Gábor Hofer-Szabó Péter Vecsernyés

Quantum Theory
and Local Causality

123
Gábor Hofer-Szabó Péter Vecsernyés
Institute of Philosophy Theoretical Physics
Research Centre for the Humanities Wigner Research Centre for Physics
Budapest Budapest
Hungary Hungary

ISSN 2211-4548 ISSN 2211-4556 (electronic)

SpringerBriefs in Philosophy
ISBN 978-3-319-73932-8 ISBN 978-3-319-73933-5 (eBook)
[Link]
Library of Congress Control Number: 2017963282

© The Author(s) 2018

 Preface

This book summarizes the results of research the authors have pursued in the past
several years on the problem of implementing Bell’s notion of local causality in
algebraic quantum field theory and relating it to such fundamental concepts as the
Common Cause Principle, Bell’s inequalities, and the EPR scenario. These results
have been presented at various workshops and department seminars. We wish to
thank the audience and the members of the Budapest Research Group, the
Budapest-Kraków Research Group, the Center for Philosophy of Science at the
University of Pittsburgh, the Munich Center for Mathematical Philosophy, the
Nagoya Winter Workshop Series, the Sidney Edelstein Center at the Hebrew
University, the Sigma Club at the London School of Economics, and the Southern
California Philosophy of Physics Group for the valuable discussions from which
the present book greatly benefited.
The results contained in this book have been published by the authors in a
number of papers. The authors gratefully acknowledge permissions to reuse
copyrighted material. A substantial part of the main text and all figures are repro-
duced from these papers:
G. Hofer-Szabó and P. Vecsernyés, “Reichenbach’s Common Cause Principle in
AQFT with locally finite degrees of freedom,” Found. Phys., 42, 241–255 (2012)
with the permission of Springer.
G. Hofer-Szabó and P. Vecsernyés, “Noncommuting local common causes for
correlations violating the Clauser–Horne inequality,” J. Math. Phys., 53, 12230
(2012) with the permission of AIP.
G. Hofer-Szabó and P. Vecsernyés, “Noncommutative Common Cause
Principles in AQFT,” J. Math. Phys., 54, 042301 (2013) with the permission of AIP.
G. Hofer-Szabó and P. Vecsernyés, “Bell inequality and common causal explana-
tion in algebraic quantum field theory,” Stud. Hist. Phil. Mod. Phys., 44 (4), 404–416
(2013) with the permission of Elsevier.
G. Hofer-Szabó and P. Vecsernyés, “On the concept of Bell’s local causality in
local classical and quantum theory,” J. Math. Phys., 56, 032303 (2015) with the
permission of AIP.
 G. Hofer-Szabó, “Relating Bell’s local causality to the Causal Markov
Condition,” Found. Phys. 45 (9) 1110–1136 (2015) with the permission of Springer.
G. Hofer-Szabó and P. Vecsernyés, “A generalized definition of Bell’s local
causality,” Synthese 193(10), 3195–3207 (2016) with the permission of Springer.
The research that the book is based on was supported by the Hungarian
Scientific Research Fund, OTKA K-108384 and K-115593.

Budapest, Hungary Gábor Hofer-Szabó

Péter Vecsernyés
 Contents

1 Local Causality: A Historical Introduction . . . . . . . . . . . . . . . . . . . . 1

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 What Is a Local Physical Theory? . . . . . . . . . . . . . . . . . . . . . . . . . . 11
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3 Locality and Causality Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4 Bell’s Notion of Local Causality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5 The Common Cause Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6 Bell’s Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
7 The EPR Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
8 A Noncommutative Locally Causal Model for the EPR
Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
9 Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Chapter 1
Local Causality: A Historical Introduction

Abstract In this chapter we briefly overview the history of local causality starting
from the early ideas on the prohibition of the action at a distance and ending with
Bell’s formulation of local causality. We state the central message of the book and
outline the content of the subsequent chapters.

Keywords Local causality · Action at a distance · John Bell

Local causality is the idea that causal influences propagate in spacetime continuously
and with velocity smaller than the speed of light. In a 1988 interview John Stewart
Bell formulated this idea as follows:
It’s the idea that what you do has consequences only nearby, and that any consequences at a
distant place will be weaker and will arrive there only after the time permitted by the velocity
of light. Locality is the idea that consequences propagate continuously, that they don’t leap
over distances. (Mann and Crease 1988)

Bell’s formulation nicely discerns three components of local causality, namely the
continuous propagation of physical influences, the decrease of the influence with
distance, and the speed of light as the limit velocity of the propagation. The three
components are logically independent. The third component has been added to the
concept of local causality with the advent of Einstein’s special theory of relativity.
The second component is a kind of precondition of doing physics: it ensures the
individuation of distant physical objects. The first component, however, has a long
and intriguing prehistory in natural philosophy in the form of the prohibition of the
action at a distance (McMullin 1989, 2002; Hesse 2005).
The core intuition is simple: a body can act only where it is present. Virtually
every major natural philosopher from Aristotle to Descartes agreed on this principle.
In Aristotle every motion that is not a self-motion of a living being is brought about
by contact with another body. In the medieval Aristotelian tradition the principle
became so firm that even God was bound by it. Aquinas in his Summa Theologica
used the principle even to argue for God’s omnipresence:
2 1 Local Causality: A Historical Introduction

No action of an agent, however powerful it may be, acts at a distance, except through a
medium. But it belongs to the great power of God that he acts immediately in all things.
Hence nothing is distant from him (Aquinas, Summa Theologica, I, q. 8, a. 1, ad. 3)

In Cartesian physics the principle simply followed from Descartes’ equating mat-
ter with extension. It appeared so self-evident that Descartes did not even make it
explicit in his Principles of Philosophy.
However, there have always been phenomena around challenging the principle.
The magnet was the prime example. From antiquity to Gilbert’s De magnete magnetic
phenomena provided a constant example for an apparently unmediated action at a
distance. Another phenomenon that seemed to involve physical action across spatial
distances was the ebb and flow of the tides. Tides correlating with the position of
the Moon counted as a strong testimony to the palpable influence of remote celestial
bodies on terrestrial affairs.
Newton’s theory of gravitation provided a coherent mathematical picture for the
tides, the planets orbiting around the sun, and many other celestial phenomena but
gave no hint as to how exactly the attraction works. Newton’s critics were agreed
that the distant unmediated action of one body on another is impossible. In a letter
to Bentley, Newton seems to agree with his critics:
That gravity should be innate, inherent and essential to matter, so that one body may act upon
another at a distance through a vacuum, without the mediation of anything else by which
their action and force may be conveyed from one to another, is to me so great an absurdity
that I believe no man who has in philosophical matters a competent faculty of thinking can
ever fall into it. (Newton 1692/2004, 102)

Nonetheless, Newton was unable to fill in the gap between attracting bodies and
neither were his followers. Some, including Huygens and Leibniz, argued that the
notion of aether has to be called on to explain away action at a distance; others,
such as Clarke, Newton’s disciple, were looking for causal agencies different from
mechanical contact action; still others, such as Berkeley, rejected intermediates, both
mechanical and nonmechanical, and left distant correlations unexplained. Newton’s
mechanics dominated the thinking in natural sciences throughout the eighteenth
century and action at a distance remained an ineliminable doctrine of the theory. It
has been echoed even in Kant’s Metaphysical Foundations of Natural Science: “The
attraction that is essential to all matter is an unmediated action through empty space
of one portion of matter on another” (Kant 1786/2004, 50; Proposition 7).
The idea that “matter cannot act where it is not” returned to physics with Faraday’s
investigations on electrical and magnetic phenomena. His field-theoretical paradigm
provided an alternative for the action-at-a-distance paradigm of the later Newtonian
tradition still prevailing on the Continent. As Maxwell writes:
Faraday, in his mind’s eye, saw lines of force traversing all space where the mathematicians
saw centres of force attracting at a distance: Faraday saw a medium where they saw nothing
but distance: Faraday sought the seat of the phenomena in real actions going on in the medium,
they were satisfied that they had found it in a power of action at a distance impressed on the
electric fluids. (Maxwell 1881, p. x)
1 Local Causality: A Historical Introduction 3

Maxwell’s theory of electromagnetism provided a coherent mathematical treatment

and also physical evidence for the reality of the fields transmitting the electric and
magnetic attraction and repulsion. Gravitation, however, resisted any attempt to be
treated within the Faraday-Maxwell field-theoretic paradigm. For that one had to
wait until Einstein’s theory of relativity.
It was the special theory of relativity that introduced what we called above the
third requirement on local causality, namely that the speed of light puts a limit on
the velocity of the propagation of physical influences. If cause precedes its effect
and the time ordering between spatially related events is relative to the different
inertial frames connected by the Lorentz transformation, then causal influences need
to proceed within the light cone. As Einstein formulates: “According to the theory
of relativity, action at a distance with the velocity of light always takes the place of
instantaneous action at a distance or of action at a distance with an infinite velocity
of transmission” (Einstein 1916/2006, 47).
Thus, the special theory of relativity provided a firm grasp on this third aspect of
local causality that could be further deployed in the criticism of the nascent theo-
ries, such as quantum mechanics. But testing local causality in quantum mechanics
was not so obvious. The reason was threefold: the new theory was probabilistic,
operational, and nonrelativistic. As for the first, the probabilistic feature of quantum
mechanics not only questioned whether local causality applies but it has even raised
serious concerns about the applicability of the very concept of causality, as the early
writings of the founding fathers attest. Second, quantum mechanics was inherently
operational in the sense that its predictions referred to measurement outcomes rather
than some physical values actually possessed by the system. Hence, it was difficult to
implement the intuition behind local causality, namely that it is something physical
that propagates continuously in spacetime. The third problem arose from the fact
that quantum mechanics was not Lorentz covariant.
Still, the idea of local causality constantly served as an aid to the formulation
and understanding of quantum mechanics. The Einstein–Podolsky–Rosen (EPR)
argument was a paradigmatic example for such a clarifying role of the principle. In
a nutshell, the EPR argument demonstrated that quantum mechanics is incomplete
in the sense that there exist certain elements of reality which are missing from the
quantum mechanical description. The argument made use of only two premises:
certain perfect correlations predicted by quantum mechanics, and local causality in
the form that the elements of reality in question cannot be causally influenced by
remote measurement procedures.
This is the point where John Bell entered the scene. In his famous 1964 paper
revisiting the EPR argument Bell showed that the assumption of local elements of
reality is incompatible with the statistical predictions of quantum mechanics. His
argument consisted of two parts: first, he recapitulated the EPR argument showing
that the existence of the elements of reality follows from local causality (or locality, as
Bell sometimes put it) and perfect (strict) correlations; second, that these elements
of reality or deterministic hidden variables lead to certain inequalities violated in
quantum mechanics. Interestingly, Bell’s first critics often missed Bell’s point and
focused only on the second part of the argument (cf. Goldstein et al. 2011; Maudlin
4 1 Local Causality: A Historical Introduction

