Quantum Logic, Conditional Probability, and Interference

by Michael Friedman

and Hilary Putnam **

Claim: The quantum logical interpretation of quantum mechanics gives us an explanation of interference that the Copenhagen interpretation cannot supply. I . Interference & Conditional Probability. With each physical quantity A of a given system, and each Bore1 subset A of the real line, quantum theory associates a projection operator EA(A) in a given Hilbert space. These projection operators form an algebraic structure (E = 1-E, E A F = EF, E V F = E F-EF) which is a partial Boolean algebra: the algebra is partial because E A F and E V F are defined (within the domain of projection operators) if E and F are compatible (EF = FE). The significance of both the association and the algebraic structure of projection operators is a matter of dispute between different interpretations of quantum theory. We will return to disputes over interpretation below; but for now let us note that for each state 9,quantum theory generates a (generalized) probability measure on the partial Boolean algebra of projection operators: Prob, (E) = (cp, Eq) - and this in turn induces a probability measure on the real line for each quantity A: Prob,A(A) = (cp, EA(A)rp). Observe that our probability measures on the partial Boolean algebra of projection operators are defined on algebraic combinations of pr6jection operators only if the operators in question are compatible, so if E and F do not commute we have no conditional probability for F given E in the usual sense - where Prob,, (F, E) = Prob, ( F A E)/Prob, (E). Nevertheless, for any two projection operators E and F quantum theory does give us a kind of non-standard conditional or transition probability for F given E : namely

* University of Pennsylvania

** Harvard University
Dialectica

Vol. 32, No 3 4 (1978)

306

Michael Friedman and Hilary Putnam

where ( P ~ is the (normalized) projection of cp onto E. This non-standard conditional probability reduces to the standard one when E and F are compatible. It is important to see (as Bub first pointed out) that (1) is responsible for the phenomenon of interference. Consider a simple 3-dimensional system (e. g., the spin-1 system described in Kochen & Specker):

The phenomenon of interference is just this: if E is not compatible with

(2) Prob, (E, a2) Prob, (a& Prob, (E, aJ Prob, (a3) but is instead proportional to (2) plus some added interference terms. Physically if we measure a2 V as - by localizing the quantity corresponding to the ortho-normal basis {al, a2, a3} to the set of corresponding eigenvalues {a2, as} - and then measure E, the distribution we get is given by Probv(n?va,,) (E). The problem of understanding the phenomenon of interference therefore reduces to the problem of understanding the non-standard conditional probability (1).
2 . The Copenhagen Interpretation. The basic idea of the Copenhagen interpretation is that quantities have well-defined values only when the system under consideration is related to an appropriate measuring apparatus or experimental arrangement: intuitively, when the system is related to a device that is capable of detecting the value in question. Moreover, the value of a quantity is well-defined only up to the precision of the relevant experimental arrangement; if the quantity is localized to A it

Quantum Logic, Conditional Probability, and Interference

3 07

doesnt necessarily follow that for some r A, the quantity has value r. A projection operator EA(A) represents the proposition A is localized to A, not The value of A is in A. Finally, it is held that incompatible projection operators cannot be simultaneously measured: no experimental arrangement can pick out two incompatible propositions. Thus, each experimental arrangement picks out a Boolean subalgebra of the partial Boolean algebra of projection operators. Relative to a particular Boolean subalgebra B (a particular experimental arrangement) every proposition corresponding to an element of B is either true or false (relative to B). We represent such a distribution of truth values by a homomorphism hB: B --f Z, . We express the relativity to experimental arrangement (measurement-dependence) characteristic of the Copenhagen interpretation by the assertion that if E E B, B then, in general

(3) hB(E) # hB(E) if B # B. (Note that the negation of (3) runs afoul of the Kochen & Specker result.) So, for example, if al(=PI) in our 3-dimensional system is an element of the 0.n. basis {ol, pz, p9} as well as the 0.n. basis { a l , az, as}

then h,(ai) (al)# h,(Pj) (al) - where B(ai) is the Boolean subalgebra generated by {a,, a 2 ,a3}, etc. For each Boolean subalgebra B, h, is a homomorphism. So i f E, F are elements of B then h, (E V F) = h, (E) V h, (F) and E V F represents the disjunction of the propositions corresponding to E, F. On the other hand, it may happen that E V F is in B but neither E nor F is in B. In this case hB is undefined on E, F and E V F does not represent a disjunction. Thus, in general, E*(A U r) is not the disjunction of EA(A) and E*(r) - only

