Belousek - Bell's Theorem
Belousek - Bell's Theorem
Belousek - Bell's Theorem
Mechanics
Darrin W. Belousek†
Department of Philosophy
†Support provided by the John J. Reilly Center for Science, Technology and Values and the
Zahm Research Travel Fund, University of Notre Dame. This research was conducted while a
Visiting Scholar in the History and Philosophy of Science Department, Cambridge University, during
the Lent term 1997. This paper is a shortened version of a chapter from my dissertation, “Ontological
Commitments and Theory Appraisal in the Interpretation of Quantum Mechanics”.
Bell's Theorem, Non-Separability and Space-Time Individuation in Quantum
Mechanics
in terms of 'separability' and 'locality' and then consider his claims that the violations of
in terms of 'non-separability' rather than 'non-locality' and that 'non-separability' implies the
1. Introduction
Don Howard has claimed that Bell's theorem and its (meta-)physical implications
can be fruitfully understood by way of the 'Separation Principle' found in Einstein's own
incompleteness argument (Howard 1985; cf. Einstein 1948). Howard gives an interpretive
distinct principles he calls 'separability' and 'locality' and shows that the Bell factorizability
condition is itself a consequence of these two principles (Howard 1989 and 1992). On this
basis he argues that the violations of the Bell inequalities by the statistical predictions of
quantum mechanics and the confirmation of those predictions in experiment can (and
and criticize his arguments for each of these claims and find his conclusions less than
compelling.
We begin with Howard's notions of 'separability' and 'locality'. Howard states the
'separability principle' "asserts that any two spatially separated systems possess their own
2
separate real states", while the 'locality principle' "asserts that all physical effects are
between systems separated by a space-like interval" (Howard 1985, 173). Regarding the
to this definition, the quantum state of a physical system would represent a definite
property of the system if and only if pλ (x | m) = 1 , that is, if and only if λ is an eigenstate
Now, beginning with Jarrett's analysis of the Bell factorizability condition into two
Howard aims to show that the factorizability condition is also equivalent to the
conjunction of what he calls the 'separability condition', which he claims follows from his
separability principle, and the 'locality condition', which he claims follows from his locality
principle (Howard 1989, 1992). Let A ( B ) be the respective outcome for an EPR-Bell
the joint state of the composite system. Then the Jarrett (1984) analysis of the
factorizability condition,
pλ12 ( A,B| a,b) = pλ1 (A|a)pλ2 (B|b) ,
S S S
(1)
yields the sets of following sets of conditions — outcome and parameter independence,
respectively:
3
pλS1 (A|a,b, B) = pλS1 (A|a,b)
pλS 2 (B| a,b, A) = pλS 2 (B| a,b)
(2)
and
pλS1 (A|a,b) = pλS1 (A| a)
.
pλS 2 (B| a,b) = pλS 2 (B| b)
(3)
following identifications:
pλ11 (A|a, b) = pλ112 (A|a,b)
S S
.
pλS 2 (B|a,b) = pλS2 (B| a,b)
2 12
(4)
Now, given these identifications (4), Howard's 'locality condition' is just the statement of
parameter independence (3) with λ 12 , λ 1 and λ 2 replacing λ where appropriate —
pλ11 (A|a, b) = pλ11 (A| a)
S S
.
pλS 2 (B|a,b) = pλS2 (B| b)
2 2
(5)
Howard claims that (5) follows from his locality principle, and this claim appears
unproblematic (but we will return to it below). So, his analysis requires further only that
(6)
where λ 12 represents the (complete) joint state of the system S12 composed of S1
and S2.
