroscopic object) and then propagates through space until (almost) all of the matter density is concentrated in a small region (see Eberhard 1989 for a proposal along these lines). In this form, GRWm would conform to the DC model, so the DC/CC underdetermination also appears in the GRW context. The realistic responses to the measurement problem therefore do not, by themselves, determine which causal model is appropriate for describing a Bell experiment. This does not sound like particularly good news for causal realism, because we now face not one but two problems of underdetermination. Furthermore, at least as far as the measurement problem is concerned, there have been many e�orts to propose experiments which could break the underdetermination, but the prospects for success are rather dim (Cordero 2001, S304-S305). It thus seems that this underdetermination is here to stay, which precludes the possibility for any causal warrant in this context.

120 � Causal Realism in the Context of Bell-Type Experiments How Should the Causal Realist Respond? The situation described here is arguably one of the clearest cases of underdetermination in science, and correspondingly, it is one of the hardest cases for causal realism and its emphasis on causal warrant. In this context, the causal realist is thus forced to endorse the fallback position sketched at the end of section 4.1 and to admit that causal warrant is not all that matters. If experimental tests are unavailable, underdetermination may be broken by evaluating the theoretical virtues of the di�erent proposals.¹� In other words: In the absence of causal warrant, we must content ourselves with theoretical warrant. But this response, reasonable though it is, might seem to undermine the main claim of the present chapter, namely that causal realism has a distinctive contribution to make, even in the face of quantum mechanical underdetermination. Retreating to theoretical warrant just seems to collapse causal realism into standard realism. Two things are to be said in response to this objection. First, although the ontological commitment of causal realism is in this case similar to the one of standard realism, they di�er in that the causal realist has a sound basis for a differentiation of the strength of his commitments: He is more strongly committed to causally warranted claims than to theoretically warranted ones. The need for such di�erentiated commitments is acknowledged even by standard realists (as shown by Psillos’s statement quoted in section 3.1), but, thanks to the distinction between causal and theoretical warrant, causal realism has a much clearer response to this need than standard realism does. It should also be noted in this context that the (causally warranted) hard core of the causal realist’s commitment di�ers from the minimalistic commitment of a constructive empiricist: As Cordero (2001, S309–S310) argues, there is a wide range of beliefs which is not a�ected by quantum mechanical underdetermination, and this range by far exceeds what the empiricist accepts as observable. The case studies in chapter 5 and section 6.3 reinforce this lesson. The second di�erence between causal realism and standard realism in the context of quantum mechanics concerns scienti�c methodology. The best way to grasp this di�erence is to �rst look at the contrast between how constructive empiricism and realism (of any kind) react to quantum mechanical underdetermination. While any realist is bothered by this situation and insists that something needs to be done about it, the constructive empiricist welcomes the multitude of

15 For the case discussed here, this task has yet to be performed. But this is now obviously no longer a problem for causal realism in particular, but for any realism which aspires to provide a complete ontology of quantum mechanics (e.g., ontic structural realism; see Esfeld 2013).

Causal Realism and Underdetermination in Quantum Mechanics

�

121

di�erent approaches to quantum theory and resists the invitation to pick out any one of them as true (van Fraassen 1991, 335-337). Now there is an interesting similarity here between the constructive empiricist and the standard realist: Both of them are prepared to rest content with a situation in which no experiment can decide between di�erent theoretical options, the former because he does not admit that such a decision is needed, the latter because he trusts that the evaluation of theoretical virtues will be able to induce the decision. By contrast, the causal realist will never be entirely satis�ed with such a solution. His top priority is the search for experimental tests, because this is the only way to generate causal warrant. As an illustration of this di�erence, let us look at how the three philosophical positions would have reacted to the research program sketched in the �nal paragraph of Einstein et al. (1935), which aimed for a completion of quantum mechanics by what we would now call local hidden variables. Prior to Bell’s (1964) work, it was not known that the empirical predictions of any such theories were incompatible with the quantum mechanical predictions,¹� and neither the constructive empiricist nor the standard realist would have been particularly worried about the fact that they could not be experimentally tested against other interpretations. By contrast, the causal realist’s prime interest would have been to �nd such experimental tests, in order to reduce (as far as possible) the redundancy of interpretational options. It thus seems to me that causal realism provides a stronger motivation for the kind of research which led to Bell’s (1964) groundbreaking result than the other two views do. This is not to say that the latter cannot acknowledge the importance of Bell’s theorem once it is in place, but their tendency to accept redundancy of explanations results in assigning to Bell’s research a lower priority than the one assigned to it by the causal realist. The example of Bell’s theorem and the ensuing experimental progress allows me to formulate causal realism’s response to the problems described in this section, a response that neither collapses into standard realism nor into constructive empiricism. Causal realism takes the way in which Bell transformed the hiddenvariable issue into an empirically testable claim to suggest a general policy regarding quantum mechanical underdetermination: Attack the problem of underdetermination at those points where there is the best prospect of being able to perform experimental tests. Seen from a theoretical point of view, these may not always

16 The actual historical situation is complicated by the fact that von Neumann (1932) had allegedly proved the impossibility of any hidden variable completion of quantum mechanics. However, Bell (1966) later exposed the limited validity of von Neumann’s proof. I therefore disregard this historical complication here.

122 � Causal Realism in the Context of Bell-Type Experiments be the most relevant aspects of the problem. But if nature does not (yet) give us answers to the questions we would like to ask, the best way to move forward may be to ask di�erent questions. The next section will show that some progress can indeed be made by following this strategy.

�.� Some Experimental Constraints on Explanations for EPR I will discuss two cases in which recent experiments have proved helpful in dealing with underdetermination. The �rst one concerns the choice between CC and DC models introduced above, the second one concerns a class of models proposed by Anthony Leggett (2003), which also inspired some experimental tests.

Testing Direct Cause Models In my discussion of the DC model in section 7.2, I showed that, if such models are to do any explanatory work, they need to assume a preferred reference frame and a �nite propagation speed of the causal in�uence within that frame. On the other hand, violations of Bell’s inequalities for space-like separated measurements (Aspect et al. 1982) imply that the propagation speed must exceed the speed of light. But by how much? Since the late 1990’s, experiments have given some substantial answers to this question, which I will now brie�y summarize.¹� The basic principle of these experiments is very simple. One performs a test of a Bell-type inequality on pairs of entangled particles in such a way that the temporal interval δt between the two measurements is as small as possible, while their spatial separation L is large. If the causal in�uence predicted by the DC model travels with a �nite speed v CI , then the EPR correlations should disappear if L/δt > v CI , because the in�uence from one wing of the experiment would then not arrive at the other wing in time. (Recall that the DC model presupposes a preferred reference frame. It is with respect to that frame that the magnitudes δt, L and v CI are to be understood.) Conversely, if the correlations do not disappear, L/δt �xes a lower bound for v CI . Such an experiment was performed in Geneva by Zbinden et al. (2001) with L = ��.� km and δt < � ps, which yields v CI >

� · ��� c, �

where c denotes the speed of light. 17 The following presentation roughly follows section 3 of Gisin et al. (2000).

Some Experimental Constraints on Explanations for EPR

� 123

For this analysis, L and δt were measured in the laboratory frame, that is, a reference frame �xed on earth. So the lower bound for v CI given here is only valid for this frame. It is, of course, far from clear that this should be the preferred frame. A less anthropocentric choice for a preferred frame would be the frame in which the cosmic microwave background radiation (CMB) is isotropic. Therefore, Scarani et al. (2000) have reanalyzed the data from the above experiment and calculated a lower bound for v CI in the CMB frame: v CI > �.� · ��� c. But still, even though there are some cosmological arguments for regarding the CMB frame as special, the relevant reference frame for measuring v CI could be a completely di�erent one. In order to address this issue, the Geneva group has more recently generalized the above results for a large class of reference frames (Salart et al. 2008; see also Cocciaro et al. 2011). The key idea for this generalization is the fact that two events which are simultaneous in some reference frame will also be simultaneous in any reference frame moving in a direction perpendicular to the line joining the two events. Then, if a Bell experiment is performed in which the two wings are oriented along the east-west axis, the earth’s rotation will ensure that all possible reference frames are scanned within a 12-hour period. In practice, it is not possible to actually scan all possible reference frames, because the alignment of the two measurement events in the earth frame would have to be optimized for each frame which is to be tested. Furthermore, the orientation of the two detectors in the Geneva experiment was not exactly east-west (see �gure 7.1). Nevertheless, Salart et al. arrived at the following general statement: For every reference frame moving relative to the earth with less than ��−� c (which is the same order of magnitude as the speed of the earth in the CMB frame), v CI > ��� c.

of the correlation between the photon detections at Satigny and Jussy (Fig. 3). The phases were controlled by the temperature of the fibre124 Causal Realism in the Context of Bell-Type Experiments based� interferometers.

t to a th angle x e A–B axis city of the

S

ene

France

Jussy

eG

E

APD

Lak

W

va

N

y

Satigny

d

18.0 km

APD

10.

m

Underground fibre length: 13.4 km Fibre coil length: 4.1 km Total fibre length: 17.5 km

m

7k

8.2 k

Geneva

Switzerland

Fibre length: 17.5 km

France

2 km

Figure 2 | Experimental The Bell source sends pairs of photons from Fig. 7.1. Experimental setup for setup. a long-distance test with (approximate) east-west orientation. A source in Geneva emits entangled photons, the which are detected simultaneously in Geneva to two receiving stations through Swisscom fibre-optic network. Satigny and Jussy. (Salart et al. 2008, 862; reprintedSatigny by permission from Macmillan Publishers The stations are situated in two villages, and Jussy, that are Ltd: Nature, © 2008) respectively 8.2 and 10.7 km from Geneva. The direct distance between them

is 18.0 km. At each receiving station, the photons pass through identically unbalanced Michelson interferometers and are detected a single-photon While all the experiments discussed so far were merely able toby give a lower bound InGaAs avalanche photodiode (APD) (id201, id Quantique). The length of for v CI , there are other experimental results which conclusively rule out some parthe fibre going to Jussy is 17.5 km. The fibre going to Satigny is only 13.4 km ticular DC models. These are models which, unlike the ones discussed up to now, long, so we added a fibre coil of 4.1 km (represented as a loop) to equalize the do not postulate a preferred global reference frame, but preferred reference frames lengths of the fibres. Having fibres with the same length allows us to satisfy depending on the experimental setup. The motivation to develop such models the condition of good alignment (r = 1). d indicates the scanning of the stems from the fact that they con�ict less starkly with the spirit of special relativphase of the interferometer at Jussy.

ity: No global structure is added to space-time, preferred frames are only invoked in connection with material objects in space-time. More precisely, these models n Publishers Limited. All rights reserved take seriously the idea that the device A which performs measurement A somehow triggers the causal in�uence from A to B postulated by the DC model. It is thus natural to measure v CI in the rest frame of A . The �rst experiment discussed above, in which the detectors (A and B ) were at rest with respect to the earth, already exempli�ed this principle, but we are now interested in the more general case, in which the two devices are in relative motion. It is then possible that in A ’s frame, A precedes B, while in B ’s frame, the time order of the two events is reversed. Such a case is called a before-before con�guration, because each apparatus (considered in its own rest frame) performs its measurement before the other apparatus does. What would the DC model have us expect to happen in such a case? Unless we allow causal in�uences to travel backwards in time, each measurement event (considered in the rest frame of the corresponding device) will be unin�uenced by the other event. Therefore, EPR correlations between the two measurements should disappear.

Some Experimental Constraints on Explanations for EPR

� 125

If we want to test this prediction in an experiment, we need to be more precise about the term “measuring device”. As is well known, di�erent interpretations of quantum mechanics have di�erent views on where exactly “measurements” take place. And since it is not feasible to set all the equipment in one wing of the experiment in motion relative to the equipment in the other wing (the necessary relative speed being in the order of 100 m/s), we need to make an assumption about which part of the apparatus is relevant for triggering the causal in�uence traveling to the other wing. Two reasonable choices have so far been tested experimentally. Zbinden et al. (2001, 3) argue that “the essential part is the irreversible process, the absorption of the photon in a solid and the transformation of its energy into heat”. Therefore, it is the absorbing part of the detector which de�nes the preferred reference frame and which needs to be set in motion in order to realize a before-before con�guration.¹� The experiment shows no disappearance of the EPR correlations for the before-before setup (Zbinden et al. 2001, 6) and thereby refutes this particular DC-model. The second option is to assume that the causal in�uence is triggered not by the absorbing detector, but by the part of the device which forces the incoming particle to “choose” between the two di�erent paths corresponding to the two possible measurement outcomes. In experiments with energy-time entangled photon pairs, such as the ones discussed here, this role is played by beam splitters. An experiment with moving beam splitters was performed by Stefanov et al. (2002).¹� Again, the before-before setting did not result in a disappearance of the quantum correlations, so this model can be ruled out as well. Bancal et al. (2012) have more recently investigated another important class of DC models, namely models satisfying the so-called no-signaling conditions. In the simple case of a two-particle system subject to an EPR experiment with settings a, b and outcomes A, B, these conditions state that the marginal distributions for the measurement outcomes on one side must be independent of the measurement setting on the other side:

18 In the actual experiment, it was not “the absorbing part of a detector” which moved relative to the rest of the equipment, but merely a (non-detecting) absorber (more concretely, a layer of black paint on a rotating wheel). This is su�cient because the interferometers used in this type of experiment have two output ports, and EPR correlations can be detected even if only one output port of each interferometer is connected to an active detector (Zbinden et al. 2001, 3). 19 As in the experiment with “moving detectors” (see the previous footnote), Stefanov et al. did not actually accelerate an ordinary beam splitter to some 100 m/s. Instead, the moving beam splitter was realized by a traveling acoustic wave in an acousto-optic modulator (Stefanov et al. 2002, 2).

126 � Causal Realism in the Context of Bell-Type Experiments � B

P(AB|ab) = P(A|a);

� A

P(AB|ab) = P(B|b).

In more practical terms, this means that no locally detectable change on the right hand side can be achieved by changing the setting on the left hand side, and vice versa. Standard quantum mechanics satis�es these conditions, as demonstrated by the no-signaling theorems mentioned in section 7.1. And since it is in the twoparticle case always possible for a DC model to reproduce the quantum mechanical predictions (by a suitable choice of the preferred reference frame and a suf�ciently high v CI ), DC models can also be made to conform to these conditions. However, this is no longer the case when four-particle scenarios are taken into account. In this context, denoting the settings by a, b, c, d, and the outcomes by A, B, C, D, respectively, the no-signaling conditions take the following form: � P(ABCD|abcd) = P(BCD|bcd), (7.3) A

and analogously for the marginal distributions for ACD, ABD, and ABC. Failure of one of these conditions results in the possibility of sending superluminal signals, as illustrated in �gure 7.2. The central result of Bancal et al. (2012) is that imposing the no-signaling conditions (7.3) on a DC model implies an inequality which is violated by quantum mechanical predictions. Moreover, the authors show that, insofar as a DC model is expected to reproduce quantum mechanical predictions whenever the measurement events can be connected by causal in�uences traveling at v CI , DC models should also violate this inequality.²� This shows that such models are incompatible with the no-signaling conditions, hence they imply the possibility of superluminal signals. Is this a problem? The answer to that question requires a closer look at the reasons for thinking that physics rules out the possibility of superluminal signaling. A �rst reason, though not a very convincing one, is the empirical fact that such

20 Due to the technical di�culties in producing and handling entangled four-particle states, an experimental test of this inequality has not yet been achieved. There is, however, not much reason to suspect that the expected violations may fail to be con�rmed. The situation is relevantly dissimilar to the case of Bell’s inequalities, where the principle of local causality could legitimately underpin the expectation that the inequalities would not be violated. (Whitaker (2012, 160) cites John Clauser as a physicist who was genuinely convinced that Bell tests were going to disprove quantum mechanics.) Here, by contrast, quantum mechanics and DC models agree in their predictions that the inequality will be violated, and there is no plausible model leading us to expect otherwise.

Some Experimental Constraints on Explanations for EPR

� 127

5

time where the um theory A� sal model d|xyw) = , see capD� marginal hat is inn C (B), measureC B ns should D quantum tions will space on of Figd B dC dD A ore one of Superluminal signaling in a four-particle scenario: Solid lines delineate the forward ated. But Fig. 7.2. FIG. 4. Let the four systems of Figure 3 lie along some spatial light cones of the four measurement events A, B, C, and D, dashed lines stand 1 for the causal 1 + BCD )+ , direction at, respectively, a distance dB = 14 (1 the ecessarily influences r 1+r postulated by the DC model. If condition (7.3) is violated, then marginal 1 1 1 ust violate correlations dC =depend (1 + ) − , d = 1 form A, where r = v/c > 1, Dsetting chosen at A. The BCD correlations can be on rthe measurement 4 1+r

2 evaluated D′ , which the forward conetof A (shaded area). This scheme andatlet themlies beoutside measured at light times A = 0, tB = tC = c+v , can therefore be used for superluminal communication from A to D′ . (Bancal et al. 2012, 869; tD = 1/r. Suppose that the correlations PR produced by a reprinted by permission from Macmillan Publishers Ltd: Nature Physics, © 2012)

e used for v-causal model are such that the BCD marginal correlations ey violate depend on the measurement x made on the first system A. If tripartite parties B and C broadcast (at light-speed) their measurement st depend signals have not been But ofthe course, the fact that something results, it will beobserved possibletotodate. evaluate marginal correlations rty. This � has not beenat possible up toDnow does not imply that it outside should be impossible . Since this point lies the future in BCD, the point ) as it is principle. The history science contains clever light-cone of Aof(shaded area), many this examples scheme of can be experimenters used for the quan- who gained access to domains that were previously thought to be inaccessible if the in superluminal communication from A to D� . Similarly, marginal principle. ABC marginal correlations depend on the measurement w w of sysmade on reason D, theyis can used formechanics superluminal A second thatbe quantum itself, communication by virtue of the no� nd on the signaling . fromtheorems, D to thedisallows point Asuperluminal signals. This is not very convincing es, these either, because the very project of looking for causal explanations in the context h party’s of Bell-type experiments presupposes that standard quantum mechanics is not ed for su- the whole story. It should then not come as a surprisetothat some ofquantum its predictions This work illustrates the difficulty modify do not match the predictions of our causal models exactly, but onlywant to a very physics while maintaining no-signaling. If we togood approximation. And maybe experiments onequantum day be able non-locality to reveal the limits keep no-signalling, it shows will that of the approximation’s validity. must necessarily relate discontinuously parts of the uniFinally, strongest motivation against countenancing signals verse the that are arbitrarily distant. This superluminal gives further comes from special relativity. The theory of relativity is usually taken to forbid weight to the idea that quantum correlations somehow sucharise signals, but, as Maudlinspacetime, (2011, 90) observes, claimthat “is often from outside in thethis sense no made storywitht we obin space and time can describe how they occur. influences Acknowledgments. We acknowledge Serge Massar and lity could Tamas V´ertesi for helpful discussions and Jonathan SilOur reman for comments on the manuscript. This work was plex relasupported by the European ERC AG Qore and SG cality and PERCENT, the European EU FP7 QCS and Q-Essence

128 � Causal Realism in the Context of Bell-Type Experiments out justi�cation and accepted without demur”. He o�ers a detailed evaluation of this claim, which I do not review here. The only question relevant for the present context is whether the possibility of superluminal signaling creates a more serious con�ict with relativity than the reality of superluminal causation argued for in section 7.1. A non-operationalist view of special relativity suggests a negative answer to that question. If one takes (as a scienti�c realist should) special relativity to describe a reality that does not depend on what human beings can or cannot do, it is hard to see how superluminal signaling could be any worse than superluminal causation, for the di�erence between a causal in�uence and a signal is just that the latter is used by some agent to communicate something. And this di�erence cannot bear much metaphysical weight, as Bell (1990b, 245) reminds us: Do we then have to fall back on ‘no signalling faster than light’ as the expression of the fundamental causal structure of contemporary theoretical physics? That is hard for me to accept. . . . The assertion that ‘we cannot signal faster than light’ immediately provokes the question: Who do we think we are?