2014). Thus, Bell’s result has been misinterpreted as excluding only deterministic
hidden variable models for quantum mechanics.
These early discussions on Bell’s results had important consequences regarding
Bell’s further work. First, in his subsequent papers Bell has developed a general
no-go theorem for local hidden variables that no longer relies on perfect correlations.
Because perfect correlations are hard to test empirically, eliminating them from the
premises lent more generality to his results. Second, these misunderstandings urged
him to give a clear formulation of the assumption featured in his derivation of the
inequalities, namely the assumption of local causality.
Throughout his career Bell has returned to the idea of local causality many times,
providing a more and more refined formulation of the concept. To our knowledge,
he addressed the question in three papers: first in his 1975, second in his 1986,
and finally in his 1990 posthumous paper. In this latter paper, entitled “La nouvelle
cuisine,” Bell formulated local causality as follows.
A theory will be said to be locally causal if the probabilities attached to values of local
beables in a space-time region V A are unaltered by specification of values of local beables
in a space-like separated region VB , when what happens in the backward light cone of V A
is already sufficiently specified, for example by a full specification of local beables in a
space-time region VC . (Bell 1990/2004, pp. 239–240)

The figure Bell attached to his formulation of local causality is reproduced in

Fig. 1.1 with the original caption. The term “beable” is Bell’s innovation and is
contrasted with the term “observable” used in quantum theory. “The beables of the
theory are those entities in it which are, at least tentatively, to be taken seriously,
as corresponding to something real.” (Bell 1990/2004, p. 234). As Bell constantly
stressed, fixing the beables of a theory is of crucial significance because local causality
as a physical constraint is legitimate only if it is formulated in terms of beables, that
is, “candidates for the description of nature,” and not in terms of any constituents
whatsoever of the theory.
One can rephrase Bell’s local causality as follows. Local causality excludes causal
processes propagating faster than the speed of light but does not exclude correlations
between spatially separated events or “beables”. Such correlations, namely can be

VA VB

VC

Fig. 1.1 Full specification of what happens in VC makes events in VB irrelevant for predictions
about V A in a locally causal theory
1 Local Causality: A Historical Introduction 5

brought about by a common cause operating in the past of the events in question.
However, fixing the past of an event at an appropriate spacetime region in a detailed
enough manner, the state of this event in a locally causal theory will be fixed once
and for all, and no other spatially separated event can have an impact on it.
It is not hard to see that Bell’s formulation of local causality above significantly
differs from his intuitive ideas expressed in his 1988 interview and quoted at the
beginning of this chapter. Recall that the three components of local causality were
continuous propagation of physical influences, the decrease of the influence with
distance, and the speed of light as the limit velocity of propagation. Now, the third
component definitely forms part of Bell’s above formulation. The second component,
as said in the preceding, is a general precondition of individuating distant objects.
The first component, however, has been altered significantly. Continuous propagation
is the idea that causal influences are mediated by intermediate physical events. A
natural interpretation of this idea is to say that what happens in the distant past is
screened-off by recent events. In an indeterministic world screening-off is taken in
the probabilistic sense: in fixing the recent past the probability of present events will
be fixed independently of what is going on in the distant past. By this interpretation
continuous propagation turns into Markovity.
Interestingly enough, in Bell’s formulation it is not the distant past that is screened-
off by the “recent events” in the region VC : the screened-off region VB is not in the
causal past of VC . What is screened-off are events happening in spatially separated
regions. The central idea of Bell’s formulation of local causality is the insight that
in a locally causal theory a detailed enough specification of the recent causal past
of an event will fully determine the very event or at least the probability of the
event. And fully determining means that no other event can contribute to fixing this
probability—be it in a spatially separated region, as in Bell’s formulation, or in the
distant past, as in the requirement of continuous propagation.
Bell’s prime merit lies in his being able to translate the philosophical intuition
lying behind local causality into easily tractable mathematical terms applicable in the
probabilistic formalism of quantum mechanics. More precisely, he was able to write
down a mathematically precise formulation of a consequence of local causality in the
context of an experiment in which measurements are performed on two subsystems
that previously have interacted but are now spatially separated. His works have then
set the scene for a whole research program in the foundations of quantum theory.
Bell’s investigations have remained an isolated research project within theoreti-
cal physics for a long time. The mainstream was working on quantum field theory
(QFT), a theory designed to unify quantum theory with the locally causal charac-
ter of relativistic field theory. Interestingly enough, Bell had never formulated local
causality in a QFT context. One can just speculate on why he did not implement
local causality in QFT.
One reason might have been that QFT has deliberately been devised to resolve the
locality issues of standard quantum mechanics. Hence, most of the physicists were
convinced that QFT is relativistically local by its very construction. Another reason
might have been that standard QFT is formulated in terms of field operators which
do not satisfy microcausality (see below) and hence have even less chance to adopt a
6 1 Local Causality: A Historical Introduction

direct ontological interpretation than the local observable operators used in algebraic
quantum field theory which at least commute if spatially separated. But for Bell it
was of crucial importance to apply the notion of local causality to beables of a theory
“corresponding to something real” out there in the world. He was so unsatisfied with
the usual formulation of QFT that in 1984 he even came up with a suggestion on
how to base QFT on beables (Bell 1984/2004). In any case, Bell never formulated
local causality in a quantum field theoretical context.
So much on history. Now, let us go back to Bell’s formulation of local causality.
Looking at it mathematically, Bell’s local causality is a probabilistic independence
relation. It states that certain events become probabilistically independent from other
events conditioned upon yet other events. This conditional independence relation,
however, is motivated by the spatiotemporal localization of the events in question.
Thus, the logical structure of Bell’s local causality is a kind of inference pattern from
spatiotemporal to probabilistic relations: if events are localized in the spacetime in
a certain way, then they are to satisfy certain probabilistic independences. But in
order to obtain a valid inference relation between spatiotemporal and probabilistic
concepts, one needs to integrate these concepts into a unified framework. Without
such a framework, one cannot account for the inferences from relations between
spacetime regions to probabilistic independences between, say, random variables.
Thus the first aim of our book is to find such a framework.
The most elaborate formalism used in quantum physics offering a general method
to connect spatiotemporal and probabilistic entities is an algebraic-axiomatic form
of quantum field theory, namely algebraic quantum field theory (AQFT) or local
quantum physics (Haag 1992). Thus, AQFT is a mathematically transparent theory
ideal for analyzing various concepts related to local causality, such as Bell’s inequal-
ities (Summers and Werner 1987a, b, 1988; Halvorson 2007), relativistic causality
(Butterfield 1995, 2007; Earman 2014; Earman and Valente 2014), or the closely
related (see below) Common Cause Principle (Rédei 1997; Rédei and Summers
2002; Hofer-Szabó and Vecsernyés 2012a, 2013a). For our ends, however, the full
formalism of AQFT would be too much. Our intention is simply to provide a mini-
mal framework that is indispensable to formulate local causality in a strict fashion
both in the classical and quantum cases. We call such a framework a local physical
theory (LPT). A LPT is based on those axioms of algebraic field theory that describe
the structure of the observable algebra. Using a translation between commutative
von Neumann algebras and σ -algebras it also can describe classical theories. Thus,
the formal structure of a LPT integrates the two most important components of a
general physical theory: a spacetime structure and an algebraic-probabilistic struc-
ture. Having the formal framework in hand, we can then accomplish the main goals
of this book, namely to formulate Bell’s original definition of local causality in a
classical LPT, to extend it in a clear-cut way to quantum LPTs, and to relate it to
other such important concepts and principles as the Common Cause Principle, Bell’s
inequalities, the EPR scenario, and other relativistic causality and locality concepts.
The central message of the book, however, is the following somewhat heterodox
statement.
1 Local Causality: A Historical Introduction 7

Local causality in the EPR scenario can be saved by adopting noncommuting events
into our ontology. In other words, one can provide a generalized noncommutative
definition of Bell’s local causality such that it does not imply Bell’s inequalities and
allows for a locally causal model for the EPR.
By extending Bell’s notion of local causality to noncommutative structures we cross
a line that Bell set for himself. As mentioned above, Bell always stressed that the
principle of local causality should exclusively be applied to beables and not to any
arbitrary ingredient of the theory. In contrast to Bell’s intentions, in this book we
use the notion of “beable” in a very permissive way also incorporating elements
of noncommutative algebras. In this respect our move gets close to the so-called
quantum logical approach to quantum mechanics (Mackey 1963; Jauch 1968; Piron
1976; Beltrametti and van Fraassen 1981). Although people in this tradition do not
use the term “beable,” they do use the terms “event,” “conjunction,” “proposition” in
a similarly permissive way. Their ambition is not to give a full-fledged ontological
interpretation of, say, the meeting of two noncommuting events, but rather to clarify
many important logical and structural questions of the theory. Our strategy is similar.
We are not able to provide a philosophically satisfactory answer to the question as
to what a noncommutative ontology is. Rather, in this book we simply test how far
we can get in providing a locally causal account of the EPR scenario by adopting
noncommutative events.
Our book is structured as follows. In Chap. 2 we introduce the mathematical formal-
ism of a LPT. Briefly, a LPT is an association of local operator algebras to spacetime
regions regulated by three principles borrowed from AQFT: isotony, microcausal-
ity, and covariance. Depending on whether the local algebras are commutative or
noncommutative, we speak about local classical theories (LCTs) or local quantum
theories (LQTs). At the end of the chapter we spend some time to motivate why we
use von Neumann algebras in the framework of LPTs.
Chapter 3 is devoted to the various causality and locality concepts in AQFT such
as causal dynamics, local primitive causality, no-signaling, independence, and sto-
chastic Einstein locality. There is an intensive discussion in the literature as to which
of these concepts properly represents local causality in QFT. We agree with Rédei
(2014) when he claims that local causality is not a single property but an intricately
interconnected web of features. In this chapter we review the various concepts in the
web.
At the beginning of Chap. 4 we present in turn Bell’s three different formulations
of local causality presented subsequently in his 1975, 1986, and 1990 papers. Then,
we analyze the key concepts featured in his third, final formulation, namely “local
beables,” “shielder-off region,” and “complete specification.” We translate them into
the framework of LPTs and provide a generalized definition of local causality. In the
same chapter we prove Proposition 1 stating that local primitive causality makes an
atomic LPT be locally causal.
In Chap. 5 we turn to Reichenbach’s Common Cause Principle. The Common
Cause Principle states that if there is a correlation between two events and there is no
direct causal (or logical) connection between the correlated events then there exists a
common cause of the correlation. We generalize this principle to the LPT framework
8 1 Local Causality: A Historical Introduction

