g : 5 Reading Natural Philosophy Essays in the History and Philosophy of Science and Mathematics EDITED BY David B. Malament OPEN COURT.Cover illustration: Newton by William Blake, ‘To order books from Open Gout, call toll-free 1-800-815-2280. ‘Open Coure Publishing Company is a division of Carus Publishing Company. © 2002 by Carus Publishing Company First printing 2002 All rights reserved. No part of this publication may be reproduced, stored in -0300. Printed and bound in the United States of America, Library of Congress Cataloging-in-Publication Data ‘Reading natural philosophy : essays in the history and philosophy of science and ‘mathematics / edited by David B. Malament pcm. Includes bibliographical references and index. ISBN 0-8126-9506-2 (alk. paper) — ISBN 0-8126-9507-0 (pbk. : alk. paper) ‘ongresses. 2. Science—Philosophy—Congresses, 1y Congresses. 4. Stein, Howard, 1929— Congresses. I. Malament, David B. Q124.6 R43 2002 509-—de21 2002066281 To Howard Stein on the Occasion of His 70 Birthday[6] Maxwell and “the Method of Physical Analogy”: Model-based Reasoning, Generic Abstraction, and Conceptual Change NANCY J. NERSESSIAN 1. Introduction ‘As my teacher at Case Western Reserve University, Howard Stein gave me 4 piece of advice—one of many for which T will always be grateful—that I recall went something like this: if you want to understand the nature of sci- fence, read the cence. To a young stadent of science, this nature of Maxwell’s “method of physical analogy” covery of the electromagnetic field equations. ‘When I frst read Maxwell T found it surprising how many commenta- tors om his work failed to take seriously what seemed to me to be the get crative role of the analogy developed in the 1861-62 paper (Maxwe 129130 Nancy J. Nersssian worst as dishonest post-hoc fabrications (Duhem 1914, 98). In this last case, Duhem claimed that Maxwell had cooked up the analogy after the fact and even falsified an equation (see also Duhem 1902) while “the results he obtained were known to him by other means” (1914, 98), ‘The “known by other means” claim is one I frequently encounter from philosophers in response to presentations of my interpretation of Maxwell, ‘When pressed as to what other means, the response is usually that the ‘equations were derived “by induction from the experiments,” with the nature of the inductive process left mysterious. Certainly experimental results played a key role in Maxwell's analy the process of deriving the equations was not what philosophers usually call “induction,” Another frequent response is the “symmetry argument,” presented in figure 1. For example, Steiner states “Once the phenomenological laws of Faraday, Coulomb, and Amptre had been given differential form, Maxwell not that they contradicted the conservation of electrical charge. ... Yet, by ering with Ampére's law adding toi the “displacement current,’ Maxwell succeeded in getting the laws actually (Steiner 1989, 458). The “tinkering” is Ampere Law: cor HE 45 Foraday Law: cut B =~ SE Abneace offre magnetic quan ofits 0 Cote ef nina ney ein of yaw ce atg-- 2, a KIGURE 1 ‘The Symmetry Account Masepell and “the Method of Physical Analogy” 131 its role in the development of the theory is insignificant generative role of physical analogy. For example, Kitcher does not discuss ‘Maxwell’ analogies but indirectly eliminates them in claiming that the aether was not a “working posit” involved in prol and prediction, but a “presuppositional posit,” wh required to make the claims of the theory true (Kitcher 1993, 174). simply be removed from Maxwell’s theory. Yet without the working pi of the existence of the aether asa continuum mechanical medium Maxwell could not have derived the equations at all* Failing to follow the well-known dictum of Einstein, S Maxwell's deeds are also dismissed, for in dismissing the role of analogy, we arc left with no evidence in Maxwel’s papers, drafts, and correspondence as to what “other ‘means” he employed. I contend that in examining Maxwell's “deeds” we see they exhibit a remarkable harmony between word and deed. And, as with Binstein and Newton, Maxwell is one scientist from whora we can learn a great deal about the nature of scientific practice by listening to his words. in coming to grips with how Maxwell employed model structing the clectomagnetic field representation we gain significant insight into a central problem of creativity: Given that we must start from possible that we ever create anything ‘tem What is needed in order to resolve both the general a problem is an explanatory account of how analogy, and more ‘what [call “model-based reasoning,” functions to generate new repres tations—representations that ultimately come to transcend the specific ‘models that were employed in their generation, 2. Cognitive-Historical Analysis My research into conceptual change in several episodes in physies led to my characterizing specific concept formation practices as forms soning is generative of concey central forms: change in science, by focusing on three al modeling, and thought exp ‘then provide an interpretation ofthe role of physical analogy inthe development of Maxwell's theory. I present unified analysis of them because they are often employed together in rea- ing episodes. For example, as wll be discussed in section 4, the idle132 Nancy J. Nersesian. ‘Maxwell's drawing of the vortex-idle wheel medium (Maxwell 1890, vol. 1, plate 7) wheel-vortex model employed by Maxwell in his derivation of the elec: ‘tromagnetic field equations and illustrated by him in figute 2 exemplifies why a unified account is needed. On my interpretation this is a visual rep resentation of an analagical model that is accompanied with instructions for simulating it correctly in thought: “Let the cureent from left to right ‘commence in AB. The row of vortices gh above AB in motion in the opposite direction to a watch. .... We shall suppose the row of vor tices Hf still at rest, then the layer of part acted on by the raw gh on their lower sid& they are fice to move, they will rotate in the same time move from right to left, of the current, and so form an induced eh 1, 477, italics in origin ‘My analysis draws from practices employed in physics, but investiga tions of other sciences by philosophers, historians, and cogniti establish that these practices are employed across the sciences (see, e.g., Darden 1980; Darden 1991; Gentner et al. 1997; Giere 1994; Giere Masowell and “the Method of Physical Analegy” 133 1988; Giere 1992; Gooding 1990; Griesemer 19912; Griesemer 1991b; Griesemer and Wimsatt 1989; Holmes 1981; Holmes 1985; Latour 1986; Latour 1987; Lynch and Woolgar 1990; Rudwick 1976; Shelley 1996; ‘Thagard 1991; Trumpler 1997; Tweney 1987; Tweney 1992). Further, although these practices are ubiquitous and significant they are, of course, not exhaustive of the practices that generate new conceptual structures. ¢ account of model-based reasoning developed here stems from an interdisciplinary method that I have called “cognitive-historical” analysis. ‘The objective of that method is to create ‘opment of science that arc informed by stuclcs of historical and ct porary scientific practices and cogaitive science investigations human cognition pertinent to these practices. When used cceptual change, the “historical” dimension of the method is required to ‘uncover the practices scientists employ and to examine these over extended petiods of time and as embedded tural contexts. The “cognitive” dime; the analysis how human cognitive capacities and limitations ec and constrain the practices of scientists, Neither