308

Michael Friedman and Hilary Putnam

when EA(A) and EA(r) are themselves well-defined. Note also that (3) holds in general even if B and B are fully compatible. Even if B c B we cant have h,(E) = h,,(E) for all E in B. For, e. g., let B be generated by a2 v a3 and suppose that h,l(an V as) = h, (aeV as). Then we also have h p (a. v a:])= h, { a i }(alV aA).It follows that h, {ai} (a,)= h, {pi} (a,), contrary to Kochen & Specker. Consequently, not only are values unstable under a change from one experimental arrangement to another incompatible experimental arrangement, they are also unstable under an extension of a non-maximal experimental arrangement to a maximal one.

{ai}

On the Copenhagen interpretation the probability distributions generated by quantum theory are seen as distributions over measurement results i. e., they are of the form Prob, h,(E*(A)) = 1): the probability that A is localized to A relative to B. (Of course, if E t B, B then Prob, (h,(E) = l j Prob, (hnl (E) = l).)Similarly, the conditional probability (1) is also seen as relativized to measurement results - i. e., it has the form Prob, (h,(F) = 1, h,, (E) = 1)where F E B and E t B. In general, F and E are incompatible and so, therefore, are B and B. But even if F and E are compatible, B and B can still differ; and, if they do, our conditional probability cannot be interpreted in the standard way (since E A F might not be in B). Only if B = B is our conditional probability classic. This point will prove important shortly. Can the Copenhagen interpretation provide any understanding of the formula (l)? Here it is essential to distinguish two cases: (a) the case where E is maximal (l-dimensional), (b) the case where E is non-maximal. Let us start with case (a), and let E be a2 in our example, say. Suppose we start with state a and then make a measurement of the quantity corresponding to the 0.n. basis {ai} - finding eigenvalue a,. What is the probability that a measurement of F will yield F? By a basic principle of the Copenhagen interpretation, the so-called eigenvalue-eigenstate link:
(4) If A is localized to A then the state cp satisfies E(A) cp = cp, we can conclude that the state after the a,-measurement is [a normalized vector in] a 2 . Therefore, after the a,-measurement the probability of F is given by Probll.(F)- and this is just what (1) requires. Note that, from the point of view of the Copenhagen interpretation, there is nothing at all mysterious about the transition from cp to a 2 . This transition simply reflects the fact that the state, just like the values of quantities, has to be relativized to an appropriate experimental arrangement (and this latter follows from the eigenvalue-eigenstate link ( 4 ) ) . The transition cp + an results from the transition B w + Bcai).

Quantum Logic, Conditional Probability, and Interference

3 09

It is case (b) that raises serious problems for the Copenhagen interpretation. Moreover, (b) is the important and interesting case: it is responsible for the phenomenon of interference and, unlike (a), it involves genuine dependence on the initial state cp. Suppose we again start with state cp and make a non-maximal measurement of our quantity - localizing it to the set of eigenvalues {a, , a3} . Finally, we measure F. What has to be explained is the fact that Prob,, (h,(F) = 1, h,, (ar V a,) = 1) = Prob, (ar v n l ) (h,(F) = l), where F E B and B is generated by a2V a3. (4) tells us that after the {a,, a,}-measurement the state is somewhere in a2 V a 3 , but it does not imply that the state is the particular vector (P~,,,~,,,) . As far as (4) is concerned the state could be anywhere in a2 V a3 ; it could even be the mixture

This latter, of course, would result in no interference, and only the state cp (ul results in the correct interference terms. In other words, unlike in case (a), the transition cp -+ c p (,,2vu,) does not follow from the transition Bm + , B( u ~ V .
(1,)

Thus, to get the correct results in case (b) the Copenhagen interpretation needs to invoke an additional principle, the so-called projection postulate:
( 5 ) A measurement localizing to A in state cp results in the state A E A ( d , . (5) is obviously much stronger than (4). It also appears to be ad hoc and objectionable in a way that (4) is not. (4) allows us to derive the transition cp --f a? from more basic and general principIes of the Copenhagen interpretation; (5) simply imposes the desired transition cp + by brute force. (4), or something like it, is needed on independent grounds in order that state-preparation be possible at all. Moreover, since (4), unlike (5), does not contain the term measurement, it can be part of alternative realistic interpretations of the theory as well. In particular, it is part of the quantum logical interpretation. (Of course, on such realistic interpretations, a system has more than one state at any given time!)