These considerations show that the violation of no-signaling may not be such a high price to pay for the proponent of a DC model, who is already committed to the existence of direct superluminal in�uences. Nevertheless, the exclusion of non-signaling DC models by Bancal et al. (2012) is a further step in the process of critically assessing the viability of DC models in general. The most recent step in this assessment is Näger’s (2013) analysis using causal graph theory, which calls into question one of the DC model’s most attractive features, namely its simplicity. While it might seem that such a model only needs to postulate a causal in�uence from one measurement event to the other in order to account for the EPR correlations, Näger’s analysis shows that this is actually not the case. In order to violate Bell’s inequalities, a causal model must at least postulate a causal in�uence from one of the measurement settings (e.g., the event denoted by S a in section 7.1) to the distant outcome (B), which does not go through the local outcome (A). This does not rule out all DC models, but it shows that they cannot be as simple as is usually thought. Let me now draw the philosophical lesson from the results discussed so far. Although experiments (and certain theoretical considerations) have ruled out some ways to �esh out the DC model, it is clear that some DC models remain viable. If one is willing to postulate a preferred reference frame and a causal in�uence propagating with a speed of v CI > ��� c within that frame, and if one is not afraid of the possibility of superluminal signaling, one can hold on to the DC idea. We thus do not have, strictly speaking, causal warrant against the DC model. Nevertheless, the experiments discussed here do have an impact on how

Some Experimental Constraints on Explanations for EPR

� 129

well the model is warranted. I shall describe this impact as raising the ontological cost of choosing the DC model. Before these experiments were done, opting for a DC model was relatively cheap. As we have seen in this section, one did not even need to postulate a globally preferred reference frame, but could do with reference frames picked out by the experimental setup. Furthermore, one only had to introduce a speed of the causal in�uence which was of the same order of magnitude as the speed of light. (In Aspect’s (1982) experiment, L and δt were of the of the order 12 meters and 5 nanoseconds, respectively, so v CI > �c in the laboratory frame.) This situation is now radically changed. Tying the preferred frame to the laboratory equipment is no longer a realistic option, as the beforebefore experiments made clear. And in any globally preferred reference frame, the causal in�uence needs to propagate with at least 10,000 times the speed of light. (Furthermore, this value is likely to increase with the availability of better experimental techniques; see Cocciaro et al. 2013.) Now one might argue that the ontological cost of postulating such a speed is not signi�cantly higher than the cost of postulating a speed which only moderately exceeds c. I admit that intuitions may be divided on this issue, but I maintain that I �nd it considerably harder to believe in a causal in�uence that could reach Alpha Centauri within four hours than in one that does so within half a year. Evaluating the ontological cost of an explanatory proposal is one way of evaluating its theoretical virtues (or lack thereof), which bears on the theoretical warrant that accrues to the proposal. The example of the DC model thus shows how considerations of causal and theoretical warrant can complement each other. The experiments discussed here a�ord some causal warrant against certain variants of the DC model, thereby signi�cantly reducing the theoretical warrant for this model in general.

The Signi�cance of Leggett-type Models As a second example of how recent experiments have furthered our understanding of the nature of quantum non-locality, I will discuss a class of models introduced by Leggett (2003) and subsequently tested by Gröblacher et al. (2007) and Branciard et al. (2007). In contrast to the previous example, I will not focus on how the experiments were actually carried out, but rather on their foundational significance, which was harshly disputed by Federico Laudisa (2008). Against Laudisa, I will argue that these experiments do indeed tell us something interesting about the nature of quantum systems. Leggett considers an EPR-type situation with polarization-entangled photon pairs. The measuring apparatus in each wing of the experiment consists of a po-

130 � Causal Realism in the Context of Bell-Type Experiments larizer (characterized by a transmission axis �a for the left wing and �b for the right wing, respectively) and a detector which registers the photon if it is transmitted (rather than absorbed) by the polarizer.²¹ Accordingly, the measurement outcomes A (left detector) and B (right detector) can take one of the two values +� and −�, depending on whether the detector does or does not register the photon. Leggett now constructs a hidden-variable theory in which each photon pair is characterized by a variable λ and a pair of polarization directions �u (left photon) and �v (right photon). The interesting di�erence between such a theory (or model) and standard quantum mechanics appears in cases which the latter describes as photon pairs of “inde�nite polarization” (e.g., in the so-called singlet state). In contrast to standard quantum mechanics, a Leggett-type model assumes that each photon pair has a de�nite polarization, such that the complete ensemble of photon pairs emitted in a series of emission events is a disjoint union of subensembles of de�nite polarization. In other words, subensembles are characterized by unique values for �u and �v. Photon pairs within a given subensemble can, however, have di�erent λ, where the probability distribution for the λ’s is given by ρ uv (λ). Due to Bell’s theorem, such a model can only reproduce the quantum mechanical predictions if it incorporates some kind of non-locality. Within Leggett’s theory, the outcomes A and B are therefore allowed to depend not only on the local, but also on the distant parameters. Thus in contrast to (7.1), we now have: A = A(λ, �u , �v , �a , �b); B = B(λ, �u , �v , �a , �b).

(7.4)

If things were kept as general as that, nothing interesting would follow from the assumption of de�nite polarizations within the subensembles. The crucial constraint is therefore the following assumption, which states that the subensemble averages of A and B (i.e., the averages over all values of λ within a given subensemble) depend only on the local variables: � ¯ =. ¯ �u , �a), A ρ uv (λ)A(λ, �u , �v , �a , �b)dλ = �(�u · �a)� − � = A( (7.5a) � ¯ =. ¯ �v , �b). B ρ uv (λ)B(λ, �u , �v , �a , �b)dλ = �(�v · �b)� − � = B( (7.5b)

Although this is a kind of locality assumption, it is weaker than assumption (7.1) from which the CHSH inequality was derived. Indeed, Leggett (2003, sec. 5) shows that a model can satisfy (7.5) and yet violate Bell’s inequalities. Therefore, Leggetttype models are not excluded by Bell tests. This raises the question whether they 21 Leggett’s results can be proved for general (elliptical) polarization (Leggett 2003, sec. 4), but for the conceptual points I am discussing here, it is su�cient to look at the simpler case of linear polarization.

Some Experimental Constraints on Explanations for EPR

� 131

are compatible with all quantum mechanical predictions. The central result of Leggett (2003) is that they are not. In analogy with Bell’s (or CHSH’s) reasoning, Leggett derives from (7.5) an inequality which is violated by certain quantum states, thus opening the way to experimental tests of his models. Performing the relevant experiments, Gröblacher et al. (2007) and Branciard et al. (2007) con�rmed the violations of Leggett-type inequalities predicted by quantum mechanics. What signi�cance do these results have? According to Laudisa (2008, 1112), none whatsoever. More speci�cally, Laudisa calls Leggett’s approach an “implausible research program” and his theories “totally irrelevant from the viewpoint of the foundations of quantum mechanics”. Before scrutinizing the arguments which lead to this scathing judgment, I should mention that Leggett’s results have more recently been generalized in several ways (Branciard et al. 2008; Colbeck and Renner 2008, 2012). It might thus seem interesting to evaluate Laudisa’s arguments in the context of these newer developments. However, if what I argue below is correct, then this is unnecessary: My basic claim is that even the original results obtained by Leggett, Gröblacher etc. have some foundational signi�cance. But it is obvious that the later, more general results have at least as much foundational signi�cance as the original, more limited ones. In other words, if Laudisa’s arguments fail with respect to Leggett’s original models, they will fail even more dramatically with respect to the more general models developed later. I therefore restrict my discussion to the original Leggett models except for some brief side remarks on later developments (see footnotes 22 and 26 below). The bulk of Laudisa’s paper is devoted to showing that Leggett and his followers (in particular the authors of Gröblacher et al. 2007) start from a mistaken interpretation of Bell’s theorem, which I shall call the LR view. They believe that Bell’s theorem is based on two independent assumptions, locality and realism, and that the violations of Bell’s inequalities therefore force us to give up at least one of these assumptions. Laudisa is not alone in criticizing this conception (Maudlin 1996, 304–305; Norsen 2007; Gisin 2012), and I fully subscribe to the criticism, which may be summarized as follows: Proponents of the LR view either fail to tell us what they mean by “realism” or they use the term to denote something which is demonstrably not an independent assumption of Bell’s theorem. If anything, realism is inferred in Bell’s derivation, not assumed. However, from the fact that the LR view is mistaken and that Legett, Gröblacher etc. seem to hold it, it does not follow that research on Leggett’s theories is misguided. This would only be the case if the development and the investigation of these theories depended on the LR conception. That there is probably no such dependence can be seen already from the fact that one of the above-mentioned

132 � Causal Realism in the Context of Bell-Type Experiments critics of the LR view (Nicolas Gisin) co-authored one of the papers reporting experimental tests of Leggett’s models (Branciard et al. 2007). More generally, the following line of thought shows that one can reject the LR conception and still be interested in Legget-type theories: Once we recognize that Bell did not assume realism for the derivation of his theorem, we see that the LR view is mistaken in suggesting that the violations of Bell’s inequalities leave us with a choice to give up either locality or realism. Instead, they simply force us to give up locality. But this leaves open the question whether there is a sense of realism which has to be given up as well. Leggett can therefore be interpreted as investigating theories which give up locality (in accordance with the correct interpretation of Bell’s theorem) but hold on to realism in the sense of condition (7.5). I do not claim that this is how Leggett himself views the signi�cance of his models. Indeed, some remarks in the �rst and the last section of Leggett (2003) indicate his sympathy for the LR inspired view that the ruling out of his non-local realistic theories supports the idea of giving up realism instead of locality. If the LR view is false (as I think it is), then such a conclusion is untenable, because Bell’s theorem already rules out all local theories, be they realistic (in whatever sense) or not. But Leggett’s problematic conclusion can easily be dissociated from his research program: The investigation of Leggett-type models need not be seen as a test between non-local realism and local non-realism (whatever that may mean). Instead, we should see it as a test between two kinds of non-local theories, those which respect (7.5) and those which do not. If one wants to stick to the LR terminology, one may call these theories non-local realistic and non-local non-realistic, respectively, but these labels are not very helpful,²² except perhaps to drive home the point that there is a sense in which the investigation of non-local realistic theories, as proposed by Leggett, is perfectly acceptable even for the opponent of the LR view. Laudisa’s second reason to question the signi�cance of testing Leggett-type theories is that their realism assumption (which he denotes by R������G&�� , the subscript referring to Gröblacher et al.) already dooms them to failure in the light of previous no-hidden-variable theorems.

22 One of the reasons why the LR terminology is unhelpful is that it obscures the fact that (7.5) is itself a kind of locality condition. This is re�ected in the more appropriate terminology introduced by Colbeck and Renner (2008), who classify Leggett’s models as models for which the hidden variables have both a local and a global part, as opposed to entirely nonlocal models. This distinction forms the basis of their generalization of Leggett’s results: As they demonstrate, all models having a nontrivial local part (not just those which satisfy the particular condition (7.5)) are incompatible with quantum mechanical predictions.

Some Experimental Constraints on Explanations for EPR

� 133

If R������G&�� were an independent assumption of any hidden variable theory, GleasonBell-Kochen & Specker would have already proved their incompatibility with quantum mechanics needless of any locality requirement. (Laudisa 2008, 1122–1123)

Unfortunately, Laudisa here falls into the very trap that often undermines the intelligibility of the LR view: the use of an insu�ciently precise notion of “realism”. The conception of realism which he attributes to Gröblacher et al. includes noncontextuality in the following sense: “The physical systems under scrutiny are endowed with preexisting properties that do not depend essentially on the measurement interactions the systems themselves may undergo” (1113). If this is understood such that individual measurement outcomes are independent of what other measurements are made simultaneously, then it is indeed a form of noncontextuality which is incompatible with quantum mechanics, as shown by the (Gleason-Bell-)Kochen-Specker theorem. (For a concise presentation of the essential part of the theorem, see Bell 1966, sec. 5.) But do Gröblacher et al. (2007) really assume this kind of non-contextuality? At �rst sight, it seems that they do, by assuming that “all measurement outcomes are determined by pre-existing properties of particles independent of the measurement (realism)” (872). But just two sentences later they add that “measurement outcomes may very well depend on parameters in space-like separated regions”. And if one looks what condition of non-contextuality they actually use in their derivation, then it turns out that it is non-contextuality on the subensemble level, as spelled out in (7.5). This is compatible with a failure of non-contextuality on the level of individual measurements, as expressed in (7.4). Since the Kochen-Specker theorem only rules out theories which are non-contextual in this latter sense, it has nothing to say about the Leggett-type theories investigated by Gröblacher et al. In fact, it is rather surprising that Laudisa should have missed the importance of distinguishing the individual level from the (sub-)ensemble level when discussing non-contextuality, because in an earlier paper (1997, 492–493) he himself used precisely this distinction to show why neither Bell’s (1964) local hidden-variable theory nor Bohm’s theory fall prey to the existing no-go theorems for non-contextual hidden-variable theories. So the theorems cited by Laudisa do not su�ce to establish the incompatibility between Leggett’s models and quantum mechanics; the incompatibility only manifests itself through Leggett’s inequality and its violation by quantum mechanical predictions. These con�icting predictions can then be tested in experiments. But again, Laudisa (2008, 1123) disputes the signi�cance of such tests. Immediately following the sentence quoted above, he writes:

134 � Causal Realism in the Context of Bell-Type Experiments But, as Bell showed, there is little signi�cance in testing against quantum theory a theory (be it local or non-local) that is supposed to satisfy a condition that we already know quantum mechanics cannot possibly and reasonably satisfy.

I am not convinced by this reasoning. Indeed, the best counterexamples to this claim are Bell’s inequalities themselves. The fact that these inequalities are violated by the quantum mechanical predictions shows that quantum mechanics “cannot possibly and reasonably satisfy” the conditions assumed for their derivation. Should we therefore conclude that Aspect’s experiments (to name just the most famous example) are of “little signi�cance”? This would amount to a dubious a priori commitment to the truth of quantum mechanical predictions in domains where quantum mechanics has not yet been tested. I suspect that Laudisa fails to appreciate the force of this counterexample because he does not properly distinguish between theoretical and experimental aspects of Bell’s theorem. This can be seen by analyzing the following extract from what Laudisa o�ers as the logical reconstruction of the Bell-CHSH argument. (The complete argument includes six steps, but we only need to look at steps 2 to 4; here “QM” stands for “the assumption of the validity of the statistical predictions of quantum mechanics” (1124), while “BI” denotes Bell’s inequalities.) 2. QM → ¬BI [Experimental fact] 3. QM [Assumption] 4. ¬BI [2, 3 Modus ponens] (Laudisa 2008, 1127; square brackets in the original)

This way of putting things strikes me as thoroughly confused. First of all, the conditional in step 2 is not an experimental fact. That quantum mechanical predictions violate Bell’s inequalities can be derived (and is in fact derived by Clauser et al. 1969) on a purely theoretical basis, without the need for any experiment. What is an experimental fact is the observed violation of Bell’s inequalities, as expressed in step 4. By presenting this step as the result of a modus ponens, Laudisa creates the false impression that one needs to assume the truth of QM (step 3) in order to conclude that Bell’s inequalities are violated. Now it is true that, as a matter of practical fact, a commitment to some well-established parts of QM (concerning, for example, the production of entangled particle pairs) is involved in performing a Bell-type experiment. And it is also true that, historically, most performers of Bell tests expected all quantum mechanical predictions to turn out correct.²³ But

23 As a relevant counterexample, let me once again mention John Clauser (see footnote 20 above).

Some Experimental Constraints on Explanations for EPR

� 135

no assumption of the correctness of QM (in the sense that would logically imply violations of Bell’s inequalities, as in Laudisa’s step 4) is involved in performing these experiments. The very point of carrying out a Bell-test is to treat the correctness of theoretical predictions regarding BI as an open question, to be decided by experiment, rather than to be derived from our preconceived assumptions. This is how the result ¬BI is obtained. As a �nal argument against Leggett’s approach, Laudisa (2008, 1129) cites the consistency of Bohmian mechanics and claims that this “directly refutes the claims of Leggett and followers”. His idea seems to be that Bohmian mechanics, by being a hidden-variable theory which is consistent with all quantum mechanical predictions (and all empirical data), somehow invalidates Leggett’s claim that the hidden-variable theories he investigates are in con�ict with some quantum mechanical predictions (and with some empirical data). But of course, Bohmian mechanics could only serve as a counterexample in this sense if it belonged to the class of Leggett-type theories. Laudisa seems to believe that it does, as he claims that “Bohmian mechanics satis�es R������G&�� ” (ibid.), but this is clearly false.²� As described above, an essential component of R������G&�� is the commitment to a de�nite polarization of each photon, giving rise to the subensembles underlying (7.5). Bohmian mechanics, by contrast, does not regard polarization as a property of individual particles, and it is therefore not a ���������G&�� theory. Put in terms of hidden variables, the di�erence between Bohmian mechanics and a Leggetttype theory is that the two theories are committed to di�erent hidden variables: position in the �rst case, polarization in the second. While it is thus unjusti�ed to view Bohmian mechanics as directly refuting Leggett’s approach, a certain tension between the two cannot be denied. In particular, from a Bohmian perspective, it amounts to “naive realism” (Daumer et al. 1997) to treat as real any property other than position. But it is at least conceivable that the Bohmian perspective—whithin which position takes priority over all other properties—is not the only possible way to construct a hidden-variable theory. More speci�cally, if we speak about photons (which not even Bohmian mechanics describes as continuously localized particles), there is no a priori reason why we should regard position as a more fundamental property than polarization. Leggett’s proposal can then be seen as an attempt to explore the empirical

24 It should be false even by Laudisa’s own lights: Since he is convinced that R������G&�� is an unreasonable assumption (1128), satisfying it would make Bohmian mechanics an unreasonable theory, which is certainly not what he wants to claim. Accordingly, the claim that Bohmian mechanics satis�es R������G&�� does no longer appear in Laudisa’s (2014a, sec. 2) recent restatement of his (2008) argument.

136 � Causal Realism in the Context of Bell-Type Experiments consequences of an alternative to the Bohmian perspective. Under this interpretation, Bohmian mechanics actually brings out the relevance of Leggett-inspired research, instead of making it obsolete, as Laudisa supposes. Let me explain this by once more highlighting the parallel between Leggett’s and Bell’s inequalities. A frequently heard complaint about Bohmian mechanics is that it is non-local. The correct response to this is to refer to the experimental violations of Bell’s inequalities, which show that non-locality is not a peculiarity of Bohmian mechanics, but an experimental fact. In a parallel fashion, experimental violations of Leggett’s inequalities furnish a reply to another complaint that is sometimes made against Bohmian mechanics, namely its non-realism with respect to all properties except position.²� Experimental tests of Leggett-type models support the Bohmian approach by demonstrating that a realism about polarization, even in the modest sense of (7.5), is in con�ict with empirical data.²� The Bohmian should therefore not join Laudisa in denouncing Leggett’s research program as irrelevant, but should rather welcome it as signi�cantly supporting his own position.²�

25 Daumer et al. (1997, 382) even call this a “frequent complaint”, and their paper is a spirited reply to it. But in contrast to my argument from experimental violations of Leggett’s inequalities, their reply presupposes the Bohmian perspective and the privileged role it assigns to positions. 26 The work of Colbeck and Renner (2008) licenses an even stronger conclusion, by no longer depending on the speci�c form of the constraint (7.5) (see footnote 22 above), and it therefore strengthens the kind of support for Bohmian mechanics described here. However, in a more recent paper, Colbeck and Renner (2012, 13) criticize Bohmian mechanics for being incompatible with a very natural freedom of choice assumption. For a response to this criticism, see Ghirardi and Romano (2013), as well as Laudisa (2014a, sec. 4). 27 In his recent response to my arguments, Laudisa (2014b) still disputes the parallel I draw between Bell’s and Leggett’s inequalities. He is right, of course, that Bell’s locality assumption has much more intuitive plausibility than the somewhat arti�cial realism assumption from which Leggett’s inequality is derived. In that sense, violations of Bell’s inequalities do indeed have more foundational signi�cance than violations of Leggett’s inequalities. But to conclude from this that Leggett’s inequalities are not worth testing is to neglect an important methodological lesson from Bell-inspired research: Since even such an intuitive assumption as locality turned out to be untenable, we should not put too much trust in our intuitions concerning the plausibility of assumptions, before we have tested them experimentally.