Howard claims also that this separability condition (6) follows from his separability
4
Howard's argument to show that (2) and (6) are equivalent runs as follows. First,
to show that (6) implies (2), we apply (6) and (4) successively to the following definition
of conditional probability:
pλ112 (A|a,b, B) ≡ pλ1212 ( A,B| a,b) pλ 122 (B|a, b)
S S S
Thus, (6) implies (2). To show that (2) implies (6), we again begin with the definition of
Thus, (2) implies (6). Therefore, the argument concludes, (2) and (6) are equivalent and,
Although the mathematics here is surely correct, one may question whether
Howard's argument constitutes strict proof that the outcome independence condition (2)
and the separability condition (6) are equivalent. Crucial to the argument is the
identifications (4), the assumption of which Howard himself does not justify. Now, the
argument establishes the equivalence of (2) and (6) only if (4) is independent of both (2)
and (6) and, moreover, only if (4) is as unquestionable here as the definition of conditional
probability. Regarding the first point, note that the very statement of the identifications
(4) implicitly assumes that there are already separate states λ 1 and λ 2 for the spatially
separated systems S1 and S2, respectively, and, hence, takes for granted (at least) clause (i)
of the separability principle which postulates the existence of such states. Thus, neither
the identifications (4) nor the locality condition (5), the statement of which assumes (4),
are fully independent of the separability principle. Howard does recognize this dependence
of the locality condition (5) upon clause (i) of the separability principle (Howard 1989,
227), but fails to see the same dependence regarding the identifications (4), the latter of
which being relevant here. The question now arises whether the identifications (4) is
independent of clause (ii) of the separability principle, which is crucial for the statement of
5
the separability condition (6). Stipulating that pλS11 (A|a, b) = pλS112 (A|a,b) and
pλS 22 (B|a,b) = pλS212 (B| a,b) assumes effectively that the marginal outcome probabilities for
S1 and S2 are independent of conditionalizing upon the states of S2 and S1, respectively.
outcome-probability measure, this amounts to the assumption that the respective states of
S1 and S2 are each independent of the states of the other system, which is just what the
separability condition (6) asserts. Thus, (4) is not independent of (6). So, while assuming
the identifications (4) when showing that the separability condition (6) implies the
outcome independence condition (2) remains unproblematic, making the same assumption
when showing the converse threatens circularity. It appears, then, that the most Howard
can prove here is that the separability condition (6) implies the outcome independence
condition (2), not that the two conditions are equivalent. However, that is in fact all that
is actually required to derive implications for the separability principle from violations of
the Bell inequality and, hence, the failure of the Bell factorizability condition (1). For if
one gives up the outcome independence condition (2) rather than the parameter
independence condition (3), as Howard argues we should, then one must still also give up
the separability condition (6), which would consequently imply that the separability
Laudisa (1995) has also called into question Howard's argument that the
also focuses on the identifications (4), but criticizes them on somewhat different grounds.
He claims that the identifications (4) are implausible, for they implicitly conflate two
distinct notions of 'state' — namely, the notions of 'value state' and 'dynamical state' —
which are generally distinguished within the 'modal' interpretation. While the value state
of a physical system is specified by which observables for the system have a definite value
at a given time and what those definite values are, the dynamical state is specified by how
the system evolves over time, where the prediction of measurement outcome probabilities
belongs primarily to the dynamical state. In the context of quantum mechanics, the
dynamical state just is the quantum state, while the value state is assigned on the basis of
6
some property rule; and in the orthodox interpretation, which Howard implicitly adopts, a
value state exists if and only if the quantum state is an eigenstate of some quantum-
mechanical observable (the Eigenstate-Eigenvalue Rule), in which case the value state is
effectively represented by the quantum state itself, but which is not the case in general.