the practices nitive factors can be known « priori, empirical research is needed in both cases. Thus cognitive-historical analyses make use of the customary range Of historical records for gaining aecess to practices and draw on and con- duct cognitive science investigations into how humans reason, and learn, In contemporary philosophical parlance, the cognitive-historical ‘method is a “naturalistic” method of analysis. Underlying the method is a “continuum hypothesis”: the cognitive practices of scientists are extensions of the kinds of practices humans employ in coping with their physical and. social environments and in problem solving of th Scientists extend and refine basic cognitive strategies cally reflects this perspective, scientific cognition of the human species, by the develoy and by the cultural development ithin the broader framework of human cognitive provides a basis fiom which to develop an epis xt moves beyond the specific case study to more general ‘lusions about the nature and function of the scientific practices. Such134 Nancy J. Nersessian placement aids in establishing that the fragments of scientific research and discovery investigated are more widely representative of scientific practices and thus acts to support drawing more general conclusions from specific aspects of case studies. That scientific cognition is contextual does not pre- clude developing generalizations from specific cases. As the cognitive ogist Edwin Hutchins has argued, “There are powerful regular: transcends the details of the not about the domain [case] specific details, they are about the nature of | cognition in human activity” (Woods 1997, 15; sce also Hutchins 1995}, ‘That there is a continuum, however, does not rule out the possiblity that there are salient differences between the scientific and ordinary cog- nce most of the research in cognitive science has been con- ‘ordinary cognition, the cognitive-historical method has to be reflexive in application. Cognitive theories and methods are drawn upon insofar as they help interpret the historical and contemporary practices, ories are evaluated as to the extent 1 scientific practices, The assumpti ts from both sides are subjected to critical evaluat ‘with corrective insights moving in both issues posed by this research into model-based reasoning field of cognitive science is to push the field toward integration and uni- are custom: analogy, imagery, 1reas in cognitive science, such as tual change, categorization, problem solving, Although it is not possible to go into the deals or give extensive ref: ‘rences within the confine ofthis paper, my account of model-based rea- soning, derives ffom extensive historical and cognitive rescarch, The other scientific 3 literatures on analogy, men- tal imagery, ima 1d constraint satisfaction processes discussed and Goel 1996; Griffith 1999; Masowell and “the Method of Physical Analegy” 135 3. Model-based Reasoning Within philosophy the identication of reasoning with argument an logic is deeply ingsined. Tiaitooal accounts of Senae nr ected the notion of reasoning prima to de angumens, Embracing modeling practices as “methods” of concepal change in science requires expanding philosophical nouons of saentic reasoning to encompass forms of centre reasoning, any of which not be reduced to an algorithm in application, are not alway prhctive of eohuons, and where good image can lead wo icomec ot Accounts have proposed abduction aa form of ceaive eaoning, bt the nature of the derlying abductive inference and hypothesis gen eration are left argely unypecfid. Examining the modeling paces scientists as forms of eaoning generative of conceptal change provides s mean of spciing the nate of ome forms of seductive rence ‘he notion of model-based ceavoning opens a set of anus about the tole of representational format (neal and external in Diferent kinds of representations suchas lngus and aplog/icoic enable diferent Kinds of op guise and formulaic representation, for evap ‘presentation is evaluated a5 being truc or false. Operations on such expressions truth preserving if the symbols are interpreted in. the properties they refer to are stable in ‘operations can be defined in with the constraints that models, diagrams, and imagistic ry senting demonstratively. The telat tion, which T “goodness of fi similar in degree to what they represent, and are thus evaluated as accurate or inaccurate. Operations on iconic representations involve transformations of the re sentations that change their properties and relations in with the constai implicit, for example, a person can do simple reasoning about what happens when a rod is bent without hav cit rule such as “given the same force a |136 Nancy J. Nersesian In cognitive psychology there is an ongoing controversy about the nature of human reasoning that parallels the issues raised about reasoning in philosophy. This is not surprising since many philosophers who adhere view have played a significant psychological view, the mental operations underlying of applying a mental logic to proposition-like represen- 8. For some time critics of this view have contended that a purely syntactical account of reasoning cannot account for significant effects of semantic information exhibited in experimental studies of reasoning (see, © Johnson-Laird 1982; Johnson-Laird 1983; Mani and Johnson-Laird 1982; McNamara and Sternberg 1983; Oakhill and Garnliam 1996; Perrig and Kintsch 1985; Wason 1960; Wason 1968). Instead, they propose adopting a hypothesis, frst proposed by Craik (1943), that in many instances people reason by earrying out thought experiments on internal porary instantiation, the “mental modeling” hypoth: westigated for numerous domains, inchuding: reasoning about causality in physical systems (se, e.g, DeKleer and Brown 1983); the role of representations of domain knovwiedge in reasoning (See, ©. Gentner and Gentner 1983); logical reasoning (sce, e.g., Johnson Laird narrative comprehension (see, e.., Perrig 1985); and induction (sec, ¢g., Holland et al. 1986). Additionally, there is considerable exper ‘mental protocol evidence collected by cognitive psychologists to support ‘mental modeling as a fundamental form of problem solving employed by contemporary scientists (see, e8, Chi, Feltovich, and Glaser 1981; ‘Clement 1989; Dunbar 1995; Dunbar 1999; Griffith, Nersessian, and Goel 1996). These studies of reasoning processes provide further support for interpreting the modeling practices exhibited in the historical records of conceptual change as indicating that mental modeling has played a role in the past episodes ‘Though “the” mental modeling hypothesis is far from unitary and in need of critical examination, my analysis of model-based reasoning, has required adopting only a “minimalist” hypothesis: that in certain problem solving tasks the situations, dane cases the reasoning performed via mental cessfal because the models and manipulative processes embody largely ct ts governing everyday real-world events. Think, for exam ple, of how people often niture through a door. to geta large chair through the door by mentally simulating turning over a geometrical Maxevell and “the Method of Physical Analegy?” 