8 Delayed-Choice Experiments and the Metaphysics of Entanglement The experiments discussed in the previous chapter mitigate to some extent the problem of underdetermination in quantum mechanics, and thereby support causal realism. At the same time, the discussion of Leggett’s models made clear that some properties (such as polarization) do not admit of a straightforward realist interpretation. In the present chapter, I will discuss a family of experiments which could be thought to push us even further in that direction, because they seem to suggest that quantum states in general do not represent real properties in the world. I will be particularly interested in the question whether entanglement relations are real physical relations, because an a�rmative answer to this question is presupposed by many authors who draw substantial metaphysical conclusions from quantum mechanics (Teller 1986; French 1989; Maudlin 1998; Esfeld 2004). The ontological status of entanglement relations is of interest for any realistic view about quantum mechanics, not just for causal realism. Accordingly, the arguments of the present chapter will not make use of the conceptual apparatus of causal realism, but will be of a more general (realist) type. Their importance for causal realism stems from the fact that the experiments to be discussed here might seem to undermine a central intuition of causal realism, namely that properties which can be detected, manipulated, and used to perform various tasks must be real. The recent �ourishing of quantum information theory—where such processes as entanglement detection, entanglement distillation, entanglement swapping, and various forms of entanglement-assisted communication play a central role—leaves little doubt that entanglement is indeed such a property.¹ Should it nevertheless turn out that there are cogent arguments against a realistic view of entanglement, surely this would be a problem for causal realism. It is therefore an important task for the causal realist to analyze the experiments purporting to furnish such arguments. To see more concretely what is at stake, let us brie�y look at the case of entanglement swapping, which I will discuss in more detail in section 8.3. In entanglement swapping, two pairs (A, B) and (C, D) of entangled particles are created by two independent sources. If one then performs the right kind of joint measurement on particles B and C, the pair (A, D) enters into an entangled state even though A and D have never interacted with each other. Interestingly, this

1 For an extensive review of the current state of play, see Horodecki et al. (2009).

138 � Delayed-Choice Experiments and the Metaphysics of Entanglement phenomenon has given rise to contradicting ontological conclusions. On the one hand, Rob Clifton (2002, S163) takes it to support a realistic view of entanglement: “It appears that there is su�cient substantiality to entanglement that it can be swapped from one pair of particles to another”. On the other hand, Richard Healey (2012, sec. 4.5) studies an entanglement-swapping experiment in which the measurement on B and C is performed after A and D have been detected. Since this seems to imply that entanglement can be transferred to a pair of particles which no longer exists, Healey (2012, 759) concludes: The delayed-choice entanglement-swapping experiment reinforces the lesson that quantum states are neither descriptions nor representations of physical reality. In particular, it undermines the idea that ascribing an entangled state to quantum systems is a way of representing some new, non-classical, physical relation between them.²

Obviously, the “delayed-choice” clause plays a central role here.³ I will therefore begin my investigation with a brief reminder of the simple and well-known delayedchoice double-slit experiment, assessing its impact on realism about the state of the quantum system (section 8.1). A more sophisticated (and more radical) version of delayed choice, the so-called quantum eraser, will be discussed in section 8.2. The case of the quantum eraser is important because it introduces the idea of sorting experimental results into di�erent subensembles, thus raising the question whether these subensembles correspond to real properties of the system. In section 8.3, I will apply these considerations to the entanglement-swapping experiment and show that if the experiment is carried out in a delayed-choice setting, no actual entanglement swapping occurs. This will lead to the conclusion that delayed-choice entanglement swapping does not undermine realism about entanglement relations.

2 Healey advances this argument in the context of his pragmatist approach to quantum theory, which I will not discuss here. Neither will I discuss the positions of those who take quantum information theory to support an epistemic or informational (as opposed to metaphysical) view of the quantum state. See Timpson (2010) for a critique of these approaches. 3 Seevinck (2006) has argued that the phenomenon of entanglement swapping by itself (even without adding the delayed-choice condition with which this chapter will be concerned) calls into question the “ontological robustness” of entanglement. See Timpson and Brown (2010) for a response to this kind of argument.

Delayed Choice in the Double-Slit Experiment �

139

�.� Delayed Choice in the Double-Slit Experiment The double-slit experiment is probably the best known illustration of the basic mystery of quantum mechanics (see section 4.4). If quantum particles (e.g., electrons) are sent through a double slit, a characteristic interference pattern appears on the screen behind the two slits. However, this pattern disappears as soon as one tries to detect through which of the two slits each electron passed. It thus seems that the electrons either behave as waves (passing through both slits and producing an interference pattern) or as particles (passing only through one slit and displaying no interference), depending on the kind of experiment we choose to perform. (In the following, I will refer to the two kinds of experimental arrangements as “DS” (for “double slit”) and “WW” (for “which way”), respectively.) This is already puzzling enough, but further puzzlement is added by the insight that the decision to perform either a DS or a WW experiment can be taken after the electron has passed through the double slit. It was John A. Wheeler (1978) who introduced this idea of a delayed choice, and he took it to imply that “the past has no existence except as it is recorded in the present”, and that “the universe does not ‘exist, out there,’ independent of all acts of observation” (41). It is not hard to see how delayed-choice experiments can lead to such antirealistic conclusions. If we think of the electron as traveling from the source to the double slit and then to the screen where it is detected, a natural question to ask is whether the electron behaved as a wave or as a particle at the time it traveled through the double slit. (That electrons are disposed to behave in either of the two ways is already known from DS and WW experiments without delayed choice.) Now if the type of experiment (DS or WW) is �xed in advance, this determines the behavior of the electron, and a unique story about its wave- or particle-like nature can be told for each type of experimental setup. However, in the delayed-choice case, the experiment-type is not yet �xed at the time the electron is at the double slit, so it seems that there is simply no fact of the matter as to whether the electron passes through both slits (as waves do) or through only one slit (as particles do). It is thus clearly impossible to tell a simple realistic story about what happens at the double slit in a delayed-choice experiment, if by “realistic” we mean that the story should attribute a de�nite, observer-independent behavior to the electron. More sophisticated realistic stories remain, of course, possible, but they do not come without a cost. In the next section, I will argue for such a story, based on the formalism of standard quantum mechanics. I will therefore not have much to say about non-standard quantum theories, such as Bohmian mechanics or the GRW theory. But I conclude the present section with some brief remarks about these two theories, in order to illustrate to what extent the delayed-choice double slit complicates the realist’s ontological commitments.

140 � Delayed-Choice Experiments and the Metaphysics of Entanglement At �rst sight, it seems that Bohmian mechanics has a straightforward answer to the question of what happens at the double slit: Being a particle theory, Bohmian mechanics clearly tells us that each electron goes only through one slit. But it also tells us that the movement of the particle is determined by the wave function, and this raises the tricky question of the ontological status of the latter. Some versions of the theory interpret the wave function as a physical entity which literally guides the particles, or they introduce a so-called quantum potential which gives rise to non-classical forces acting on them. But as already mentioned in section 7.2, there is now a tendency among Bohmians to regard the wave function no longer as a physical entity, but merely as a component of the law according to which the particles move. However, in their detailed analysis of delayed-choice experiments from a Bohmian perspective, Hiley and Callaghan (2006a) manage to avoid a commitment to very bizarre particle trajectories only by relying explicitly on the physical reality of the wave function and the quantum potential, so there is reason to doubt that these entities can really be cut o� from the ontology of Bohmian mechanics without a loss. An alternative way to tell a realistic story about the double slit is given by the GRW theory, which adds spontaneous collapses of the wave function to the Schrödinger evolution. I already mentioned in section 7.2 that this theory is also ambiguous in its ontological commitments. In the GRWm version, the wave function describes a matter density in space. Again, this seems to suggest a straightforward solution to our problem: The matter density, being a spatially extended �eld, (almost) always passes through both slits and collapses to a particle-like object only upon interaction with the detecting screen. The fact that the experimental setup can be chosen after the matter wave has passed the double slit then poses no particular problem. However, according to this story, the result of a WW experiment must be regarded as outright illusory: Even though it looks as if the electron went only through one slit (the experiment telling us which one), the fact is that it went through both. An even more severe illusion takes place according to the GRWf version of the theory, which is merely committed to the existence of some �ashes in space-time. In this picture, contrary to what we might take as an unquestionable truth about any double-slit experiment, nothing at all travels from the source to the screen.

�.� The Quantum Eraser In the experiments discussed so far, the DS/WW decision is taken after the electron has passed through the double slit, but it obviously has to be taken before the electron is detected. Using a quantum eraser (�rst proposed by Scully and Drühl

The Quantum Eraser

� 141

Fig. 8.1. The quantum eraser thought experiment. (Scully et al. 1991, 115; reprinted by permission from Macmillan Publishers Ltd: Nature, © 1991)

1982, experimental realization by Walborn et al. 2002), even this restriction can be removed. Consider the thought experiment by Scully et al. (1991) depicted in �gure 8.1: In a double-slit experiment with atoms, we place a micromaser cavity in front of each slit. The cavities are designed such that excited atoms passing through them inevitably decay into the ground state by emitting a photon. So for © 1991 Nature Publishing Grou each atom, the corresponding photon emitted in one of the cavities provides us with WW information. However, this information can be “erased” by opening the shutters which separate the two cavities from a thin-�lm photodetector placed between them. Since this “detector wall” does not discriminate between photons coming from one or the other cavity, the WW information is lost and a situation resembling the original DS con�guration is reestablished. So the experimenter has two options: He can either leave the shutters in place and detect which of the cavities contains the photon, thereby obtaining WW information, or he can open the shutters, allowing the photon to interact with the detector wall without yielding WW information. Note that he can (in principle) decide between these two options after the atom is detected at the screen. But one might ask how this can really be a choice between a DS and a WW scenario, given that the two scenarios should lead to radically di�erent distributions of atoms on the screen (displaying interference fringes in one case but not in the other). Surely the pattern on the screen can not be changed retroactively? Well, in

142 � Delayed-Choice Experiments and the Metaphysics of Entanglement a certain sense, it can. To see how, a closer look at the quantum mechanical description of the atom-photon-system is necessary.� If we denote the photon state by |�� or |��, depending on whether the photon is in cavity 1 or 2, and the spatial wave function of an atom coming through one of the two slits by ψ� (x) and ψ� (x), respectively, the state of the system after the atom has passed the slits is � �

|Ψ �(x) = √ [|��ψ� (x) + |��ψ� (x)].

(8.1)

The probability density �Ψ(x)�� associated with this state has vanishing interference terms, because |�� and |�� are orthogonal to each other. Therefore, the distribution of atoms on the screen shows no interference fringes, which is what we expect for a WW experiment. Now let us see what happens if the shutters are opened to erase the WW information. As mentioned above, the detector wall does not discriminate between the |�� and the |�� state. However, it does (maximally) discriminate between the symmetric and the antisymmetric superposition states � �

|+� = √ [|�� + |��]

� and |−� = √ [|�� − |��]. �

As a consequence, the detector records only photons in the |+� state and ignores photons in the |−� state. Introducing the corresponding symmetric and antisymmetric states of the atom � ψ± (x) = √ [ψ� (x) ± ψ� (x)], � we can rewrite (8.1) as � �

|Ψ �(x) = √ [|+�ψ+ (x) + |−�ψ− (x)].

(8.2)

Since this is just another way of expressing the same state |Ψ �(x), the probability distribution �Ψ(x)�� is still the one corresponding to the WW setup. But if we restrict our attention to those atoms for which the photodetector records a photon, the contribution from |−�ψ− (x) vanishes and the probability distribution becomes P+ (x) = |ψ+ (x)|� =

� |ψ� (x) + ψ� (x)|� , �

which simply corresponds to the result of a DS experiment, displaying the usual interference fringes. Conversely, selecting atoms for which the detector does not record a photon yields P− (x) = |ψ− (x)|� , which corresponds to the complementary 4 For the following, I adopt the notation of Englert et al. (1999).

The Quantum Eraser �

143

“anti-fringe” interference pattern (see �gure 8.1b). So each atom can be assigned to one of the four subensembles “1”, “2”, “+” and “−” in the following way: By measuring the photons in the WW con�guration (shutters closed) we classify the atoms into the subensembles 1 and 2, by measuring the photons in the DS con�guration (shutters open), we carry out a +/− classi�cation. To assess the metaphysical signi�cance of delayed choice in this context, we now need to ask to what physical property (if any) the sorting into subensembles corresponds. Focusing for the moment on subensembles 1 and 2, the answer seems obvious: All atoms in subensemble 1 went through the �rst slit, the atoms in subensemble 2 through the second one. Englert et al. (1999, 328) endorse this view, but they add an antirealistic twist: “The ‘. . . went through . . . ’ is not a statement about the atom’s past”. This reinterpretation of everyday language is motivated by their “minimalistic interpretation” of the quantum state, which I will not further discuss (see footnote 2 above). In a closely related proposal, Ulrich Mohrho� (1999) invokes a kind of retrocausation, according to which the present determines the past of the atoms. (Notice the similarity to Wheeler’s above-mentioned view that the past’s existence depends on its being recorded in the present.) This commitment to retrocausation renders Mohrho�’s “reality-of-phenomena point of view” rather unattractive, but one might be willing to accept this consequence. What one should not accept is Mohrho�’s (1999, 332) claim that “retrocausation is a necessary feature of any realistic interpretation of the quantum formalism”. He reaches this conclusion by (allegedly) showing that the alternative “reality-ofstates point of view" is not viable.� Here is his argument: Adherents to the reality-of-states view thus �nd themselves faced with a dilemma. If they . . . deny the possibility of retrocausation, they must insist that it is only as if the atom had traveled through the �rst cavity or only as if it had been retroactively furnished with a de�nite phase relation. They cannot say that the atom really was in the state ψ� (or ψ� , or ψ+ , or ψ− , as the case may be). And so they �nd themselves compelled to foreswear realism and embrace operationalism. And if they stick to realism, they will have to drop the as if ’s and accept the reality of retrocausation. (ibid.)

5 I should mention that Mohrho� (1996) has earlier defended a “reality-of-states point of view” himself. Therefore, I cannot claim originality for much of what I will have to say in the rest of this section. In response to criticism by Englert et al. (1999), Mohrho� (1999) later abandoned this view in favor of a “reality-of-phenomena point of view”, which may be seen as a kind of compromise between the minimalistic interpretation advocated by Englert et al. and realism about the quantum state. The details of this debate do not matter here, because I am not attempting to evaluate the �nal positions of either of the two parties, but am only concerned with the claim that quantum state realism is not viable, a claim which seems to be common ground between Englert et al. and Mohrho�’s more recent view.

144 � Delayed-Choice Experiments and the Metaphysics of Entanglement But the �rst thing a reality-of-states view should take seriously is the fact that |Ψ � in (8.1) and (8.2) is an entangled state of the atom-photon system, so it is clear from the start that none of the four (pure) ψ-states can be ascribed to the atom alone, as long as the system is in state |Ψ �. It is true that this commits the realist to Mohrho�’s as-if -statements (compare my remarks on the GRWm story in section 8.1), but to say that reality di�ers from what measurements seem to reveal is very di�erent from saying (as the operationalist would) that there is no reality beyond measurements. The second thing a reality-of-states view should take seriously is state reduction. After a measurement on one of the two particles, standard quantum mechanics� no longer describes the atom-photon system by |Ψ �, but by a separable state. A realistic view of the quantum state suggests that this change of description corresponds to a real physical change. This implies that the metaphysical signi�cance of the sorting of atoms into subensembles depends on the temporal ordering of the measurements.� If the photon is measured prior to the atom’s arrival at the screen, the subensembles correspond to real properties, because the photon measurement brings about the state reduction |Ψ �(x) → |i�ψ i (x),

i ∈ {�, �, +, −},

(8.3)

so that each atom actually is in one of the ψ i states prior to its hitting the screen. But if the time order of the two measurements is reversed, the atom never is in any one of these states, because (8.3) does not occur. Instead, the atom’s arrival at the screen (at a location x� ) results in a state reduction of the form |Ψ �(x) → [α |�� + β |��]ϕ x� (x),

(8.4)

where ϕ x� (x) is a spatial wave function well localized at x� .� In this case, assigning the atom to a subensemble depending on the result of the photon measurement implies nothing about the physical state of the atom, whether past or present.� 6 As is well known, there is no satisfying characterization of “measurement” within standard quantum mechanics (Bell 1990a). The reality-of-states view presented here is compatible with di�erent solutions to this problem, e.g., the GRW theory or an Everett-type approach. In the latter case, state reduction is to be understood as a splitting of worlds due to environment-induced decoherence. ome additional remarks on how this plays out in the experiments discussed here will be given in footnotes 9 and 12. 7 This becomes problematic in relativistic settings, testifying to the unsolved problem (familiar from chapter 7) of reconciling quantum non-locality with relativity. I am here only interested in the non-relativistic case. 8 For details about the coe�cients α and β, see Englert et al. (1999, eq. 8). Of course, we could equally well express the photon state in the |+�, |−� basis. 9 Although I have derived this result within a view that takes state reductions as real events, it is interesting to note that the result is not peculiar to such a view. Hiley and Callaghan (2006b)

Delayed-Choice Entanglement Swapping

�

145

But then, isn’t the appearance of de�nite (WW or DS) patterns within the subensembles somewhat miraculous? I do not think so. Given that the atom and the photon formed an entangled system up to the moment of the atom’s detection, it is not so surprising that we can obtain interesting patterns by correlating the location of the atom’s detection with the result of a posterior photon measurement. But this correlating is something the experimenter needs to do; the correlation is no longer “there”, once the transition (8.4) has occurred.

�.� Delayed-Choice Entanglement Swapping We can now apply the foregoing considerations to the process of entanglement swapping, �rst proposed by Yurke and Stoler (1992; see also Żukowski et al. 1993) and experimentally realized by Pan et al. (1998). In the simplest case, this involves entangled pairs of two-state systems, which can be conveniently described by introducing the four so-called Bell states: � �

|ψ± � = √ [|��|�� ± |��|��],

� �

|ϕ± � = √ [|��|�� ± |��|��].

Now consider two independent sources, each one emitting a particle pair in the state |ψ− �. Denoting the two pairs by (A, B) and (C, D) respectively, the state of the complete system is given by |Ψ � = |ψ− �AB |ψ− �CD .

(8.5)

analyze the situation from a Bohmian perspective, which does not regard state reductions as real, but instead assigns a de�nite trajectory to each atom. Their conclusion is analogous to mine, namely that the retrospective sorting into subensembles does not correspond to di�erences in the trajectories of the atoms. From an Everettian perspective, there is a sense in which the subensembles never have any metaphysical signi�cance: Strictly speaking, according to Everettian quantum mechanics, superpositions never disappear, hence if the atom started out as part of an entangled system, it never enters into one of the ψ i states. But in that sense, no experience ever has metaphysical signi�cance, because our experiences keep telling us that the world around us is in some unique, nonsuperposed state, while it really is not. Everettians usually prefer a less drastic formulation, according to which our experience is not so much illusory as merely incomplete: The macroscopic world we experience then counts as perfectly real, but it is only one branch among in�nitely many others which we do not experience. With such a branch-relative notion of reality in place, the metaphysical signi�cance of the measurement subensembles can be assessed in exactly the same way as in a theory with state reduction. If the photon is measured �rst, branching occurs (due to decoherence), and within each branch, the atom really (in the branch-relative sense) enters into one of the ψ i states. By contrast, if the measurement of the photon is delayed, the atom never is in such a state (not even in the branch-relative sense), because no branching occurs until the atom hits the screen.

146 � Delayed-Choice Experiments and the Metaphysics of Entanglement This is obviously a separable state, re�ecting the fact that the two pairs are mutually independent. But now suppose that particles B and C are sent to the same location, where a Bell measurement is performed on them, such that their joint state is projected onto one of the four Bell states |ψ± �BC , |ϕ± �BC .¹� To see how this a�ects particles A and D, we rewrite equation (8.5) by expressing |Ψ � in the basis given by the Bell states of the pairs (A, D) and (B, C): |Ψ � =

� �� + |ψ �AD |ψ+ �BC − |ψ− �AD |ψ− �BC − |ϕ+ �AD |ϕ+ �BC + |ϕ− �AD |ϕ− �BC . (8.6) �

Since the Bell states are orthogonal to each other, the Bell measurement on the (B, C) pair projects the state |Ψ � onto an entangled state of the (A, D) pair, for example: � �ψ+ |BC |Ψ � = |ψ+ �AD , � and analogously for the other Bell states. Thus particles A and D emerge as an entangled pair, although they never interacted with each other. As Asher Peres (2000) points out, this procedure can be carried out in a delayed-choice mode, such that the decision to perform a Bell measurement on the (B, C) pair may take place after any measurements on the (A, D) pair.¹¹ But since particles A and D only become entangled with each other if the (B, C) measurement is actually performed, it seems that “entanglement is produced a posteriori, after the entangled particles have been measured and may no longer exist” (Peres 2000, 139). We have seen at the beginning of the present chapter that Healey takes this to undermine the idea that entanglement is a physical relation. To support his view, he o�ers the following reductio ad absurdum: To hold onto that idea in the context of this experiment would require one to maintain not only that which entanglement relation obtains between a pair of photons at some time, but also whether any such relation then obtains between them, depends on what happens to other independent systems later, after the pair has been absorbed into the environment. (Healey 2012, 759)

There is a clear parallel between this argument and the discussion in section 8.2, and this might seem to force me into accepting Healey’s conclusion. In section 8.2, my unwillingness to accept retrocausation led me to reject the claim that the atom really went through either one of the slits (even if a WW measurement seems to tell

10 This is an idealized description. In practice, a Bell measurement is unable to identify all of the four Bell states. For technical reasons, experiments usually focus on the singlet state |ψ− � (Pan et al. 1998; Jennewein et al. 2002). 11 The proposal has recently been experimentally realized (Ma et al. 2012).

