18
On the Extension of Beth's Semantics of Physical Theories Bas C. van Fraassen Philosophy of Science, Vol. 37, No. 3. (Sep., 1970), pp. 325-339. Stable URL: http://links.jstor.org/sici?sici=0031-8248%28197009%2937%3A3%3C325%3AOTEOBS%3E2.0.CO%3B2-Y Philosophy of Science is currently published by The University of Chicago Press. Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/ucpress.html. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is an independent not-for-profit organization dedicated to and preserving a digital archive of scholarly journals. For more information regarding JSTOR, please contact [email protected]. http://www.jstor.org Tue Jun 19 06:30:05 2007

On the Extension of Beth's Semantics of Physical Theories

Embed Size (px)

Citation preview

On the Extension of Beth's Semantics of Physical Theories

Bas C. van Fraassen

Philosophy of Science, Vol. 37, No. 3. (Sep., 1970), pp. 325-339.

Stable URL:

http://links.jstor.org/sici?sici=0031-8248%28197009%2937%3A3%3C325%3AOTEOBS%3E2.0.CO%3B2-Y

Philosophy of Science is currently published by The University of Chicago Press.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available athttp://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtainedprior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content inthe JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained athttp://www.jstor.org/journals/ucpress.html.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.

JSTOR is an independent not-for-profit organization dedicated to and preserving a digital archive of scholarly journals. Formore information regarding JSTOR, please contact [email protected].

http://www.jstor.orgTue Jun 19 06:30:05 2007

Philosophy of Science September, 1970

ON THE EXTENSION OF BETH'S SEMANTICS OF PHYSICAL THEORIES*

BAS C. VAN FRAASSEN1

University of Toronto

A basic aim of E. Beth's work in philosophy of science was to explore the use of formal semantic methods in the analysis of physical theories. We hope to show that a general framework for Beth's semantic analysis is provided by the theory of semi- interpreted languages, introduced in a previous paper. After developing Beth's analysis of nonrelativistic physical theories in a more general form, we turn to the notion of the 'logic' of a physical theory. Here we prove a result concerning the conditions under which semantic entailment in such a theory is finitary. We argue, finally, that Beth's approach provides a characterization of physical theory which is more faithful to current practice in foundational research in the sciences than the familiar picture of a partly interpreted axiomatic theory.

1. Beth's program. In his work in philosophy of science, E. W. Beth's aim was to apply the methods of formal semantics (as developed by Tarski et al.) to the analysis of theories in the natural sciences. When he wrote his book Natuurphilosophie 131, he was still of the common opinion that the semantics of a physical theory is constituted by a set of 'correspondence rules' linking theoretical with observation language. Just prior to publication, however, he added the following note to his chapter on the logical structure of quantum mechanics:

On rereading the material, which was written several years ago, I lean to the opinion that it is possible to give a semantic construction to the logic of quantum mechanics with the help of Hilbert space. .. .The interpretation [in terms of experimental results] is therefore not analogous to semantics. ([3], p. 133)

Beth then developed this point of view, which is related to those of von Neumann, Birkhoff, Destouches, and Weyl, in three articles (121 [4] [5]).2 As we shall suggest, it is a point of view which has close affinities to much contemporary foundational work in physics.

Beth's opinion, in which I concur, is that his semantic approach represents a much more deep-going analysis of the structure of physical theories than the

* Received May, 1969. This study was supported in part by NSF grant GS-1566. I also wish to express my debt to

Dr. F. Suppe, University of Illinois, for stimulating discussion. His doctoral thesis [27] develops a point of view closely related to Beth's.

The new statement by Suppes of his approach ([28] [29]) shows important agreement with this point of view notwithstanding earlier apparent differences.

326 BAS C . VAN FRAASSEN

axiomatic and syntactical analysis which depicts such a theory as a symbolic calculus interpreted (partially) by a set of correspondence rules. However, Beth did not present his new analysis in a general form, but only through specific examples. My hope is that a general framework for his approach is provided by the theory of what I have called "semi-interpreted languages" ([31] [32]). Sections 3-6 will provide a general exposition, except for the theorem in section 6, without much attention to what is and what is not Beth's.

Sections 2 and 7 are concerned with more general issues in philosophy of science, the former to introduce our point of view, and the latter to show its relation to current questions and problems.

2. The language of science. The language of science, as Carnap notes, is "mainly a natural language . . . with only a few explicitly made conventions for some special words or symbols" ([6], p. 241). Of course, there are many technical terms, and much use of mathematical language. Yet we do not have here a case of an artificially constructed symbolic language, but a naturally grown "variant of the pre-scientific language, caused by special professional needs" (loc. cit.). And the technical terms are most often old nontechnical terms given a new role. Thus, at some point the part of our language which concerns wave motion in liquids was taken over, almost bodily, to provide a new, technical way of talking about sound. There is, of course, a reason why this part of the language (rather than, say, the well-developed, sophisticated way in which horse lovers talk about the modes of motion of horses) was adapted for this new role. This is a subject which has been much discussed in recent years, for it concerns the use of analogies and models in theory constr~ction.~