and point out the similarities and differences between the principle and Bell’s local
causality.
Chapter 6 collects the most important concepts and some of the representative
propositions concerning Bell’s inequalities in the general C ∗ -algebraic setting and
in the special LPT framework.
In Chap. 7 threads get intertwined: Bell’s local causality, the common cause prin-
ciple, and Bell’s inequalities all meet in the notorious EPR scenario. In this chapter
we introduce the EPR scenario and show (Proposition 2) that the violation of Bell’s
inequalities does not block the implementation of the EPR situation in a locally
causal LQT because in a locally causal LQT Bell’s inequalities cannot be derived.
In Chap. 8 we explicitly construct a locally causal LQT for the EPR. The model
is the 1 + 1 dimensional local quantum Ising model. In this chapter we show that the
model is not only locally causal in Bell’s sense but also able to implement four pairs
of events correlating in the singlet state.
In Chap. 9 we summarize the results of the book and investigate their philosophical
consequences. We argue that embracing noncommuting beables in our ontology
significantly extends our explanatory sources in accounting for correlations. We also
examine what price we need to pay for abandoning classicality in order to preserve
local causality.
Our book fits nicely into a recent research line on a deeper conceptual and formal
understanding of Bell’s notion of local causality. Travis Norsen’s illuminating paper
on local causality (Norsen 2011) or its relation to Jarrett’s completeness criterion
(Norsen 2009), the paper of Seevinck and Uffink (2011) aiming at providing a “sharp
and clean” formulation of local causality, or Henson’s (2013b) paper on the relation
between separability and Bell’s inequalities all attest to this renewed interest in local
causality. We comment on the points of contact with these papers. For a list of our
own papers on which this book is based see the Preface.

References

E.G. Beltrametti, B.C. van Fraassen (eds.), The Logic of Quantum Mechanics (Addison Wesley,
Reading Mass, 1981)
J.S. Bell, Beables for quantum field theory, (CERN-TH. 4035/84, 1984); (reprinted in Bell, 2004,
173–180)
J.S. Bell, La nouvelle cuisine, in Between Science and Technology, ed. by J. Sarlemijn, P. Kroes
(Elsevier, 1990) (reprinted in Bell, 2004, pp. 232–248)
J. Butterfield, Vacuum correlations and outcome independence in algebraic quantum field theory,
in Fundamental Problems in Quantum Theory, Annals of the New York Academy of Sciences,
Proceedings of a conference in honour of John Wheeler, ed. by D. Greenberger, A. Zeilinger
(1995) pp. 768–785
J. Butterfield, Stochastic Einstein locality revisited. Brit. J. Phil. Sci. 58, 805–867 (2007)
J. Earman, No superluminal propagation for classical relativistic quantum fields. Stud. Hist. Phil.
Mod. Phys. 48, 102–108 (2014)
J. Earman, G. Valente, Relativistic causality in algebraic quantum field theory. Int. Stud. Phil. Sci.
28(1), 1–48 (2014)
References 9

A. Einstein, in Relativity: The Special and the General Theory, 2016, (Penguin Classics, 2006)
S. Goldstein, T. Norsen, D.V. Tausk, N. Zanghi, Bell’s theorem. Scholarpedia 6(10), 8378 (2011)
R. Haag, Local Quantum Physics (Springer, Heidelberg, 1992)
H. Halvorson, Algebraic quantum field theory, in Philosophy of Physics, vol. I, ed. by J. Butterfield,
J. Earman (Elsevier, Amsterdam, 2007), pp. 731–922
J. Henson, Non-separability does not relieve the problem of Bell’s theorem. Found. Phys. 43, 1008–
1038 (2013b)
M.B. Hesse, Forces and Fields: The Concept of Action at a Distance in the History of Physics
(Dover Publications, Mineola, 2005)
G. Hofer-Szabó, P. Vecsernyés, Reichenbach’s Common Cause Principle in AQFT with locally
finite degrees of freedom. Found. Phys. 42, 241–255 (2012a)
G. Hofer-Szabó, P. Vecsernyés, Noncommutative Common Cause Principles in AQFT. J. Math.
Phys. 54, 042301 (2013a)
J.M. Jauch, Foundations of Quantum Mechanics (Reading Mass, Addison Wesley, 1968)
I. Kant, in Metaphysical Foundations of Natural Science, 1786, transl. M. Friedman (Cambridge
University Press, Cambridge, 2004)
G.W. Mackey, Mathematical Foundations of Quantum Mechanics (Benjamin, New York, 1963)
C. Mann, R. Crease, John Bell, particle physicist (Interview). Omni 10(8), 84–92 (1988)
T. Maudlin, What Bell did. J. Phys. A: Math. Theor. 47, 424010 (2014)
J.C. Maxwell, Treatise on Electricity and Magnetism, (Oxford, 1881)
E. McMullin, “The Explanation of Distant Action: Historical Notes,” in Philosophical Conse-
quences of Quantum Theory, ed. by J.T. Cushing, E. McMullin (University of Notre Dame Press,
Indiana, 1989)
E. McMullin, The origins of the field concept in physics. Phys. Perspect. 4, 13–39 (2002)
I. Newton, Isaac Newton: Philosophical Writings (Cambridge University Press, Cambridge, 2004)
T. Norsen, Local causality and Completeness: Bell vs. Jarrett. Found. Phys. 39, 273 (2009)
T. Norsen, J.S. Bell’s concept of local causality. Am. J. Phys 79, 12 (2011)
C. Piron, Foundations of Quantum Physics (Benjamin, Reading Mass, 1976)
M. Rédei, Reichenbach’s Common Cause Principle and quantum field theory. Found. Phys. 27,
1309–1321 (1997)
M. Rédei, A categorial approach to relativistic locality. Stud. Hist. Phil. Mod. Phys. 48, 137–146
(2014)
M. Rédei, J.S. Summers, Local primitive causality and the Common Cause Principle in quantum
field theory. Found. Phys. 32, 335–355 (2002)
M. P. Seevinck, J. Uffink, Not throwing our the baby with the bathwater: Bell’s condition of local
causality mathematically ‘sharp and clean’, in Explanation, Prediction, and Confirmation The
Philosophy of Science in a European Perspective, vol. 2, ed. by D. Dieks, W.J. Gonzalez,
S. Hartmann, Th. Uebel, M. Weber (2011) pp. 425–450
S.J. Summers, R. Werner, Bell’s inequalities and quantum field theory, II: Bell’s inequalities are
maximally violated in the vacuum. J. Math. Phys. 28, 2448–2456 (1987b)
S.J. Summers, R. Werner, Bell’s inequalities and quantum field theory, I: General setting. J. Math.
Phys. 28, 2440–2447 (1987a)
S.J. Summers, R. Werner, Maximal violation of Bell’s inequalities for algebras of observables in
tangent spacetime regions. Ann. Inst. Henri Poincaré Phys. Théor. 49, 215–243 (1988)
Chapter 2
What Is a Local Physical Theory?

Abstract In this Chapter we introduce the mathematical formalism of a local physi-

cal theory (LPT). Briefly, a LPT is an association of local operator algebras to space-
time regions subjected to the requirements of isotony, microcausality and covariance,
all borrowed from algebraic quantum field theory. Depending on whether the local
algebras are commutative or noncommutative, we call a LPT a local classical theory
(LCT) or a local quantum theory (LQT). At the end of the chapter we motivate the
application of von Neumann algebras in the LPT framework.

Keywords Local physical theory · von Neumann algebra

As stated in the Introduction, Bell’s notion of local causality presupposes a framework

that treats spatiotemporal and probabilistic entities in a common formalism. In this
chapter we develop such a framework, called local physical theory. Instead of jumping
directly to the full-fledged definition, we proceed here “inductively,” unfolding step
by step the various notions featuring in the definition of a local physical theory.
Having listed these features we formulate the exact definition only at the end of the
chapter.
The central idea of a local physical theory is the association of local operator
algebras to spacetime regions regulated by the physically motivated requirements
(Haag 1992):
1. Isotony. Let M be a globally hyperbolic spacetime1 and let K be a covering
collection2 of bounded, globally hyperbolic subspacetime regions of M such
that (K, ⊆) is a directed poset under inclusion ⊆. The net of local observables is
given by the isotone map K  V → A(V ) to unital C ∗ -algebras; that is, V1 ⊆ V2

1 By a spacetime we mean a connected time-oriented Lorentzian manifold. A spacetime M is called

globally hyperbolic if M contains a Cauchy hypersurface, which is by definition a subset S ⊂ M

such that each inextendible timelike curve in M meets S at exactly one point. (See Pfäffle 2009
and references therein.)
2 For all x ∈ M there exists V ∈ K such that x ∈ V .
12 2 What Is a Local Physical Theory?