Now one might think that we could derive (5) from more basic principles after all. E*(A) and [the 1-dimensional subspace spanned by] cpEA(A) are compatible, so between them we already have a well-defined conditional probability in the classical sense. Further, it is easy to see that Prob, (cp
E~(A)

, EA(A))= 1 = Prob, (EA(A),cp ~ ~ ( d , ) .

3 10

Michael Friedman and Hilary Putnam

Therefore, if we measure EA(A) we can conclude that we will find cpE*(d) with probability 1. But there is one and only one state that gives us this result: namely, (pEA(d) itself. Consequently, the state after the E*(A)-measurement is cpEA(,,, . Q. E. D. This reasoning is fallacious (from the point of view of the Copenhagen interpretation itself), as can be seen from the points about different but compatible Boolean subalgebras noted above. To return to our example, it is true that if we make a maximal measurement including both cp(u,vu,) and a2 V a3 we can infer that they are correlated with probability 1. After such a measurement we are able to conclude that the probability of F is given by Prob, ( a l V (F). This is just to repeat case (a); but it doesnt help us in the least with case (b)! It by no means follows that if we make a non-maximal measurement of apV a3 - the case under consideration - we will get the same result. Let B be the (non-maximal) algebra generated by a2 V a 3 , B* an algebra containing both (p(u2vuy) and a2V a3, and B an algebra containing F. What the above argument shows is that Prob, (hB*((p(u2vu~)) = 1, hB. (a2 V a,)= 1) = 1, and so Prob,o (hB (F) = 1, h,. (az V a3))= Prob(a: v aJ) (hB (F) = 1). What we want, however, is Prob, (hB(F) = 1, h,, (a. Va3) = 1) = Prob, (a?V a y ) (hB(F) = l), and this does not follow from the above fact about the algebra B*. Indeed, h,, uJ) ) is not even defined. In other words, we have to distinguish (from the point of view of the Copenhagen interpretation itself) conditional probability within a single Boolean subalgebra from conditional probability beetwen different Boolean subalgebras. From the fact that Prob,o (hB*(9 (a,Va3) ) = 1, h,, (a2V a3) = 1 ) = 1 it does not even follow that Prob, (hB. (cp ) = 1, hgt (a. V as) = 1) = 1. We conclude that the Copenhagen interpretation is quite incapable of accounting for case (b) without some a d hoc additional principle like (5).
3. The Quantum Logical Interpretation. The quantum logical interpretation is realistic in that it assumes, contrary to the Copenhagen interpretation that quantities always have well-defined values. Such measurementindependent values are perfectly precise; so if a quantity is localized to A it does follow that for some r t z A, the quantity has value r. A projection operator E*(A) represents the proposition The value of A is in A - and this is true no matter which experimental arrangement (if any) is under consideration. There is no relativity to experimental arrangement and no need, therefore, to use the term measuring apparatus in formulating the

Quantum Logic, Conditional Probability, and Interference

311

basic laws of the theory. Moreover, whether or not incompatible projection operators can be simultaneously measured, they do have simultaneous values Incompatible propositions can be true simultaneously. The algebraic structure of projection operators has a very different significance than on the Copenhagen interpretation. On the Copenhagen interpretation the partial Boolean algebra of projection operators is split up into various Boolean subalgebras, and common elements of different Boolean subalgebras are not identified (3). On the quantum logical interpretation there is no such splitting up into Boolean subalgebras: common elements of different Boolean subalgebras are identified (measurementindependence). Instead, we view the algebraic structure of projection operators as a direct copy of the truth-value relations between propositions about the system. What does this mean? (i) An atomic proposition of the form A E A correspond to the projection operator EA(A).(ii) Propositional combinations of compatible propositions correspond to partial Boolean algebra combinations of the corresponding component projection operators. (iii) Propositional combinations of incompatible propositions are undefined. (iv) If E,, is the projection operator corresponding to P and E, then