There are currently two general approaches to the formal study of language. One is syntactic and axiomatic, the other semantic in orientation. Within the first approach, a language or language game has as rational reconstruction a syntactic system plus an axiomatized deductive theory formulated within that syntax. In the second, the rational reconstruction consists in a syntactic system plus a family of interpretations of that syntax. In either case the construction may aptly be called an 'uninterpreted language': in neither case are the nonlogical terms provided with a specific interpretation, though in both cases the possible such interpretations are delimited to some extent (any interpretation must satisfy those axioms, any interpretation must fall within the described class). There are natural interrelations between the two approaches: an axiomatic theory may be characterized by the class of interpretations which satisfy it, and an interpretation may be characterized by the set of sentences which it satisfies; though in neither case is the characteriza- tion unique. These interrelations, and the interesting borderline techniques provided by Carnap's method of state-descriptions and Hintikka's method of model sets, would make implausible any claim of philosophical superiority for either approach. But the questions asked and methods used are different, and with respect to fruitful- ness and insight they may not be on a par in specific contexts or for special purposes. So we may reasonably hope to explore the semantic approach to the language of

Cf., e.g. [26], [27], 1291, Ch. 2.

science, and we shall begin with a brief account of our general position (developed in [31]).

Our view, to state it succinctly, is that in natural and scientific language, there are meaning relations among the terms which are not merely relations of extension. When a particular part of natural language is adapted for a technical role in the language of science, it is because its meaning structure is especially suitable for this role. And this meaning structure has a representation in terms of a model (always a mathematical structure, and most usually some mathematical pace).^ This language game then has a natural formal reconstruction as an artificial Ianguage the semantics of which is given with reference to this mathematical structure (called a "semi-interpreted language" in 1311; see section 3 below).

Before entering into the details of this kind of reconstruction, it may be helpful to discuss what is valid in such a language. First of all, of course, the admissible interpretations are such that those statements which are true in virtue of logic are true-if that familiar notion is applicable, i.e. if the language has among its expres- sions some which are meant to express the usual logical operations. Secondly, the meaning relations referred to above are such that certain logically contingent statements will always be true, in virtue of the meanings of the terms which occur in them. In other words, the mathematical structure with reference to which the language is partly interpreted plays a role in determining validity, and we may say in such a case that a statement is analytic or holds a priori in a broad sense.

There are many obvious and simple examples of this. In the case of simple discourse about color hues, the mathematical structure in question is the color spectrum, a segment of the real line. In the case of temperature it is the temperature scale being used, also a segment of the real number continuum. Thus

(1) Whatever is scarlet, is red.

holds a priori because the region of the colour spectrum assigned to the predicate "scarlet" is contained in the region of the spectrum assigned to "red," and

(2) Nothing is warmer than itself.

holds a priori because "is warmer than" is represented by the relation < (which is irreflexive) on the temperature scale.

That the language of science has such an inherent meaning structure-which may change in its historical development, however-has long been argued by Wilfrid Sellars ([24], [25], Chs. 4, 10, 1 I think that providing a formal representation such as we have attempted (intuitively here and more rigorously in [31]) helps to spell out this conception somewhat further. But on the other hand, I do not think this would be enough to make the view philosophically cogent; in addition, we must

The word "model" has many uses; the present sense is that found in discussions of the role of models in scientific theory, and differs from the sense in which it is used in formal semantics (and hence, below).

Hutten's point of view seems also in basic agreement with our own; compare "A model is a possible semantic interpretation; it is a picture of a situation which shows the semantic rules but does not state them explicitly. . . ." ([13], p. 120).

328 BAS C. VAN FRAASSEN

leave the abstract level of syntax and semantics, and provide a pragmatic counter- part to truth ex vi ternzinorum. In [31], [32] I have argued that the work of Sellars and Maxwell provides us with such a pragmatic retrenchment.

This general view concerning the structure of the language of science might perhaps be accompanied by quite different formal representations. Yet I think that the limitations of the axiomatic method are such that the semantic approach is the correct general approach here. For example, Carnap does not deny that principles like (I) and (2) hold; however, he feels that they can be made explicit in a set of "meaning postulates" to be laid down besides the axioms proper of the physical theory, from which they can be sharply distingui~hed.~ I must admit that I have not found myself equally capable of drawing such a sharp distinction between "meaning postulates" and "empirical postulates," beyond the rough and ready criterion that the meaning postulates are those not made explicit by the physicist. The divergence in approach may, however, be too fundamental to be accessible to simple direct arguments, or may not represent a disagreement but merely a difference in per- spective. We shall attempt to show the feasibility of our approach in a more detailed exploration before arguing its advantages.

3. Physical systems and physical theories. Like Beth we shall here address ourselves to the formal structure of notzrelativistic theories in physics, leaving the extension to the relativistic case for later. A physical theory then typically uses a mathematical model to represent the behavior of a certain kind of physical system. A physical system is conceived of as capable of a certain set of states, and these states are represented by elements of a certain mathematical space, the state-space. Specific examples are the use of Euclidean 2n-space as phase-space in classical mechanics and Hilbert space in quantum mecl~anics.~ To give the simplest example, a classical particle has, at each instant, a certain position q = (q,, q,, q,) and momentum p = (p,, pY, p,), SO its state-space can be taken to be Euclidean 6-space, whose points are the 6-tuples of real numbers (q,, q,, q,, p,, p,, p,).