137 structure approximating the configuration of the chair through vatious rotations. The task is made easier ysical chair isin front of the reasoner acting to support the structure in imagination, In the case of sci- ence where the situations are more removed from human sensory experi ence and the assumptions more imbued with theory, there i less a that a simulative reasoning, process, even if carried out correctly, wi success, Clearly scientists create erroneous models—revision and ev the gen- worked out, as Newton did with the concept of gravitation and Maxwell id with the concept of electromagnetic fed. To explain how model-based reasoning could be generative of tual change in scence requires a fandamental revision of the undets ings of concepts, conceptual structures, conceptual change, and reas customarily employed explicitly in philosophy and at le other science studies fields. It is not posible to pro details and arguments in the confines ofthis paper: account will be developed here. Hopefully it will bes reader to understand the interpretation provided of physical analogy.” A basic ingredient of he revisi sentation of a concept as providing sets of constra bers of classes of models. Concept formation an generating new, and modifying existing, is to view the repre- 9f radical conceptual chi ighly effective means of in constructing and tran138 Nancy J. Nersesian tation, but often, and especially in eases of conceptual change, no direct, analogy exists and construction of an initial model itself is required. In these cases the source domain(s) provides constraints that are used together with those provided by the target problem to create the initial as well as subsequent models (Nersessian 1992a; Nersessian 1999; Nersessian 2000). Evaluation of the analogical modeling process i in terms of how ‘well the salient constraints of a model fit che salient constraints of a target problem, with key differences playing a significant role in further model sgencration (Griffith 1999; Grifith, Nersessian, and Goel 1996). There is and the syntactic, semantic, and pragmatic constraints operating, on these processes (see, €g., Gentner 1983; Gentner 1989; Gentner et al. 1997; and Holyoak 1983; Holyoak and Thagard 1990). Although no cognitive theory is able to handle the complexity of the Maxwell case, the literature does agree with my analysis in that analogies are not “merely” guides to reasoning but form the creative heart of the reasoning processes in which they are employed. There is also widespread agreement (on criteria for good analogical reasoning, drawn from psychological stud- ies of productive and nonproductive use of analogy and formulated by Gentner (Gentner 1983; Gentner 1989): 1, “structural focus”: preserves al systems, 2. “Structural consistency”: isomorphic mapping of ‘objects and relations, and 3. “systematiity”: maps systems of intercon- nected relationships, especially causal and mathematical relationships. Constraints in both the target and source domains are domain-speci to be understood in the reasoning process at a suficient level tral objective is to create a model that is of the same kind with respect to salient dimensions of the target phenomena one is trying to represent. Thus, although an instance of a if a reasoning process are a8 representing fea phenomena. The relation between the generic model and the specific instantiation is similar to the type-token distinction used in logic. Generality in representation is achieved by interpreting the components of the representation as referring to object, property, relation, or behavior types rather than tokens of these. One cannot draw or imagine a “triangle Maszvell and ‘he Method of Plysical Analogy” 139 in general” but only some specific instance of a triangle. However, in con- sidering what it has in common with all triangles, humans have the ability to view the specific triangle as lacking specificity in its angles and sides. In considering the behavior of a physical system such as a spring, again one often draws or imagines a specific representation. However, to consider ‘what it has in common with all springs, one needs to reason as though it cific spring be generic. ‘The kind of creative reasoning employed in conceptual involves not only applying. generic abstract forming them during the reasoning, process. the generic level is a significant reasoning process in analogical in conceptual chenge which often requires recognition of potential larities across disparate domains, and abstraction and integration of i mation from these. There are many significant examples of generic abstraction in conceptual change in science. In the domain of cl mechanics, for example, Newton can be interpreted as employing ge abstraction in reasoning, about the commonalities among the 1 planets and of projectiles, mnabled his formulating a u ‘matical representation of their motions. The models he employed, under- stood generically, represent what is common among the m example, that within the cont jecties can both be represented as point masses. After Newton, the model of gravitational force served as a generic modi t-at-distance forces for those who tried to bring all forces into the scope of Newtonian mechanics. AA variety of perceptual resources can be employed in mo focus on the use of the visual modality sinc of conceptual change across the sciences tal imagery that provides evidence that hum imaginative combinations and transformations that mimic perceptsal spa tial transformation (Kosslyn 1980; Shepard and Cooper 1982). These sim- ulations are hypothesized to take place using internalized con assimilated during perception. Other research indicates that people use varM40 Nancy J. Nevesian nus kinds of knowledge of physical situations in imaginary simulations. or example, when objects are imagined as separated by a wall, the spatial transformations exhibit latency time consistent with having simulated ‘moving around the wall rather than through it. There are significant dif- ferences between spatial transformations and transformations req ‘causal and other knowledge contained in scientific theories. Although gery in problem solving is scant, recently cognitive sci soning of scientists, such as Bohr, who claim other words pictures, in reasoning. The conflation of mental imagery wi pictures-in-the-head stems from the fact that we presently lack an adequate ‘means for expressing the notion ofa representational format that is neither picturelike nor linguistic. External visual representations provide support for the processes of constructing and reasoning with a mental model. They aid significantly in organizing cognitive activity during reasoning, such as fixing attention on the salient aspects of a model enabling retrieval and storage of salient information, and exhibiting salient interconnections, fas structural and causal, in appropriate co-location. Further they facil- construction of shared mental models within a community and trans portation of scientific models out of the local milieu of their construction. ‘As used in mode in physics, visual representations te in modeling phenom including providing abstracted and idealized representations of aspects of phe embodying aspects of theoretical models. For example, ear construction of an electromagnetic field concept, the imag sd reason the patterns of iron filings surrounding a magnet (figure 3) substantiates that later in his development of the feld conc istic model functioned as the embodiment of a dynamical ‘theoretical model of the transmission and interconversion of forces, generally, through stresses and strains in, and various motions of, the lines (Gooding 1981; Gooding 1990; Nersessian 1984b; Nersessian 1985; Tweney 1985; and Tweney 1992). But, as I have argued, the visual representation ‘Maxwell presented of the idle wheel-vortex model was intended as an embodiment of an imaginary system, displaying @ generic dynamical rela Masowell and “the Method of Physical