Introduction Fifty years after the publication of Bell's theorem, there remains some controversy regarding what the theorem is telling us about quantum mechanics, and what the experimental violations of Bell inequalities are telling us about the world. This chapter represents my best attempt to be clear about what I think the lessons are. In brief: There is some sort of nonlocality inherent in any quantum theory, and, moreover, in any theory that reproduces, even approximately, the quantum probabilities for the outcomes of experiments. But not all forms of nonlocality are the same; there is a distinction to be made between action at a distance and other forms of nonlocality, and I will argue that the nonlocality needed to violate the Bell inequalities need not involve action at a distance. Furthermore, the distinction between forms of nonlocality makes a difference when it comes to compatibility with relativistic causal structure. The Bell locality condition is a condition which, if satisfied, renders possible a completely local account of the correlations between outcomes of spatially separated experiments. Bell's theorem tells us that the probabilities that we compute from quantum mechanics do not admit of such an account. Modulo auxiliary assumptions that, in my opinion, should be noncontroversial, this yields the conclusion that there must be something nonlocal about any theory that yields the quantum mechanical probabilities or anything close to them. One way to violate the Bell locality condition – and this is the way that any deterministic theory must do it – is via straightforward action at a distance. However, we should also consider the possibility of a stochastic theorywith probabilistic laws that involve irreducible correlations. The sort of nonlocality inherent in such a theory involves relations between events that are significantly different from causal relations as usually conceived. Perhaps surprisingly, such a theory, unlike its deterministic cousins, need not invoke a distinguished relation of distant simultaneity. This makes a difference for the situation of the two sorts of theories vis-a-vis relativity. Though a deterministic theory can achieve the appearance of being relativistic, at the level of observable phenomena, it can do so only by introducing a preferred foliation and with it a causal structure at odds with relativistic causal structures. Things are otherwise with dynamical collapse theories.
Bell's theorem is a conditional statement about theoretical models specified by a complete physical state
λ. It shows that models satisfying measurement independence and a stochastic Bell-locality (factorizability) condition obey Bell inequalities. Bell experiments, however, do not directly observe
λ or the
λ-conditioned probabilities appearing in that condition. They record analyzer settings, detector outcomes, and experimental metadata, from which they estimate operational joint distributions, marginals, correlations, and Bell statistics. We clarify the logical relationship between these two levels. The
λ-conditioned factorization identity cannot be directly evaluated at fixed
λ from the operational record alone. Nevertheless, by marginalizing over
λ under measurement independence, the model class implies observable inequalities that can be empirically violated. A Bell-inequality violation therefore excludes the corresponding Bell-local model class under the assumptions required for the inequality, but it does not by itself identify which latent assumption---measurement independence, parameter independence, or outcome independence---failed, nor does it uniquely determine an ontological interpretation. To illustrate this distinction concretely, we employ the Unified Quantum Mechanics formulation, in which the physical state is real, and the complex wavefunction is representational. We show that all four formulations of the dynamics---Schr\"odinger, Hilbert-transform, complex second-order, and real second-order---require the same two real initial functions
u0=u(0) and
u1=∂tu(0); the complex initial state is merely a repackaging of them, not additional physical information. We formulate the distinction as a three-level hierarchy: operational data, model-class exclusion, and ontological interpretation. The hierarchy separates experimental loopholes, which concern assumptions linking an actual experiment to an ideal Bell model, from the conceptual underdetermination of latent-variable conditional probabilities, which remains even in a loophole-free experiment. The result clarifies what Bell experiments establish without claiming that Bell-local models are empirically untestable. A Monte Carlo simulation illustrates the distinction between operational Bell-test results and latent-state factorization.
Bell's theorem is 50 years old. Still there is a controversy about its implications. Much of it has its roots in confusion regarding the premises from which the theorem can be derived. Some claim that a derivation of Bell's inequalities requires just locality assumption, and nothing more. Violations of the inequalities are then interpreted as ``nonlocality'' or ``quantum nonlocality''. We show that such claims are unfounded and that every derivation of Bell's inequalities requires a premise -- in addition to locality and freedom of choice -- which is either assumed tacitly, or unconsciously, or is embedded in a single compound condition (like Bell's ``local causality''). The premise is equivalent to the assumption of existence of additional variables which do not appear in the quantum formalism (in form of determinism, or joint probability for outcomes of all conceivable measurements, or ``additional causes`'', or ``hidden variables'', ``complete description of the state'' or counterfactual definiteness, etc.). A certain irony is that perhaps the main message of violation of Bell's inequalities is that our notion of locality should be based on an operationally well-defined no-signalling condition, rather than on local causality.
It is shown that the following statements about a quantum correlation experiment are mutually equivalent. (1) There is a deterministic hidden-variables model for the experiment. (2) There is a factorizable, stochastic model. (3) There is one joint distribution for all observables of the experiment, returning the experimental probabilities. (4) There are well-defined, compatible joint distributions for all pairs and triples of commuting and noncommuting observables. (5) The Bell inequalities hold.
Review and preview In chapters 6 and 7, I described some links between locality and the treatment of probabilities in models of the EPR–Bohm experiment. From chapter 6, the main lesson was that, given the strict correlations of quantum mechanics, factorizability of the two-time probabilities is equivalent to two-time determinism. I also noted there that an adequate factorizable theory – one that avoids Bell's theorem – is possible only if λ-independence fails. In chapter 7, I argued that weak symmetry, local determination, and independent evolution together entail deterministic results and weak deterministic transitions. In chapter 8, I discussed the relationships among the locality conditions of chapter 6 and 7, Lorentz-invariance, and ‘metaphysical’ locality, especially local causality. I have by no means tried to give general answers to the questions raised there. Indeed, part of my thesis is that there is no general answer to be had. Nonetheless, one can use the results of chapters 6 and 7 to investigate the status of various interpretations. Most obviously, they can be used to evaluate whether a given theory is local: if a theory can be shown to violate one or more of the types of determinism entailed by the locality conditions of chapters 6 and 7, then that theory must be non-local in some sense.
Relativistic theory and experiments teach us that causal influences are constrained by the speed of light. Deterministic theories incorporate this lesson in
functional terms: whether an event occurs or not is functionally determined by events that occur in its immediate causal past. In this paper we suggest a natural generalization of this constraint that applies to probabilistic theories: the
probability that an event occurs is to be functionally determined by the
probability of the event-occurrences in its immediate causal past. This constraint differs from Bell's criterion of local causality and turns out to imply it only under an additional assumption that requires history to be, in a sense,
irrelevant. Bell-inequality-violating quantum phenomena are thus arguably
not in tension with relativity, locality, "statistical independence", nor with a supposed "realism". What they instead show is that history is, in a sense,
relevant.