implies that A(V1 ) is a unital C ∗ -subalgebra of A(V2 ). The quasilocal algebra A

is defined to be the inductive limit C ∗ -algebra of the net {A(V ), V ∈ K} of local
C ∗ -algebras.3
Sometimes additivity, which is a stronger property than isotony, is also required
for the net of observables: A(V1 ) ∨ A(V2 ) = A(V1 ∪ V2 ); V1 , V2 , V1 ∪ V2 ∈ K,
where ∨ refers to the generated C ∗ -algebra in A.
2. Microcausality (also called Einstein causality) is the requirement that local alge-
bras belonging to spacelike (i.e., causally) separated regions commute, that is,
A(V  ) ∩ A ⊇ A(V ), V ∈ K, where primes denote spacelike complement and
algebra commutant, respectively.
3. PK -covariance. Let PK be the subgroup of the group P of global isometries of
M leaving the collection K invariant. A group homomorphism α : PK → Aut A
is given such that the automorphisms αg , g ∈ PK of A act covariantly on the
observable net: αg (A(V )) = A(g · V ), V ∈ K.
Here the possible spacetimes spread from Minkowski spacetime through stationary
spacetimes to generic globally hyperbolic ones where no global Killing vector field
exists. Choosing the collection K in such a way that every V ∈ K contains only a
finite number of elements of K, one can consider theories with locally finite degrees of
freedom, that is, with finite-dimensional local algebras. Otherwise the local algebras
themselves are infinite-dimensional. We would like to treat classical and quantum
theories on an equal footing as far as possible. The difference between the two is
that the quasilocal algebra of a local classical theory is required to be commutative
whereas that of a local quantum theory is required to be noncommutative.
Thus, microcausality fulfills trivially in local classical theories. On the other hand,
in local quantum theories it is usually required that the quasilocal algebra is “highly
noncommutative” and the local algebras are “fat enough.” This is assured by algebraic
Haag duality, which is a stronger requirement than microcausality:
4.Q Algebraic Haag duality. A(V  ) ∩ A = A(V ), V ∈ K.
Clearly, Haag duality is inherently connected to the noncommutativity of the observ-
able algebra. In the case of commutative local algebras Haag duality would imply
that A(V ) = A for any V ∈ K; that is, the net structure of local algebras would
be completely lost. To avoid this trivial net structure in local classical theories, one
requires less than Haag duality:
4.C Intersection property for spacelike separated regions. The intersection property

A(V1 ) ∩ A(V2 ) = A(V1 ∩ V2 ); V1 , V2 , V1 ∩ V2 ∈ K (2.1)

holds for spacelike separated regions V1 , V2 ∈ K; that is, A(V1 ) ∩ A(V2 ) =

A(∅) := C 1A for them.

3 This formulation is a special case of the general category theoretical formulation of AQFTs in
curved backgrounds (Brunetti and Fredenhagen 2009). Namely, the functor from globally hyperbolic
spacetimes to unital C ∗ -algebras is restricted to the full subcategory induced by the object M and
the (sub)collection K of its subobjects.
2 What Is a Local Physical Theory? 13

In the case of local quantum theories this property follows from Haag duality and
primitive causality (see below) if the net is additive and the quasilocal algebra is a
factor; that is, its center is trivial: A ∩ A = C 1A .4 Note that the intersection property
(2.1) is not required for disjoint pairs V1 , V2 ∈ K if they are causally connected,
V1 ∩ (J+ (V2 ) ∪ J− (V2 )) = ∅ inasmuch as it would forbid subalgebra constituents
from the past to rearrange themselves in the future. Requiring (2.1) generally would
contradict primitive causality which, as we show, makes the dynamics deterministic.
Different physical realizations of a single local theory are given by unitary inequiv-
alent representations π : A → B(H) of the quasilocal C ∗ -algebra A by bounded
operators B(H) on a (separable) Hilbert space H. Inequivalent representations can
be produced from essentially different states φ : A → C through GNS construction.
Representations are required to be locally faithful not to lose local observables. Once a
particular representation is chosen, one can consider the natural von Neumann algebra
extension of the local algebras by taking weak closures N (V ) := π(A(V )) , V ∈ K.
5. Representation. A locally faithful representation π : A → B(H) is chosen where
a (strongly continuous) unitary representation U : PK → B(H) implements
α : PK → Aut A. The local and quasilocal observables are extended as N (V ) :=
π(A(V )) , V ∈ K and AH := ∪V ∈K N (V ) ⊂ B(H), respectively.
It is easy to see that the net {N (V ), V ∈ K} of local von Neumann algebras given
above also obeys isotony, microcausality, in the sense that π(A(V  )) ∩ B(H) ⊇
N (V ), V ∈ K, and PK -covariance. Because we concentrate on local and causal
properties we do not consider further requirements on the representation π , for
example, how a vacuum representation can be characterized and is chosen among
the allowed representations.5
Here we briefly comment on the use of von Neumann algebras as local algebras in
local classical theories. The crucial point is the link between von Neumann algebras
and σ -algebras. Every element S ⊂  of a σ -algebra (, ) determines a projection
χ S in the abelian ∗ -algebra F(, C) of complex functions on , namely, χ S is the
characteristic function of the subset S ∈ . In general, we translate local σ -algebras
(, ) to local commutative operator algebras generated by projections χ S , S ∈  in
the function algebra F(, C). This abundance of projections is, however, the reason
why the local operator algebras cannot be represented by generic commutative C ∗ -
algebras in a local classical theory. Namely, a commutative unital (nonunital) C ∗ -
algebra, according to the Gelfand duality, is isomorphic to the algebra of complex-
valued continuous functions (vanishing at infinity) on a (locally) compact Hausdorff

4 Let V1 , V2 ∈ K be spacelike separated regions. Due to Haag duality and additivity of the net

A(V1 ) ∩ A(V2 ) = A(V1 ) ∩ A(V2 ) = (A(V1 ) ∨ A(V2 )) = A(V1 ∪ V2 ) . (2.2)

Because V1 ∪ V2 always contains a Cauchy surface if V1 and V2 are spacelike separated bounded
spacetime regions, we arrive at A(V1 ∪ V2 ) = A due to primitive causality. Therefore A(V1 ) ∩
A(V2 ) = A(V1 ∪ V2 ) = A ∩ A =: Center A.
5 However, to stay within the quasiequivalence class of the representation π one considers only

states in the folium of π (Haag 1992), that is, normal states of π(A) which lead to locally normal
states, that is, normal states, by restricting them to the local von Neumann algebras N (V ), V ∈ K.
14 2 What Is a Local Physical Theory?

topological space. However, unless the topology is discrete, such algebras generally
do not contain nontrivial projections at all. Therefore one is to consider the subclass
of commutative von Neumann algebras in local classical theories as local operator
algebras that are not only rich enough in projections, but are also generated by them.
The paradigmatic case of a commutative von Neumann algebra is the space
of complex-valued essentially bounded measurable functions L ∞ (, , μ) on the
σ -finite measure space (, , μ). This Neumann algebra is generated by the sub-
class {χ S , S ∈ } of characteristic functions on , and acts on the separable Hilbert
space L 2 (, , μ) by multiplication. This subclass of characteristic functions, or
equivalently, the sets of their supports form the σ -algebra (, ) of classical events.
The lattice operations and the algebra operations relate to one another as follows:
χ S∧T = χ S χT , χ S∨T = χ S + χT − χ S χT . This σ -algebra, however, is not the most
general σ -algebra one can imagine, because not every σ -algebra can be equipped by
a σ -finite measure μ. Nevertheless, they give us a rich enough set of examples for
classical theories. The probability measure p on the corresponding σ -algebra (, )
can be provided by any normal state φ on the von Neumann algebra L ∞ (, , μ)
by pφ (S) := φ(χ S ), S ∈ .
It is a further question as to what kind of local σ -algebras can correspond
to local classical theories, for example, to classical field theories on the space-
time M with configuration space F M := { : M → F} and with field values
F = Rn , Cn , for example. The maximal σ -algebra of classical events one can imag-
ine is (F M , P(F M )) given by the power set P(F M ) of the set of field configurations.
One also needs narrower σ -algebras in tune with the net structure of the theory. This
is done by taking local equivalence classes of those configurations that have the
same field values on a given region V ∈ K. Two field configurations ,
∈ F M
are said to be locally V -equivalent, ∼V
, if |V =
|V . The isotone net struc-
ture {(F M , (V )), V ∈ K} of unital σ -subalgebras (V ) ⊂ P(F M ) can be given
by the “cylindrical subsets” of F M corresponding to the image sets of canonical
projections Z V : P(F M ) → P(F M ), V ∈ K, which map a set S of configurations
onto the corresponding union of V -equivalence classes of configurations in S:

P(F M )  S  → Z V (S) := { ∈ F M |∃
∈ S : |V =
|V } ∈ (V ) := Z V (P(F M )).
(2.3)

Clearly, the net {(F M , (V )), V ∈ K} — or {(V ), V ∈ K}, for short — is PK -
covariant. The hard and unsolved problem is to give a probability measure on the
σ -algebra (F M , P(F M )) or on a meaningful σ -subalgebra of it. We can avoid this
conundrum by choosing a locally finite covering of M, that is, choosing a subnet
Km ⊂ K in a way that every V ∈ Km contains only a finite number of elements
of Km , and restricting the field configurations to be piecewise constant on regions
corresponding to minimal elements in Km . The power set of this configuration space
m
F S , where S m denotes the set of minimal elements in Km , can also be mapped into
m
local σ -algebras (F S , m (V )), V ∈ Km as before in (2.3). Although the maximal
local σ -algebra m (V m ) of a minimal region V m ∈ S m is isomorphic to the power
set P(F) of field values, one can restrict them to the Borel σ -subalgebra of P(F).
2 What Is a Local Physical Theory? 15

Then a generic local σ -algebra m (V ), V ∈ Km is isomorphic to a finite product of

the copies of corresponding Borel σ -subalgebras, because V is covered by a finite
subset of S m . We can simplify the situation further by restricting the field values F
to a finite set.
Last but not least, we stress that the projections χ S , S ∈ (V ) in the local von
Neumann algebras do not possess a direct spacetime localization: they project to
subsets of F M and not to those of M.
Inspired by the above considerations, we define a local physical theory as
Definition 1 A local physical theory (LPT) is a net {N (V ), V ∈ K} of local von
Neumann algebras associated with a directed poset K of globally hyperbolic bounded
regions of a globally hyperbolic spacetime M. The net satisfies isotony, microcausal-
ity, PK -covariance, and the intersection property for spacelike separated regions. If
the local von Neumann algebras are commutative, we speak about a local classical
theory (LCT); if they are noncommutative, we speak about a local quantum theory
(LQT).
The framework of LPTs provides the natural context in which Bell’s notion of local
causality can be formulated.