Analogy” un FIGURE 3 araday’s drawing of the lines of force surrounding a bar magnet (Faraday 1839-55, vol. 1, Plate 1) tive reasoning would involve constructing a model and using tacit and explicit knowledge to produce new states from it. Constructing a thought experimental model requires understanding the salient constraints govern- Of entities or processes in the model and the possible causal, | and fisnctional relations among them, Conducting a simulation ‘or explicit understanding of the constraints governing how those kinds of things behave and interact and how the relations change. A simulation creates new states of a system being modeled, in turn creates or makes evident ‘Changing the conditior of a model enables inference the way that a system can behave. Because the sim the same constraints of the system it represents, 8 4 simulation with a model enables ted narrative repos142, Nancy J. Nersesian videdl a significant means of eflecting conceptual change within a scient community. Accounting for the generative role of this form of model- based reasoning begins with examining how these thought-experimental narratives support modeling processes, and by means of cognitive-histori- ‘eal analysis infers that the original experiment involves a similar form of ‘model-based reasoning. What needs to be determined is: (1) how a narra tive facilitates the consteuction of a model of an experimental situation in ‘thought and (2) how one can reach conceptual and empirical conclusions by mentally ing the experimental processes, From a mental modeling perspective, the function of the narrative form. of presentation of a thought experiment would be to guide the reader in ‘constructing a mental model of the situation described by it and to make inferences through simulating the events and processes dey thought-experimental model can be construed as a form of “discourse” ‘model studied by cognitive scientists, for which they argue that the oper: ations and inferences are performed not on propositions but on the con structed model (see, ¢g., Johnson-Laird 1982; Johnson-Laird 1989; ,,and Greenspan 1989; Perrig and Kintsch 1985). Unlike ofthe scientific thought experi significant abstractions, wi dimensions of the model and in recognizing the situation as prototypical ental consequences are seen to go beyond narrative is presented in a polished form that “works, it an effective means of getting comparable mental mod bers of a community of scientists. Undoubtedly some experimental yn and tweaking goes on in the original re ‘ive construction, although accounts ofthis process are scientists. ‘Although some kinds of mental mo jons, those derived fgom thought-experimei employ static tific and general world knowledge to make constrained transformations from one possible physical state to the next. Much of the information ‘employed in these transformations is tacit. Thus, expertise and learning play a crucial role in the practice. The thought-experimental reasoning processes link the conceptual and the experiential dimensions of huma cognitive processing (see also Gooding 1992). Thus, the constructed uation inherits empirical force by being abstracted both ftom our experi- Maxovell andl “the Method of Physical Analogy” 43 ences and activities in the world and from our knowledge, conceptualiza- tions, and assumptions of it. In this way, the data that derive from thought ‘expetimenting have empirical consequences and at the same time pinpoint the locus of the needed conceptual reform. The derived understanding forms the basis of further problem-solving, efforts to construct cally adequate conceptualizatic In summation, there are several key ingredients common to the various -based reasoning considered in this section. The mod 4. Maxwell’ ‘A Cognitive-Historical Interpretation 41 The “Method of Physical Analogy” mage of the geometrical fea studying geometry by this i, we have our minds engaged to fancy ourselves geometers ‘Maxwell in the ranks of cognitive-historical anal ful. In investigating a new area in science, Maxwell asserted that one begins process ofaaa Naney J. Nersessian ‘can grasp them” (Maxwell 1855-56, process requires a “method of investigation, which allows the every step to lay hold of clear physical conception, without being com- mitted t0 any theory founded on they physical scence from which that conception is borrowed so neither drawn aside from the 1or carties beyond truth by fy partial makes each of id.). As Howard has pointed ou nse is an isomorphism, an equivalence of fo: could be applied, but through the discovery of an equivalence of form between the dynamical structure of certain mechanical relations and cer tain electromagnetic relations, he was able to construct the requisite struc- tures, This kind of model-based reasoning process has the potential to genuinely new representational structures, change. It does not matter whether the 1 the models do or do ‘mechanically conceivable. ing to the classes of phe mon to mechanics and processes Maxwell abst mathematical form of genetic dynamical structure, the mathematical representation of the el ” That ‘class of mechanism, in other words, of the these provided by the arget problem to re- ate the imaginary analog that served as ‘Maxwell also employed several imagistic represent ure 2, which we discussed previously, and those in figure 4, which were Masoell and “the Method of Physical Analogy” 145 4 “Maxwell 1890, vol. 1, p. 460, figs. 1-3, FIGURE 4 accompanied by text for the motion of the vortices in the planes above and below during the period, what litte draft material ex published work. As I have argued elsewhere (Nersesi 1992a) I believe there is sufficient evide reasoning processes. We can look at him as attempting to lead his audience through his own reasoning processes as a rhetorical move to help his col- leagues to understand the new field representation,46 Nancy J. Nevsesian 42 The Initial Model ‘Maxwell constructed a mathematical re ¢ field concept over the course of sey rs. I will focus on “On physical lines of force” (Maxwell 1861-6: Ww the relevance of tha analysis for “A dynamical theory of the electromagnetic field” (Maxw 1864). In part 1 of the 1861-62 paper, the mathematical representation of various magnetic phenomena derives from a vortex-fluid model of the aether. Maxwell began tion for the electromag he contiguous parts of sions differen in the medium” (1861-62, 454). F greatest pressure and a the axis of least pressure, Maxwe ‘magnetic phenomena are stresses ‘magnetic aether, which transmi through the space surrai bodies and charges. Given this fone can assume a is I relations chat hold in the don and elastic solids, and those that hold in the domain of electromagnetism. Thus, con jum mechanics can serve as a source domain for constructing models. problem is to determine which relations, and determination of these ‘was guided by constraints drawn fiom the domain of electromagnetism. As specific constraints on stresses that could account for magnetism, Maxwell considered (1) a tension in the direction of the lines of force and (2) a pressure greater in the equatorial than in the axial direction. These con straints are consistent with the geometrical configurations of the magnetic lines of force and Faraday’s interpretation of them as resulting from lateral repulsion and longitudinal attraction. The magnetic constraints specify a configuration of forces in the medium and this configuration, in explained as resulting from the centrifugal forces of vortices in the medi with axes parallel to the lines of force. ‘The centrifugal