Besides the state-space, the theory uses a certain set of measurable physical magizitudes to characterize the physical system. This yields the set of elementary statements about the system (of the theory) :each elementary statement U formulates a proposition to the effect that a certain such physical magnitude m has a certain value r at a certain time t. (Thus we write U = U(m, r, t), or U = U(m, r) if we abstract from variation with time, or U = U(t) if we wish to emphasize dependence on time.)

Whether or not U is true depends on the state of the system: in some states m has the value r and in some it does not. This relation betvieen states and the values of physical magnitudes may also be expressed as a relation between the state-space and the elementary statements. For each elementary statement U there is a region h(U) of the state-space H such that U is true if and only if the system's actual state is represented by an element of h(U). (We also say that these elements satisfy

See, for example, [6] ,Appendix B. Other terms for "state-space" are "phase-space" and "system space" (Weyl).

ON THE EXTENSION OF BETH'S SEMANTICS OF PHYSICAL THEORIES 329

U; thus in the case of the classical particle, (q,, q,, q,, p,, p,, p,) satisfies "The X-component of momentum is r" if p, = r).

The mapping h (the satisfaction function) is the third characteristic feature of the theory. It connects the state-space with the elementary statements, and hence, the mathematical model provided by the theory with empirical measurement results. This follows because, as we have said, the elementary statements concern measur- able physical magnitudes. We do not have in mind by this an operationalist identification of meaning. The exact relation between U(m, r, t) and the outcome of an actual experiment is the subject of an auxiliary theory of measurement, of which the notion of "correspondence rule" gives only the shallowest characteriza- tion. But this is not a subject into which we intend to go deeper here (cf. Suppes t281).

The elementary statements pertaining to a given kind of physical system con- stitute the set of well-formed formulas of a simple kind of formal language. This set of formulas, together with the state-space and the satisfaction function, con- stitutes what elsewhere we have called a semi-interpreted language. Before in- vestigating the formal structure of this simple variety of languages, however, we wish to discuss certain fundamental questions about the notion of state, and the physical laws describing the behavior of such a system.

4. The role of the time variable. We have said that elements of the state-space represent possible states of the system. This is not unambiguous, however, for sometimes "state" is used in such a way that a system (though undisturbed) has different states at different times, and sometimes such that a system remains in the same state unless it is subject to interaction. In the second case, the magnitudes m may change even though the system remains in the same state. In our representation this amounts to: in the former case the satisfaction function is not itself a function of time, but the 'location' of the system in state-space changes with time; in the latter case, the satisfaction function must be time dependent. It must also be noted that if the first alternative is embraced, one finds that the term "state" is then also used to designate the function $ such that +(t) is the state (in the first sense) of the system at t. So in that case the usage of "state" is not unequivocal, though it is not confusing.

These distinctions are easily illustrated with reference to quantum mechanics, since Schrodinger embraced the first alternative and Heisenberg the second. For this reason, the physicist speaks of the Schrodinger picture and the Heisenberg picture (cf. 1171, pp. 100-101; [9], pp. 35-36; [14], p. 44). In the case of quantum mechanics, the states of the system are represented by the elements (state-vectors) of a Hilbert space. For each measurable physical magnitude ("observable") m there is a linear operator M on this space, such that m has the value r if and only if Mx = rx, where x is the state-vector representing the state of the sy~tern .~ Thus we have

(3) If U = U(m, r) then h(U) = {x:Mx = rx)

8 If M x = rx, we call r an eigenvalue of M and x the corresponding eigenuector.

330 BAS C. VAN FRAASSEN

The difference between the two pictures appears when we make explicit the time dependence of the magnitude m, writing U = U(m, r, t). In the Schrodinger picture, the state-vector x is a function of the time. Writing x, for the value of x at t, we have

(4) Schrodinger picture: If U = U(m, r, t) then h(U) = {x,:Mx, = rxt).

In the Heisenberg picture not the state-vector but the operator representing the magnitude m is time dependent. Writing M, for the value of M at t, we have

( 5 ) Heisenberg picture: If U = U(m, r, t) then h(U) = {x:M,x = rx).

Finally we may note that the elements of the Schrodinger state-space are functions from the time (the whole real number continuum or a segment thereof) into an 'instantaneous state-space', which one may take to contain the set h(U) of equation (3). Each Schrodinger state describes an 'orbit' in this 'instantaneous state-space'. The opcrators in the Schradinger picture operate on the instantaneous states.

Formally speaking, one might of course identify the Heisenberg state-vectors with the Schrodinger state-vectors, and regard the two pictures as but two frames of reference in a single Hilbert space. (Heisenberg's matrix mechanics resolves the state-vector on axes of definite energy, wave mechanics on axes of definite position.) This is what allowed Dirac's, and von Neumann's, unification of the theory ([12], section B). It also shows at once that we have here but several variants of a single concept of state, and that we can use one or the other as convenience or fancy dictates.