Laudisa claims that, given such a distinction, what is required of a 'separable' state-
description is only that it assign spatially separated systems S1 and S2 separate (i.e.,
distinct) value states, not separate dynamical states (in the sense of the separability
condition). Thus, Howard's separability condition (6), which requires separate dynamical
states, demands too much, and, hence, the identification of dynamical states in (4) need
not hold for a state-description that is separable in terms of value states. And, in that case,
without the crucial identifications (4), the separability condition (6) is not equivalent to the
One could further develop this criticism as follows. The separability principle,
generally in terms of value, rather than dynamical, states (i.e., in terms of definite
properties rather than probabilities). But, in order to maintain the equivalence of the
separability condition (6) and the outcome independence condition (2), the separability
condition must be defined in terms of dynamical states. Thus, whereas separate states in
the separability principle would refer to the value states of the spatially separated systems
S1 and S2, the separability condition (6) would refer to the dynamical states λ 1 and λ 2 of
S1 and S2, respectively; and, because separate (i.e., distinct) value states does not
necessarily imply separate dynamical states (in the sense of the separability condition), the
inference from the separability principle to the separability condition would be rendered
non sequitur, for the latter could fail (with respect to dynamical states) while the former
The above criticism is valid given a general value/dynamical state distinction. Such
makes it clear, then, that Howard's claims that the separability principle implies the
separability condition and that the separability condition (6) is equivalent to the outcome
7
independence condition (2) are specific to an orthodox interpretation of 'state' and thus
need not hold generally. This point will be of importance when assessing the implications
Leaving aside such objections until the next section, we turn next to assessing
Howard's claim that we must (or, at least, should) give up the separability condition (4) in
the face of violations of the Bell inequality by the statistical predictions of quantum
mechanics. His claim rests upon the argument that violation of the locality condition (5) is
incompatible with both the special theory of relativity and reasonable methodological
turn.
Howard's argument that violation of the locality condition is incompatible with the
special theory of relativity begins with the claim only that the separability condition is
suspect regarding violation of the Bell inequality, for violation of the locality condition
violations of locality and the special theory of relativity. But, in the summary of the
Whereas before his claim was only that violations of the locality condition present a
violations of the former imply violations of the latter; he thus concludes that the only
possible way to explain the violation of the Bell inequality by the statistical predictions of
quantum mechanics in a way that is compatible with the special theory of relativity is to
Crucial to this sudden shift of modality in Howard's argument is the claim that
constraints'. Regarding what such locality constraints entail, Howard uses this phrase
8
throughout the argument to refer consistently to "constraints on superluminal signaling"
(Howard 1992, 306). Thus, the crucial claim here is just that violation of the locality
But this claim, as it stands, is incorrect; for parameter dependence by itself does not
(Shimony 1984). Therefore, Howard's claim notwithstanding, the locality condition does
stronger assumption than is needed to rule out super-luminal signaling (i.e., non-
controllability will do). So, his conclusion here is simply non sequitur.
This also opens up an objection to his claim that the locality condition follows from
the locality principle, the latter of which explicitly prohibits super-luminal signaling.
Because violations of the locality condition do not necessarily permit signaling at all and,
hence, do not necessarily imply failure of the locality principle, one could well argue that
the locality principle does not by itself imply the locality condition. Something more — a
Principle of Common Cause, say — is still needed. Combined with the objection
concerning the connection between the separability principle and separability condition
made in the previous section, this would undercut completely Howard's initial claim,
namely, that the Bell factorizability condition is a consequence his separability and locality
principles; for although the factorizability condition does follow from the locality and
separability conditions (5) and (6), respectively (given the identifications (4)), there
principle.
Again leaving this objection aside, the second argument that Howard marshals in
favor of giving up the separability condition rather than the locality condition is that
abandoning the locality principle (as a consequence of giving up the locality condition)
concerning the relation of separability and locality to the formulation and testing of
theories (cf. Einstein 1948). The separability principle is necessary to the formulating of
9
physical theory because, if separability were to fail so that we could not individuate
could refer unambiguously would be the entire universe itself. And the locality principle is
necessary to the testing of physical theory because, if locality were to fail so that we could
not screen off distant influences, we could not be sure of the reliability of measurement
results (Howard 1989, 245–6). Where I disagree with Howard concerning the
primarily in regard to which should be privileged in this respect when a choice between
them is forced.