Delayed-Choice Entanglement Swapping

�

147

us so). Should it then not also lead me to reject the claim that the (A, D) pair really either was or was not in an entangled state prior to the (B, C) measurement? But such an inde�niteness seems incompatible with the view that entanglement relations are real. This result would be particularly troubling in view of the fact that the notion of an entangled state played a crucial role in the account I defended in section 8.2. Yet, a closer look reveals that the parallels between the two cases do not threaten realism about entanglement. Rather, they can be exploited to refute Healey’s argument. In section 8.2, I showed that a delayed measurement of the photon results in a sorting of atoms into subensembles which do not correspond to any physical properties of the atoms. Precisely the same thing can happen in entanglement swapping: The Bell measurement on the (B, C) pair allows us to sort the (A, D) pairs into four subensembles corresponding to the four Bell states. Without delayed choice, this has physical signi�cance, because each (A, D) pair actually is in such a state after the (B, C) measurement.¹² But if the (A, D) measurements precede the (B, C) measurement, the (A, D) pair never is in any of these states. This is entirely compatible with the fact that evaluating the (A, D) measurements within a certain subensemble shows Bell-type correlations, just as the subensembles in section 8.2 showed interference or WW patterns. Therefore, far from being committed to any indeterminacy about entanglement (or any backward-in-time in�uences), a realistic view of the quantum state yields a perfectly clear assessment of what happens in entanglement swapping: If the (B, C) measurement occurs at a time the complete system is still in state |Ψ �, it confers entanglement on the (A, D) pair, if it occurs at a later time, it does not.¹³ To sum up: I have argued that delayed-choice experiments do not undermine a realistic view of the quantum state. In the case of the double slit, we saw that they merely undermine a simplistic realism which unre�ectively identi�es the result of a WW experiment with a statement about which slit the particle went through. The quantum eraser and the case of delayed-choice entanglement swapping required a more careful treatment, because one needs to get clear about the metaphysical signi�cance of the subensembles appearing in these experiments.

12 In an Everettian context, “really” here again has to be understood in a branch-relative sense. Analogously to the situation described in footnote 9, the important contrast is that in the delayedchoice case, the (A, D) pair never (not even in a branch-relative sense) is in a pure state. 13 Megidish2013 have recently reported a rather spectacular variant of a delayed-choice entanglement-swapping experiment, in which the photons A and D not only never interact, but never even coexist. My analysis straightforwardly applies to this case as well: Since the complete system is not in state |Ψ� at the time the (B, C) measurement occurs (in fact, there is no fourparticle system at any time in this experiment), no actual entanglement swapping takes place.

148 � Delayed-Choice Experiments and the Metaphysics of Entanglement Once this is achieved, a straightforward reality-of-states story can be told for these seemingly troubling cases. This does not, of course, exclude non-realism about the quantum state. But it seems to me that the empirical success of quantum mechanics gives us at least some prima facie reason to view quantum states as describing an independent reality. The non-realist then needs an argument for the claim that this is a mistake. Wheeler, Englert et al., Mohrho� and Healey all think that delayed-choice experiments furnish such an argument. This I have shown not to be the case. The same dialectic applies, more speci�cally, to the metaphysics of entanglement. The various things physicists can do with entanglement support the intuition that there must be some reality to it. Against this, Healey argues that delayedchoice entanglement swapping implies an indeterminacy of entanglement which is incompatible with realism. Having shown that the realist can avoid such an indeterminacy, I conclude that we can, until further notice, consider entanglement as a real relation.

9 Particle Physics without Particles? On Causal Realism in Quantum Field Theory Most experiments in particle physics are performed at high energies. This forces us to go beyond the framework of non-relativistic quantum mechanics and to turn to relativistic quantum theories, or more speci�cally, quantum �eld theories (QFT). The puzzle that arises in the wake of this transition is that these theories, which are supposed to furnish the theoretical description of particle physics, militate against an interpretation in terms of particles. Sections 9.1 and 9.2 will review the most important theorems challenging the particle notion. Section 9.3 will then show how the causal realist can respond to these challenges. Finally, section 9.4 connects the discussion of QFT with some more general metaphysical issues confronting causal realism.

�.� Against Localizability: Malament’s Theorem and Its Generalizations In spite of the non-locality discussed in chapter 7, we have up to now regarded particles as the kind of entities which can have local e�ects: The pre-measurement state of a pair of particles in an EPR experiment may be non-local, but it manifests itself upon measurement as two clearly separate detection events, each one localized in one of the wings of the experiment. In short, quantum mechanical particles are in general not localized, but they are localizable. If we now turn to relativistic quantum theory, even localizability becomes problematic. There are several theoretical results supporting this claim, and I will not discuss all of them. To grasp the basic idea, it will be su�cient to look at a fundamental theorem by David Malament (1996) in some detail (but not in complete mathematical rigor) and to then brie�y discuss some generalizations proposed by Hans Halvorson and Rob Clifton (2002). I will add some remarks on speci�c aspects of the theorems, but I defer the discussion of their general ontological impact until section 9.3. To formalize the notion of localizability for a single particle, Malament (1996, 2–5) considers what he calls “spatial sets”, that is, bounded open subsets ∆ of spatial hyperplanes in Minkowski spacetime M. If the Hilbert space H is the state space of the system under study, then the detection of the particle in ∆ corresponds to a projection operator P ∆ on H. Localizability then means that the following condition is satis�ed:

150 � Particle Physics without Particles? On Causal Realism in Quantum Field Theory (1) Localizability: For two disjoint subsets ∆� and ∆� of a single hyperplane, P ∆� P ∆� = P ∆� P ∆� = �. Intuitively, this amounts to saying that the particle cannot be detected in two disjoint spatial regions at a given time. There are three more conditions necessary to derive Malament’s theorem. Two of them are related to the unitary representation a �→ U(a), where a is a translation vector in M and U(a) a unitary operator on H. We then require: (2) Translation Covariance: Let ∆ + a denote the set that results from translating the spatial set ∆ by the vector a. Then P ∆+a = U(a)P ∆ U(−a), for all ∆ and a. (3) Energy Condition: For any timelike vector a in M, the Hamilton operator H(a) given by U(ta) = e−itH(a) , t ∈ R has energy spectrum bounded from below.

These are very natural assumptions for any quantum theory. However, a reason to question the applicability of (2) will be discussed below. Finally, special relativity requires that measurements at spacelike separation do not in�uence each other’s statistics. This is captured by the condition that projectors associated with spacelike separated sets commute with each other: (4) Locality: For two spacelike separated sets ∆� and ∆� , P ∆� P ∆� = P ∆� P ∆� . As Malament (1996, 5n5) shows, failure of this condition would imply that the act of performing a measurement at ∆� would in�uence the statistics for a measurement at ∆� in a way that could be exploited for superluminal signaling. Notice that this is a weaker locality condition than the one used in the derivation of Bell’s inequalities discussed in chapter 7. The non-locality implied by the violation of Bell’s inequality does not allow superluminal signaling and is therefore compatible with locality in the sense of condition (4). Having formulated these four conditions, Malament (1996, 6–9) proves the following Theorem: If the structure (H, a �→ U(a), ∆ �→ P ∆ ) satis�es conditions (1)–(4), then

Against Localizability: Malament’s Theorem and Its Generalizations � 151

P∆ = �

(9.1)

for all spatial sets ∆. In other words: Given the conditions mentioned above, the probability of detecting the particle in any spatial set is 0. Consequently, the concept of a detectable particle is incompatible with the conjunction of conditions (1)–(4). Malament (1996, 1) takes this to support the following “received view”: In the attempt to reconcile quantum mechanics with relativity theory, . . . one is driven to a �eld theory; all talk about ‘particles’ has to be understood, at least in principle, as talk about the properties of, and interactions among, quantized �elds.

One way to avoid this conclusion is to question the reasonableness of Malament’s assumptions (1)–(4). Halvorson and Clifton (2002, 5–7) discuss and dismiss several such attempts. I will here focus my attention on Je�rey Barrett’s (2002) objection, which, in the light of recent theoretical developments, deserves a more favorable comment than the one given by Halvorson and Clifton. Barrett’s claim is that the metaphysical impact of Malament’s theorem depends crucially on how one goes about solving the quantum measurement problem. More speci�cally, taking state reductions seriously (as, e.g., a GRW-type solution to the measurement problem does) implies that the dynamics of the quantum state is in general not unitary, so condition (2) is violated (Barrett 2002, 177). Halvorson’s and Clifton’s reply to this objection has two parts. The �rst one is as follows: Existing (non-relativistic) collapse theories take the empirical predictions of quantum theory seriously. . . . However, in the present case, Malament’s theorem shows that any quantum theory predicts that if there are local particle detections, then act-outcome correlations are possible at spacelike separation. Thus, if a collapse theory is to reproduce these predictions, it too would face a con�ict between localizability and relativistic causality. (Halvorson and Clifton 2002, 6)

The authors here obviously take conditions (2) and (3) as necessary components of any quantum theory; otherwise, they could not simply claim that the occurrence of local particle detections implies the possibility of spacelike act-outcome correlations (or, in more formal terms: that (1) and ¬(9.1) together imply ¬(4)). In this sense, the GRW theory is not, strictly speaking, a quantum theory. But insofar as it aspires to reproduce the empirical predictions of quantum mechanics, it seems that the theory is subject to Malament’s theorem and therefore has to forgo either localizability or relativistic causality. However, Roderich Tumulka’s (2006)

152 � Particle Physics without Particles? On Causal Realism in Quantum Field Theory relativistic version of the GRW theory is a clear counterexample to this claim: Its fundamental entities, the �ashes, are as localizable as anything can be, and relativistic causality is ensured by the theory’s manifest Lorentz covariance (see, however, Esfeld and Gisin 2014). Relativistic �ashy GRW also furnishes a partial answer to the second part of Halvorson’s and Clifton’s criticism: Furthermore, noting that Malament’s theorem requires unitary dynamics is one thing; it would be quite another thing to provide a model in which there are localizable particles— at the price of non-unitary dynamics—but which is also capable of reproducing the wellcon�rmed quantum interference e�ects at the micro-level. (7)

It would not be completely accurate to cite Tumulka’s model as ful�lling these requirements, because it clearly is not a theory of localizable particles. But neither is it a �eld theory, which shows that the above-mentioned “received view” does not follow from Malament’s theorem. I now turn to generalizations of Malament’s theorem. The �rst one starts with the observation that, beside assumptions (1)–(4), the theorem relies on a further tacit assumption, namely that there is no preferred inertial reference frame and thus no absolute velocity (Halvorson and Clifton 2002, 8). Combining Malament’s theorem with a theorem by Hegerfeldt (1998), Halvorson and Clifton manage to derive a theorem which no longer needs to ban absolute velocities, but only adds a very weak probability conservation requirement to Malament’s four explicit conditions. A further generalization can be achieved by relaxing the assumption that position measurements must be represented by projection operators. Instead, they can be represented by so-called e�ects, that is, operators A satisfying �ψ, Aψ� ∈ [�, �] for any unit vector ψ ∈ H. The class of projectors is a proper subclass of the class of e�ects. The physical signi�cance of the generalization from projectors to e�ects is the following: While projectors localize states in �nite regions of space with sharp boundaries, e�ects can be seen as representing “unsharp localization”. Building on an earlier result by Busch (1999), Halvorson and Clifton (2002, 16) now show that a theorem analogous to Malament’s holds even for unsharp localization. Until now, we were concerned with the localization of a single particle. Turning to many-particle systems, we associate a number operator N ∆ with every spatial set ∆, representing the number of particles in ∆. Instead of condition (1) in Malament’s theorem, we now require (Halvorson and Clifton 2002, 19): (1a) Additivity: For two disjoint subsets ∆� and ∆� of a single hyperplane, N ∆� + N ∆� = N ∆� ∪∆� .

Against Localizability: Malament’s Theorem and Its Generalizations �

153

(1b) Number conservation: There is a well-de�ned global number operator N on H, and U(a)NU(−a) = N for any timelike translation a of M. Malament’s theorem then generalizes to the multi-particle case in the following way: Theorem: If the structure (H, a �→ U(a), ∆ �→ N ∆ ) satis�es conditions (1a–b), (2)–(4)¹ and if there are no absolute velocities, then N∆ = �

(9.2)

for all spatial sets ∆. Let me conclude this discussion with three remarks on Malament’s theorem and its generalizations. Firstly, all these theorems are exclusively concerned with noninteracting states. This is obvious for the one-particle case, but it also holds for the multi-particle case, since condition (1b) is violated for interacting states. However, this does not mean that the prospects for the particle concept are any better in interacting theories, as will be seen in the next section. Secondly, the mathematical apparatus needed to derive these theorems does not go beyond standard quantum mechanics. In particular, the derivations do not make use of algebraic quantum �eld theory (AQFT).² This di�erentiates them from similar results based on the Reeh-Schlieder theorem (Redhead 1995) and from the results to be discussed in the next section. Nevertheless, there is a close relation to the AQFT approach, which also begins by laying down some general conditions supposed to hold for any relativistic quantum theories, and then goes on to derive some consequences from them. Therefore, the critique of AQFT to be discussed in section 9.3 equally applies to the theorems introduced here. Finally, one might wonder how the above results are compatible with the manifestly particle-like phenomena observed in experiments. Here, Halvorson and Clifton (2002, 20–22) turn to AQFT to explain how the illusion of detecting localizable particles can arise while there are, strictly speaking, no such things. One

1 with the obvious substitutions P ∆ → N ∆ in (2) and (4) 2 I should mention that the proof of Malament’s theorem relies on a lemma by Borchers (1967), which Kuhlmann (2009, sec. 4.3) counts among the results of AQFT. This is justi�ed in the sense that Borchers derives his lemma in the context of investigating the C* -algebra generated by all local observables. However, the lemma itself (Theorem III.1 in Borchers 1967) is independent of this investigation and its proof does not employ any methods or concepts speci�c to AQFT.

154 � Particle Physics without Particles? On Causal Realism in Quantum Field Theory of the basic ideas of AQFT is the correspondence between bounded open subsets O ⊂ M and the subalgebras R(O) of observables measurable in O. The above results then translate into the statement that particle detections cannot be represented by operators in R(O). However, for any δ > �, a particle detection can be represented by an operator C for which there is an operator C′ ∈ R(O), such that �C − C′ � < δ. This means that, even though there are no localizable particles in the strict sense, particle detections can always be simulated to a high degree of accuracy by local measurements.

�.� Against Countability: Unruh E�ect and Interacting Fields Beside localizability, a second key feature associated with the concept of a particle is discreteness or countability: Particles come in well-de�ned numbers. As with localizability, already non-relativistic quantum mechanics forces us to acknowledge that particles are not countable in quite the same way as classical objects. Classically, one way of counting a given set of objects is to assign a label to each one of them (e.g., “�rst”, “second”, . . . ). The standard reading of quantum statistics has it that this is not possible for quantum objects (French 2011a, sec. 2; see sec. 3–4 for dissenting views). Nevertheless, quantum mechanical particles are countable in the sense that a unique (cardinal) number can be assigned to any system, telling us of how many particles (or quanta, as one usually calls these non-individuals) the system consists. For some cases, the same is possible in quantum �eld theory, by interpreting the eigenvalues of the number operator N as the number of quanta in a certain state, which we already did implicitly when introducing conditions (1a) and (1b) in the previous section. But this type of countability in QFT encounters a number of problems. The most obvious one is that, since N is an operator on a vector space F (called Fock space), the state vectors |Ψ � ∈ F will in general not be eigenvectors of N. This means that in general, physical states will not have a de�nite particle number. However, this problem is not very serious. It is one of the strengths of QFT that it is able to describe creations and annihilations of particles, and in this sense, inde�nite particle numbers in QFT are no more worrying than inde�nite positions or momenta in non-relativistic QM. The real problems for countability arise from the fact that there may not even be a well-de�ned number operator for a given system, either because there is no Fock space in which to describe it, or because its Fock space is not unique. I will �rst discuss the failure of uniqueness for free �elds, associated with the so-called Unruh-e�ect, and then turn to the failure of existence for interacting �elds, which is a consequence of Haag’s theorem.

Against Countability: Unruh E�ect and Interacting Fields

� 155

Non-uniqueness of the Fock Representation for Free Fields In order to describe the non-uniqueness problem, I need to introduce some elements of the Fock space formalism. Very roughly, Fock space is obtained from a given one-particle Hilbert space H� by building the direct sum of all the n-particle Hilbert spaces: ∞ � F= Hn , n=�

where H� ≡ {αΩ, α ∈ C} represents the vacuum state, and Hn for n ≥ � is the nfold (symmetric or antisymmetric) tensor product of H� with itself. One can then introduce the so-called creation and annihilation operators a* (f ), a(f ) ; and the number operator

f ∈ H�

. � * N= a (f j )a(f j ),

(9.3)

j

where {f j } is an orthonormal basis of H� . Informally, a* (f ) applied to an arbitrary state |Ψ � ∈ F adds a state-f -quantum to |Ψ �, while a(f ) removes such a quantum from |Ψ �. If there is no such quantum present in |Ψ �, then a(f )|Ψ � = �. In particular, a(f )Ω = � for all f ∈ H� . (9.4)

It immediately follows that Ω is an eigenstate of N with eigenvalue 0, justifying the term “vacuum”. Let us now focus on the concrete case in which F is used to describe a (noninteracting) system of spinless massive bosons (a more general discussion of the following is given by Clifton and Halvorson 2001, sec. 3.1). The orthonormal basis {f j } of H� used in the de�nition (9.3) is then given by the positive frequency solutions of the Klein-Gordon equation. The crucial point now is that the splitting of solutions into positive and negative frequencies depends on the time parameter used, and this in turn depends on the kind of reference frame that is considered. Therefore, di�erent reference frames give rise to a di�erent number operator N, so there does not seem to be a unique way of counting quanta. More concretely, where an inertial observer in Minkowski spacetime detects a vacuum, a uniformly accelerated observer may detect an arbitrarily large number of quanta (called Rindler quanta). This is (a very informal statement of) the Unruh e�ect. As it turns out, it is not so easy to give a less informal account of the Unruh e�ect, partly because di�erent phenomena are associated with that name (see Earman 2011 for a review). But this latter problem need not concern us here, since we

156 � Particle Physics without Particles? On Causal Realism in Quantum Field Theory are not interested in the Unruh e�ect per se, but only as an argument against the particle concept. (In fact, some of the authors I will discuss do not use the term “Unruh e�ect” at all.) A more serious di�culty arises when trying to make precise what one observer can say about a state de�ned in the Fock space pertaining to the other observer. To facilitate discussion, I will use an index M to mark the objects associated with the inertial (Minkowski) frame and an index R to mark the objects pertaining to the accelerated (Rindler) frame. The problem just mentioned can then be seen, for example, in Paul Teller’s (1995, 111) discussion of Rindler quanta: States with a de�nite number of Minkowski quanta are superpositions of states with di�erent numbers of Rindler quanta. In particular, [the Minkowski vacuum state |Ω M �] is a superposition of Rindler quanta states, including states for arbitrarily large numbers of Rindler quanta.

This presupposes that the Rindler number operator N R can be de�ned on the Minkowski Fock space FM . Now for quanta with a speci�ed wave function f , it is possible to de�ne Rindler creation and annihilation operators a*R (f ), a R (f ) (and �nite combinations thereof) on FM . But the total number operator N R , as de�ned by (9.3), is an in�nite sum of operators, which is not de�nable on FM , because if the number of degrees of freedom goes to in�nity, FM and FR host unitarily inequivalent (even disjoint) representations of the canonical commutation relations. To speak (as Teller does) of |Ω M � as a superposition of eigenstates of N R is therefore inadmissible (Clifton and Halvorson 2001, 446). To improve on Teller’s unsatisfactory formulation of the Unruh e�ect, Clifton and Halvorson (2001, sec. 4.2) construct representation-independent probabilities, which permit to formulate statements about the eigenvalues of N R for states in FM , even though, as we just saw, an expression like N R |Ω M � is not mathematically well-de�ned. Exploiting the fact that, in AQFT, states need not be described by vectors in a speci�c Hilbert space, but can be described abstractly as linear functionals on the Weyl algebra W, Clifton and Halvorson derive the following conclusion: P Ω M (N R ∈ [�, n]) = � for all n ∈ N,

where the expression on the left hand side denotes the probability to �nd an eigenvalue in the interval [�, n] for the Rindler number operator in the Minkowski vacuum state. The fact that this probability is zero implies that the number of Rindler quanta in the Minkowski vacuum is > n for any n ∈ N. Provided that Clifton’s and

Against Countability: Unruh E�ect and Interacting Fields

� 157

Halvorson’s construction is sound, this makes precise the above informal statement of the Unruh e�ect.³ Arageorgis et al. (2003) are not convinced that this tells us anything about the reality of Rindler quanta. They observe that the strategy pursued by Clifton and Halvorson is not limited to the Rindler representation, but can be extended to a large class of other representations, so if we accept that the Minkowski vacuum contains an in�nite number of Rindler quanta, we must also accept that it contains an in�nite number of all kinds of other quanta. As Arageorgis et al. (2003, 192) put it, we would then have to conclude “that the Minkowski vacuum is a very crowded place indeed”. They take this to suggest that Clifton’s and Halvorson’s result is “an artifact of the peculiarity of disjoint representations, not a guide to the particle contents of the Minkowski vacuum” (ibid.). While the alleged crowdedness of the Minkowski vacuum casts some doubt on Clifton’s and Halvorson’s formulation of the Unruh e�ect, it obviously does not salvage the particle concept. If anything, the availability of a multitude of di�erent particle number operators beside N M and N R makes things worse. A similar comment can be made in the context of Laura Ruetsche’s (2011, chap. 9; see also Arageorgis et al. 2002, 180) characterization of the Unruh effect in terms of incommensurable particle notions. According to Ruetsche, it is not the case that Jill (an accelerated observer) detects quanta in (unaccelerated) Jack’s vacuum state, because Jack’s and Jill’s particle notions are incommensurable: “There is no state to which Jack and Jill attribute di�erent particle contents because any state to which one’s particle notion applies is a state to which one the other’s doesn’t” (205). Again, this disquali�es the statements of the Unruh effect which we have considered so far, but the case against particles persists (or is even strengthened), because incommensurability shows that there are physically relevant states (namely the states in FR ) which cannot be characterized in terms of the particle notion associated with FM (Ruetsche 2011, sec. 9.9 and 10.1). It thus seems that, regardless of how exactly the Unruh e�ect is formulated, it poses a threat to the countability of particles. However, this threat depends on a premise I have not yet discussed, namely the requirement that the particle number assessment of an inertial (Minkowski) observer is not to be privileged over

3 This way of stating the problem which the Unruh e�ect poses for a quanta interpretation of QFT also undermines the resolution Teller (1995, 111) proposes for this problem. He claims that, since |Ω M � is not an eigenstate of N R , there are no actual Rindler quanta in the Minkowski vacuum, “only a propensity for detection of one or another number of Rindler quanta by an accelerating detector”. But Clifon’s and Halvorson’s result shows that there is in fact no propensity to detect anything less than a number of n + � Rindler quanta in the Minkowski vacuum (for any n ∈ N).