(6) r P -+ Q 1 holds for all systems of the kind in question if E, E, In other words, the algebraic structure of projection operators supplies us with a set of (non-statistical) laws about the system. This interpretation compares with the Copenhagen interpretation as follows: (i) Atomic propositions are treated similarly, except that they always have the stronger form The value of A is in A. (ii) Algebraic combinations of compatible projection operators always correspond to propositional combinations. E*(A U r) is always the disjunction of E*(A) and E A (r)- since E A (A) and E A (r) are always themselves well-defined. (iii) Incompatible propositions can be true simultaneously, but their propositional combinations simply dont exist (states-of-affairs are not closed under propositional combination)! (iv) Most laws of the form (6) do not hold on the Copenhagen interpretation. E. g., on the quantum logical interpretation the truth of r a , or a37 follows from the truth of b, ; but, on the Copenhagen interpretation, if b, is true then the disjunction of a, and a3 is not well-defined. Consequently, the quantum logical interpretation says more than the Copenhagen interpretation. It immediately follows from (i) - (iv) that quantum mechanical propositions obey a non-classical logic. First, there are classical contradictions that are always true for some quantum mechanical systems! Following section 7 of Kochen & Specker, consider the proposition

<

312 (7)
aE A

Michael Fnedrnan and Hilary Putnam

(a,

where V denotes exclusive disjunction (a V b if (a or b) and not-(a and b)) and the conjunction is taken over the (finite) set A of 0.n. triples in Kochen & Speckers algebra D. (7) is a classical contradiction - since there is no homomorphism into Z, that assigns it 1 - but it corresponds to the element 1 in the partial Boolean algebra of projection aperators. (Thus, we interpret the Kochen & Specker proof as a proof that logic is non-classical. Note that (7) asserts that all 0.n. triples in D have precise values. This assertion is true when interpreted quantum logically, but inconsistent when interpreted classically. Similarly, although there is no homomorphism h: D ---f Z2, there is a map h: D - t Z, such that h(a V b) = 1 if h(a) = 1 or h(b) = 1, etc. But h(b)-h(a)h(b), etc. we cant conclude from this that h(a V b) = h(a) This only follows if or is classical.) Second, there are classical inferences that fail because the conclusion is not defined even though the premises are true. The chief example of this phenomenon is the distributive law: (ar or as) and b21 can be true, but r(a, and bJ or (as and b 2 ) l is not defined. What does all this mean for us - as rational believers in a quantum logical world? First, of course, we will refrain from asserting propositional combinations of incompatible propositions. Second, our degrees of belief or subjective probabilities will respect the algebraic structure of proposiE, then Pr(P) Pr(Q). tions: if E, Third, not only will we refrain from asserting propositional combinations of incompatible propositions, we will not simultaneously assert individual incompatible propositions. To see this, idealize assertion as the assignment of probability 1, and recall that Gleasons theorem states that all (generalized) probability functions on the partial Boolean algebra of projection operators are generated by quantum mechanical (pure or mixed) states. It follows that incompatible propositions cannot simultaneously have probability 1. (Actually, there are exceptions to this involving partially compatible propositions.) Thus the quantum logical interpretation agrees with the Copenhagen interpretation that incompatible propositions cannot be simultaneously asserted (with the exceptions noted above); but it gives a completely different explanation of this phenomenon. According to the Copenhagen interpretation, propositions can be meaningfully asserted only in the context of the appropriate experimental arrangement, and no experimental arrangement (it is alleged) can measure incompatible propositions simultaneously. According to the logical interpretation, experimental arrange-