force of a vortex would cause it to expand equatorially and contract longitudinally. Each dipolar in that it rotates in opposite directions at its extremities. the geometric constraints on the lines of force ate satisfied by the shape of the vortex, which is wider the farther it is from its or lines become farther apart as they approach their midpoints. ex motion supplies a causal process that is capable of producing the con- figuration of the lines of force and the stresses in and among them. Figure 5 is my representation of such a vortex drawn from Maxwell's description, The vortex-fluid model is also consistent known experimental results: ( angles to each ot Maxovell anil “she Method of Physical Analogy” 47 FIGURE 5 Vortex segment ized light passed through a diamagnetic substance is rotated by magnetic The system of infinitesimal vortices does not correspond to any kno physial system. Maxwell constructed it ro serve as the bass for driving ‘mappings between known 4 mechanies and those thought to produce electromagnetic phenomena. The Iatical expressions for the magnetic phenomena are derived from the ‘mathematical formula for the stresses in the vortex-ffuid model by substi- tution. The vortex-fluid model is “generic” in that it is to be understood as satisfying constraints that apply to the syper of entities and processes that can be considered as constituting either domain. The model represents the class of phenomena in each domain that are capable of producing the spe- cific configurations of stresses. The modeling process Maxw throughout the analysis went as follows. Fist he constructed a model rep- resenting a specific mechanism, Then he treated the dynamical properties and relations generically by abstracting. features common to the mechani- cal and the electromagnetic classes of phenomena. He proceeded to for- ulate the mathematical equations of the generic model and substituted in the electromagnetic variables. In part 1 of the analysis Maxwell used the mathematical properties of the limiting case ofa single vortex to derive formulas for tions consistent with the only give the highlights of jal relations due to variation in (i grad v)] + [nvx ( Se denotes the cross pro ich we now call the general mechanical1s. Naney J. Nersesian stress tensor). He then constructed the electromagnetic version by map- ping quantitative properties as stipulated that the quantities related to the velocity of the vortex (ct = v1, B = vim, and 7 = vn, with 1, 1m, m the direction cosines of the axes of the vortices) be mapped to the ‘components of the force acting on a unit magnetic bar pointing north the magnetic a contemporary notation is designated as “SET, is here related to the velocity gradient of the vortex at the surface. ‘The quantity ‘pis taken to represent the magnetic permeability, thus relat- 0 the mass of the medium, The quantity ‘WED represents the mag- netic induction, Substituting the magnetic quantities, Maxwell rewrote the first term of ‘the mechanical stress tensor for the magnetic system as H(1/4x (div wH)) (equation 7, p. 459). He followed the same procedure of constructing a ‘mapping between the model and the magnetic quantities and relations to rewrite all of the components of the stress tensor for magnetism. The resulting lectromagnetic stress tensor represents the resultant force on an clement of the magnetic medium due to its internal stress. The four com ponents of the mechanical stress tensor, as interpreted for the electroms netic medium, are F = [H(1/4n (div wE1))] + ) the force of magnetic action component is required itions and map- pings, except i Out that he also derived an expression relating cur- rent density to the circulation of the magnetic field around the ‘current-carrying wite j = 1/47 (cuel H) (equation 9, p. 462). This equa .w he had derived from iderations in the frst paper of this series of three (1855-56, lerivation given here still did not provide a mechanism con’ necting current and magnetism, 43 Introulucing Ile Wheels Maxwell had been able to provide a mathematical form [ation for magnetic induction, paramagnetism, and diamagnetism through modeling these phenomena by means of a nonexistent, but mechanically conceivable dynamical system. A mechanics fluid model ed Maxwell to a means of repe between magnetism and electricity. He began part 2 by st Purpose was to inquire into the connection between the magnetic vortices Masowell and “the Method of Physical Analogy” 9 FIGURE 6 Cross section of model of vortex-fluid meditim and current. Thus he could no longer simply consider a single ge tex in his analysis. Hee admitted a serious problem with the mode! he “found great difficulty in conceiving of the existence of vortices medium, side by side, revolving in the same direction” (468). Figure 6 my drawing of a cross section of the vortex-fluid model as described by “Maxwell. By imagining the motion of the vortices in this gure, it bee ‘evident that direct contact between co in that there will be fiion and, thus, jamming. Further, since they are going in the same direction and, thus, at points of contact they would be ‘going in opposite directions, in the case where they are revolving at the same rate, the whole mechanism machine mechanics this kind of problem is solved by the “idle wheels.” On that basis he proposed to enhance his imaginary model by supposing that “a layer of particles, acting as idle whet between each vortex and the nex “Maxwell stipulated that the particles would evolve in place in di ‘opposite to the vortices without slipping oF tou seutive vortices poset a prok ayers of the vortice simplified calculations, but is inconsistent with the ‘mechanical constraint that the vortices have elasticity, and would be clim-150 Nancy J. Newesian inated in part 3. Figure 2 is Maxwell’s rendering of the idle wheel-vortex model. ‘The diagram shows a cross section of the medium. The vortex cross sections are represented by hexagons rather than ciscles, presumably to provide a better representation of how the particles are packed around the vortices, with the three-dimensional dodecahedsa approximating to spheres in the limit. ‘The idle wheel-vortex model is a hybrid constructed from two source domains: fluid dynamics and machine mechanics. As discussed earlier, to combine salient entities and processes fzom two disparate domains requires abstraction of these to a sufficient level. My explanation of how generic abstraction could have led to the introduction of the idle-wheel par ‘Modes of Connection: FIGURE 7 Introducing idle wheels via generic modeling Masowelland ‘ihe Method of Physical Analagy” 1st wheels from the vortex-fluid model (A). ‘The generic model of wheels reminded him of specific mechanical systems conti ‘gears (B). He noticed an analogy between the vortices and the gears ( ‘but how this analogy would provide a new mode of connection for the vor tices was not immediately evident. Next, from the model of the machine ‘gears he abstracted the generic mode! of idle wheels (D), and then further abstracted that model into the generic model of dy ly smooth con- he instantiated the generic model of tigated, and between che because the of the same common. This class includes electric and magnetic assumptions of Maxwell’s treatment. Thus, in the analy action discussed below, the idle wheel-vortex mechanism is not the cause of electromagnetic induction; it represents the casal structure Of that kind of process. Thro fart 2 Maxwell provided analogies ‘with machinery as specific me hhad derived between the idle-wheel par establish for the reader that there are real