5. The representation of physical laws. In the case of a nonrelativistic theory, the function of a law is to describe the behavior of the kind of physical system with which the theory deals: to describe the possible states of which it is capable, its normal evolution through time when undisturbed, and its behavior in interaction. We shall therefore proceed in accordance with the traditional threefold distinction between laws of coexistence, laws of succession, and laws of interaction. Our discussion will first address itself to nonstatistical laws (section 5.1) and then to statistical laws (section 5.2).

5.1 Nonstatistical laws. What has traditionally been called a law of coexistence is a condition limiting the class of physically possible states. The most familiar example is the Boyle-Charles ideal gas law whicli functionally relates the volume, pressure, and temperature of a gas, in such a way that, given any two of these magnitudes at a time t, the value of the third at t is uniquely determined. It has the form

where R is a constant, the "gas constant." What this means exactly is that at any given time the values of P, V, and Ta re related by equation (6). Hence if we use triples of real numbers (p, v, t) to represent the possible thermodynamic states of the gas, the law says that (p , v, t ) represents a physically possible state only if pc = Rt. Laws of coexistence select the physically possible subset of the state-space.

Of course, when the law of coexistence is general enough, it may be incorporated in the definition of the state-space. In our example, we may take the state-space of the ideal gas to be a region in Cartesian 3-space all of whose points satisfy (6)-and not the whole 3-space. There is nothing right or wrong about this procedure-and this is just to say that there is no objective distinction within physics between "meaning postulate" and "empirical postulates."

Still using the instantaneous state picture, we can say that similarly, laws of succession select the physically possible trajectories in the state-space. As an example, we can take a simple harmonic oscillator: a classical particle moving along a straight line subject to a force which is proportional to the distance from, and directed to, a fixed point 0 on that line. Using q and p to range over the values of its position on the line and its momentum along that line, its motion is described by equations of the form

At each time t , the particle has a state described by the values of q and p; that is, a state represented by a point in the q-p plane (its state-space or phase-space). In time, the particle describes an 'orbit' or 'trajectory' in this plane, described by the equation

1~ = - - q 2 +-p2 = constant 2n 2m

(Ebeing the mechanical energy). The law of succession is expressed by this equation (equivalently, by the previous equation) and the physically possible trajectories are the curves which satisfy this equation.1°

We consider this subject briefly from a more abstract point of view. Let x, y, . . . range over the state-space. We shall call the system deterministic if and only if there is for each non-negative real number t a unique transformation U, such that

(9) If the state of the system at t' is represented by x, then its state at t' + t is represented by U,(x).

We clearly have here a one-parameter semi-group of transformations with Uo as identity element and group-multiplication defined by

This is called the dynamical semi-group of the system ([16], 1 ; [34], 80-81; [20], p. 418). If in addition the system is time-reversible, each operator Ut has a unique inverse U,-l = U-,. In that case the set of transformations forms a group, the dynamical group. Thus a system subject to a diffusion process (say, heat-flow) has a dynamical semi-group governing the evolution of its state, which is not a group

Though this may be a histovicnl distinction; before a certain stage in the development of the theory, the law may not yet be an inherent principle of the language game.

lo Thus if we use "state" in the sense of a function 4such that $( t )is the state (in our present sense) at t , then this law of succession appears with the formal character of a law of coexistence. Speaking generally, laws describe conditions on what is physically possible.

332 BAS C. VAN FRAASSEN

(cf. [lo], pp. 233-234). But the simple harmonic system of our example has its motion governed by a group of transformations. We may state this condition equivalently in the form: there is only one physically possible trajectory through each point in the state-space.

When the laws of the system are given by differential equations, we are essentially given the injinitesimal generators of the (semi-)group ([I 11, 6.1). The elements U, of the dynamical (semi-)group can then usually be found through integration. For the harmonic oscillator of our example, the infinitesimal transformation U,,(q, y) = (q +dq, p +dp) is therefore given by

(In the second form, this is easily generalized to more complex systems (cf. [3], p. 94-95).)

Finally, we turn to the laws of interaction. Theoretically the most satisfying approach to interaction is no doubt the procedure of regarding the interacting systems as constituting a complex system. Thus an aggregate of n gravitationally interacting masses forms an n-body system subject to the same basic mechanical laws as individual bodies. Also, if two quantum mechanical systems have their (Schrodinger) states represented by 1 $,(T)) and 1 $,(T)) in Hilbert spaces H1 and H z , the complex system they constitute has a state-vector / $lz(~)) in the product space H1@ H2.

So in principle, the study of interaction of systems reduces to the study of syste~ns simpliciter. But this approach is often not feasible, especially if the interaction of the system studied is not with a like system but with a quite different kind of system. This leads to the typical problems of perturbation theory, scattering theory,ll and the theory of systems with input.