References

R. Brunetti, K. Fredenhagen, Quantum field theory on curved backgrounds, in Quantum Field The-
ory on Curved Spacetimes, Concepts and Mathematical Foundations, Lecture Notes in Physics,
vol. 786, ed. by C. Bär, K. Fredenhagen (Springer, Berlin, 2009)
R. Haag, Local quantum physics (Springer, Heidelberg, 1992)
F. Pfäffle, Lorentzian manifolds, in Quantum Field Theory on Curved Spacetimes, Concepts and
Mathematical Foundations, Lecture Notes in Physics, vol. 786, ed. by C. Bär, K. Fredenhagen
(Springer, Berlin, 2009)
Chapter 3
Locality and Causality Principles

Abstract This Chapter provides a brief overview of the interconnections between

the various causality and locality concepts in algebraic quantum field theory such as
causal dynamics, primitive causality, local primitive causality, no-signaling, selective
and nonselective measurements, local determinism, stochastic Einstein locality.

Keywords Causal dynamics · Local primitive causality · No-signaling

Stochastic Einstein locality

With the rise of the field-theoretic paradigm in quantum physics an intensive debate
has begun as to exactly how quantum field theory implements the intuitive notion of
local causality (also referred to as relativistic causality or relativistic locality). Some
argued that local causality is secured in quantum field theory by the very construction
of the theory. Others took the position that the presence or absence of entanglement
across spatially separated regions really matters when a theory is gauged against local
causality (Clifton and Halvorson 2001). Still others argued that Lorentz covariance
is the essential feature that implements local causality in a theory (Ruetsche 2011).
Earman and Valente (2014) on the other hand regarded local primitive causality (see
below) as the condition that represents local causality. Finally, Rédei (2014) took the
position that local causality is not a single property but an intricately interconnected
web of features.
We agree with Rédei. Our aim is to explore these interrelated concepts by starting
from one specific point of the web, namely from Bell’s local causality. We show
how this notion can be implemented in quantum field theory, or in our terminology,
into a LPT. We also see how our position is related to the above positions. Before
turning to the definition of local causality, however, in the present chapter we give
a brief overview of the various locality and causality principles and their relation to
one another.
Let us start with causal dynamics or, equivalently, causal time evolution. The
term “causal” here puts a restriction on the dynamics: the observables (classical
or quantum) in a region can depend only on observables in the causal past of the
18 3 Locality and Causality Principles

region. The motivation for causal dynamics comes from classical field theory on a
globally hyperbolic spacetime where a global time parameter can be chosen. If the
field equations of the theory are symmetric hyperbolic partial differential equations
(see Geroch 2010), then there exists an initial value formulation of the theory in the
form: given the initial values on (a piece of) a Cauchy surface, the time evolution
equation provides a unique solution in the domain of dependence1 of (that piece
of) the Cauchy surface. It is this restriction of the complete influences of the initial
values to the domain of dependence that makes the dynamics of the theory causal,
inasmuch as it forbids superluminal propagation (see Earman 2014).
This type of causal dynamics has three additional basic properties: It is defined
within a classical theory. It is Markovian in the sense that the causal past of the domain
of the initial conditions does not count in the evolution. It is deterministic, because
fixing the (expectation) values of the observables at a certain time, the dynamics
provides unique (expectation) values of the observables in the future (or in the past)
within the domain of dependence of the initial values. We show that the properties of
a LPT, classical or quantum, are not strong enough to provide us a causal dynamics.
An additional property, called primitive causality, will ensure the dynamics in a
LPT to be deterministic in the above sense; and only a stronger property, called local
primitive causality, will ensure the dynamics in a LPT to be not only deterministic but
also causal. It turns out that in the absence of local primitive causality the causality of
the dynamics on the observables is meaningless. In the absence of primitive causality
the notion of an initial state on the observables is already missing, because a state
on the quasilocal algebra involves the prescription of the state on the proper Cauchy
surface subalgebras for all time slices t ∈ R. Expectation values in a generic state of
such LPTs are hardly expected to show any causal properties; a causal way of state
extension for Cauchy surface subalgebras is needed (cf ; Hofer-Szabó and Vecsernyés
2015). However, Bell’s local causality condition, as a simultaneous requirement for
states and observables with specific support can be formulated and applied even in
the absence of primitive causality or a causal state extension.
In the case of stationary spacetimes, that is when a global timelike Killing vector
field exists, a natural dynamics exists in LPTs on the observables, the covariant
dynamics: The one-parameter isometry group T  (R, +) of M generated by the
global timelike Killing vector field is mapped onto a one-parameter automorphism
group {αt , t ∈ T } of the quasilocal observable algebra A acting covariantly on the
net (Requirement 3). In the case of a generic globally hyperbolic spacetime M no
global timelike Killing vector field exists, therefore there is no natural dynamics on
the observables in LPTs. However, a foliation {St , t ∈ R} of M by Cauchy surfaces
exists, which is indexed by a global time parameter. Such a foliation will lead to a
dynamics on the observables if the observable algebra corresponding to any of the
Cauchy surfaces already exhausts the quasilocal observable algebra; that is, primitive
causality holds:

1 The domain of dependence D(S ) of a (piece of) a Cauchy surface S consists of those points in M
for which any causal curve containing them intersects S .
3 Locality and Causality Principles 19

6. Primitive causality. For any covering collection K(S) ⊆ K of any Cauchy surface
S, one has AK(S) = A.
A covering collection K(St ) ⊆ K of the Cauchy surface St determines a subal-
gebra AK(St ) of AH . Let us define the Cauchy surface algebra ASt of St by the
injective limit algebra of a decreasing net of subalgebras corresponding to decreas-
ing coverings [see Brunetti and Fredenhagen (2009) for details]. Thus, in the case
of primitive causality any subalgebra AK(S) , hence any Cauchy surface subalge-
bra AS is equal to the whole quasilocal algebra A. Therefore, the injective algebra
morphisms corresponding to embeddings of globally hyperbolic Cauchy surface
coverings into M become isomorphisms and one also obtains algebra isomorphisms
ιt : ASt → A, t ∈ R between the Cauchy surface algebras and the quasilocal alge-
bra. Then the isomorphism αt  ,t := ι−1 t  ◦ ιt : ASt → ASt  provides the Cauchy
time evolution isomorphism, that is, the dynamics on the observables, between the
Cauchy surface algebras corresponding to time slices t and t  in the chosen foliation.
In the presence of a covariant dynamics the two dynamics coincide: αt  ,t = αt  −t if
the chosen foliation of M by Cauchy surfaces is compatible with the action of the
global time translation isometry group of M.
But this is not the only role of primitive causality. It makes the (covariant) dynam-
ics on the observables deterministic. Because a state on a single Cauchy surface
algebra AS , that is, a prescription of “initial (expectation) values,” fixes already the
state on the whole quasilocal algebra A, the expectation values of the observables
at arbitrary times can be given uniquely in terms of the (covariant) time evolution
automorphisms of the observable algebra A and the “initial” state.
Although the dynamics {αt  ,t , t  , t ∈ R} is deterministic, it is not necessarily
causal. That is the deterministic dynamics per se does not ensure that

αt  ,t (A(Vt )) ⊂ A(Vt  ), Vt  := St  ∩ (J+ (Vt ) ∪ J− (Vt )), Vt ⊂ St ; t, t  ∈ R, (3.1)

where Vt := V ∩ St for some V ∈ K and J+ (Vt ) ∪ J− (Vt ) is the causal cone of Vt ,

that is, the union of its causal future and causal past. The (deterministic) dynamics
on the observables meeting the requirement (3.1) is called causal dynamics on the
observables. It means that the “propagation” of local observable algebras under the
dynamics respects the causal cone structure of the underlying spacetime. It also
ensures that the state on a local algebra ιt  (A(Vt  )) fixes the state on a local algebra
ιt (A(Vt )), if Vt is in the domain of dependence of Vt  .
The local and stronger version of primitive causality is
7. Local primitive causality. For any globally hyperbolic bounded subspacetime
regions V ∈ K, A(V  ) = A(V ).2

V  ∈
2 If / K this requirement would mean that by extending K by the globally hyperbolic bounded
subspacetime regions V  , V ∈ K and defining A(V  ) := A(V ) one obtains an extended net of
local algebras satisfying isotony, microcausality, and covariance.
20 3 Locality and Causality Principles

Local primitive causality entails not only primitive causality but also the causality
requirement (3.1) of the dynamics: given Vt and Vt  as in (3.1) local primitive causality
and isotony (Requirement 1) leads to A ⊃ ιt  (A(Vt  )) = ιt  (A(Vt )) ⊃ ιt (A(Vt )).
We note that if a net satisfies Haag duality for all bounded globally hyperbolic
subspacetime regions V ∈ K, then it also satisfies local primitive causality for them:

A(V ) = A(V  ) ∩ A = A(V  ) ∩ A = A((V  ) ) ∩ A = A(V  ), V ∈ K. (3.2)

Conversely, requiring Haag duality only for causally complete regions (i.e., for
regions V ∈ K satisfying V  = V ) and local primitive causality for all V ∈ K
Haag duality follows for all V ∈ K:

A(V ) = A(V  ) = A((V  ) ) ∩ A = A(V  ) ∩ A = A(V  ) ∩ A. (3.3)

What can we say in the absence of primitive causality? In the case of a generic
globally hyperbolic spacetime there is no Cauchy dynamics {αt,t  , t, t  ∈ R} on
the observables and the Cauchy surface proper subalgebras ASt , t ∈ R are not
necessarily isomorphic. In the case of stationary spacetimes a covariant dynamics
{αt , t ∈ R} ⊂ Aut A does exist but the isomorphic Cauchy surface subalgebras
ASt , t ∈ R remain proper subalgebras of A. Their intersection can even be trivial.
Therefore there is no point in speaking about causality of the covariant dynamics
because local subalgebras “propagate” into completely new local subalgebras of A.
Moreover, the covariant dynamics is not deterministic in this case; that is, the covari-
ant dynamics and the “initial” state φs : ASs → C do not fix for t = s the expectation
values of the isomorphic but not identical proper subalgebras ASt of A. Hence, either
one prescribes the state for the whole quasilocal algebra A or an extension of the
initial state φs from ASs to A is needed. In the first case no property forbids a generic
state to reveal acausal properties. However, in the latter case properly chosen causal
restrictions on the state extension procedure may lead to a subclass of states obey-
ing causal properties (Hofer-Szabó and Vecsernyés 2015). Unfortunately, we do not
know how to do such a state extension in the case of a LQT. However, in LCTs,
where conditional probabilities of local observables have a meaning and provide
local extensions of a state, a state extension procedure can be interpreted in terms
of a stochastic dynamics, where the mentioned conditional probabilities are given
by the transition probabilities of the underlying stochastic process. To this end there
is no need for a covariant dynamics on the classical observables either. Of course,
this would ensure the isomorphisms of the image σ -algebras of the random variables
on the different Cauchy surfaces in the underlying stochastic process, however, a
stochastic process can be defined without such isomorphisms.
In this book we do not address the problem of local causality in stochastic theories.
For a specific model see Hofer-Szabó and Vecsernyés (2015). Rather, we briefly
review in the rest of the chapter some further relativistic causality principles present
in the literature and their relations to (local) primitive causality. These principles are
formulated in a quasilocal algebra AH generated by an isotone (Requirement 1) net
{N (V ), V ∈ K} of local von Neumann algebras.
3 Locality and Causality Principles 21