158 � Particle Physics without Particles? On Causal Realism in Quantum Field Theory the assessment of an accelerated (Rindler) observer. Without this requirement, we could just dismiss Rindler quanta as unphysical and insist that the eigenvalues of N M are the real particle numbers. The no-privilege requirement is usually justi�ed by the intuition that the presence or absence of particles is not the sort of thing that should depend on an observer’s state of motion (unlike, e.g., the presence or absence of inertial forces in Newtonian mechanics). But there are reasons to question the physical relevance of the Rindler representation, and this might imply that the Minkowski representation holds a privileged status after all. I will return to this point in section 9.3. Meanwhile, I take my discussion of the Unruh e�ect to license a somewhat mixed conclusion: On the one hand, we may agree with Arageorgis et al. (2003, 166), who “do not think that the demotion of the particle concept is supported in any straightforward and unproblematic way by the Unruh e�ect”. On the other hand, the problem that N is not unique must be taken seriously and does not seem to admit of an easy answer.

Non-existence of the Fock Representation for Interacting Fields If interacting states in a quantum �eld theory are to be interpreted as consisting of countable particles, the Hilbert space HI in which they are represented must be equipped with a total particle number operator N. In other words, the interacting system must have a Fock representation. We will now see that this may not be attainable. For concreteness, let us consider bosonic scalar �elds ϕ j , which, along with their conjugate momentum �elds π j satisfy (1) Equal Time Canonical Commutation Relations (ETCCR): [ϕ j (x, t), π j (x ′ , t)] = iδ(x − x ′ ); [ϕ j (x, t), ϕ j (x ′ , t)] = [π j (x, t), π j (x ′ , t)] = �. The index j will be used to distinguish interacting (I) from free (F) �elds. The most obvious way to obtain a Fock representation of the ETCCR for interacting �elds would be to express the interacting �elds in terms of free �elds, for which a Fock representation is known to exist.� This presupposes identifying (up

4 This is the underlying idea of the usual textbook approach to scattering theory, invoking the interaction picture. There, the Hamilton operator of the system is split into free and interacting parts, H = H F + H I , with H F generating the time evolution of operators and H I the time evolution of states. In particular, �eld operators evolve according to H F , which is to say that the �elds are

Against Countability: Unruh E�ect and Interacting Fields

� 159

to unitary equivalence) the space HI of interacting states with the (Fock) space FF of free states. A necessary condition for this to be possible is that free �elds (ϕ F , π F ) and interacting �elds (ϕ I , π I ) are related in the following way: (2) For some time t, there is a unitary transformation V(t) : FF → HI such that ϕ I (x, t) = V(t)ϕ F (x, t)V −� (t),

π I (x, t) = V(t)π F (x, t)V −� (t).

To demonstrate that this condition is problematic, I will now formulate a version of what is known as Haag’s theorem (a more general formulation and discussion of the theorem is given by Earman and Fraser 2006). For this purpose, I introduce three more (very plausible) assumptions. (3) Poincaré Covariance: Poincaré transformations (a, Λ) are represented by unitary operators U j (a, Λ) on HI and FF , such that U j (a, Λ)ϕ j (x) = ϕ j (Λx + a),

j ∈ {I, F }, a ∈ M, Λ ∈ O(�, �).

(4) Poincaré Invariant Vacua: There exist unique normalizable vacuum states |Ω j � such that: U j (a, Λ)|Ω j � = |Ω j �.

(5) No states of negative energy exist.

Haag’s Theorem: If the �elds ϕ I and ϕ F , de�ned on HI and FF , respectively, satisfy conditions (1)–(5), then ϕ I is a free �eld. As our original intention has been to describe an interacting �eld by means of ϕ I , we have now arrived at a contradiction. And since it is a legitimate requirement on any relativistic quantum theory to satisfy the basic assumptions (1) and (3)–(5), we can conclude that (2) has to go. Hence, the two representations of the ETCCR on HI and FF cannot be unitarily equivalent to each other, and the described strategy to obtain a Fock representation for an interacting �eld fails (Fraser 2008, sec. 3). The failure of this strategy does not, however, imply that there cannot be a Fock representation for interacting �elds (Earman and Fraser 2006, sec. 7). Maybe we just need a more sophisticated strategy. Doreen Fraser considers the following two proposals for such strategies:

free. This is what makes the widely used perturbative calculation of the scattering matrix possible. The intriguing question now is why the textbook approach works so well even though the idea underlying it is, as we will shortly see, untenable. See Earman and Fraser (2006, sec. 5) for further discussion and a possible answer to this question.

160 � Particle Physics without Particles? On Causal Realism in Quantum Field Theory 1.

2.

Instead of trying to use the Fock representation for the free �eld to represent an interacting �eld, one could try to use the method by which the Fock representation for the free �eld was constructed and apply it to an interacting �eld. Fraser identi�es two problems invalidating this approach. First, the �eld operators obtained in this procedure di�er for di�erent inertial reference frames. In other words, they are not Lorentz covariant (Fraser 2008, 850). The frame dependence arises from splitting the solutions of the relevant �eld equation into positive and negative frequency parts, precisely as in the case of the Unruh e�ect considered above. But there the di�erence was between an inertial reference frame and an accelerated frame, which could be seen as a reason to privilege the former point of view over the latter. Here, however, the di�erence is between two inertial frames, so such a reply is not available.� Second, due to the non-linearity of the interacting �eld equations, it is not clear how even to construct the one-particle Hilbert space H� , which, as discussed above, is the starting point for constructing a Fock space (851–852). Another way to arrive at a quanta-friendly Hilbert space representation of the ETCCR would be to de�ne the representation by the formal (Fock-like) properties it is required to have, namely the existence of creation and annihilation operators and of a normalizable vacuum state vector, characterized by (9.4). It can be shown that these conditions uniquely specify the sought-after representation (up to unitary equivalence). Furthermore, this representation does have a well-de�ned total number operator N, which seems to assure countability (Fraser 2008, 852–854). However, the eigenstates of N do not have the necessary properties to sustain a quanta interpretation. In particular, the vacuum state is (in the presence of non-trivial interactions) not Poincaré invariant, in violation of condition (4) above. As for one-particle states, their expectation values for the energy operator do not correspond to the values given by relativistic one-particle dynamics. Hence, it is not justi�ed to interpret the eigenvalues of N as numbers of quanta (854–855).

Having dismissed these two strategies, Fraser (2008, 856) turns to what she calls “a last-ditch attempt to save the quanta interpretation of QFT”. This is the idea, proposed by Jonathan Bain (2000, 394), that the existence of a number operator for asymptotic states in a scattering experiment is su�cient for a particle interpretation of interacting �elds. Intuitively, asymptotic states are the states of a

5 Of course, considerations as the ones in section 7.2 might serve as a motivation to privilege a certain inertial frame, but this would amount to abandoning the project of constructing a genuinely relativistic quantum theory, thus rendering moot much of the content of the present chapter.

Defending Localizability and Countability �

161

system “long before” and “long after” the interaction, so they can be regarded as free and be given a Fock representation. That no Fock representation is available for the system at intermediate (non-asymptotic) times is, according to Bain, irrelevant for a particle interpretation. Instead, he suggests that “particle” be understood as “a system that minimally possesses an asymptotic state”. But then, Fraser (2008, 857) is right to object that “particles” in this sense, whatever they are, are not countable in the presence of interactions, and it is doubtful whether they do (at �nite times) possess any other particlelike properties.

�.� Defending Localizability and Countability The previous two sections showed that the idea of localizable and countable entities is hard to reconcile with the principles of relativistic quantum theory. In other words, relativistic quantum theory seems unable to accommodate a particle ontology.� This by itself is not a problem for causal realism, because causal realism has no a priori commitment to particles. Granted, the causal realist embraces a realist stance towards particle physics, but he may well acknowledge that the scienti�c discipline identi�ed by that name is actually about something other than particles. However, the no-go results of sections 9.1 and 9.2 are in con�ict with at least two a posteriori commitments of causal realism. The �rst one appeared in section 5.3, where I argued for realism about neutrinos. One of the essential properties I attributed to neutrinos in this context was their ability to produce an inverse beta-decay in the detector. Since the detector has a �nite volume, this amounts to saying that neutrinos are localizable in �nite spatial regions, in contradiction to Malament’s theorem.� The second con�ict between causal realism and the theorems discussed here arises from the case study in section 6.3: In order to give some content to the atomic hypothesis, it was necessary to attribute certain properties

6 As mentioned in section 9.1, the received view takes this to imply that QFT should be interpreted in terms of a �eld ontology. But this conclusion is premature, not only because the quantum �eld is relevantly di�erent from classical �elds which might be interpreted realistically (Teller 1995, chap. 5; 2002), but also because the arguments reviewed in the previous section undermine a �eld ontology just as much as they undermine a particle ontology (Baker 2009). Instead of favoring a �eld ontology, these results rather seem to tell us that there is at present simply no convincing proposal for an ontology of QFT. 7 As the �nal paragraph of section 9.1 shows, no contradiction arises if localizability is understood merely as “giving rise to local measurement results”. But a realist commitment to localizability implies that the localization must not depend on anyone performing a measurement, so causal realism cannot rest content with such a minimal notion of localizability.

162 � Particle Physics without Particles? On Causal Realism in Quantum Field Theory to atoms and molecules, and the most important of these was countability. In sum, causal realism is de�nitely committed to localizable and countable entities, and the question is how this is compatible with the alleged absence of such entities in relativistic quantum theories.

Solving the Problem by Privileging Causal Warrant? A seemingly easy way out of this predicament presents itself in the observation that the no-go theorems in question are only theoretically warranted, while the above-mentioned case studies provide causal warrant for localizability and countability. Since it is a basic tenet of causal realism that causal warrant trumps theoretical warrant (cf. the quote from Suárez 2008 at the beginning of chapter 4), why not just dismiss the theoretical objections against localizability and countability as irrelevant in the light of manifest experimental evidence to the contrary? This seems to be the approach taken by Edward MacKinnon (2008), although he does not formulate it in the language of causal realism. MacKinnon questions the relevance of the no-go theorems on the grounds that the standard model of particle physics simply assumes localizable (or even localized) and countable particles, and that this assumption is justi�ed by the standard model’s empirical success, regardless of the theoretical problems associated with the particle concept.� Even as a proponent of causal realism, I see two problems with this response. First, one of the lessons of chapter 7 was that in the context of quantum mechanics, considerations of causal warrant alone will not get us very far in discerning how to explain the phenomena. I argued that these considerations are of value, but that they need to be combined with an evaluation of theoretical virtues. The mutual dependence of causal and theoretical considerations is probably even more important in the context of quantum �eld theory, and a premature dismissal of theoretical results puts the constructive interplay between theory and experiment in jeopardy. Second, if there really is theoretical warrant against localizability and countability, then it is simply false to claim causal warrant for these properties. At the beginning of chapter 4, I characterized theoretical warrant as what is generated by an inference to the best explanation. But if there is a good explanation of the relevant phenomena which excludes localizability and countability, then an alternative explanation in terms of localizable and countable entities can-

8 MacKinnon’s account thus emphasizes “the di�ering roles that ‘particle’ plays in experimental and theoretical contexts” (456). For a detailed investigation of this distinction, see Falkenburg (2007, chap. 6).

Defending Localizability and Countability

�

163

not claim any causal warrant, because it falls short of non-redundancy (section 4.1). Therefore, if the causal realist wants to maintain that localizability and countability are causally warranted, he actually has to deny that the no-go results are even theoretically warranted. In other words, he has to show that localizability and countability are non-redundant in the sense that no satisfactory explanation of the phenomena of particle physics can do without them. This requires a critical examination of the theoretical framework within which the no-go theorems are derived. Before turning to that task, I brie�y mention a di�erent way of disputing the relevance of the no-go theorems, proposed by Bain (2011). Bain’s criticism is not motivated by the empirical success of the standard model, but by the worry that the way in which the theorems mathematically represent the necessary conditions for localizability and countability (namely via the existence of a local number operator N ∆ and of a unique total number operator N) may be inappropriate for the relativistic domain. In other words, the fact that relativistic quantum theories do not admit local/unique total number operators may not have to be taken as an argument against particles, but may simply express the inadequacy of representing the particle concept by means of such operators. The question then is how else localizability and countability should be represented. In the absence of any concrete proposal on how to answer that question, I do not at present see much interest in pursuing this line of argument.

Underdetermination again: Di�erent Approaches to Quantum Field Theory Fraser (2009, 544) describes QFT as furnishing “a genuine case of the classic problem of underdetermination of theory by all possible evidence”. She discusses three variants of QFT, distinguished by the way in which they respond to the in�nite terms appearing in the quantum �eld theoretical treatment of interacting systems: the in�nitely renormalized, the cuto� and the formal variant, respectively. The �rst variant is mathematically ill-de�ned and rather ad hoc, and it has not played a signi�cant role in the philosophical debate. I will therefore disregard it in the following discussion. The second variant is the one favored by the majority of contemporary theoretical physicists, employing sophisticated renormalization methods to tame the in�nities and to arrive at highly successful empirical predictions. David Wallace (2006) defends the conceptual respectability of this approach, which he calls Lagrangian quantum �eld theory. I will instead (and in accordance with Wallace 2011) use the term conventional quantum �eld theory, or CQFT, where the C can be taken to stand either for conventional or for cuto�. The third variant approaches the problem of in�nities by looking for a rigorous axiomatic basis for

164 � Particle Physics without Particles? On Causal Realism in Quantum Field Theory quantum �eld theory, in order to avoid the appearance of any mathematically illde�ned objects in the derivation of empirical predictions. In the form of algebraic quantum �eld theory (AQFT), this approach has attracted most attention in the recent philosophical debate on QFT. All the arguments discussed in sections 9.1 and 9.2 are to some extent linked to this approach, even though (as noted at the end of section 9.1) not all of them explicitly depend on the AQFT formalism. The three variants of QFT are, according to Fraser (2009, 541) “empirically indistinguishable in the sense that the sets of scattering matrix elements generated by the three variants can be brought into arbitrarily close agreement”. Since the three variants di�er substantially on the theoretical level, we seem indeed to be confronted with a case of underdetermination. The connection to the above question whether localizability and countability are causally warranted is the following: CQFT is the theoretical basis for the standard model, and as such, it can be viewed as explaining the phenomena of particle physics in terms of localizable and countable entities.� But if Fraser is right about underdetermination in QFT, then localizability and countability are redundant, because there is a rival explanation (based on AQFT) which excludes them. However, there is an obvious reason to be suspicious of Fraser’s case for underdetermination: There is (as yet) no AQFT model for an interacting system in four spacetime dimensions, a point emphasized by Wallace (2011, 120): Being uncharitable, [AQFT] makes no (non-falsi�ed) empirical predictions whatsoever, because the world we live in manifestly has (a) interactions and (b) four (or more) spacetime dimensions, and we have no physically realistic interacting �eld theories in four dimensions. Even being charitable: the only empirical predictions of AQFT are general results (the spinstatistics theorem, the CPT theorem, etc.¹�) which are also derivable (by the usual standards used in theoretical physics) in CQFT (perhaps only as extremely good approximations, depending on whether the world is Poincaré-covariant at the fundamental level).

It is therefore not correct to say that CQFT and AQFT are empirically indistinguishable, because only the former makes concrete empirical predictions about

9 It should be noted, however, that the implementation of localizability in CQFT is not a trivial matter; see Wallace (2006, sec. 5) for details. Wallace remarks that his treatment “is rather similar in character” (68) to Halvorson’s and Clifton’s explanation of the appearance of local measurement results mentioned at the end of section 9.1, but the crucial di�erence is that his account does not invoke an axiomatic notion of measurement (cf. footnote 7 above). 10 Kuhlmann (2010, 1630) mentions two other successful applications of AQFT, but even he has to admit that they are “not as strong as a direct predictive success for realistic cases, which one would like to see”.

Defending Localizability and Countability

�

165

real physical systems.¹¹ Consequently, there is not really a redundancy of explanations in this case, and the claim that localizability and countability are causally warranted stands unrefuted. Fraser is, of course, aware of this objection. In her reply to Wallace, she admits that only a restricted underdetermination claim has been veri�ed. According to this claim, CQFT and AQFT are empirically indistinguishable yet distinct theories in all cases in which models of AQFT have been constructed (Fraser 2011, 129). By contrast, the unrestricted underdetermination claim, stating that CQFT and AQFT are empirically indistinguishable and distinct in all cases (including the one of prime interest, namely interacting QFT in four spacetime dimensions), “is not at present veri�able” (ibid.). Nevertheless, Fraser claims that her case for underdetermination is su�cient to undermine the validity of inference to the best explanation in the context of CQFT. (The following quote is directed against a speci�c claim of CQFT, namely that “�eld degrees of freedom are frozen out at su�ciently short lengthscales” (Wallace 2011, 120). But Fraser’s argument is general, and it therefore also a�ects the question of how well localizability and countability are warranted.) If there is to be a link between explanatory power and truth, the set of alternative explanations considered must be su�ciently broad for it to be plausible that an approximately true explanation is among the candidate explanations. In [Wallace’s] case, it seems that the class of candidate theories under consideration is being illegitimately restricted. Ideally, the class of candidate theories should include not just the theories that we possess right now, but all possible theories. Clearly, in general, it is impractical to hold inference to the best explanation arguments to this high standard. However, in the present case we possess additional information about the theories that we possess right now and their empirical consequences: there exists a proposed theory (AQFT) about which we do not yet know the empirical consequences which (I would contend) would provide a superior account of the empirical data if it does turn out to have the same empirical consequences as the other theory (CQFT). In these circumstances, it seems wise to refrain from applying inference to the best explanation arguments. (Fraser 2011, 133)

11 As alluded to at the end of section 9.1, this problem not only pertains to AQFT in the narrow sense, but also to results derived in the same spirit, such as Malament’s theorem and its generalizations: All these theorems describe the (unrealistic) case of non-interacting systems. Halvorson’s and Clifton’s (2002, 20) reply is that “this hardly indicates a limitation on the generality of our conclusion, since Haag’s theorem . . . entails that interacting models of [relativistic QFT] have no number operators—not even a global number operator”. But this reply only works under the assumption that Haag’s theorem (a result of AQFT proper) applies to real systems, which is questionable as long as there is no realistic model satisfying the axioms of AQFT.