Specifically, the abstract theory of systems with input (and possibly output) is one of the newest subjects of mathematical research ([19], 8, 1). A system with input has a state-space H a n d a nonempty set of inputs D. These inputs change the state of the system in a manner described by the 'next-state function'

We shall write I(x) = y for S(x, I ) = y. If n inputs are supplied successively, the resultant sequence of inputs is itself an input, whose effect is defined by

(13) I 1 . . . I n ( x ) = x n + l

where x, = x, x,+, = I,(xj) for j = 1, 2, . . .,n. Thus if we accept also that the environment can leave the state unchanged, the inputs form a semi-group with identity :

The work by Lax and Phillips [15] develops scattering theory via the notion of the dynam- ical group {Ut}.

(14) D has a member I, such that I,(x) = x for all x in H; if I,, I, belong to D, SO does an element I,, such that I12(x) = 12(11(x)).

In special cases, the inputs of an abstract machine may be a group, namely, if the action of the environment can reverse its previous actions. The theory can also be extended to the effect of a continuous or compact input applied throughout a time interval; see for example the study of 'compact acts' in [I].

5.2 Staristical laws. The statistical analogue of a law of coexistence is a statement of a priori probabilities (also called, somewhat less misleading perhaps, "initial probabilities"). That is9 instead of dividing the (logically possible) states into the physically possible and the physically impossible, it assigns a probability to each state. An example of such a law is the Boltz~nann hypothesis that each microstate of a gas has equal probability (cf. the discussion of initial probability metrics in [7], Ch. XL or [22], section 10).

Formally speaking, this amounts to the specification of a probability measure on the state-space (in [22], Reichenbach used the term "probability metric"). If U = U(m, r) is an elementary statement, andp the probability measure in question, we would expect p to be defined for the set h(U),and the law of coexistence would have the general form:

(15) The probability that the physical magnitude m has the value r equals p(h(U)).

(One qualification is necessary here: "m has the value r" must here be understood as "the system is in an eigenstate of nz corresponding to value r," not as, e.g., "measurement of m would yield value r." This distinction is important only in the nonclassical case in which not all states are eigenstates of m.)

Similarly, laws of succession have as statistical analogue statemei~ts of transition probabilities. As example, let us consider a finite Markov chain. We have here a finite state-space, for which the law of succession is given by describing a transition probability matrix M = [Mtj], with 0 < Mi, < 1 and 2,Mi,= 1 (cf. [19]). The law then has the general form :

(16) Given that the system is in state xi at time t, the probability that it is in state xi at t + At equals Mi,.

Just as nonstatistical laws of coexistence can be used to redefine the state-space in such a way that then the law becomes analytic, its statistical analogue can be used to redefine the state-space; the problem of describing the behavior of the system is then essentially reduced to the nonstatistical case, as we shall see.

Suppose that the original state-space is H; for convenience suppose H to be finite again. Let the elements of H be indexed as {x,, x,, . . .) and let p be the probability measure on H. We can now switch our attention to the new state-space H" whose elements are the vectors

334 BAS C. VAN FRAASSEN

The new 'rule of correspondence' reads that a system is in state v if and only if the probability that it is in state xi equals pi(i = 1,2, .. .). If nothing special is known about the system, it is then in the state v such thatp, = p({xi}). But our information might be more definite, say enough to locate the system at a(0) = (1, 0, 0, . . .) at time t. Then the law of succession (16) uniquely determines its location at t + At, as the vector v(1) = (MI,, MI,, . ..). In other words, the law of succession now assigns a unique orbit in HPto the system for all times (counted in intervals of At) from t on: with respect to Hpthe system is deterministic. The law can now be written, using the notation v(m) = (pl(m), p,(m), . ..) to designate the state at t + mat.

(17) p,(m) = 2p,(m - I)Mij, or equivalently, a

Using the notation Mnto denote the nth power of the matrix M, this means that we have here a dynamical semi-group (U,), n ranging over the non-negative integers, defined by

This semi-group has U, = MO, the identity matrix, as identity element. We have only indicated informally what the satisfaction function comes to in

the case of H". Denoting it as hP, we have

(19) hP(U) is the set of vectors v = <pl,p2,...) in H" such that pi # 0 only if xi E h(U).

This condition amounts to: the probability that the system is in a state satisfying U equals 1.

The present case must not be thought to cover quantum theory, where we can indeed formulate a principle somewhat like (19), but where we do not have an 'underlying' state-space of totally determinate states.12 Finally, we have not yet considered statistical laws of interaction; but after the preceding, the extrapolation is perhaps obvious.

6. Semantic reIations among elementary statements. In general, the state-space has a certain mathematical (specifically, topological) structure, and the satisfaction sets

h(U) : U an elementary statement

are defined in terms of this. Thus the basic principle of von Neumann and Birk- hoff's 'quantum logic' is that every elementary statement about a quantum-mechanical system has as satisfaction set a subspace of the state-space.13 The

l2 If the eigenvectors of M are xj = I rf) and x = 2,qx j then the probability that upon measurement the observable m represented by M is found to have the value I.' equals cf*c, when the system is in state x. Concerning the dissimilarities to the present case, see for example 1211.

l3 A semantic analysis of the main systems of quantum logic was given in my paper [30].