Let {Ak }k∈K ⊂ N (V A ) be a decomposition of the unit, that is, a set of mutu-
 orthogonal projections in the local von Neumann algebra N (V A ) such that
ally
k Ak = 1. The corresponding nonselective projective measurement is defined
as a map T{Ak } : AH → AH

T{Ak } (X ) := Ak X Ak , X ∈ AH . (3.4)
k∈K

Being a unit-preserving completely positive map (even a conditional expectation)

T{Ak } maps states to states via

φ → φ{Ak } := φ ◦ T{Ak } . (3.5)

The following causality principle requires that projections (quantum events) located
in spatially separated regions should be insensitive to such a change of states.
8. No-signaling. Let V A , VB ∈ K be spacelike separated. For any decomposition
of the unit {Ak }k∈K ⊂ N (V A ) and projection B ∈ N (VB ), and for any locally
faithful and normal state φ : AH → C, we have

φ{Ak } (B) = φ(B) (3.6)

No-signaling follows from microcausality (Requirement 2). Schlieder (1969) showed

that the converse is also true: if no-signaling holds for a decomposition of the unit
{Ak }k∈K and a projection B for all normal states of a local von Neumann algebra
with support containing V A , VB , then [Ak , B] = 0 for all k ∈ K . Being equivalent to
microcausality no-signaling trivially fulfills in LCTs. Although it is formulated as a
requirement for states, it gives a restriction for the structure of the local algebras.
Instead of non-selective projective measurements (3.4) one can also consider
selective projective measurements using a single local projection A ∈ N (A):

T A (X ) := AX A, X ∈ AH , (3.7)

which defines a completely positive but not unit preserving map T A : AH → AH .

The generated state transition

φ ◦ TA φ ◦ TA
φ → φ A := = , (3.8)
φ(A) (φ ◦ T A )(1)

sometimes called Lüders projection (Lüders 1950), provides another causality

requirement:
9. Independence. For any projections A ∈ N (V A ) and B ∈ N (VB ) such that
V A , VB ∈ K are spacelike separated regions, and for any locally faithful and
22 3 Locality and Causality Principles

normal state φ, we have3

φ A (B) = φ(B). (3.9)

In the case of microcausality (Requirement 2), (3.9) implies that φ(AB) = φ(A)φ(B);
that is, φ becomes a product state by restricting it to the subalgebra generated by
N (V A ) and N (VB ). Hence, it is a too strong assumption, which is violated in LQTs,
for example, by any entangled state. Of course, it is also violated also in the case of
superluminal correlations.
In general, (completely) positive maps T : A → A on a C ∗ -algebra A with
the property 0 < T (1)  1 can be considered as generalized  measurements or
operations. They are called inner if T has the form T := i Ad K i with K i ∈ A.
If the K i -s are mutually orthogonal projections one speaks about projective (inner)
operations. Operations with T (1) = 1 and T (1) < 1 are called nonselective and
selective operations, respectively. If A is a von Neumann algebra one usually requires
T to be normal. If A = B(H) this means that T is σ -weakly continuous. See, for
example Werner (1987), and references therein.
A net satisfying local primitive causality (Requirement 7) also satisfies:
10. Local determinism. (Earman and Valente 2014) If φ|A(V ) = φ  |A(V ) for any two
states φ and φ  and for any globally hyperbolic spacetime region V ∈ K, then
φ|A(V  ) = φ  |A(V  )
and consequently it also satisfies
11. Stochastic Einstein locality. Let V A , VC ∈ K such that VC ⊂ J− (V A ) and V A ⊂
VC . If φ|A(VC ) = φ  |A(VC ) holds for any two states φ and φ  on A then φ(A) =
φ  (A) for any projection A ∈ A(V A ).
Microcausality alone does not entail local primitive causality. Because micro-
causality is equivalent to no-signaling and local primitive causality represents no-
superluminal propagation (Earman and Valente 2014), it is therefore an interest-
ing question whether there exist nets that satisfy local primitive causality but vio-
late microcausality. Usually the translation covariant field algebra extension of the
observables F ⊃ A, in which the localized and transportable endomorphisms—
the Doplicher–Haag–Roberts morphisms—of the observables can be implemented,
serve such examples: Although local field algebras are defined to be relatively local
to observables

F(V ) := A(V  ) ∩ F, V ∈ K, (3.10)

local field algebras corresponding to spacelike separated regions do not commute in

general, hence microcausality fails. (For example, in the field algebra of the local
quantum Ising model there are field operators with spacelike separated supports

3 Butterfield(1995, Eqs. 3.6 and 3.7) and Earman and Valente (2014, Sect. 7.2) called (3.6) and
(3.9) parameter independence and outcome independence, respectively (Shimony 1986). For the
difference between parameter independence, where φ in (3.6) is conditioned on the common cause,
and no-signaling, where φ is unconditioned, see Maudlin (2002) and Norsen (2011).
3 Locality and Causality Principles 23

that anticommute.) However, local primitive causality does hold in the net of field
algebras, because V  = V  and hence

F (V ) := A(V  ) ∩ F = A(V  ) ∩ F = A((V  ) ) ∩ F =: F (V  ), V ∈ K. (3.11)

Thus, for such a net of local (field) algebras no-signaling is violated whereas no-
superluminal propagation holds.
In the subsequent chapters we work within the framework of a LPT. When speak-
ing about deterministic dynamics, we also assume Requirements 6–7.

References

R. Brunetti, K. Fredenhagen, Quantum field theory on curved backgrounds, in Quantum Field

Theory on Curved Spacetimes, Concepts and Mathematical Foundations, vol. 786, Lecture Notes
in Physics, ed. by C. Bär, K. Fredenhagen (Springer, Berlin, 2009)
J. Butterfield, Vacuum correlations and outcome independence in algebraic quantum field theory,
in Fundamental Problems in Quantum Theory, Annals of the New York Academy of Sciences,
Proceedings of a conference in honour of John Wheeler, ed. by D. Greenberger, A. Zeilinger
(1995), pp. 768–785
R. Clifton, H. Halvorson, Entanglement and open systems in algebraic quantum field theory. Stud.
Hist. Philos. Mod. Phys. 32(1), 1–31 (2001)
J. Earman, No superluminal propagation for classical relativistic quantum fields. Stud. Hist. Philos.
Mod. Phys. 48, 102–108 (2014)
J. Earman, G. Valente, Relativistic causality in algebraic quantum field theory. Int. Stud. Philos.
Sci. 28(1), 1–48 (2014)
R. Geroch, Faster than light? (2010). arXiv:1005.1614
G. Hofer-Szabó, P. Vecsernyés, On the concept of Bell’s local causality in local classical and
quantum theory. J. Math. Phys. 56, 032303 (2015)
G. Lüders, Über die Zustandsänderung durch den Messprozess. Ann. Phys. 443, 322 (1950)
T. Maudlin, Quantum Non-Locality and Relativity (Malden, Massachusetts, 2002)
T. Norsen, J.S. Bell’s concept of local causality. Am. J. Phys. 79, 12 (2011)
M. Rédei, A categorial approach to relativistic locality. Stud. Hist. Philos. Mod. Phys. 48, 137–146
(2014)
L. Ruetsche, Interpreting Quantum Theories (Oxford University Press, Oxford, 2011)
S. Schlieder, Einige Bemerkungen über Projektionsoperatoren (Konsequenzen eines Theorems von
Borchers). Commun. Math. Phys. 13, 216–225 (1969)
A. Shimony, Events and processes in the quantum world, in Quantum Concepts in Space and Time,
ed. by R. Penrose, C. Isham (Oxford University Press, Oxford, 1986), pp. 182–203
R. Werner, Local preparability of states and the split property in quantum field theory. Lett. Math.
Phys. 13, 325–329 (1987)
Chapter 4
Bell’s Notion of Local Causality

Abstract At the beginning of this Chapter Bell’s different formulations of local

causality will be reviewed. Next, we analyze the three key concepts featuring in
Bell’s final formulation, namely “local beables”, “shielder-off region”, and “complete
specification.” We translate them into the LPT framework and provide a generalized
definition of local causality. Finally, we relate shielder-off regions to d-separating
sets in a Bayesian network and prove that local primitive causality renders an atomic
LPT to be locally causal.

Keywords Bell’s local causality · Local beables · Shielder-off region · Complete

specification · Bayesian networks · Atomicity

Local causality is one of the most important notions that Bell introduced into the
foundations of quantum mechanics. To our knowledge, it crops up three times in
Bell’s writings and gains a more and more refined formulation. Bell addressed local
causality for the first time in his “The Theory of Local Beables” delivered at the
Sixth GIFT Seminar in 1975 (Bell 1975/2004, p. 54); later in a footnote added to
his 1986 paper, “EPR correlations and EPW distributions,” intending to clean up the
first version (Bell 1986/2004, p. 200); and finally in the most elaborate form in his
posthumously published, “La nouvelle cuisine” (Bell 1990/2004, pp. 239–240). It is
this latter formulation that has been considered in the literature as the definition of
local causality. In what follows we also use this third formulation but first we briefly
overview the previous two.
Bell’s first formulation of local causality reads:
Consider a theory in which the assignment of values to some beables  implies, not nec-
essarily a particular value, but a probability distribution, for another beable A. Let p(A|)
denote1 the probability of a particular value A given particular values . Let A be localized
in a space-time region A. Let B be a second beable localized in a second region B separated
from A in a spacelike way. (Fig. 4.1.) Now my intuitive notion of local causality is that events

1 For the sake of uniformity throughout the book we slightly changed Bell’s notation and figures.
26 4 Bell’s Notion of Local Causality

A B

Fig. 4.1 Bell’s (1975) first figure illustrating local causality

A B

C1 C2

Fig. 4.2 Bell’s (1975) second figure illustrating local causality

in B should not be “causes” of events in A, and vice versa. But this does not mean that the two
sets of events should be uncorrelated, for they could have common causes in the overlap of
their backward light cones. It is perfectly intelligible then that if  in (4.1) does not contain
a complete record of events in that overlap, it can be usefully supplemented by information
from region B. So in general it is expected that

p(A|, B) = p(A|) (4.1)

However, in the particular case that  contains already a complete specification of beables
in the overlap of the light cones, supplementary information from region B could reasonably
be expected to be redundant.
Let C2 denote a specification of all beables, of some theory, belonging to the overlap of
the backward light cones of spacelike regions A and B. Let C1 be a specification of some
beables from the remainder of the backward light cone of A, and B of some beables in the
region B. (See Fig. 4.2.) Then in a locally causal theory

p(A|C1 , C2 , B) = p(A|C1 , C2 ) (4.2)

whenever both probabilities are given by the theory. (Bell 1975/2004, p. 54)

We comment on the terminology Bell is using in his definition below in detail.