<

<

Quantum Logic, Conditional Probability, and Interference

313

ments have nothing to do with the matter. The failure of simultaneous assertability follows from the assumed algebraic relations between propositions via Gleasons theorem. Note also that Gleasons theorem only holds if we identify common propositions in different Boolean subalgebras, contrary to (3). If we relativize propositions to Boolean subalgebras as the Copenhagen interpretation does, Gleasons theorem doesnt apply: there are many more probability functions (e. g., there are dispersion-free measures). Finally, although incompatible propositions cannot be simultaneously asserted on the quantum logical interpretation, there exist incompatible propositions that are simultaneously true - and we can assert that (although we have to formulate it carefully: e. g., as r 3 q and 3 pil, but not r 3 (ai and pJ1) ! Let us now observe that the algebra of propositions given by (6) is very rich indeed. Not only does it determine a co-assertability relation as above, it actually determines the entire probabilistic structure of quantum theory. Consider first our basic statistical laws of the form Prob,(E) = (cp, Ecp). Although we havent stressed this fact so far, these laws are also conditional probability statements: they ascribe a probability to E given state cp. (Thus, (1) is really a doubly conditional probability.) Furthermore, in accordance with the eigenvalue-eigenstate link (4)we interpret the. [ 1-dimensional subspace spanned by the] state cp as itself representing a proposition. Our basic laws give the probability of a (maximal or non-maximal) proposition E conditional on a maximal proposition cp. Suppose, then, that we are given the truth of cp. By Gleasons theorem there is one and only one probability measure on the algebra of propositions consistent with this information i. e., such that Pr(rp) = 1 - and this is of course Prob,(E) = (cp, Ecp). (Note that this is true in classical mechanics as well. The difference is that all such classical mechanical probability functions are trivial. This is because maximal propositions generate ultrafilters in the algebra of classical mechanical propositions.) What now of our doubly conditional probability (l)?Case (a) is just a repetition of the singly conditional case. Suppose we are given cp. Our probability function must therefore be Prob,(F). Suppose we are then given a?, say. There is, of course, no conjunction of cp and a2 ; nor can we simultaneously assert rp and a 2 . All we can do is adjust our probability function to the new information: i. e., we change from ProbJF) to Prob,,(F). But there is one and only one conditional probability Prob,(F, ) such that our new probability function results from the old one by conditionalization, and that is given by (1). Thus, once again we get the same results as the Copenhagen interpretation with a different explanation. On the Copenhagen

314

Michael Friedman and Hilary Putnam

interpretation the change from Prob,(F) to Prob,,(F) results from the physical transition from one experimental arrangement B, to a second experimental arrangement Blair. On the quantum logical interpretation the change from Prob,(F) to Prob,,(F) represents no physical change at all (we could simply change our minds!). Rather, it depends once again on the constraints on possible probability functions determined by the algebraic structure of propositions. But the really interesting case is (b). Can the quantum logical interpretation account for (b) without invoking an additional ad hoc postulate like (5)? Suppose we are given cp and then given a2 Va3. Since a2 V a 3 is nonmaximal we cant mimic the above reasoning: there are an infinite number of distinct probability functions such that Pr(a2 V a3) = 1. The right one, Prob, ( a , V a y ) (F), is just one among these. However, on the quantum logical interpretation we have some additional information: namely, that if cp is true then a? V a3 and cp(auva3) are equivalent! To see this, define equivalence between quantum mechanical propositions in the usual way (a b) if (a and b) or (not-a and not-b), and observe that

Since (P(,,~,:,) and apV a3 are equivalent on the assumption of cp, we can a3) ). But this latter is conclude that Prob,(F, a2 V a3) = Prob,(F, cp v (F), and this gives us our already defined by case (a) as Prob, ( a ~a3) desired case (b). Hence, with no additional principles beyond (6), the quantum logical interpretation provides us us with a genuine explanation of interference:

Quantum Logic, Conditional Probability, and Interference

315

given the initial state cp, the disjunction a2V a3 is equivalent to the superposition cp ( a r V a j ) . Moreover, there is no way in which the Copenhagen interpretation can use the equivalence (8) to provide a parallel explanation. It is true that if h,cp(cp) = 1 then hBcp(cp ( a , V u y ) -(a2 V as)) = 1 as well. But -(a2 V a3)]cannot be interpreted as the equivalence of the procp is undefined on cp (u2va,) and a2V a3, ~p (ar v a3) and a2 V a3. h, ~ositions
[cp

) h~cp(az V a d Similarly, when we move to the new experimental arrangement B(aLv uJ) CP ( a , v u J is still not defined and no equivalence with a2 V a3 can be asserted. By contrast, on the quantum logical interpretation all relevant propositions are always well-defined and the crucial equivalence is always true (on the assumption of cp). Conclusion: Relativization to Boolean subalgebras (experimental arrangements) results in a significant loss of information information that can only be recovered by an ad hoc postulate like (5).
so h,cp(cp
(a.v
al)
9

(a2 V a d ) # h, cp (cp

REFERENCE
S. Kochen and E. P. Specker, The Problem of Hidden Variables in Quantum Theory, in Journal of Mathematics and Mechanics 17 (1967), 59-67.

Dialectica

Vol. 32,No 3 4 (1978)