(topological) structure of the state-space is then a major determining factor of the logic of the theory, in the sense which we shall now explain.

Suppose that for a given kind of physical system, the theory specifies a set E of elementary statements, state-space H, and satisfaction function h. We call the triple L = (E, H, h) a semi-interpreted language (cf. [31]). A model for L is a couple M = (loc, X), where X is a system of the kind in question14 and 'loc' is a function assigning a location in H to X. The formal semantics of L has as truth-definition :

(20) An elementary statement U is true in M = (loc, X) if and only if loc (X) E h(U).

The following are standard notions of formal semantics:

(21) U is a valid sentence (IFU) in L if and only if U is true in every model of L.

(22) A set X c E semantically entails U(X IF U) in L if and only if U is true in every model .M of L such that every member of Xis true in M.

The logic pertaining to a language is essentially a syntactic description of the set of valid sentences and the semantic entailment relation in that language.

A basic question concerning these notions is the question: is semantic entailment jinitary, in the sense :

(23) semantic entailment in L is jinitary if and only if X IF U in L only if Y I t - U in L for some finite subset Y of X; for all sets of sentences X and sentences U.

At first sight, it might seem very unlikely that semantic entailment would be finitary in an elementary language associated with a theory dealing with continuous physical quantities or with continuous temporal processes. For example, the information that in the interval a < t < b, the magnitude m has the value r = f(t) would seem to go considerably beyond any finite set of statements of the form U(m, r,, ti), a < t, < b. But, on the one hand, there may be a finite set of such conditions determining the function f relative to the laws of the theory, and on the other hand, given the value of m at t, the values of m at times earlier than t may be irrelevant to the system's behavior at later times.

One of Beth's contributions to formal semantics was to study its notions by means of topological methods. We shall here use a generalization of the method he developed to prove a theorem concerning finitary entailment.15 Two preliminary definitions are needed.

l4 The theory may mis-describe the systems with which it is concerned, but presumably it is concerned with some instantiated kind of system. For our present purposes however it would make no difference if we took X to be anything whatever.

l5 In the standard case in which L has negation such that -- U is true in M if and only if U is not true in M, this is equivalent to a model-theoretic compactness theorem. In quantum logic this kind of negation is not present. For another application and further references to Beth's work, see [33].

336 BAS C. VAN FRAASSEN

(24) Ajilter on a set S is a nonempty family of nonempty subsets of S closed under intersection and superset formation; an ultrajilter on S is a filter on S not properly contained in any other filter on S.

(25) An ultraproduct of V G H (in L = (E, H, h)) generated by an ultra-jilter F on V is an element u E H such that u E h(U) iff h(U) n VE F.

(26) Theorem. Semantic entailment is finitary in L = (E, H, h) if H admits all ultraproducts, that is, if for every subset V of H and ultrafilter Fon V, Hcontains an ultraproduct of V generated by F.

Proof. Suppose that H admits all ultraproducts, and that no finite subset Xm of X c E semantically entails U. We need now to show that X It- Udoes then not hold in L.

Let V be the set of points u in H such that U is not true in the model M = (loc, Y) for which loc (Y) = u. For each finite subset Xm of X define:

Fm= {u E V: u satisfies each member of Xm)

Then no set Fmis empty, for we know that no finite subset Xm of X semantically entails U(so there must be a point urn which satisfies all of Xm but not U). Secondly, the intersection of any finite family F,,, . . .,Fmnis a set F, defined as above, where X, is the finite subset of X which is the union of X,,, . . .,X,,. Therefore the family F = {F,) forms a 'filter-base' on V, and by a basic theorem concerning filters, is contained in an ultrafilter F on V.

Since H admits all ultraproducts, it contains a point u' such that

u' E h(U') if and only if V nh(U1)E Ffor all elementary statements U'.

This means that u' $ h(U), for V nh(U) is empty and hence cannot belong to F. But let U' E X. Then Xi = {U') is a finite subset of X, and

F,= {u E V: u satisfies U') = V nh(U1)

belongs to F. Hence u' satisfies all elementary statements in X. Taking M to be the model (loc, Y) such that loc (Y) = u', we see that every

member of X, but not U, is true in M. Hence X It U does not hold, as we meant to prove. Q. E. D.

We have so far disregarded the occurrence of the time variable in the elementary statements. This we can make explicit in various ways, corresponding to the ways in which variation with time was regarded in section 4 above. We shall here explore one particular way, which shows significant interconnections with the subjects of modal logic and the logic of chronologicalpropositions (cf. [23] [32]).

Let the state-space H in L = (E, H, h) represent instantaneous states, and read "loc (X) = u" to mean that the state of X at time t = 0 is u. Let the system have the dynamic semi-group {U,), t > 0. Then we clearly have (20) still holding as our truth-definition, in the presence of the further clause:

if we now introduce 0U and Ci U to mean that U(t) is true for some, respectively for all, values of t, L satisfies

(28) h ( 0 U)= (u E H: U,(u) E h(U) for some t)

(29) H(D U) = {u E W :U,(u) E h(U) for all t)

We have made a complete transition, formally speaking, to the logic of the modali- ties. These are natural modalities in the sense of [32], and we have there proved that if (U,) is a semi-group we obtain an &type system, and if it is a group, with t ranging over non-negative numbers as well, an &-type system. (If we view physical modality as not concerning possible transitions, but as determined by the laws of coexistence, we get an &-type modality of course.)