physical syst the generic relations, In ordinary mechanisms, 5 governor for stear ‘out of place. The steady current pro a current in it. The dynamical relations between the vortices and the idle wheels serve to model the constraints governing the dynamical relations between electric currents and magnetism,152 Nancy J. Nersessian Going through the piece of the analysis in which Maxwell derived the ‘equations for the translational motion of the particles in the imaginary sys ‘em will help us to understand more fully the relationship between the ‘model and the electromagnetic system, There is a tangential pressure between the surfaces of the spherical particles urfaces of the vor tices, treated here as approximating rigid pseudospheres. So conceived, these vortices appear to be incor the vortices in part 1 which should require the vortices to be clastic, but this would not be addressed until the analysis of static electricity in patt 3 ‘Maxwell derived the equation for the average flux density of the particles as a function of the circumferential velocity of the vortices p = 1/2 (p curl ¥) (equation 33, p. 471) and noted that it i of the same form as the equa- tion relating current density and magnetic field intensity j = 1/4 (curl H) (equation 9), the form of Ampéte’s law for closed circuits he had derived fed to make the eqy that “it appears there! is represented by the transference of the between the neighboring vortices” (471). That is represented by particles are free ‘experience resistance and W: ‘with current, A continuo connects current with the rotating, es exert on the particles. Maxwell went on to show that n is consistent with the equations he had derived in part 1 for the distribution and configuration of the magnetic lines of force around a steady current (eq 15-16, p. 464). Although I won’t go through the derivations, Maxwell derived the Jaws of electromagnetic induction in two parts. Again we see the role of Mesewell and “the Method of Physical Anslagy” 153 clin the derivation, since the two cases are diffe in the idle wheel-vortex system. In the first case, the mecha ‘municating rotational velocity in the medium accounts for induction of currents by starting or stopping a primary cucr switching current off and on in a conduct ‘current in a nearby conducting loop. A decrease or increase in curves cause a corresponding change in the velocity of the adjacent vortices. st mechanically inducing a current orient ‘ageees with experimental results, In the second cas induced by motion of a conductor across the ), Maxwell used considerations pertaining, to direction opposite connecting the ends of the wire, The case of ope part 3. 44° The Displacement Current” and Inconsistent Signs id of part 2, Maxwell had gi phenomena. The fall m: field was constructed in part 3 with Mo ity. Lwill focus on next paper on the how he eliminated itin 1873 a deeper appreciation of the tial representation of stati the basis of the idle wheel-vortex identify charge with the accumu154 Nancy J. Nersesian ary of dielectric and conducting media. Thus, itis puzzling why Maxwell did not immediately proceed tothe type of analysis he ultimately presented and he provides no clues of his path to that solution. Siegel (1991) pre- sents a detailed and plausible analysis of the nature of the problems ‘Maxwell would have encountered in tying to construct an account ofthe interfce between conducting and diclectrie media given the specific mech- anism of the model The solution to these problems was to make the vor- tices elastic. So, the idle wheel-vortex model was modilied in part 3 by again considering its plausibility asa mechanical syste. In part 2, the sy5- tem contains vortex cells of rotating fluid separated by particles very small in comparison to them. ‘To simplify the calculations forthe transmission of rotation fom one cell to another va the tangential action between the sur- face ofthe vortices and the particles, Maxwell had assumed the vortices to be rigid. But he now noted that in order forthe rotation to be transmit ted from the exterior to the interior parts of the cell, the cell material needs to be elastic. And, although he does not comment on it, conceiving (of the molecular vortices ax preudosphecal blobs of elastic material would also give them the right configuration on rotation (figure 5), and thus climinate the inconsistency of the rigid vortices with the geometrical con- straints of part 1 for magnetism. "He began by noting the constraint that “electric tension” associated with a charged body is the ‘whether produced from static or eueent electricity. produce ether et (490) and a dielectric to an clastic membrane which does sage of a fluid, but “transmits the pressure of the fuid on one side to that the other” (491). In the process of electrostatic induction, electricity can be viewed as “displaced” within a molecule of a dielectric, 60 that o side becomes cule to mole hybrid model ‘made al discussion. I introduced before hout the model, electrostatic induc lacement is similar these calculations cannot properly ‘Maxwell claimed that the dis tion can be likened to a curr “the commencement of a curre iven the mod electrostatic force produc the vortices caus- ing a slight translational motion of the idle-wheel particles, which is prop- agated throughout the dielectric medium, Masovell and “the Method of Physical Analogy” 155 placement ‘and the disy shat whi given here can be accounted for clastic resorting force is opposite in ‘we keep in mind that jentation to the ns expressed by the current.” Ieis uanslational motion of the particles which co rent Thus, in its intial derivation, the “displacement ct con a specific mechanical process. We can sec this in the following way. Recall the difference Maxwell specified between conductors and dielectrics when he first introduced the idle-wheel particl tor, they are free to move from vortex to vortex. In a dielecttc, they can ‘only rotate in place, In electrostatic induction, then, the particles can only bbe urged forward by the elastic distortion of the vortices, bu out of place. This motion is similar to that ol current.” But, their motion “does not amount to a current, because when. it has attained a cectain value it constant” (491). That ticles do not actually move ov action, instead they accumulate, creating regions of stress. Since they are free to flow, they must react back on the vortices with a force to restore their position, The system reaches a certain level of stress and remains there. ‘Charge’, then, preted as the excess of tension in the dictec: tric medium created by the acct the model, “displacement cu what bothered so many of the readers of the Treatise, cal model is no longer employed. As we will see below, lems for Maxwell in his 1864 analysis, Since the vortices are now elastic and since are free to move, the current produced by the particle per unit area) must elasticity. So Maxwell corr 0 the rotation of the vortices, the closed current of equation with the derive “displacement current” actually ope itcuial current? He coupled this equati tinuty for charge, which inks current and charge,186 Naney J. Nereian an expression linking charge and the electric field, © = 1/4n (K2 div B) (equation 115, p. 497), which is equivalent to p = - div D. This la oks similar to what we now call Coulomb’s law except for wo at turn out to be highly salient for understanding Maxwell's rea the form of this equation and the modified equat rent (equation 112) again den tes Maxwell’s field conception of ‘current and charge: intezpreted left to right, charge and curr the