Howard privileges the locality principle over the separability principle, and for this
which, though neither a priori certain nor uniquely determined by the phenomena, serve as
constructive theories, which articulate the models (or “build up a picture”) of physical
systems that (in Einstein's view) are necessary for understanding phenomena. Einstein
gives the law of the conservation of energy and the second law of thermodynamics as
examples of such empirical-regulative principles and statistical mechanics and the kinetic
might construct to explain the collective behavior of gases, such a model must obey the
law of conservation of energy and the second law of thermodynamics. Now, Einstein
himself considered the special theory of relativity to be a principle theory and thus took the
and, while Einstein himself never makes any characterization of his own ‘Separation
Principle’ in such categories, Howard places the separability principle in the constructive
10
While appeal to authority and proof by default is no argument, we will challenge this
claim.
First, it is not clear that Einstein himself did believe that a given constructive
For in his view, it is constructive theories, not principle theories, that comprise the “most
important class of theories” (Einstein 1919); and the primacy of constructive theories is
due to their essential role in fulfilling the aim of physical theory, namely, understanding the
phenomena. Second, even if Einstein would agree with Howard that a given constructive
construction; for it does not make any prescription for filling in the content a specific
physical model of a certain phenomenon, but rather asserts a generic constraint on all
phenomenon. In this respect separability plays a role in theory construction similar to that
of the principle of conservation of energy and the second law of thermodynamics. And,
principle, for it does not necessarily follow from what Einstein calls the 'special principle of
relativity'. The latter for Einstein consists precisely of the two original postulates —
namely, the invariance of the speed of light in all inertial frames and the covariance of the
laws of physics under all Lorentz transformations (Einstein 1919). As Maudlin (1994) has
which is what Howard's locality principle asserts. Thus, violation of the locality principle
(via, e.g., superluminal signaling) need not imply violation of the 'special principle of
relativity'. So, the supposed privileged status of the locality principle as an empirical-
regulative principle cannot be justified on the basis of its being a strict consequence of the
‘special principle of relativity’ and is thus open to question. Such considerations aside, I
think one has here two empirical-regulative principles and not, as Howard suggests, one
11
empirical-regulative and one constructive principle, in which case one must make a further
ask to begin with whether they are equally necessary to the formulation and testing of
physical theory. First, consider locality; is it, as both Einstein and Howard (1989, 246)
Assume that it is, that if locality failed and the world were in fact 'non-local', then any
physical theory would be untestable in principle. And suppose that Bohmian mechanics is,
in fact, true; because Bohmian mechanics is 'non-local' in the sense that it fails to satisfy
the locality condition, this would effectively be to assume that the world itself is non-local
in that sense. But, if the world were non-local, then Bohmian mechanics would not be
quantum mechanics in that it makes all the same statistical predictions as standard quantum
experimentally, they are so with respect to both standard quantum mechanics and
Bohmian mechanics alike. Thus, in this sense at least, Bohmian mechanics is just as
testable as standard quantum mechanics. But, then, if Bohmian mechanics is testable, then
the world itself cannot be non-local, for by hypothesis locality is a necessary condition of
the testability of any physical theory. And if the world itself is local, then Bohmian
mechanics must be false. So, if locality is a necessary condition of the possibility of testing
physical theories, then we have the following conclusion: if Bohmian mechanics is true,
then it is not testable; if it is testable, then it is not true. The claim, then, is effectively that
no non-local physical theory can be both true and testable in principle, and that seems
clearly too strong a constraint on physical theory construction, for it would rule out a
priori all viable (i.e., logically consistent, empirically adequate) non-local theories.