7. The comsept of physisal theory. The most familiar reconstruction of scientific theories depicts them as comprising (a) a symbolic language, (b) a set of postulates formulated in that language, and (c) a set of correspondence rules which (partially) interpret that language. The postulates are meant to comprise all the required mathematics, the 'meaning' postulates, the the empirical hypothesis peculiar to the theory. Supposedly, theoretical reasoning of the scientist can be represented by deductive arguments in this symbolic language. This reconstruction played a central role in the philosophical analysis of physical geometry, for example by Reichenbach and Carnap, and it appears to fit physical geometry very well. But this reconstruction seems to represent a very shallow analysis of the structure of physical theories im general. More important, even if this reconstruction fits all scientific theories in principle, it does not present anything like a faithful picture of actual foundational work in physics. I do not just mean that physicists formulate their theories in natural (though technical, mathematical) language, or that the idea of a correspondence rule makes light of the complexities of the mire and blood of measurement theory. Those familiar criticisms I think important, but more important yet seems the following: in the foundations of physics we do indeed find that theories are characterized by 'postulates,' but this signifies only a very superficial similarity to the traditional axiomatic development of geometry.

We may illustrate this with the familiar postulate set for quantum mechanics, found in one form or another in most texts (we quote from Mandl [17]):

Postulate I. A system with n degrees of freedom is completely specified by the normed state vector Gt(ql, ...,q,,). .. .

Postulate 11.To every observable corresponds a Hermitian operator. ... These 'postulates' resemble nothing so much as the description of a state-space together with a mapping correlating the state-space with the elementary statements about measurable physical magnitudes.

Beth's approach does not require or presuppose the complete formalization of the theory under analysis, He does not view physical theory as, ideally, a kind of Principia Mathematica cum nonlogical postulates. His approach takes into account the essential role of models in science. In Beth's account, the mathematics is not part of the physical theory, but is used to construct the theoretical framework.

338 BAS C. VAN FRAASSEN

The theoretical reasoning of the physicist is viewed as ordinary mathematical reasoning concerning this framework. All this seems to me to be much more nearly in accordance with actual foundational work in physics. Finally, Beth's approach makes possible the use of formal semantic concepts and methods. The earlier syntactic approach certainly led to much insight into the structure of scientific theories. But perhaps it is not fanciful to hope that the semantic approach, so fruitful in foundational work in logic and mathematics, will also prove fruitful in the philosophy of science.

REFERENCES

[I] Bednarek, A. R. and Wallace, A. D., "Finite Approximants of Compact Totally Dis- connected Machines," Mathematical Systems Theory, vol. 1, 1967, pp. 209-224.

121 Beth, E. W., "Analyse Sdmantique des Theories Physiques," Synthese, vol. 7, 1948149, pp. 206-207.

[3] Beth, E. W., Natuurphilosophie, Noorduyn, Gorinchem; 1948. 141 Beth, E. W., "Semantics of Physical Theories," Synthese, vol. 12, 1960, pp. 172-175. 151 Beth, E. W., "Towards an Up-to-date Philosophy of the Natural Sciences," Methodos,

vol. 1, 1949, pp. 178-185. [6] Carnap, R., Meaning and Necessity, 2nd edit., University of Chicago Press, Chicago,

1956. 171 d'Abro, A., The Rise of the New Physics, Dover, New York, 1951. [8] Ginsberg, S., "Some Remarks on Abstract Machines," Transactions of the American

Mathematical Society, vol. 96, 1960, pp. 400-444. [9] Green, H. S., Matrix Mechanics, Noordhoff, Groningen, 1965.

[lo] Griinbaum, A., Philosophical Problems of Space and Time, Knopf, New York, 1963. 1111 Hall, G. G., Applied Group Theory, American Elsevier, New York, 1967. 1121 Hanson, N. R., "Are Wave Mechanics and Matrix Mechanics Equivalent Theories?," in

Current Issues in the Philosophy of Science (eds. H. Feigl and G. Maxwell), Holt, Rinehart and Winston, New York, 1961, pp. 401-425.

1131 Hutten, E. H., "On Semantics and Physics," Proceedings of the Aristotelian Society, vol. 49, 1948149, pp. 115-132.

1141 Kaempfi'er, F. A., Concepts in Quairtunz Mechanics, Academic Press, New York, 1965. [IS] Lax, P. D. and Phillips, R. S., Scattering Theory, Academic Press, New York, 1967. [16] Mackey, G. W., The Mathematical Fou17dations 0.f Quantum Mechanics, Benjamin, New

York, 1963. [17] Mandl, F., Quantum Mechalzics, Academic Press, New York, 1957. 1181 Moore, E. F., "Gedanken-Experiments on Sequential Machines," Automata Studies,

Annals of Mathematics Studies, no. 34, pp. 129-153. [19] Morse, P. M., "Markow Processes," in Notes on Operations Research 1959, M.I.T. Press,

Cambridge (Mass.), 1959, pp. 84-105. 1201 Nemyckii, V. V., Topological Problerrzs of the Theory of Dynamical Systems, American

Mathematical Society Translation no. 103 (1954); reprinted in American Mathematical Society Tra~zslatio~zs, Series 1, vol. 5, 1962.