feld. Second, the ign is not part ofthe contemporary equation, odel because the elastic restoring force exe the vortices by the particles and the electromotive force have opp entation. Through what can be interpreted simply as a substitutc in equation 104 the equations in this paper are consistent, ‘Turning to the 1864 paper, we can see how the specific interpretation of the displacement current created problems. In this paper, Maxwell re erived the ficld equations without explicit reference to the me: model. c yznetic dynamical properties nergy. Here ight acthe 1-62 analysis, where the close agreement of the velocity of transverse vibratic cal media led him to state “we can scarcely avoi nsverse unslulations of the same medium whi slectric and magnetic phenomena” (italics in the original, interpret Maxwel an clastic medium t provided by the model In the 1864 paper Maxwell for propagating electromagnetic action tucated the aether as a generic ek mber) and thus saw no reason for multiply ing acthers. Blastic systems can receive and store energy in two forms, what ‘Maxwell called the “energy of motion,” or kinetic energy, and the “energy tension,” of potential energy. He identified kinetic energy with mag. polarization and potential energy with electric polarization. Figure 8 tes my interpretation of how the 1861-62 analysis enabled hitn to do this through generic abstraction from the model, Although in 1864 Masovell and “the Method of Plysical Analogy” 157 ‘Maxwell is thinking of the actherial medium in more abstract, dynamical terms, vestiges of the earlier specific mechanical mod: .own to have remained in his thinking and this created a probl the current and charge equations. genetically in terms of the stress created by lectzicity, so E ~ KD and coupling the equation for the total curr 1/4 (eutl H) - AD /At with the equation of continuity ap/at + div j= 0, yields p - div D. However, Maxwell wrote the equivalent of p ~~ div D as the “equation of free electricity” form of the 1861-62 analysis. So the complete interpretation is that Maxwell continued to think of charge as associated oe + ve Rational Motion Rotation of Magnetic Vo tale wheel particles FIGURE 8 Identifying energy components via generic modeling,158 Nancy J. Neresian with the acc ‘mechanism of idle-wheel particles through the specific thus with the reactive force that is ori- ented away from the accumulation point. However, the mathen cquations of the ge dium require that it be pointing toward the stress point and so clearly require p = div D. ‘There are few existing drafts of Maxwells published papers, but fortu nately there are a couple of pages of a draft pertaining to this derivation.29 ‘These reveal that Maxwell had some confusion about how to think fent and charge without the medium. In the draft equation for curren ‘Maxwell wrote the components of “electric resistance,” in other words, the electromotive force required to keep the current flowing through a ‘conductor, a8 pointing in the opposite direction, as would have been the ‘case in the mechanical medium, but in the published paper, these are writ ‘ten correctly. Inthe draft equation for “statcal electricity” the components are written with the negative sign as above. Maxwell's handwriting ean be interpreted as indicating that he struggled with this result In writing the first two components Maxwell made the equals sign three times the length used in other samples of his handwriting and he actually blends the equals, sign into the minus sign, Only for the final component is the equals sign cof regular length and clearly separated fom the minus sign. As noted above, itis also writ the published paper.!2 ‘Maxwell never discussed this inconsistency and then in the Treatise, again wy is gone. Althouj speculation on my part, given how Maxwell collects all of hi hhard to imagine that he di fency. I believe he kept the Coulomb equation p sp: 561) in 1864 because he had not figured out how to con ceive of charge ge from the specific mecha- nism through which he had been able initially to represent it. In the ‘Treatise charge is abstracted from the notion of stress and re ddue to accumulating particles and the flow of electricity through « medi of flow. In conduction the current flow stress is created at points of discontinuity, such as where a charging capac’ itor and a dielectric meet, The generic notion of charge as associated with clastic stress is Maswell and “ihe Method of Physical Analegy” 159 5. Further Reflections: Model-based Reasoning and Conceptual Change My analysis has shown how Maxwell was able to formul clectromagnetic field by abstracting from specific mechanical mode! I properties and relations that continuum-m certain machine mechanisms, and electromagnetic syst ‘mon. In their mathematical and relations were separated the models through which they had been rendered concrete. TI inconsistencies in the models could be ignored. The gencric ial relationships represented by the imaginary systems of the mod. ved as the basis from which he abstracted a mathematical structure sal m analysis in ‘ofthe magnetic: that at an. sure of the creative reasoning empl representational change. IF Maxw ymagnetic field in the way domains, he constr lations of greater generality than generic abstraction from mech a system of abstract laws th applied to systems yield the laws of a dynamical system the fone that is inconsistent with the laws of the mechanical which its mathematical structure was abstracted. For Maxwell, Newtonian ‘mechanics and general (Lagrangian) dynamics were thought to be coex tensive, In artiving at the general dynamical form of the field equations160 Nancy J. Nersesian starting from considerations of the specific mechanisms of the models, he could think that he had demonstrated the possibility of a “mechanical explanation” for the electromagnetic phenomena. Bu the Lagrangian for: ‘malism provides no information about the underiying system. Maxwell did not, however, “turn away ftom mechanical models” (Siegel 1991, 54). A complete explanation would need to provide an account of the actual ‘mechanisms in the aether that create the phenomena, and Maxwell fully expected such an account to be forthcoming. The laws he had formulated constraint on any acceptable mechanism. Maxwell saw him- ‘Newton had in formulating the ing] the forces w wovided a satisfactory assessment of the role Maxwell’s analogies played in the development of hs electromagnetic theory but I thank him for inspiring, me to try, the 1861-62 paper. Hess's positio to characterize because she does see analogy as playing a sig Inypothesis formation and theory interpretation in general, However, her ieory pastes over the analysis in the 1861-62 paper—which I arguing contains the main generative work—entirely (Hesse 1973). Although I cannot elaborate here, I believe the Masovelt and “te Method of Physical Analogy” 161 them, and would lke ta have them regarded by others, not as creations of thought ‘but as given realities” (Einstein 1973, 264). 6, See especially leters of November 13, 1854; May 18, 1855, pp. 8 and 11; and September 13, 1855, pp. 1718. 7. Thave written this equation, which Maxwell wrote in rejected this alternative in the calculations for part 2 and this specific feature of the idle wheel-vortex version of the model, 9. See Nersessian 1984, 82 and Siegel 1991, 112. 10. Add MS 7655, V, ¢/8. Universiy Library, 11. See Siegel 1991, 174-75 for 12. Harman mistakenly sys this eg the 1864 paper (Harman [Heimann] 1995, 1 REFERENCES of World View From Faraday His Theory of Light Application of it to lecromagnetitm, Seudier in the History and Philomphy of Science 4 ( 107-64. 