So, we must ask again whether locality is an absolutely necessary condition for the
evidently testable in practice, at least to the extent that it makes all the same statistical
12
suggests that what is minimally required for testing physical theory is only that physical
unambiguous as is ever possible or practically necessary. To make the claim that locality
testing physical theory in general, one must show that such effective 'localizability' is
impossible under any conditions for any physical theory that represents physical systems
via state-descriptions that fail to satisfy the locality condition; or to make the claim that a
must show that the specific way in which the state-descriptions of the theory fail to satisfy
those conditions implies that physical systems as represented by the theory cannot be
effectively 'localized' under any conditions. Neither Einstein nor Howard have shown that
such is the case in general or in particular. And I take the evident in-practice testability of
Bohmian mechanics to strongly suggest, if not imply, that the claim of the in-principle
Let us suppose, though, that Einstein and Howard are correct in claiming that
locality then distinguished from separability in this respect such that locality is privileged
over separability? Einstein himself did not think so. For in that regard he made a claim for
separability similar to that he made for locality: "Neither does one see [if separability were
to fail] how physical laws could be formulated and tested without such clean separation [of
spatially distant physical systems]" (Einstein 1948, 321, my translation). But Howard,
while acknowledging that "some scheme for individuating systems is necessary in order to
formulate and test scientific theories", goes on to diverge from Einstein's (supposed) view
that the separability principle is itself necessary. And the reason he gives for disagreeing
with Einstein's view here is not unlike the one we have just given for disagreeing with both
Einstein and Howard regarding the necessity of locality for testing physical theory:
Agreeing with [Einstein] on this last point would entail one's declaring the quantum
theory, which violates the separability principle, to be, in effect, a fundamentally
incoherent theory.... But this is a step I do not feel compelled to take. (Howard 1989,
246)
13
So, separability and locality do appear after all to be on equal footing with regard to the
Where, then, does this leave us regarding the relative methodological status of the
separability and locality principles? Privileging either principle over the other is ultimately
a matter of theory selection, one decided by normative criteria, not by logic or experiment.
That is, to privilege one principle over the other is effectively to choose one theory over
another, a theory which satisfies that principle but not the other. Einstein himself made a
“The advantages of the constructive theory are completeness, adaptability, and clearness,
those of the principle theory are logical perfection and security of the foundations.”
(Einstein 1919). This suggests that in his view — if strict adherence to both a well-
precludes the construction of a coherent and adequate model or ‘picture’ (and, hence,
and that would cohere with his view that all theoretical scientific concepts and principles
conventional and so always revisable and provisional (Einstein 1933 and 1936).
Of course, which choice is appropriate in any given context would depend upon
one’s relative evaluation of the respective virtues or advantages of the available options for
theory modification, considered with respect to both the empirical evidence and the aims
privilege locality over separability, as Howard claims we should, would be to choose, say,
standard quantum mechanics over Bohmian mechanics, which are equally well-supported
by the available empirical evidence; and the ultimate reason for doing so would lie not in
the individual merits of locality relative to separability, but rather in the appraisal of
normative criteria of theory selection. Indeed, Howard reveals himself on this point by
14
remarking that the potential fertility of giving up the separability principle is chief among
Finally, let us now suppose along with Howard that it is indeed the separability
condition (6), rather than the locality condition (5), that fails in the violations of the Bell
inequality by the statistical predictions of quantum mechanics and, hence, that it is the
separability principle, rather than the locality principle, that must be sacrificed. The
question at hand, then, is what implications follow for the separability principle as a
We begin by noting that the denial of the separability principle presents (at least) a
two-fold ambiguity. First is the non-separability of states — either there are no separate
(i.e., distinct) states for spatio-temporally separated systems or the joint state of spatio-
temporally separated systems is not completely determined by their separate states; second
condition for individuating systems. Now, regarding the implication of the violation of the
separability condition for the separability principle, it is clear that violation of the
definition of 'state'). Whether it implies further the non-separability of systems is not clear,
and Howard defers this question by way of a footnote, but never answers it directly. He
does, though, express his view that the non-separability of systems is also implied by the
separable systems and, in particular, by the following remark: "We confront here a radical
physical holism at odds with our classical intuitions about the individuation of systems and
states..." (Howard 1989, 228, emphasis added). In any case, whether that is his express
view or not, it is this question that is at stake here. For violation of the Bell factorizability