166 � Particle Physics without Particles? On Causal Realism in Quantum Field Theory This is reminiscent of Stanford’s argument from unconceived alternatives, discussed in chapter 6. (Although Fraser herself does not explicitly draw this connection, Smeenk and Myrvold (2011, 78) do.) Stanford and Fraser give di�erent reasons why we should worry about unconceived (or not yet entirely conceived) alternatives to present theories, but they agree that these worries invalidate at least some cases of inference to the best explanation. My argument against Stanford was that we should expect an ontological continuity between present and future theories in the sense that causally warranted claims will be retained in the future. Now in the present case, what we already know about AQFT seems to exclude such a continuity. In fact, Fraser (2009, 560– 561) discusses two reasons why we should not expect an approximate agreement on matters of ontology between the present theory (CQFT) and a future realistic model of AQFT. One of these reasons concerns countability, or, as Fraser puts it, the existence of quanta: “A theory according to which quanta exist is not approximately equivalent to a theory according to which quanta do not exist” (560). To this, Wallace replies that if a future interacting AQFT is to be capable of accounting for the empirical evidence available in particle physics, it must admit quanta in at least some approximate sense. And this restores approximate equivalence between CQFT and AQFT, because a theory according to which quanta exist “is approximately equivalent to a theory according to which quanta do approximately exist” (Wallace 2011, 123). One may worry about Wallace’s notion of approximate existence, and I will address some of these worries in the next section. For the moment, I just note that this notion is probably no more problematic than the notion of approximate truth which appears in most statements of scienti�c realism, at least of the non-fundamentalist kind I advocated in chapter 1. Another important remark in this context is that even if we were to adopt fundamentalism, it is unclear how seriously we should take the lessons of AQFT, because the absence of gravity in the AQFT framework makes it an unlikely candidate for the fundamental theory. Here is how Wallace (2011, 120–121) puts the point: Whatever our sub-Planckian physics looks like (string theory? twistor theory? loop quantum gravity? non-commutative geometry? causal set theory? something as-yet-undreamed-of?) there are pretty powerful reasons not to expect it to look like quantum �eld theory on a classical background spacetime. As such, what QFT (of any variety) says about the nature of the world on lengthscales below ∼ ��−�� m . . . doesn’t actually tell us anything about reality.

This brings me to Fraser’s second reason for denying ontological continuity between CQFT and AQFT: Due to the �nite cuto� introduced in CQFT’s treatment of divergent integrals, this theory has only a �nite number of degrees of freedom, and therefore does not admit unitarily inequivalent representations, in con-

Defending Localizability and Countability �

167

trast to AQFT (Fraser 2009, 560). But this is an inaccurate characterization of CQFT, as Wallace (2011, 123) points out. It is true that CQFT introduces a cuto� in the short-distance (or equivalently, high-energy/momentum) domain, but longdistance divergences are usually not treated in this way. Here, as in AQFT, algebraic methods are applied, such that QFTs de�ned on spatially in�nite manifolds do have in�nitely many degrees of freedom, and hence unitarily inequivalent representations. The disagreement between Fraser and Wallace then comes down to the question whether the inequivalent representations associated with the shortdistance regime have any ontological signi�cance. And the answer to that in turn depends on the ontological project one is engaged in. This is probably the most fundamental di�erence between Fraser’s and Wallace’s approaches: By “interpretation” I mean the activity of giving an answer to the following hypothetical question: “If QFT were true, what would reality be like?” In contrast, the interpretive question that Wallace focuses on is “Given that QFT is approximately true, what is reality (approximately) like?” (Fraser 2009, 558)

I �nd Fraser’s “hypothetical question” rather uninteresting, because I agree with Wallace (2011, 122) that “we have excellent reasons to think that . . . QFT is not true”. These reasons are, of course, precisely the Planck-scale related arguments mentioned above. Fraser does not explicitly reject these arguments, but downplays their force. Commenting on the analogy between condensed matter physics and CQFT, suggested by the fact that the same renormalization group methods are applied in both disciplines, she claims that the belief in an actual breakdown of the �eld concept at small lengthscales is justi�ed only in the former case, because there we have independent evidence for a discrete (atomic) structure becoming relevant at small distances. In the latter case, she claims that “we do not possess evidence of this sort that QFT breaks down at short distance scales” (2011, 134). By “evidence of this sort”, she obviously means empirical evidence (she mentions Perrin’s experiments in support of the atomic hypothesis), implying that the evidence for a breakdown of QFT below the Planck scale is merely theoretical. But this amounts to a depreciation of theoretical warrant not unlike the one exempli�ed by MacKinnon, which, if it were granted, would save us the trouble of entering into all the AQFT-fuelled no-go considerations in the �rst place. I conclude that Fraser’s arguments against ontological continuity between AQFT and CQFT are inconclusive and that therefore her case for underdetermination is not strong enough to undermine the causal realist’s commitment to localizability and countability.

168 � Particle Physics without Particles? On Causal Realism in Quantum Field Theory Possible Responses to the Unruh E�ect The attempts by Halvorson, Clifton and Ruetsche to give a precise formulation of the Unruh e�ect, discussed in section 9.2, are carried out in the framework of AQFT. To that extent, the above critique of AQFT also diminishes the force of the argument against countability based on the Unruh e�ect. But it does not completely remove it, because the fact that di�erent reference frames give rise to inequivalent Fock space representations (and thus to di�erent ways of counting quanta) does not itself depend on the AQFT formalism. Further work is therefore needed to respond to the challenge posed by the Unruh e�ect. As noted in section 9.2, a basic assumption of this challenge is that the Minkowski representation is not to be privileged over a Rindler representation. From the perspective of causal realism, this assumption is questionable. Setting aside astroparticle physics, the spacetime regions in which we perform particle physics experiments are, to a very good approximation, Minkowskian, so we have overwhelming causal warrant for the countability of Minkowski quanta. And we do not have the same kind of warrant for Rindler quanta. In fact, it is not even clear how a Rindler detector (i.e., a detector su�ciently accelerated to detect Rindler quanta in a Minkowski vacuum) is to be modeled theoretically (Arageorgis et al. 2003, sec. 11; Earman 2011, sec. 7). In other words, we do not have a clear notion of what it would mean to actually count Rindler quanta, and this implies, in the terminology of section 4.2, that the countability of Rindler quanta is not a material property. Consequently, the Rindler quanta hypothesis violates the criterion of material inference and it therefore lacks causal warrant. Note that in this case (unlike the case of MacKinnon’s proposal discussed above), the theoretical warrant for Rindler quanta does not threaten the causal warrant for Minkowski quanta, because the two kinds of quanta do not pertain to di�erent explanations of the same phenomena, but to the explanations of di�erent phenomena (measurements of a Minkowski detector vs. measurements of a Rindler detector). The threat to countability does, in this case, not arise from a redundancy of explanations, but from the fact that countability becomes observer-dependent. This threat is defused by the insight that the particle number assessments of the two observers are not on a par, but that one of them is privileged by being causally warranted, while the other one is not. There is a second possible response to the Unruh e�ect, which does not depend on the notion of causal warrant. I take it to be ultimately unsuccessful, but I mention it here because the way in which it fails holds an important lesson for the �nal section of this chapter. The core of this response is that there are not only empirical, but also theoretical reasons to privilege the Minkowski representation over a Rindler representation. More precisely, Rindler states lack some features which

Concluding Remarks on Realism, Fundamentalism, and QFT � 169

could legitimately be regarded as a prerequisite for physical relevance (Arageorgis et al. 2003, sec. 6; Ruetsche 2011, sec. 10.3). One key feature in this respect is the ful�llment of the Hadamard condition, which ensures that the stress-energy observable has a well-de�ned expectation value. The problem with imposing such restrictions on physical reasonableness is that not all successful applications of QFT respect them. For example, non-Hadamard states play an important role in the prediction of Hawking radiation (Ruetsche 2011, 238–239). Going beyond the context of the Unruh e�ect, this lesson generalizes to more basic features than the Hadamard condition and to less exotic applications than Hawking radiation. Ruetsche (2011, sec. 10.4) discusses the example of coherent states, which are indispensable in many applications of quantum optics, in particular in laser theory. Yet they are in a certain sense unphysical, because they con�ict with Poincaré symmetry. More precisely, coherent states are eigenstates of a “phase operator”, but such an operator is not de�nable in the Poincaré-invariant Fock representation of the canonical commutation relations. Assessing physical relevance on theoretical grounds is therefore a delicate matter. As we will see below, this has an important consequence for scienti�c realism in the context of QFT. These remarks should make it clear that I do not claim that privileging one particular representation is always a viable (or even desirable) strategy. I accept that some of QFT’s explanatory tasks can only be performed if unitarily inequivalent representations of the canonical commutation relations are taken into account. What I claimed here is that in the case of the Unruh e�ect, the causal warrant associated with the particle concept of one of the representations gives us good reason to privilege this representation, but there may not be such a reason in other cases. However, this would only undermine my defense of countability if it could be shown that in these other cases, the availability of inequivalent representations threatens the particle concept at least as seriously as in the case of the Unruh e�ect. As we have seen in section 9.2, even the Unruh e�ect does not straightforwardly lend itself to a precisely formulated argument against countability, and I do not know of a similarly developed argument against countability stemming from an instance of inequivalent representations for which my privileging strategy would fail.

�.� Concluding Remarks on Realism, Fundamentalism, and QFT The previous section sought to establish that localizability and countability are real (causally warranted) properties, although they may not be fundamental. As

170 � Particle Physics without Particles? On Causal Realism in Quantum Field Theory we have seen, Wallace expresses the same commitment by saying that quanta do approximately exist. In an earlier paper (2001), he spoke of “particle” as an emergent concept. No matter in what language the idea is couched, it faces a crucial objection: For this to be a viable response, the cogency of the distinction between fundamental and less fundamental entities must be defended and a case must be made for admitting additional, non-fundamental entities into our ontology. (Fraser 2008, 858)

I have tried to make such a case by defending a non-eliminativist version of scienti�c realism. I recognize, however, that I may not have provided a su�cient reply to all the aspects of Fraser’s objection. This has to do with the fact that the main focus of my work has been on the epistemological project of defending scienti�c realism against the skeptical attacks of present-day antirealism. The metaphysical project of defending a non-eliminativist realism against an eliminativist one has only been a side issue. Let me therefore restate my case for non-fundamental entities, before mentioning some loose metaphysical ends which will have to be tied up in future work on causal realism. The main point is familiar from section 1.2, but it can be spelled out more concretely in the light of the foregoing discussion on QFT. Scienti�c realism should be interested in QFT because it is by far the most successful description of the subatomic realm. But as we have seen, the variant of the theory which yields this success (CQFT) is explicitly non-fundamental, and the variant which claims to be fundamental (AQFT) does not really make good on that claim (because it disregards gravity), nor is it really successful (because it does not have a realistic model). Therefore, none of the entities appearing in QFT can be considered as fundamental. This makes it highly misleading for Fraser to speak of “additional, non-fundamental entities”: The ontology of our best (present) science simply does not contain any fundamental entities, to which the non-eliminativist could be accused of adding something (presumably something super�uous). Until we have a truly fundamental theory, the contrast between eliminativism and noneliminativism is not, as Fraser suggests, the contrast between a parsimonious and an in�ationary ontology, but between an empty and a non-empty one. The eliminativist might reply that an empty ontology (combined with the hope of some day �nding a theory which �lls it) is preferable to the in�ationary ontology of the non-eliminativist, because the latter trivializes the notion of existence. We encountered this argument in section 2.2 in the form of Gelfert’s worry that a Hacking-style entity realist would have to admit all kinds of dubious entities (holes, excitons and other quasi-particles) into his ontology. There I argued that these entities may not be any more dubious than what Gelfert (2003, 257) calls “ba-

Concluding Remarks on Realism, Fundamentalism, and QFT

� 171

sic substantive entities” (electrons etc.), and I referred the reader to the present chapter, which indicates that electrons and other so-called elementary particles can at most claim the same kind of non-fundamental reality as holes and excitons. So once again, restricting ontological commitment to an elite class of supposedly basic entities is not a viable option; whoever wants a non-empty ontology may have to acknowledge excitons along with electrons. But where does this proliferation of entities stop? If we accept that particles, quasi-particles and all kinds of other non-fundamental entities exist (or, to adopt Wallace’s terminology, approximately exist), are we not forced to also accept the existence (or approximate existence) of such really dubious entities as caloric and phlogiston? Based on the causal realism developed in the previous chapters, I do not think so. With regard to caloric, I remarked that it is false to invest it with even derivative reality (note 11 in section 4.2) and I suggested that this could have been known already in the heyday of caloric theory (section 6.2). An analogous argument about phlogiston (or dephlogisticated air) is given by Chakravartty (2007, 55–56). If these arguments generalize to other cases, then the ontology of causal realism need not include any straightforwardly problematic entities. Ruetsche (2011, chap. 15) gives yet another argument for approaching QFT in a non-fundamentalist (and consequently non-eliminativist) spirit, namely that the way in which QFT achieves its various explanatory and predictive successes does not support the idea of a single, uni�ed, fundamental ontology. This was illustrated by the example of coherent states mentioned at the end of section 9.3: If we ask whether there is such a thing as a phase observable, the answer is “yes” for coherent states, but “no” for states in a Poincaré-invariant representation, yet both kinds of states have undeniable physical signi�cance. Drawing on this and other examples, Ruetsche anatomizes what she calls a “scandal” for scienti�c realism. Brie�y put, she concludes that the virtues which realists usually take to speak for the truth of a theory do not support any single interpretation of QFT. Instead, di�erent applications support di�erent (sometimes con�icting) interpretations. This resonates with causal realism’s reservations about theoretical virtues (section 4.1) and with Cartwright’s idea that a theory’s success in accurately describing the phenomena may con�ict with its unifying power (section 4.3). More importantly, since Ruetsche works in an AQFT context, her argument shows that a non-fundamentalist approach to QFT need not depend on an anti-AQFT attitude as the one advocated by Wallace or MacKinnon. The two lines of argument can thus be viewed as complementary: While Wallace’s case against eliminativism focuses on the eventual failure of QFT (below the Planck scale), Ruetsche shows that even the success of QFT does not support an eliminativist reading of it. By accepting the reality of non-fundamental entities, the non-eliminativist incurs the burden of explaining how the di�erent domains of reality are related to

172 � Particle Physics without Particles? On Causal Realism in Quantum Field Theory each other. I take it that this is what Fraser means when she requires that “the cogency of the distinction between fundamental and less fundamental entities must be defended”. Causal realism does not have much to say about this issue, because it primarily characterizes its objects by describing their causal properties, not by reducing them to more fundamental entities. Nevertheless, the case for causal realism could be strengthened by showing how the particle concept relates to the fundamental ontology. Di�erent directions of investigation are conceivable. For example, one could build upon ideas by Wallace (2001; 2006, sec. 5), trying to ground localizability and countability in underlying quantum �eld theoretical properties. However, given the shortcomings of a �eld ontology for QFT (see footnote 6 above), this by itself will not be enough. More will need to be said about the quantum measurement problem and about quantum non-locality, and doing so may result in moving away from the QFT framework. We already witnessed an example for this move in section 9.1, where we saw that a GRW-type theory will not satisfy all the conditions which Halvorson and Clifton consider essential for any future relativistic quantum theory. A similar remark will apply to future Bohmian versions of QFT. In such a context, the no-go theorems discussed in this chapter will lose their relevance, so countability and localizability may turn out to be fundamental after all. (In the Bohmian scenario, the fundamental entities will presumably even be continuously localized, not just localizable.) Of course, in their current state, GRW-type and Bohm-type approaches are nowhere near reproducing the predictions of CQFT, but this is a feature they share with AQFT. So none of these directions should be excluded from consideration. For the time being, causal realism therefore has to leave open the question whether particles are fundamental or not, and has to content itself with the statement that particles are real. And in response to the question what particles are, the causal realist cannot point to a neatly axiomatized theory from which one might hope to extract a precise metaphysical description of the nature of particles. What he can do is characterize them as localizable and countable entities, and for further theoretical details refer to the somewhat messy apparatus of CQFT. As in the case of atoms discussed at the end of chapter 6, our knowledge about particles is substantive, though not exhaustive. Nevertheless, there is a lack of speci�city here, which parallels the lack of speci�city we already encountered in the results of chapters 7 and 8. For example, the exclusion of DC models and Leggett-type models as explanations of quantum non-locality (section 7.3) left the question of how the correlations are to be explained unanswered. And the argument for the reality of entanglement relations (section 8.3) did not say much about what these relations are. Again, the reason for this lack of speci�city is causal realism’s focus on refuting skeptical challenges fueled by underdetermination (chapter 7) or an alleged indeterminacy of the past

Concluding Remarks on Realism, Fundamentalism, and QFT

� 173

(chapter 8). But in its future development, causal realism may well turn from these epistemological questions to more metaphysical ones. Let me conclude by formulating this plan in terms of the examples discussed in the previous chapters: Having made a reasonable case that there are such things as neutrinos, atoms and entangled states, causal realism can turn to the task of spelling out more precisely what they are.

Bibliography Achinstein, P. (2001). The Book of Evidence. New York: Oxford University Press. Achinstein, P. (2002). Is there a valid experimental argument for scienti�c realism? The Journal of Philosophy 99, 470–495. Alai, M. (2014). Novel predictions and the no miracle argument. Erkenntnis 79, 297–326. Allori, V., S. Goldstein, R. Tumulka, and N. Zanghì (2008). On the common structure of Bohmian mechanics and the Ghirardi-Rimini-Weber theory. British Journal for the Philosophy of Science 59, 353–389. Alvarez, L. W. (1937). Nuclear K electron capture. Physical Review 52, 134–135. Andrade, E. N. d. C. (1964). Rutherford and the Nature of the Atom. Garden City NY: Anchor Books. Arageorgis, A., J. Earman, and L. Ruetsche (2002). Weyling the time away: The non-unitary implementability of quantum �eld dynamics on curved spacetime. Studies in History and Philosophy of Modern Physics 33, 151–184. Arageorgis, A., J. Earman, and L. Ruetsche (2003). Fulling non-uniqueness and the Unruh effect. Philosophy of Science 70, 164–202. Aspect, A., J. Dalibard, and G. Roger (1982). Experimental test of Bell’s inequalities using timevarying analyzers. Physical Review Letters 49, 1804–1807. Bain, J. (2000). Against particle/�eld duality: Asymptotic particle states and interpolating �elds in interacting QFT (or: Who’s afraid of Haag’s theorem?). Erkenntnis 53, 375–406. Bain, J. (2011). Quantum �eld theories in classical spacetimes and particles. Studies in History and Philosophy of Modern Physics 42, 98–106. Baker, D. J. (2009). Against �eld interpretations of quantum �eld theory. British Journal for the Philosophy of Science 60, 585–609. Bancal, J.-D., S. Pironio, A. Acín, Y.-C. Liang, V. Scarani, and N. Gisin (2012). Quantum nonlocality based on �nite-speed causal influences leads to superluminal signalling. Nature Physics 8, 867–870. Barrett, J. A. (2002). On the nature of measurement records in relativistic quantum �eld theory. See Kuhlmann et al. (2002), pp. 165–179. Beck, G. and K. Sitte (1933). Zur Theorie des β-Zerfalls. Zeitschrift für Physik 86, 105–119. Bell, J. S. (1964). On the Einstein-Podolsky-Rosen paradox. Physics 1, 195–200. Reprinted in Bell (2004), pp. 14–21. Bell, J. S. (1966). On the problem of hidden variables in quantum mechanics. Reviews of Modern Physics 38, 447–452. Reprinted in Bell (2004), pp. 1–13. Bell, J. S. (1990a). Against ‘measurement’. In A. I. Miller (Ed.), Sixty-two years of uncertainty: historical, philosophical, and physical inquiries into the foundations of quantum mechanics. New York: Plenum Press. Reprinted in Bell (2004), pp. 213–231. Bell, J. S. (1990b). La nouvelle cuisine. In A. Sarlemijn and P. Kroes (Eds.), Between Science and Technology, pp. 97–115. Amsterdam: Elsevier Science Publishers. Reprinted in Bell (2004), pp. 232–248. Citations refer to the reprint. Bell, J. S. (2004). Speakable and Unspeakable in Quantum Mechanics (2nd ed.). Cambridge: Cambridge University Press. Berkovitz, J. (2008). Action at a distance in quantum mechanics. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Winter 2008 ed.). http://plato.stanford.edu/archives/ win2008/entries/qm-action-distance/.