[21] Putnam, H., "A Philosopher Looks at Quantum Mechanics," in Beyond the Edge of Certainty (ed. R. G. Colodny), Prentice-Hall, Englewood Clifi's, 1965, pp. 75-101.

1221 Reichenbach, H., The Direction of Time, University of California Press, Berkeley, 1956. 1231 Rescher, N., "On the Logic of Chronological Propositions," Mind, vol. 75 (N.S.), 1966,

pp. 75-96. 1241 Sellars, W., "Counterfactuals, Dispositions, and the Causal Modalities," Minnesota

Studies in the Philosophy of Science, vol. 11, 1957, pp. 225-308. [25] Sellars, W., Science, Perception and Reality, Humanities Press, New York, 1963. [26] Spector, M., "Models and Theories," British Journal for the Philosophy of Science, vol. 16,

1965, pp. 121-142. 1271 Suppe, F., Meaning and Use of Models in Mathematics and the Exact Sciences, Disserta-

tion, University of Michigan, 1967.

[28] Suppes, P., "What is a Scientific Theory ?," in Philosophy of Science Today, (ed. S. Morgen- besser), Basic Books, New York, 1967, pp. 55-67.

[29] Suppes, P., Set-Theoretical Structures in Science, forthcoming (mimeographed, Stanford University, 1967).

[30]van Fraassen, B. C., "The Labyrinth of Quantum Logic," presented at the Philosophy of Science Association First Biennial Meeting, Pittsburgh, October 1968.

[31] van Eraassen, B. C., "Meaning Relations among Predicates," Nous, vol. 1, 1967, pp. 161-179.

[32] van Fraassen, B. C., "Meaning Relations and Modalities" Nous, Vol. 3, 1969,pp. 155-167. [33] van Fraassen, B. C., "A Topological Proof of the Lowenheim-Skolem, Compactness, and

Strong Completeness Theorems for Free Logic," Zeitschrift fur math. Logik und Grund- lagen der Math., vol. 14, 1968, pp. 245-254.

[34] IVeyl, H., The Theory of Groups and Quantum Mechatzics (tr. 8.P. Robertson), Methuen, London. 1931.

You have printed the following article:

On the Extension of Beth's Semantics of Physical TheoriesBas C. van FraassenPhilosophy of Science, Vol. 37, No. 3. (Sep., 1970), pp. 325-339.Stable URL:

http://links.jstor.org/sici?sici=0031-8248%28197009%2937%3A3%3C325%3AOTEOBS%3E2.0.CO%3B2-Y

This article references the following linked citations. If you are trying to access articles from anoff-campus location, you may be required to first logon via your library web site to access JSTOR. Pleasevisit your library's website or contact a librarian to learn about options for remote access to JSTOR.

[Footnotes]

3 Models and TheoriesMarshall SpectorThe British Journal for the Philosophy of Science, Vol. 16, No. 62. (Aug., 1965), pp. 121-142.Stable URL:

http://links.jstor.org/sici?sici=0007-0882%28196508%2916%3A62%3C121%3AMAT%3E2.0.CO%3B2-1

References

23 On the Logic of Chronological PropositionsNicholas RescherMind, New Series, Vol. 75, No. 297. (Jan., 1966), pp. 75-96.Stable URL:

http://links.jstor.org/sici?sici=0026-4423%28196601%292%3A75%3A297%3C75%3AOTLOCP%3E2.0.CO%3B2-L

26 Models and TheoriesMarshall SpectorThe British Journal for the Philosophy of Science, Vol. 16, No. 62. (Aug., 1965), pp. 121-142.Stable URL:

http://links.jstor.org/sici?sici=0007-0882%28196508%2916%3A62%3C121%3AMAT%3E2.0.CO%3B2-1

http://www.jstor.org

LINKED CITATIONS- Page 1 of 2 -

NOTE: The reference numbering from the original has been maintained in this citation list.

31 Meaning Relations among PredicatesBas C. van FraassenNoûs, Vol. 1, No. 2. (May, 1967), pp. 161-179.Stable URL:

http://links.jstor.org/sici?sici=0029-4624%28196705%291%3A2%3C161%3AMRAP%3E2.0.CO%3B2-N

32 Meaning Relations and ModalitiesBas C. van FraassenNoûs, Vol. 3, No. 2. (May, 1969), pp. 155-167.Stable URL:

http://links.jstor.org/sici?sici=0029-4624%28196905%293%3A2%3C155%3AMRAM%3E2.0.CO%3B2-R

http://www.jstor.org

LINKED CITATIONS- Page 2 of 2 -

NOTE: The reference numbering from the original has been maintained in this citation list.