condition, assuming the locality condition is not violated, implies the impossibility of the
15
First, note that the non-separability of states, which is implied by the violation of
the separability condition, itself presents a further two-fold ambiguity; for one could deny
the separability of states in either of two senses: either there are no separate (i.e., distinct)
separated systems is not completely determined by their separate states. One could deny
the separability of states in the second sense without denying it in the first. In this case,
there would be separate or distinct states for each system, but such states would not
determine uniquely the joint state of the composite system they comprise when represented
example of a physical theory which denies separate states in the second sense but not in
the first. For, as mentioned above, the 'modal' interpretation introduces a general
distinction between the value and dynamical state of a system; by denying the 'only if'-
prepared in a joint quantum or dynamical state are always assigned distinct value states —
namely, definite values for observables defined in terms of the basis vectors of the bi-
orthogonal representation of the quantum or dynamical state (cf. Kochen 1985 and Dieks
1989) — but such distinct value states do not in general determine uniquely the joint
quantum or dynamical state (except in the case of a factorizable quantum state), thereby
violating clause (i) of the separability principle (as formulated by Howard in terms of
dynamical states). Thus, it is only the failure of the separability of states of spatially
separated systems in the second sense — namely, the failure of such separate or distinct
(value) states to determine uniquely the joint (dynamical) state of a composite system —
that follows necessarily from the violation of the separability condition. This leaves open
the possibility of the existence of separate or distinct states for spatially separated systems,
conditional probability measure and his interpretation of 'separate states' in the sense of
determining a joint probability measure as the product of the two separate measures, then
to have separate states would just be to have a joint product state; in that case, to deny the
16
separability of states would be to deny it in both senses at once, which would eliminate the
ambiguity (by effectively denying a general value/dynamical state distinction). But, one
surely need not accept Howard's orthodox view here; and if one does not, then the non-
separability of states is left open to interpretation regarding its significance for the space-
warranted only if the non-separability of states is denied in the first sense — namely, the
the second sense. For as long as separate or distinct states for spatially separated systems
exist, or at least are possible, then the space-time individuation of such systems in terms of
those states also remains possible. Therefore, because the violation of the separability
condition does not necessarily imply the non-existence of separate or distinct states for
spatially separated systems (i.e., non-separability of states in the first sense), the violation
of the separability condition by itself does not necessarily (i.e., independently of a given
interpretation) imply the non-separability of systems and, a fortiori, does not necessarily
Again, though, let us deny a general value/dynamical state distinction and, hence,
grant that separate or distinct states for spatially separated systems would not exist if the
separability condition were violated. Does such an assumption imply the non-separability
mechanical systems? The answer, I think, is still 'no'. For what has been assumed here is
at most the non-existence of separate or distinct quantum states for spatially separated
systems, which is surely the case for 'entangled' systems (e.g., pairs of spin–1/2 particles
prepared in a singlet state). To take the non-separability of states (in both senses) to imply
can be 'separable' only with respect to their joint quantum state. And such an assumption
17
state-description of such systems which includes the non-separable quantum state yet
independently of the quantum state) to each system in terms of which they could be
individuated. And one cannot appeal here to Bell's theorem for support; for violation of
the Bell inequality alone implies at most that any (hypothetical) supplemental-variable
state-description compatible with the statistical predictions of the quantum state cannot
reproduce factorizable probabilities, not that distinct supplemental-variable states for such
So, the conclusion that Bell's theorem implies, via the violation of the separability
requires, over and above granting methodological privilege to the locality principle over
the separability principle, that one assume further both that spatially separated systems
having a non-separable joint dynamical or quantum state fail to possess separate or distinct
value states for quantum-mechanical observables (or that such value states do not exist in
general) and that the quantum state-description is complete in principle. Thus, the
conclusion does not hold generally, but rather is peculiar to an orthodox interpretation.
From the point of view of the Bell factorizability condition under Howard's analysis, then,
for any viable interpretation that denies (at least) one of these three claims.
Bibliography
I. Born.
Dieks, D. (1989), “Quantum Mechanics Without the Projection Postulate and its
227–232.
pp. 270–276.
18
Einstein, A. (1936), “Physik und Realität”, The Journal of the Franklin Institute
Bell Experiments", in A. van der Merwe, et al. (eds.), Bell's Theorem and the Foundations
Jarrett, J.P. (1984), “On the Physical Significance of the Locality Conditions in the
Laudisa, F. (1995), "Einstein, Bell, and Nonseparable Realism", British Journal for
19