176 � Bibliography Bird, A. (2007). Nature’s Metaphysics: Laws and Properties. Oxford: Oxford University Press. Bogen, J. and J. Woodward (1988). Saving the phenomena. The Philosophical Review 97, 303– 352. Bohm, D. (1952). A suggested interpretation of the quantum theory in terms of ‘hidden’ variables. I and II. Physical Review 85, 166–193. Bohm, D. and B. Hiley (1993). The undivided universe: An ontological interpretation of quantum theory. London: Routledge. Borchers, H. J. (1967). A remark on a theorem of B. Misra. Communications in mathematical Physics 4, 315–323. Boyd, R. (2010). Scienti�c realism. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Summer 2010 ed.). http://plato.stanford.edu/archives/sum2010/entries/scienti�crealism/. Branciard, C., N. Brunner, N. Gisin, C. Kurtsiefer, A. Lamas-Linares, A. Ling, and V. Scarani (2008). Testing quantum correlations versus single-particle properties within Legget’s model and beyond. Nature Physics 4, 681–685. Branciard, C., A. Ling, N. Gisin, C. Kurtsiefer, A. Lamas-Linares, and V. Scarani (2007). Experimental falsi�cation of Leggett’s nonlocal variable model. Physical Review Letters 99, 210407. Busch, P. (1999). Unsharp localization and causality in relativistic quantum theory. Journal of Physics A: Mathematical and General 32, 6535–6546. Cartwright, N. (1983). How The Laws of Physics Lie. Oxford: Clarendon Press. Cartwright, N. (1999). The Dappled World: A Study of the Boundaries of Science. Cambridge: Cambridge University Press. Chakravartty, A. (2007). A Metaphysics for Scienti�c Realism: Knowing the Unobservable. New York: Cambridge University Press. Chakravartty, A. (2008). What you don’t know can’t hurt you: Realism and the unconceived. Philosophical Studies 137, 149–158. Chakravartty, A. (2011). Scienti�c realism. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Summer 2011 ed.). http://plato.stanford.edu/archives/sum2011/entries/ scienti�c-realism/. Chalmers, A. (2011). Drawing philosophical lessons from Perrin’s experiments on Brownian motion: A response to van Fraassen. British Journal for the Philosophy of Science 62, 711–732. Chandrasekhar, S. (1943). Stochastic problems in physics and astronomy. Reviews of Modern Physics 15, 1–89. Chang, H. (2003). Preservative realism and its discontents: Revisiting caloric. Philosophy of Science 70, 902–912. Chang, H. and N. Cartwright (1993). Causality and realism in the EPR experiment. Erkenntnis 38, 169–190. Clarke, S. (2001). Defensible territory for entity realism. British Journal for the Philosophy of Science 52, 701–722. Clarke, S. and T. D. Lyons (Eds.) (2002). Recent Themes in the Philosophy of Science: Scienti�c Realism and Commonsense, Volume 17 of Australasian Studies in History and Philosophy of Science. Dordrecht: Kluwer Academic Publishers. Clauser, J. F., M. A. Horne, A. Shimony, and R. A. Holt (1969). Proposed experiment to test local hidden-variable theories. Physical Review Letters 23, 880–884.

Bibliography � 177 Clifton, R. (2002). The subtleties of entanglement and its role in quantum information theory. Philosophy of Science 69, S150–S167. Clifton, R. and H. Halvorson (2001). Are Rindler quanta real? Inequivalent particle concepts in quantum �eld theory. British Journal for the Philosophy of Science 52, 417–470. Cocciaro, B., S. Faetti, and L. Fronzoni (2011). A lower bound for the velocity of quantum communications in the preferred frame. Physics Letters A 375, 379–384. Cocciaro, B., S. Faetti, and L. Fronzoni (2013). In search of superluminal quantum communications: recent experiments and possible improvements. Journal of Physics: Conference Series 442, 012005. Cohen, R. S., R. Hilpinen, and Q. Renzong (Eds.) (1996). Realism and Anti-Realism in the Philosophy of Science. Dordrecht: Kluwer Academic Publishers. Colbeck, R. and R. Renner (2008). Hidden variable models for quantum theory cannot have any local part. Physical Review Letters 101, 050403. Colbeck, R. and R. Renner (2012). The completeness of quantum theory for predicting measurement outcomes. Preprint, submitted August 20, 2012. arXiv:1208.4123v1. Cordero, A. (2001). Realism and underdetermination: Some clues from the practices-up. Philosophy of Science 68, S301–S312. Crane, H. R. (1948). The energy and momentum relations in the beta-decay, and the search for the neutrino. Reviews of Modern Physics 20, 278–295. Daumer, M., D. Dürr, S. Goldstein, and N. Zanghì (1997). Naive realism about operators. Erkenntnis 45, 379–397. Devitt, M. (1991). Aberrations of the realism debate. Philosophical Studies 61, 43–63. Devitt, M. (1997). Realism and Truth (2nd ed.). Princeton: Princeton University Press. Devitt, M. (2011). Are unconceived alternatives a problem for scienti�c realism? Journal for General Philosophy of Science 42, 285–293. Dorato, M. (1988). The world of the worms and the quest for reality. Dialectica 42, 171–182. Dorato, M. and M. Esfeld (2010). GRW as an ontology of dispositions. Studies in History and Philosophy of Modern Physics 41, 41–49. Duhem, P. (1954). The Aim and Structure of Physical Theory. Princeton: Princeton University Press. Dürr, D., S. Goldstein, and N. Zanghì (1997). Bohmian mechanics and the meaning of the wave function. In R. Cohen, M. Horne, and J. Stachel (Eds.), Experimental Metaphysics: Quantum Mechanical Studies for Abner Shimony, Volume 193 of Boston Studies in the Philosophy of Science, pp. 25–38. Dordrecht: Kluwer Academic Publishers. Earman, J. (2011). The Unruh e�ect for philosophers. Studies in History and Philosophy of Modern Physics 42, 81–97. Earman, J. and D. Fraser (2006). Haag’s theorem and its implications for the foundations of quantum �eld theory. Erkenntnis 64, 305–344. Eberhard, P. H. (1989). A realistic model for quantum theory with a locality property. In W. Schommers (Ed.), Quantum Theory and Pictures of Reality, pp. 169–215. Berlin: Springer. Egg, M. (2009). What is Kitcher’s real realist really a realist about? Conceptus 94, 107–119. Egg, M. and M. Esfeld (2014). Non-local common cause explanations for EPR. European Journal for Philosophy of Science 4, 181–196. Eigler, D. M. and E. K. Schweizer (1990). Positioning single atoms with a scanning tunnelling microscope. Nature 344, 524–526.

178 � Bibliography Einstein, A. (1905). Über die von der molekularkinetischen Theorie der Wärme geforderte Bewegung von in ruhenden Flüssigkeiten suspendierten Teilchen. Annalen der Physik 17, 549–560. Einstein, A., B. Podolsky, and N. Rosen (1935). Can quantum-mechanical description of physical reality be considered complete? Physical Review 47, 777–780. Ellis, B. (1987). The ontology of scienti�c realism. In P. Pettit, R. Sylvan, and J. Norman (Eds.), Metaphysics and Morality: Essays in Honour of J.J.C. Smart, pp. 50–70. Oxford: Basil Blackwell. Ellis, C. D. and W. A. Wooster (1927). The avarage energy of disintegration of radium E. Proceedings of the Royal Society A 117, 109–123. Elsamahi, M. (1994). Could theoretical entities save realism? In PSA 1994: Proceedings of the Biennial Meetings of the Philosophy of Science Association, Volume 1, pp. 173–180. University of Chicago Press. En�eld, P. (2008). Review of Exceeding Our Grasp. British Journal for the Philosophy of Science 59, 881–895. Englert, B.-G., M. O. Scully, and H. Walther (1999). Quantum erasure in double-slit interferometers with which-way detectors. American Journal of Physics 67, 325–329. Esfeld, M. (2004). Quantum entanglement and a metaphysics of relations. Studies in History and Philosophy of Modern Physics 35, 601–617. Esfeld, M. (2010). Physics and causation. Foundations of Physics 40, 1597–1610. Esfeld, M. (2013). Ontic structural realism and the interpretation of quantum mechanics. European Journal for Philosophy of Science 3, 19–32. Esfeld, M. and N. Gisin (2014). The GRW flash theory: A relativistic quantum ontology of matter in space-time? Philosophy of Science 81, 248–264. Everett, H. (1957). “Relative state” formulation of quantum mechanics. Reviews of Modern Physics 29, 454–462. Fahrbach, L. (2011). How the growth of science ends theory change. Synthese 180, 139–155. Falkenburg, B. (2007). Particle Metaphysics: A Critical Account of Subatomic Reality. Berlin: Springer. Fermi, E. (1934). Versuch einer Theorie der β-Strahlen. I. Zeitschrift für Physik 88, 161–177. Feyerabend, P. (1975). Against Method: Outline of an Anarchistic Theory of Knowledge. London: New Left Books. Fine, A. (1986). Unnatural attitudes: Realist and instrumentalist attachments to science. Mind 95, 149–179. Fine, A. (1991). Piecemeal realism. Philosophical Studies 61, 79–96. Fine, A. (1996). The Shaky Game: Einstein, Realism and the Quantum Theory (2nd ed.). Chicago: University of Chicago Press. Franklin, A. (1996). There are no antirealists in the laboratory. See Cohen et al. (1996), pp. 131–148. Franklin, A. (2004). Are There Really Neutrinos? An Evidential History. Boulder (Colorado): Westview Press. Fraser, D. (2008). The fate of ‘particles’ in quantum �eld theories with interactions. Studies in History and Philosophy of Modern Physics 39, 841–859. Fraser, D. (2009). Quantum �eld theory: Underdetermination, inconsistency, and idealization. Philosophy of Science 76, 536–567. Fraser, D. (2011). How to take particle physics seriously: A further defence of axiomatic quantum �eld theory. Studies in History and Philosophy of Modern Physics 42, 126–135.

Bibliography � 179 French, S. (1989). Individuality, supervenience and Bell’s theorem. Philosophical Studies 55, 1–22. French, S. (2011a). Identity and individuality in quantum theory. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Summer 2011 ed.). http://plato.stanford.edu/archives/ sum2011/entries/qt-idind/. French, S. (2011b). Metaphysical underdetermination: Why worry? Synthese 180, 205–221. Frigg, R. and I. Votsis (2011). Everything you always wanted to know about structural realism but were afraid to ask. European Journal for Philosophy of Science 1, 227–276. Gelfert, A. (2003). Manipulative success and the unreal. International Studies in the Philosophy of Science 17, 245–263. Ghirardi, G., A. Rimini, and T. Weber (1980). A general argument against superluminal transmission through the quantum mechanical measurement process. Lettere Al Nuovo Cimento 27, 293–298. Ghirardi, G., A. Rimini, and T. Weber (1986). Uni�ed dynamics for microscopic and macroscopic systems. Physical Review D 34, 470–491. Ghirardi, G. and R. Romano (2013). About possible extensions of quantum theory. Foundations of Physics 43, 881–894. Giere, R. N. (1988). Explaining Science: A Cognitive Approach. Chicago: University of Chicago Press. Gisin, N. (2012). Non-realism: Deep thought or a soft option? Foundations of Physics 42, 80– 85. Gisin, N., V. Scarani, W. Tittel, and H. Zbinden (2000). Optical tests of quantum nonlocality: From EPR-Bell tests towards experiments with moving observers. Annalen der Physik 9, 831–841. Goldstein, S. (2001). Bohmian mechanics. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Winter 2001 ed.). http://plato.stanford.edu/archives/win2001/entries/qmbohm/. Gröblacher, S., T. Paterek, R. Kaltenbaek, Č. Brukner, M. Żukowski, M. Aspelmeyer, and A. Zeilinger (2007). An experimental test of non-local realism. Nature 446, 871–875. Gross, A. G. (1990). Reinventing certainty: The signi�cance of Ian Hacking’s realism. In PSA 1990: Proceedings of the Biennial Meetings of the Philosophy of Science Association, Volume 1, pp. 421–431. Hacking, I. (1983). Representing and Intervening: Introductory Topics in the Philosophy of Natural Science. Cambridge: Cambridge University Press. Hacking, I. (1989). Extragalactic reality: The case of gravitational lensing. Philosophy of Science 56, 555–581. Halvorson, H. and R. Clifton (2002). No place for particles in relativistic quantum theories? Philosophy of Science 69, 1–28. Harker, D. (2013). How to split a theory: Defending selective realism and convergence without proximity. British Journal for the Philosophy of Science 64, 79–106. Hartmann, S., C. Hoefer, and L. Bovens (Eds.) (2008). Nancy Cartwright’s Philosophy of Science. New York: Routledge. Hausman, D. M. and J. Woodward (1999). Independence, invariance and the causal markov condition. British Journal for the Philosophy of Science 50, 521–583. Healey, R. (2012). Quantum theory: A pragmatist approach. British Journal for the Philosophy of Science 63, 729–771.

180 � Bibliography Hegerfeldt, G. C. (1998). Instantaneous spreading and Einstein causality in quantum theory. Annalen der Physik 7, 716–725. Hiley, B. J. and R. E. Callaghan (2006a). Delayed-choice experiments and the Bohm approach. Physica Scripta 74, 336–348. Hiley, B. J. and R. E. Callaghan (2006b). What is erased in the quantum erasure? Foundations of Physics 36, 1869–1883. Hitchcock, C. (1992). Causal explanation and scienti�c realism. Erkenntnis 37, 151–178. Hoefer, C. (2003). For fundamentalism. Philosophy of Science 70, 1401–1412. Horodecki, R., P. Horodecki, M. Horodecki, and K. Horodecki (2009). Quantum entanglement. Reviews of Modern Physics 81, 865–942. Jennewein, T., G. Weihs, J.-W. Pan, and A. Zeilinger (2002). Experimental nonlocality proof of quantum teleportation and entanglement swapping. Physical Review Letters 88, 017903. Jensen, C. (2000). Controversy and Consensus: Nuclear Beta Decay 1911-1934. Basel: Birkhäuser. Kant, I. ([1787] 1855). Critique of Pure Reason. London: Henry G. Bohn. Translated by J. M. D. Meiklejohn. Kitcher, P. (1989). Explanatory uni�cation and the causal structure of the world. In P. Kitcher and W. C. Salmon (Eds.), Scienti�c Explanation, Volume XIII of Minnesota Studies in the Philosophy of Science, pp. 410–505. Minneapolis: University of Minnesota Press. Kitcher, P. (1993). The Advancement of Science. New York: Oxford University Press. Kitcher, P. (2001). Real realism: The Galilean strategy. The Philosophical Review 110, 151–197. Kuhlmann, M. (2009). Quantum �eld theory. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Spring 2009 ed.). http://plato.stanford.edu/archives/spr2009/entries/ quantum-�eld-theory/. Kuhlmann, M. (2010). Why conceptual rigour matters to philosophy: On the ontological signi�cance of algebraic quantum �eld theory. Foundations of Physics 40, 1625–1637. Kuhlmann, M., H. Lyre, and A. Wayne (Eds.) (2002). Ontological Aspects of Quantum Field Theory. London: World Scienti�c Publishing. Kuhn, T. S. (1962). The Structure of Scienti�c Revolutions. Chicago: University of Chicago Press. Kukla, A. (1993). Laudan, Leplin, empirical equivalence and underdetermination. Analysis 53, 1–7. Ladyman, J. (2014). Structural realism. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Spring 2014 ed.). http://plato.stanford.edu/archives/spr2014/entries/structuralrealism/. Ladyman, J., I. Douven, L. Horsten, and B. C. van Fraassen (1997). A defence of van Fraassen’s critique of abductive inference: Reply to Psillos. The Philosophical Quarterly 47, 305–321. Laudan, L. (1981). A confutation of convergent realism. Philosophy of Science 48, 19–49. Laudan, L. and J. Leplin (1991). Empirical equivalence and underdetermination. The Journal of Philosophy 88, 449–472. Laudisa, F. (1997). Contextualism and nonlocality in the algebra of EPR observables. Philosophy of Science 64, 478–496. Laudisa, F. (2008). Non-local realistic theories and the scope of the Bell theorem. Foundations of Physics 38, 1110–1132. Laudisa, F. (2014a). Against the ‘no-go’ philosophy of quantum mechanics. European Journal for Philosophy of Science 4, 1–17. Laudisa, F. (2014b). On Leggett theories: A reply. Foundations of Physics 44, 296–304.

Bibliography � 181 Leggett, A. J. (2003). Nonlocal hidden-variable theories and quantum mechanics: An incompatibility theorem. Foundations of Physics 33, 1469–1493. Lyons, T. D. (2002). Scienti�c realism and the pessimistic meta-modus tollens. See Clarke and Lyons (2002), pp. 63–90. Lyons, T. D. (2005). Toward a purely axiological scienti�c realism. Erkenntnis 63, 167–204. Lyre, H. (2010). Why quantum theory is possibly wrong. Foundations of Physics 40, 1429–1438. Ma, X.-s., S. Zotter, J. Kofler, R. Ursin, T. Jennewein, Č. Brukner, and A. Zeilinger (2012). Experimental delayed-choice entanglement swapping. Nature Physics 8, 480–485. MacKinnon, E. (2008). The standard model as a philosophical challenge. Philosophy of Science 75, 447–457. Maddy, P. (2007). Second Philosophy. New York: Oxford University Press. Magnus, P. D. (2006). What’s new about the new induction? Synthese 148, 295–301. Magnus, P. D. (2010). Inductions, red herrings, and the best explanation for the mixed record of science. British Journal for the Philosophy of Science 61, 803–819. Malament, D. B. (1996). In defense of dogma: Why there cannot be a relativistic quantum mechanics of (localizable) particles. In R. Clifton (Ed.), Perspectives on Quantum Reality. Non-relativistic, relativistic, and �eld-theoretic, pp. 1–10. Dordrecht: Kluwer Academic Publishers. Massimi, M. (2004). Non-defensible middle ground for experimental realism: Why we are justi�ed to believe in colored quarks. Philosophy of Science 71, 36–60. Maudlin, T. (1996). Space-time in the quantum world. In J. T. Cushing, A. Fine, and S. Goldstein (Eds.), Bohmian Mechanics and Quantum Theory: An Appraisal, Volume 184 of Boston Studies in the Philosophy of Science, pp. 285–307. Dordrecht: Kluwer Academic Publishers. Maudlin, T. (1998). Part and whole in quantum mechanics. In E. Castellani (Ed.), Interpreting Bodies: Classical and Quantum Objects in Modern Physics, pp. 46–60. Princeton: Princeton University Press. Maudlin, T. (2011). Quantum Non-Locality and Relativity (3rd ed.). Chichester: Wiley-Blackwell. Maxwell, J. C. ([1873] 1955). A Treatise on Electricity and Magnetism, Volume II. London: Oxford University Press. Mayo, D. G. (1986). Cartwright, causality, and coincidence. In PSA 1986: Proceedings of the Biennial Meeting of the Philosophy of Science Association, Volume 1, pp. 42–58. Mayo, D. G. (1996). Error and the Growth of Experimental Knowledge. Chicago: University of Chicago Press. McArthur, D. (2003). Is Pickering’s “pragmatic realism” viable? Dialectica 57, 71–88. McMullin, E. (1991). Comment: Selective anti-realism. Philosophical Studies 61, 97–108. Meitner, L. and W. Orthmann (1930). Über eine absolute Bestimmung der Energie der primären β-Strahlen von Radium E. Zeitschrift für Physik 60, 143–155. Mohrho�, U. (1996). Restoration of interference and the fallacy of delayed choice: Concerning an experiment proposed by Englert, Scully, and Walter. American Journal of Physics 64, 1468–1475. Mohrho�, U. (1999). Objectivity, retrocausation, and the experiment of Englert, Scully, and Walther. American Journal of Physics 67, 330–335. Moon, A. (2012). Warrant does entail truth. Synthese 184, 287–297. Morrison, M. (1990). Theory, intervention and realism. Synthese 82, 1–22. Morrison, M. (1994). Causes and contexts: The foundations of laser theory. British Journal for the Philosophy of Science 45, 127–151.

182 � Bibliography Musgrave, A. (1989). Noa’s ark—Fine for realism. The Philosophical Quarterly 39, 383–398. Musgrave, A. (1996). Realism, truth and objectivity. See Cohen et al. (1996), pp. 19–44. Näger, P. M. (2013). Causal graphs for EPR experiments. Preprint, submitted August 5, 2013. http://philsci-archive.pitt.edu/9915/. Nola, R. (2002). Realism through manipulation, and by hypothesis. See Clarke and Lyons (2002), pp. 1–23. Norsen, T. (2007). Against ‘realism’. Foundations of Physics 37, 311–340. Norton, J. D. (2007). Causation as folk science. In H. Price and R. Corry (Eds.), Causation, Physics, and the Constitution of Reality: Russell’s Republic Revisited, pp. 11–44. Oxford: Clarendon Press. Nye, M. J. (1972). Molecular Reality: A Perspective on the Scienti�c Work of Jean Perrin. London: Macdonald. Pais, A. (1986). Inward Bound: Of Matter and Forces in the Physical World. New York: Oxford University Press. Pan, J.-W., D. Bouwmeester, H. Weinfurter, and A. Zeilinger (1998). Experimental entanglement swapping: Entangling photons that never interacted. Physical Review Letters 80, 3891– 3894. Pauli, W. (1979). Scienti�c correspondence with Bohr, Einstein, Heisenberg a. o., Volume 1: 1919-1929. New York: Springer. Pauli, W. (1984). Zur älteren und neueren Geschichte des Neutrinos. In Physik und Erkenntnistheorie, pp. 156–180. Braunschweig: Vieweg. Pauli, W. (1985). Scienti�c correspondence with Bohr, Einstein, Heisenberg a. o., Volume 2: 1930-1939. Berlin: Springer. Peres, A. (2000). Delayed choice for entanglement swapping. Journal of Modern Optics 47, 139–143. Perrin, J. (1910). Brownian Movement and Molecular Reality. London: Taylor and Francis. Reprinted in M. J. Nye (1986): The Question of the Atom, Los Angeles: Tomash Publishers, pp. 507–601. Citations refer to the reprint. Pierson, R. and R. Reiner (2008). Explanatory warrant for scienti�c realism. Synthese 161, 271–282. Pontecorvo, B. ([1946] 2000). Inverse β process. See Winter (2000), pp. 23–28. First published as Report PD-205 (National Research Council of Canada, Division of Atomic Energy). Citations refer to the reprint. Post, H. (1975). The problem of atomism. British Journal for the Philosophy of Science 26, 19–26. Psillos, S. (1999). Scienti�c Realism: How Science Tracks Truth. London: Routledge. Psillos, S. (2005). Scienti�c realism and metaphysics. Ratio 18, 385–404. Psillos, S. (2008). Cartwright’s realist toil: From entities to capacities. See Hartmann et al. (2008), pp. 167–194. Psillos, S. (2009). Knowing the Structure of Nature: Essays on Realism and Explanation. Basingstoke: Palgrave Macmillan. Psillos, S. (2011). Moving molecules above the scienti�c horizon: On Perrin’s case for realism. Journal for General Philosophy of Science 42, 339–363. Putnam, H. (1975). Mathematics, Matter and Method. Cambridge: Cambridge University Press. Quine, W. V. O. (1951). Two dogmas of empiricism. The Philosophical Review 60, 20–43. Recami, E. (1978). How to recover causality in special relativity for tachyons. Foundations of Physics 8, 329–340.