1986. The Heuristic Role of Maxwell's Mech lectromagnetic Phenomena. Studies in she History and 7415-27. Chi, M. TFL, PJ. Feltovie ‘Representation of Physics Problems by Experts and Novices. Cag 5:121-82, (Clement, J. 1989. Learning Via Model Construction and. Of Creativine: Auesment, Theory, and Research, I Model of losphy of Science Sele Hondboot ied by G. Gloves, R. Darden, L. 1980. “Theory Construction in Genetics.” In Scientific Discovery: Case Studies, edited by T. Nickles. Dordeecht: Reidel162 Naney J. Neresian ——. 1991. Theory Change in Science: Strategies from Mendelian Genstcr. New York: Oxford Universi DeKleer, J.,and J. 8. Brown. 1983, Assumptions and Amb ‘Mental’ Models. In Mental model, edited by D. G. Lawrence Erlbaum. Duhem, P. 1902. Les uhéries électriques de J. Clerk Maxwell: Eeude histarique et ei igue, Pats: A. Hermann & Cie. 1914. The Aim and Structure of Physical Theory. Translated by P,P. Weinet ‘New York: Atheneum, Dunbar, K. 1995, “How Scientists Really Reston: Scientific Reasoning in Real- ‘world Laboratories.” In The nature of Insight edited by R. J. Steenberg and J se." In Model-based J. Nersessian, and “Structure-mapping: A Theoretical Framework for Analogy” 7155-70. 1e Mechanisms of Analogical Learning.” In Similarity and soning, edited by 8. Vosniadou and A. Ortony. New York: Cambridge University Pres. Gentner, D.,§. Brem, R. W. Feruson, A. B. Markman, B. B. Levidow, P. Wolff Reasoning and Conceptual (Case Study of Johannes Keplez” The Journal ofthe Learning Se 340, Gick, M.L,, and K. J. Holyoak. 1980, “Analogical Problem Solving.” Caynitine Prychlapy 12: 306-355, ——. 1983. Schema Induction and Analogical Transfe:” Cygnitine Pocholeny 15: 1-38. Giere, R. 1994, “The Cognitive Structure of Scientific Theories.” Philsephy of Seience 61:276-96. “Final Steps to the Field Theory: Faraday’s Study of e—s—«—s—=—=sr Plsical ences 11:231-75, 990. Experiment and the Making of Meaning: Human Age iemtifie Observation and Esperiment. Dordkecht: Kluwer, 1992, “The Procedural Turn: Or Why Did Farsday’s ‘Thought Mascwell and “the Method of Physical Analogy” 163 ——. 1991b, “Muse Scientific Diagrams be Bliminable? The Case of Path n What the Philsepy of Biology 1, Essays for David ll, edited by M. Ruse. Dordrecht: Kluwer. Gaiffih, ‘T. W. 1999. “A Computational Theory of Generative ‘Scientific Reasoning.” Ph.D., College of Computing, Georgia Institute of “Technology, Atlant Griffith, TW. NJ. Nersessian, and A. Goel, 1996, “The Role of Generic Models in Conceptual Change.” Val. 18 of Proceeding of the Cegnitve Science Socesy, 312-17. Hillsdale, NJ: Lawrence Erlbaum. 12000. “Punction-follows-form Transformations in Scientiic Problem Solving.” Vol. 22 of Precedings of the Cagnitive Science Society, 196-201, Hillsdale, NJ: Lawrence Erlbaum, Harman (Heimann), P. M. 1993. The Scientific Lettrs and Papers of James Clerk ‘Masovell 3 vols. Vol. 2. Cambridge: Cambridge University Press. Hegarty, M. 1992. “Mental Animation: Infersing Motion from Static Diagrams of ‘Mechanical Systems.” Journal of Experimental Prycholagy: Learning, Memory, ‘Inference, Learning, and Discovery. Ca 1981. “The Fine Structure of Scientific 1996. Mentat Leaps: Analy in Creative Toyghe. Cambridge, Mass MIT , 1995. Cygnition in she Wild. Cambridge, Mass. bridge, Mass: MIT Press. Foundations of Cognitive Science, edited by M. Posner. Cambridge, Mass: MIT Press Kitcher, P1993. The Advancement of Science. New York: Oxford U Kossyn, §. M. 1980. Image and Mind. Carbide, Mass: Harvard164 Naney J. Nerssian Larmor, J. ed. 1937. The Origins of Clerk Mavell’s Electric Ideas. Cambridge, ‘Mass: Cambridge University Pres, Latour, B. 1986. “Visualisation and Cognition: Thinking with Eyes and Hands.” Knowledge and Society 61-40. ‘Cambridge, Mass: Harvard University Press. Mani, K. and FN. Johnson-Laird, 1982. “The Mental Representation of Spatial Descriptions.” Memory and Ci 10:181-87. ‘Maxwell, J. C. 1855-56. *On Faraday's Lines of Force.” In Scientific Papers ——- 1856. “Are There Real Analogies in Nature?” In The Life of James Clerk { by L. Campbell and W. Garrett, London: Mac “On Physical Lines of Force.” D. Niven. Cambridge: Cambridge Unive Ian and Co, entific Papers edited by W. “on Aci aa Distance In Seni Pa 18906. Tie Sense Paper of Jomes Cork Maxwell Ete by WD, vols. Cambridge: Cambridge Univesity. McNamara. and RJ Srerberg 1983, "Mental Model of Word Meng” Journ of Vera earings Viral Bayon 22489°74 Morrow, D.., G.H. Bove, snl S. 1. Grecnpen. 1989. "Updaing Stuaon ‘Models During Narasve Comprcheasion.” Journal of Memory nd Language 28:292-312. I a ” ace Nenenian, N. J. 1984s, “Acter/Or: ‘Th ing Meaning in Scientific Tories. Dordrecht: Martinus Nihoff/Kluwer Academic Publishes. 1985. “Faraday Faraday Reditcovered: Esays on the Life and Work of Michnc! Faraday, edited by D, ©. Gooding and F. A. J. L. James. London: Macmillan 1988. “Reasoning from Imagery and Analogy in Scien Formatio 1992. “How Do. Scie ‘Conceptual Change in Science in the Philsophy of Science, edited by R. Giere. Minneapolis: Univetsty of Minnesota Pres. 1992b. “In the Theoretcian’s Laboratory: Thought Experimenting as ‘Mental Modeling.” PA 1992, 1995. “Opening the Black Box: Cognitive Science and the History of ence.” Osiris 10 (Constructing Knowledge in the History of Science, ed. A, ——. 1999. "Model-based Reasoning in Conceptual Change.” In Model-based Reasoning in Scienifi Discovery, edited by L. Magnani, N. J. Nersessian, nd P.Thagard. New York: Kluwer Academic/Plenum Publishers, Masowell and “the Method of Physical Analegy” 165 2000. “Abstraction via Generic Modeling in Concept Formation in Science.” In Carvecting the Model: Abstraction and Healizatin in Science, 35, “Propositional and Situational Representations ‘Memary and Language 24:508-518, MJ. S. 1976. “The Emergence of a Visual Language for Geological Science.” History of Science 14:149-96. Schwartz, D. L., and J. B. Black, 1996, “Analog Imagery in Mental Model soning: Depictive Models.” Cognitive Pycholagy 30:154-219, Shepard, RN. L. A. Cooper. 1982. Mental Images and ‘Their ‘Transformations. Cambridge, Mass: MIT Press. Siegel, D. 1986. “The Origin of Displacement Current.” Historical Sudies im the Pipsical Sciences 1799-145, 1991. Innovation in Maxwells Electromagnetic Theory. Cambridge: (Cambridge University Press, 970. “On the Notion of I, and Bey ial and Philosophical Perspectives on Science, edited by R. Hi Stuewer polis: University of Minnesota Press. 1976. “On Action Ata Distance: Metaphysics and Method in Newton and “Maxwell” Unpublished tal, Yale University. 7 ine Period Following Maxwell” In by G..N. Cantor and M. J. 8. Hodge. Cambridge: ‘rumpler, M. 1997. “C fof Representa verging Images: Techniques of nt f Sodiam-channel Proteins in Nerve Juernal ofthe Hi 0:55-89. Tweney, R. D. 1988. “Faraday’s Discovery of Induction: A Coy ‘in Faraday Rediscovered, edited by D. Gooding and P. A «is Scicntfic Thinking?” Unpublished manuscript 1992, “Stopping Time: Paraday and the Scientific Creation of Perceptual Onder” Pris 29:149-64. Failure to Eliminate Hypotheses in a Conceptual erimensal Pacholgy 32:109-123. in a Conceptual Rask—A. by PC. Wason and PN.166 Nancy J. Nevsesian ‘Computer Medium: Bcological Perception and Aiding Human Cognition.” In ‘Te Ecology of Hursan-machine Stems edited by J Bach etal. Hillsdale, N.J. Lawrence Eelbaam, ACKNOWLEDGMENTS 1 appreciate the comments received on earlier versions of this paper from Abner Buchwald, and other participants in the colloquium presented st the forthe History of Science and Te