Bibliography � 183 Redhead, M. (1995). More ado about nothing. Foundations of Physics 25, 123–137. Reiner, R. and R. Pierson (1995). Hacking’s experimental realism: An untenable middle ground. Philosophy of Science 62, 60–69. Reines, F. (1996). The neutrino: From poltergeist to particle. Reviews of Modern Physics 68, 317–327. Reines, F., C. L. Cowan, F. B. Harrison, A. D. McGuire, and H. W. Kruse (1960). Detection of the free antineutrino. Physical Review 117, 159–174. Resnik, D. B. (1994). Hacking’s experimental realism. Canadian Journal of Philosophy 24, 395–412. Roush, S. (2005). Tracking Truth: Knowledge, Evidence, and Science. New York: Oxford University Press. Roush, S. (2010). Optimism about the pessimistic induction. In P. D. Magnus and J. Busch (Eds.), New Waves in Philosophy of Science, pp. 29–58. Basingstoke: Palgrave Macmillan. Ruetsche, L. (2011). Interpreting Quantum Theories: The Art of the Possible. Oxford: Oxford University Press. Ruhmkor�, S. (2011). Di�culties for the problem of unconceived alternatives. Philosophy of Science 78, 875–886. Saatsi, J. (2009). Review of Exceeding Our Grasp. Metascience 18, 355–363. Salart, D., A. Baas, C. Branciard, N. Gisin, and H. Zbinden (2008). Testing the speed of ‘spooky action at a distance’. Nature 454, 861–864. Salmon, W. C. (1984). Scienti�c Explanation and the Causal Structure of the World. Princeton: Princeton University Press. Salmon, W. C. (1998). Causality and Explanation. New York: Oxford University Press. San Pedro, I. (2012). Causation, measurement relevance and no-conspiracy in EPR. European Journal for Philosophy of Science 2, 137–156. Sankey, H. (2008). Scienti�c Realism and the Rationality of Science. Aldershot: Ashgate Publishing. Scarani, V., W. Tittel, H. Zbinden, and N. Gisin (2000). The speed of quantum information and the preferred frame: Analysis of experimental data. Physics Letters A 276, 1–7. Scully, M. O. and K. Drühl (1982). Quantum eraser: A proposed photon correlation experiment concerning observation and “delayed choice” in quantum mechanics. Physical Review A 25, 2208–2213. Scully, M. O., B.-G. Englert, and H. Walther (1991). Quantum optical tests of complementarity. Nature 351, 111–116. Seevinck, M. (2006). The quantum world is not built up from correlations. Foundations of Physics 36, 1573–1586. Sellars, W. (1963). Science, Perception and Reality. London: Routledge and Kegan. Shapere, D. (1993). Astronomy and antirealism. Philosophy of Science 60, 134–150. Sklar, L. (2003). Dappled theories in a uniform world. Philosophy of Science 70, 424–441. Smeenk, C. and W. Myrvold (2011). Introduction: Philosophy of quantum �eld theory. Studies in History and Philosophy of Modern Physics 42, 77–80. Stanford, P. K. (2001). Refusing the devil’s bargain: What kind of underdetermination should we take seriously? Philosophy of Science 68, S1–S12. Stanford, P. K. (2006). Exceeding Our Grasp: Science, History, and the Problem of Unconceived Alternatives. New York: Oxford University Press. Stanford, P. K. (2009a). Author’s response. Metascience 18, 379–390.

184 � Bibliography Stanford, P. K. (2009b). Scienti�c realism, the atomic theory, and the catch-all hypothesis: Can we test fundamental theories against all serious alternatives? British Journal for the Philosophy of Science 60, 253–269. Stefanov, A., H. Zbinden, N. Gisin, and A. Suarez (2002). Quantum correlations with spacelike separated beam splitters in motion: Experimental test of multisimultaneity. Physical Review Letters 88, 120404. Suárez, M. (2007). Causal inference in quantum mechanics: A reassessment. In F. Russo and J. Williamson (Eds.), Causality and Probability in the Sciences, pp. 65–106. London: London College. Suárez, M. (2008). Experimental realism reconsidered: How inference to the most likely cause might be sound. See Hartmann et al. (2008), pp. 137–163. Suárez, M. (2013). Interventions and causality in quantum mechanics. Erkenntnis 78, 199–213. Sutton, C. (1992). Spaceship Neutrino. Cambridge: Cambridge University Press. Teller, P. (1986). Relational holism and quantum mechanics. British Journal for the Philosophy of Science 37, 71–81. Teller, P. (1995). An Interpretive Introduction to Quantum Field Theory. Princeton: Princeton University Press. Teller, P. (2002). So what is the quantum �eld? See Kuhlmann et al. (2002), pp. 145–164. Teller, P. (2004). How we dapple the world. Philosophy of Science 71, 425–447. Thomasson, A. (2012). Categories. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Summer 2012 ed.). http://plato.stanford.edu/archives/sum2012/entries/categories/. Timpson, C. G. (2010). Information, immaterialism, instrumentalism: Old and new in quantum information. In A. Bokulich and G. Jaeger (Eds.), Philosophy of Quantum Information and Entanglement, pp. 208–227. Cambridge: Cambridge University Press. Timpson, C. G. and H. R. Brown (2010). Building with quantum correlations. Quantum Information Processing 9, 307–320. Tumulka, R. (2006). A relativistic version of the Ghirardi-Rimini-Weber model. Journal of Statistical Physics 125, 821–840. van Fraassen, B. C. (1980). The Scienti�c Image. Oxford: Clarendon Press. van Fraassen, B. C. (1991). Quantum Mechanics: An Empiricist View. Oxford: Clarendon Press. van Fraassen, B. C. (2009). The perils of Perrin, in the hands of philosophers. Philosophical Studies 143, 5–24. von Neumann, J. (1932). Mathematische Grundlagen der Quantenmechanik. Berlin: Springer. Votsis, I. (2007). Review of Exceeding Our Grasp. International Studies in the Philosophy of Science 21, 103–117. Votsis, I. (2011). The prospective stance in realism. Philosophy of Science 78, 1223–1234. Walborn, S. P., M. O. Terra Cunha, S. Pádua, and C. H. Monken (2002). Double-slit quantum eraser. Physical Review A 65, 033818. Wallace, D. (2001). Emergence of particles from bosonic quantum �eld theory. Preprint, submitted December 23, 2001. arXiv:quant-ph/0112149v1. Wallace, D. (2006). In defence of naiveté: The conceptual status of Lagrangian quantum �eld theory. Synthese 151, 33–80. Wallace, D. (2008). Philosophy of quantum mechanics. In D. Rickles (Ed.), The Ashgate Companion to Contemporary Philosophy of Physics, pp. 16–98. Aldershot: Ashgate Publishing. Wallace, D. (2011). Taking particle physics seriously: A critique of the algebraic approach to quantum �eld theory. Studies in History and Philosophy of Modern Physics 42, 116–125.

Bibliography � 185 Wheeler, J. A. (1978). The ‘past’ and the ‘delayed-choice’ double-slit experiment. In A. R. Marlow (Ed.), Mathematical Foundations of Quantum Theory, pp. 9–48. New York: Academic Press. Whitaker, A. (2012). The New Quantum Age: From Bell’s Theorem to Quantum Computation and Teleportation. Oxford: Oxford University Press. Wilson, J. (2007). Newtonian forces. British Journal for the Philosophy of Science 58, 173–205. Winter, K. (Ed.) (2000). Neutrino Physics (2nd ed.). Cambridge: Cambridge University Press. Woodward, J. (2003). Making Things Happen: A Theory of Causal Explanation. Oxford: Oxford University Press. Worrall, J. (1989). Structural realism: The best of both worlds? Dialectica 43, 99–124. Wray, K. B. (2013). The pessimistic induction and the exponential growth of science reassessed. Synthese 190, 4321–4330. Yurke, B. and D. Stoler (1992). Bell’s-inequality experiments using independent-particle sources. Physical Review A 46, 2229–2234. Zbinden, H., J. Brendel, N. Gisin, and W. Tittel (2001). Experimental test of nonlocal quantum correlation in relativistic con�gurations. Physical Review A 63, 022111. Żukowski, M., A. Zeilinger, M. A. Horne, and A. K. Ekert (1993). “Event-ready-detectors” Bell experiment via entanglement swapping. Physical Review Letters 71, 4287–4290.

Index Achinstein, P., 94 Alpha particle, 3, 6 Alvarez, L., 72 Angular momentum, conservation of, 68 Antirealism, 4–5, 10–14, 19, 24, 33–35, 46, 60, 71–72, 79–80, 82, 137, 139, 148 Arageorgis et al., 157–158 Argument from coincidence, 93–96, 98 Aspect, A., 105, 129, 134 Atomic hypothesis, 85, 92–101, 161, 167 Atomic nucleus, 3, 67–69, 72, 83 Avogadro’s number, 94–101 Bain, J., 160, 163 Barrett, J. A., 151 Before-before experiment, 124–125, 129 Bell (or EPR) correlations, 105–118, 122–128, 147 Bell states, 145–146 Bell, J. S., 105, 121, 128, 133 Bell-type experiment, 65, 105–114, 122–130, 134, 149 Bell-type inequalities, 105–109, 126, 130, 134–136, 150 Beta decay, 67–69, 73–77 – Fermi’s theory of, 69–72 – inverse, 72–73, 76–77 Bogen, J., 32 Bohm, D., 105, 133 Bohmian mechanics, 65, 117–119, 135–136, 139–140, 145, 172 Bohr, N., 46, 68–70, 76 Borchers’ lemma, 153 Boyd, R., 4 Brown, H. R., 138 Brownian motion, 92–100 Callaghan, R. E., 140, 144 Caloric, 58, 88, 171 Cartwright, N., 15, 19, 23, 30–32, 49, 51–55, 60–63, 93–94, 98, 116, 171 Causal knowledge, 85–89 Causal realism, see Realism

Causal warrant, see Warrant Causation – counterfactual account, 55, 107–110 – interventionist (or manipulability) account, 51, 57, 76, 110–113 – relata of, 56–57 – superluminal, 106–111, 114–116, 122–129 Chakravartty, A., 8, 19, 56–58, 82–91, 171 Chalmers, A., 93, 98–100 Chang, H., 83, 88, 116 CHSH inequality, see Bell-type inequalities Clarke, S., 50, 77 Clauser, J., 126, 134 Clifton, R., 138, 149–153, 156–157, 164–165, 168, 172 Colbeck, R., 132, 136 Collapse of the wave function, see State reduction Common cause (CC), 93, 107–119 Common sense realism, see Realism Constructive empiricism, 6, 11, 33, 39, 42, 71, 101, 120–121 Cordero, A., 118–120 Core causal description, 86–87 Countability, 101, 154–169, 172 Cowan, C., 72–78 Delayed-choice experiment, 138–148 Devitt, M., 8–9, 23, 29, 84–85 Direct cause (DC), 110–119, 122–129 Double-slit experiment, 61, 64–65, 139–145 Duhem, P., 46, 62 Eigler, D., 101–102 Einstein, A., 97, 105 Electron, 21, 26–30, 56–57, 61, 64–65, 67, 139, 171 Electron capture, 72 Eliminative inference, 79–80, 102 Eliminativism, see Fundamentalism Ellis, B., 23 Empirical adequacy, 11, 42, 50, 59–61, 65–66, 68, 71, 90, 97, 99

188 � Index Energy/momentum conservation, 68–70, 75–77 En�eld, P., 82 Englert et al., 143 Entanglement, 137–138, 144, 146–148 Entanglement swapping, 137–138, 145–148 Entity realism, see Realism EPR correlations, see Bell correlations Ether, 69, 83, 86–87, 91 Everett, H., 117 Everettian quantum mechanics, 117, 144–145, 147 Explanation – causal, 30–32, 49–66, 92–93, 100, 105, 127 – deductive-nomological, 31, 54 – theoretical, 30–32, 49–56, 60–63 – uni�cationist account, 54, 61 Explanationism, 19–20 Extrapolation from common sense to scienti�c realism, 6–8, 11–13, 41 Fahrbach, L., 85 Fermi, E., 67, 69–72 Feyerabend, P., 22 Fine, A., 22, 33–46, 96 Fock space representation, 154–161, 168 Franklin, A., 39 Fraser, D., 159–161, 163–172 Friedman, M., 54 Fundamentalism, 15, 17, 166–172 – causal, 58–59 – moderate vs. eliminative, 16 Galilei, G.; Galilean strategy, 6 Geiger, H., 3 Geiger-Marsden experiment, 3 Gelfert, A., 27–28, 170 Giere, R., 23 Gisin, N., 132 Gröblacher et al., 131–133 Gravitational force, 58, 61 Gross, A., 24, 26–27 GRW theory, 118–119, 139–140, 144, 151–152, 172 Haag’s theorem, 159, 165 Hacking, I., 19–31, 44, 102

Halvorson, H., 149–153, 156–157, 164–165, 168, 172 Harker, D., 82–83 Hausman, D. M., 112–113 Healey, R., 138, 146–148 Higgs boson, 39–40 Hiley, B. J., 140, 144 Hitchcock, C., 64–66 Hoefer, C., 15–16 Holism, see Non-separability Inference to the best explanation (IBE), 29–32, 42, 49–50, 54–56, 162, 165 Inference to the most likely cause (IMLC) vs. inference to the best theoretical explanation (IBTE), 49–51, 54–60, 62–65, 93 Instrumentalism, 10–11, 14, 33–45, 80, 83, 117 Kitcher, P., 6–7, 12, 19, 54, 88 Klein, O., 68 Kochen-Specker theorem, 133 Kuhlmann, M., 153, 164 Kuhn, T., 22 Laudan, L., 11–12, 60, 70, 79 Laudisa, F., 129–136 Laws – theoretical vs. phenomenological, 31–32, 108 – truth of, 31, 55, 60 Leggett, A. J., 129–135 Leggett-type inequalities, 131, 133, 136 Leplin, J., 60, 70 Localism, 36–37, 40 Localizability, 149–154, 161–167, 172 Lorentz invariance, 115 LR view, 131–133 Lyons, T., 9 MacKinnon, E., 162, 167, 168, 171 Maddy, P., 94 Magnus, P. D., 81–82 Malament’s theorem, 149–153, 161, 165 – generalized, 153

Index Manipulability, 21, 23–30, 50, 102, 110–113, 137 Marsden, E., 3 Massimi, M., 56 Material inference, 50, 54–58, 68, 90–91, 97, 99, 168 Maudlin, T., 106–110, 127 Maxwell, J. C., 86–87, 91 Mayo, D., 94–95 McArthur, D., 35 McMullin, E., 38 Measurement problem, 106–109, 117–119, 172 Millikan, R. A., 26 Minimal interpretation of mathematical structure, 83–84, 89–90 Miracle argument, 7 Mohrho�, U., 143–144 Morrison, M., 25, 51, 53 Musgrave, A., 16, 29, 34, 35 Natural epistemological attitude, 7 Natural ontological attitude (NOA), 22, 33–46 Neutrino, 67–78, 161 Neutron, 67, 73 New induction (NI), 79–85, 92, 99 No-signaling, 109, 111, 125–128, 150 Nobel prize, 71 Non-contextuality, 133 Non-locality, 65, 105, 130, 132, 149–150, 172 Non-redundancy, 50–53, 63, 68, 70, 75–76, 90–91, 97, 99, 105, 114, 163 Non-separability, 111, 116, 119 Norton, J., 58–59 Novel prediction, 72, 82, 164 Observability, 6, 20, 31, 40–43, 45, 59, 120 Ontological pluralism, 15 Osiander, A., 46 Particle physics, 9, 15–16, 39–40, 55, 67, 72, 149, 153, 161–172 Pauli, W., 67–69 Peres, A., 146 Perrin, J., 85, 92–101, 167 Pessimistic (meta-)induction, 12, 19–21, 26, 79, 82

� 189

Phlogiston, 15–16, 171 Photon, 124–125, 129, 135, 141–145 Pierson, R., 62 Pontecorvo, B., 75 Post, H., 72 Principle of fairness (POF), 40–43, 45 Problem of unconceived alternatives (PUA), see Unconceived alternatives Property – auxiliary, 87–97 – causal, 26–29, 56–57, 69, 172 – detection, 57, 84, 87–97 – material (or detectable) vs. formal, 57–58, 62, 90, 168 Psillos, S., 8–9, 12–14, 17–19, 35, 36, 42, 53, 71, 82–84, 86–88, 94, 120 Putnam, H., 7 Quanta, see Countability Quantum chromodynamics, 110 Quantum electrodynamics (QED), 56–57 Quantum eraser, 140–145 Quantum �eld theory (QFT), 149–151, 154–172 – algebraic (AQFT), 153–154, 156, 164–172 – conventional (CQFT), 163–172 Quantum information theory, 137–138 Quantum measurement problem, see Measurement problem Quasi-particle, 27–28, 170 Realism – as an assumption of Bell’s theorem, see LR view – causal, 49, 53, 58–59, 62–65, 67, 72, 90–101, 108–114, 120–121, 137, 161–163, 168, 171–173 – common sense, 4–8, 13–18, 46, 59, 90, 117 – dimensions of, 8–9, 21–23 – entity (or experimental), 19–32, 44–46, 49, 53, 56, 62–63, 94, 170 – purely axiological, 9, 15, 22 – real, 7 – scienti�c, 3–23, 28, 33–44, 52–53, 58–60, 71–72, 80–85, 90, 117, 120–121, 128, 137, 166, 171 – selective, 19, 33, 37, 44, 82 – structural, 19–20, 58, 120

190 � Index Reductive empiricism, 10–11 Reeh-Schlieder theorem, 153 Reference – causal theory of, 35, 44 – causal-descriptivist account, 86 Reiner, R., 62 Reines, F., 71–78 Relativity – general, 58 – special, 80, 114, 124, 127, 144, 150 Renner, R., 132, 136 Retrocausation, 143, 147 Rindler quanta, see Unruh e�ect Roush, S., 84–85, 93–97, 100–101 Ruetsche, L., 157, 168–171 Ruhmkor�, S., 85 Rutherford, E., 3–4, 83 Saatsi, J., 82 Salmon, W. C., 93–94, 99 Sankey, H., 8 Schweizer, E., 101–102 Scienti�c realism, see Realism Scully, M. O., 141 Seevinck, M., 138 Sellars, W., 20, 60 Shapere, D., 24–25 Skepticism, 4–5, 42, 52, 77, 80–81 – partial, 10–12 Sklar, L., 16 Social constructivism, 5 Speed of causal influence, 122–124, 129 Stanford, P. K., 13–14, 17–18, 52, 71, 79–102, 166 State reduction, 117–119, 144–145, 151

Suárez, M., 22, 44, 49–54, 63–66, 110, 112–113 Svedberg, T., 96 Teller, P., 15, 16, 156–157 Theoretical virtues, 50, 52–53, 61, 71, 120–121, 129, 162, 171 Thomson, J. J., 26, 83 Timpson, C. G., 138 Truth – approximate, 14, 81, 166 – theories of, 10, 34–35 Unconceived alternatives, 13–14, 17, 71, 79–85, 91–93, 97–102, 166 Underdetermination of theory by evidence, 12, 52, 105, 114–122, 163–167 – recurrent, transient (RTU), 79–84, 92 Unobservable entity, see Observability Unruh e�ect, 155–158, 168–169 van Fraassen, B. C., 11–12, 33, 39, 41–42, 46, 59–65, 93, 115, 121 von Neumann, J., 121 Votsis, I., 83–84 Wallace, D., 118, 163–172 Warrant, 22, 32, 62–63 – causal vs. theoretical, 49–67, 70–72, 77–78, 90–92, 97–100, 114, 120–121, 129, 162–169 Wheeler, J. A., 139 Which-way experiment, 139–145 Woodward, J., 32, 51, 57, 112–113 Worrall, J., 19, 33