Here just note that in his screening-off condition (4.2) Bell takes into account the
whole causal past of the events in question. He does not assume some kind of
Markovity rendering superfluous of the remote past regions below a certain Cauchy
4 Bell’s Notion of Local Causality 27

surface. His second formulation of local causality can be regarded as a step towards
this Markovian direction:
The notion of local causality presented in this reference [namely in Bell (1975/2004)] in-
volves complete specification of the beables in an infinite space-time region. The following
conception is more attractive in this respect: In a locally-causal theory, probabilities attached
to values of local beables in one space-time region, when values are specified for all local
beables in a second space-time region fully obstructing the backward light cone of the first,
are unaltered by specification of values of local beables in a third region with spacelike
separation from the first two. (Bell 1986/2004, p. 200)

Bell’s second version is in a footnote; it is very succinct and contains no figure.

The new element is the phrasing “space-time region fully obstructing the backward
light cone of the first.” This idea gets a more precise exposition in Bell’s third, final
formulation of local causality:
A theory will be said to be locally causal if the probabilities attached to values of local
beables in a space-time region V A are unaltered by specification of values of local beables
in a space-like separated region VB , when what happens in the backward light cone of V A
is already sufficiently specified, for example by a full specification of local beables in a
space-time region VC . (Bell 1990/2004, pp. 239–240)

The figure Bell attached to this formulation is reproduced in Fig. 4.3 with the
original caption. Bell elaborates on his formulation as follows.
It is important that region VC completely shields off from V A the overlap of the backward
light cones of V A and VB . And it is important that events in VC be specified completely.
Otherwise the traces in region VB of causes of events in V A could well supplement whatever
else was being used for calculating probabilities about V A . The hypothesis is that any such
information about VB becomes redundant when VC is specified completely. (Bell 1990/2004,
p. 240)

The notions featured in Bell’s formulation have become the target of intensive dis-
cussion in the philosophy of science (see Norsen 2009, 2011). Here we concentrate
only on three terms, namely local beables, complete specification, and shielding-off.
Local beables. The notion “beable” is Bell’s neologism and is contrasted with the
term “observable” used in quantum theory. “The beables of the theory are those

VA VB

VC

Fig. 4.3 Full specification of what happens in VC makes events in VB irrelevant for predictions
about V A in a locally causal theory
28 4 Bell’s Notion of Local Causality

entities in it which are, at least tentatively, to be taken seriously, as corresponding to

something real” (Bell 1990/2004, p. 234); or elsewhere:
The beables of the theory are those elements which might correspond to elements of reality,
to things which exist. Their existence does not depend on “observation.” Indeed observation
and observers must be made out of beables.
I use the term “beable” rather than some more committed term like “being” or “beer” to recall
the essentially tentative nature of any physical theory. Such a theory is at best a candidate
for the description of nature. (Bell 1984/2004, p. 174)

The clarification of what the “beables” of a theory are, is indispensable in order to

define local causality inasmuch as “there are things which do go faster than light.
British sovereignty is the classical example. When the Queen dies in London (long
may it be delayed) the Prince of Wales, lecturing on modern architecture in Australia,
becomes instantaneously King” (p. 236).
Beables are to be local: “Local beables are those which are definitely associ-
ated with particular space-time regions. The electric and magnetic fields of classical
electromagnetism, E(t, x) and B(t, x) are again examples” (p. 234).
Complete specification. Local beables are to “specify completely” region VC in order
to block causal influences arriving at V A from the common past of V A and VB . [For
the question of complete versus sufficient specification see Norsen (2011), Seevinck
and Uffink (2011), Hofer-Szabó (2015a).]
Shielding-off. “It is important that region VC completely shields off from V A the
overlap of the backward light cones of V A and VB .” Why is that so? Why is local
causality not required for such regions VC as depicted in Fig. 4.4, for example? The
reason for that is the following. If VC is localized as in Fig. 4.4, then the spacetime
region above VC in the common past of the correlated events may contain stochastic
events (with determined probabilities by the complete specification on the region VC )
which can establish a correlation between A and B in a classical stochastic theory.
The “shielding-off” condition is required just to exclude this case.
But if this is the reason, then why not also allow for regions VC as depicted in
Fig. 4.5? Allowing for shielding-off regions that intersect with the common past is
indeed a possible interpretation of Bell’s term “shielding-off.” We return to this point

VA VB

VC

Fig. 4.4 A not completely shielding-off region VC

4 Bell’s Notion of Local Causality 29

VA VB

VC

Fig. 4.5 An intersecting and completely shielding-off region VC

in Remark 4 below. [For the relation between the localization of the region VC and
the causal Markov condition see Hofer-Szabó (2015b).]
How do Bell’s three terms above “translate” into the framework of LPT? Let us see
them again in turn.
Local beables. In a classical field theory beables are characterized by sets of field
configurations. In our local algebraic framework local equivalence classes of field
configurations, namely configurations having the same field values on a given
spacetime region, generate local σ -algebras, as explained in Chap. 2. The elements
of local σ -algebras capture all the beables of the theory; moreover they also provide
a localization for them. Translating σ -algebras into Abelian von Neumann algebras
one can use a common language for classical and quantum theories: local beables
in a region V ∈ K are elements of the local von Neumann algebra N (V ), which is
Abelian for a classical and non-Abelian for a quantum theory.
We note here that our use of beables and hence our upcoming definition of local
causality transcends Bell’s ideas in an important sense. Events in a noncommutative
local algebra can readily be interpreted operationally as measurement outcomes but
can hardly be something ontological. It is true that at one point, namely in his Subject
and object, Bell gives some thought to the idea whether beables can be associated
with quantum mechanical observables:
Could not one just promote some of the “observables” of the present quantum theory to the
status of beables? The beables would then be represented by linear operators in the state
space. The values which they are allowed to be would be eigenvalues of those operators. For
the general state the probability of a beable being a particular value would be calculated just
as was formerly calculated the probability of observing that value. (Bell 1973/2004, p. 41)

But the program envisioned here by Bell and later called the maximal beable ap-
proach (Halvorson and Clifton 1999) aims to identify as beables only a commutative
subalgebra of observables that can enjoy determinate values simultaneously and not
the full noncommutative algebra. Many interpretations of quantum mechanics can
be seen as a variant of this approach. Collapse interpretations specify the maximal
beable algebra for a system subject to the measurement of a non-degenerate observ-
able. Bohmian interpretation picks a preferred observable and generates the maximal
30 4 Bell’s Notion of Local Causality

Abelian algebra by the spectral projections of this observable. Modal interpretations

(Clifton 2000; Dieks 2002) use the so-called eigenstate-eigenvalue link and iden-
tify the beables of the theory with the spectral projections of the density operator
associated with the quantum state.
Contrary to these interpretations, we do not put any constraint whatsoever on local
algebras that can stand for “beables.” We think that at this level of generality where
neither the dynamics, nor the measurement procedure, nor any other features of
the system are specified, we cannot determine which elements of the local algebras
should be regarded as local beables. [For some problems of the maximal beable
approach in nonatomic von Neumann algebras see Ruetsche (2011), Chap. 8.] On
the other hand, we do think that treating the classical and quantum case parallel
manner sheds light on the general structural position of local causality in LPTs and
its relations to other locality principles. We come back to this liberal understanding
of “beables” in Chap. 9.
Complete specification. Complete specification of field configurations in a given
spacetime region means that one specifies the field values to a prescribed value in
the given spacetime region; that is, one specifies the corresponding local equivalence
class (a cylindrical set) of a single configuration. In probabilistic language com-
plete specification is translated to a probability measure having support on this local
equivalence class of the single specified configuration. More precisely, complete
specification is a procedure bringing about a change of the probability measure on
the whole σ -algebra and resulting in the probability measure restricted to the local
σ -algebra in question that has support on the local equivalence class of the single
specified configuration. In the Abelian von Neumann language this corresponds to a
change of the original state that results in a pure state on the local von Neumann alge-
bra in question with value 1 on the projection corresponding to the local equivalence
class of the single specified configuration. However, we would also like this change
of states to be as local as possible. Therefore we translate a “complete specification
of beables in a region V ∈ K” as a change of state

φ◦T
φ(X ) → φT (X ) := (4.3)
(φ ◦ T )(1)

by a completely positive map T on the quasilocal observables obeying the properties:

P1 : the restriction of φT to the local algebra N (V ) is pure,
P2 : BT (1) = T (B) = T (1)B hold for local observables B supported in V  .
Concerning property P1 we note that von Neumann algebras in B(H) that have a
separating vector in H, irrespective of being Abelian or non-Abelian algebras, do
not possess a pure normal state (Clifton and Halvorson 2001). This is the case, for
example, in AQFTs with type III local von Neumann algebras. Thus starting from a
(locally) normal state φ on them, a normal operation T leads to a (locally) normal
state φT that cannot be pure. There are certain ways to circumvent this problem
(neither of them being fully satisfactory). (1) One can use a nonnormal operation to
get a pure state for the local von Neumann algebra. In this case, however, one jumps
4 Bell’s Notion of Local Causality 31

into a different quasiequivalence class of representations of observables that we just

wanted to avoid by considering only (locally) normal states for the local von Neumann
algebras. (2) In the case of type III (hence non-Abelian) local von Neumann algebras
one can also assume the split property [see, e.g. Werner (1987) and references therein]
and use the (atomic) type I intermediate von Neumann algebra to provide a pure state,
hence a “full specification,” for a somewhat larger local observable algebra supported
in a somewhat larger local region.2
Concerning property P2 we note that weakly localized operations in V (Werner
1987) obey property P2 for all elements B ∈ N (V ) ⊇ AH (V  ) by definition. More-
over, if T is normal and AH = B(H) then every  weakly localized operation T with
respect to V ∈ K is inner in N (V ); that is, T = i Ad K i with K i ∈ N (V ).
In a general LPT, we do not know how to characterize the operations that result
in a state obeying properties P1 and P2 , but in the case of atomic (type I) local von
Neumann algebras it is almost trivial: one has to do a selective projective measure-
ment defined in (3.7) by an atom (a minimal projection) C in the local algebra N (V )
that induces the change of states φ → φC defined in (3.8).
Shielding-off. Finally, a shielding-off region in a LQT (see Fig. 4.3) can be defined
as VC ∈ K satisfying the three localization requirements:
L1 : VC ⊂ J− (V A )
L2 : V A ⊂ VC
L3Q : VC ⊂ VB
that is, VC should be in the causal past of V A ; it has to be “wide enough” such that
V A is in the domain of dependence of VC , and it should be spacelike separated from
VB .
In a LCT a shielding-off region intersecting with the common past (see Fig. 4.5)
is allowed, and requirement L 3Q can be replaced by the weaker requirement:
LC3 : J− (VC ) ⊃ J− (V A ) ∩ J− (VB ).
If VC is contained in an infinitely thin Cauchy surface, requirement L C3 coincides
with requirement L 3Q .
Given the above interpretation of the terms “local beables,” “complete specification,”
and “shielding-off,” we are now in the position to formulate Bell’s notion of local
causality in the framework of LPTs:
Definition 2 Let an LPT be represented by a net {N (V ), V ∈ K} of von Neumann
algebras. Let A ∈ N (V A ) and B ∈ N (VB ) be a pair of projections supported in
spacelike separated regions V A , VB ∈ K. Let φ be a locally normal and locally faithful
state on the quasilocal observables establishing a correlation φ(AB) = φ(A)φ(B)
between A and B. Let T be an operation on the quasilocal observables obeying prop-
erties P1 and P2 . Finally, let VC ∈ K be a spacetime region defined by requirements
L 1 , L 2 , and L 3Q /L C3 . The LPT is called locally causal if for any such quintuple
(A, B, φ, T , VC ) the following screening property holds:

2 We thank Yuichiro Kitajima for drawing our attention to these points.

32 4 Bell’s Notion of Local Causality

φT (AB) = φT (A)φT (B). (4.4)

Remarks:
1. If the local algebras of the net are atomic,3 the states φT in Definition 2 can be
replaced by the state φC given by (3.7)–(3.8), where C ∈ A(VC ) is an arbitrary
atomic event, that is, a minimal projection. This converts (4.4) into the screening-
off condition of the well-known form:
φ(C ABC) φ(C AC) φ(C BC)
= . (4.5)
φ(C) φ(C) φ(C)

2. In the commutative case Definition 2 is just Bell’s local causality. In LCTs (4.5)
can be written into the standard conditional form

p(AB|C) = p(A|C) p(B|C), (4.6)

or into the equivalent asymmetric form

p(A|BC) = p(A|C) (4.7)

sometimes used in the literature [e.g., in Bell (1975/2004, p. 54)]. We come back
to “noncommuting beables” in Chap. 8.
3. Here we briefly comment on a definition of local causality recently given by
Henson (2013b). Henson’s definition differs from ours in several respects. First,
Henson formulates local causality in terms of σ -algebras. Using the recipe given
in Chap. 2 to convert σ -algebras into Abelian von Neumann algebras, this dif-
ference can be easily dissolved. Second, the Henson definition applies only
to atomic σ -algebras: his screening-off condition is equivalent to (4.5). Our
more general screening condition (4.4) applies both to noncommutative and to
nonatomic local algebras. Third, in Henson’s definition the screener-off region
VC is not localized according to requirements L 1 , L 2 , and L 3Q /L C3 . It is an un-
bounded region, a “suitable past” of V A and VB .4 In our opinion, here Henson
follows Bell’s first formulation of local causality given in Bell (1975/2004, p.
54), where the screener-off regions are identified with the complete, unbounded
causal past of the correlated events. Our definition, on the other hand, is based on
Bell’s last, operationally more desirable definition provided in Bell (1990/2004,
pp. 239–240), where the screener-off regions are only bounded Cauchy segments
of the unbounded past regions.5

3 Which is typically not the case in a general AQFT.

4 Where the term “suitable past …has been left open deliberately. …It could be …the ‘mutual past’

…the ‘joint past’ or the past of one of the regions but not the other” (Henson 2013b, p. 1015). For
an argument for, against, and again for not specifying the screener-off region see Henson (2005,
2013a), Rédei and San Pedro (2012), respectively.
5 Cf. also Bell (1986/2004, p. 200): “The notion of local causality presented in this reference [namely

in Bell (1975/2004)] involves complete specification of the beables in an infinite space-time region.
4 Bell’s Notion of Local Causality 33

In his paper Henson shows that the lack of separability (additivity, in our lan-
guage; see Chap. 2) does not block the derivation of Bell’s inequalities. As we
show, this result is in complete agreement with ours: additivity is not required in
our book; hence it plays no role in the derivation of Bell’s inequalities in LCTs.
4. There is a symmetric independence relation closely related to Bell’s asymmetric
local causality requirement. The idea comes from the theory of hyperbolic differ-
ential equations describing deterministic causal processes. As said above, in the
case of hyperbolic differential equations there exists an initial value formulation
of the theory: given the initial values on a segment of a Cauchy surface, the time
evolution equation provides a unique solution in the domain of dependence of
the segment. This also means that one can freely fix the data on two (spatially
separated) segments of a Cauchy surface, VC and VD , and solve the initial value
problem independently in the future domain of dependence regions D + (VC ) and
D + (VD ). Thus, if V A ⊆ (D + (VC ) \ VC ) and VB ⊆ (D + (VD ) \ VD ), then the
Cauchy data in VC will fix the values in V A independently of the values in VB .
Indeterministic causal processes (see Hofer-Szabó and Vecsernyés 2015, Sect.
3) inherit this feature of deterministic causal processes. Here one can indepen-
dently fix the values in regions VC and VD such that the stochastic evolution
within D + (VC ) and D + (VD ) will be probabilistically independent. Thus, fixing
the value in VC will fix the probability of the values in V A ⊆ (D + (VC ) \ VC )
irrespective of the values in VB . And this is exactly what Bell’s local causality
requires.
5. Above we raised the question as to why the shielder-off regions featuring in the
definition of Bell’s local causality need to separate V A from the common past
of V A and VB , as depicted in Figs. 4.3 and 4.5, and cannot be “pushed back” in
the remote past as in Fig. 4.4. Or to put it formally, why the shielder-off regions
should be characterized by the criteria L 1 , L 2 , and L 3Q /L C3 . Here we claim (for the
details see Hofer-Szabó 2015b, 2018) that the above localization of the shielder-
off regions can be nicely translated into the theory of Bayesian networks, and
it turns out that the shielder-off regions conform in a well-defined sense to the
d-separating sets in Bayesian networks or m-separating sets in mixed graphs.
(For causal graphs, d-separation, and m-separation see Pearl 2000; Glymour et
al. 2000; Richardson and Spirtes 2002; Sadeghi and Lauritzen 2014.)
The translation goes as follows. Let K be a finite subset of the covering collection
K. The collection K gives rise to a mixed graph G as follows. Let the vertices
of G be the regions V in K and let there be a directed edge from V1 to V2
(V1 , V2 ∈ K ) if there is a future directed causal curve γ from a point p1 ∈ V1 to
a point p2 ∈ V2 such that γ does not intersect any other region in K , except for
V1 and V2 . The vertices V1 and V2 are connected by a bidirected edge if there
are directed edges both from V1 to V2 and from V2 to V1 .
This “translation manual” uniquely assigns a causal graph G to a collection K .
The graph G is based on the causal structure of the spacetime M. However, being

The following conception is more attractive in this respect.” And then comes the new definition
based on bounded regions.
34 4 Bell’s Notion of Local Causality

a kind of coarse-graining, the graph also depends sensitively on the collection

K . Choosing “small” pointlike regions, the resulting graph will comprise many
edges because the regions do not causally shield one another, whereas the graph
coming from collections with “fat” overlapping regions will contain fewer edges.
On the two extreme ends of the scale one obtains the “densest” graph G max and
the “sparsest” graph G min . Namely, by shrinking the regions in K to pointlike
regions one ends up with G max where there is an edge pointing from V1 to V2
whenever V1 ∩ J− (V2 ) = ∅. By “blowing up” the regions in K such that they
causally shield one another, one obtains G min where there is no edge pointing
from V1 to V3 if V1 → V2 and V2 → V3 .
The type of the graph also greatly depends on the regions in K . For certain
collections one ends up with directed acyclic graphs; for other collections one
obtains mixed (not necessarily acyclic) graphs. In Chap. 8 we present a covering
collection of the 1+1-dimensional Minkowski spacetime which is composed
of minimal double cones of unit diameter. This covering collection leads to a
directed acyclic graph. In the 2+1-dimensional Minkowski spacetime, however,
one can choose a collection consisting of a finite number of double cones such
that the resulting graph will be a cyclic graph composed of one single cycle:
V1 → V2 → . . . Vn → V1 . Thus, not just the “density” but also the type of the
causal graph sensitively depends on the collection K itself.
Here we do not intend to provide a systematic treatment as to which collections
lead to which type of graphs. Our aim is simply to state a proposition proven in
Hofer-Szabó (2018):
Let V A and VB be two spacelike separated spacetime regions in K . Call a set
{Vi } ⊂ K a shielder-off set of regions for V A if ∪i Vi is a shielder-off region for
V A characterized by the criteria L 1 , L 2 , and L C3 . (We take the classical criterion
L C3 instead of the quantum criterion L 3Q because Bayesian nets are classical.)
Consider the graph G resulting from K containing the vertices V A and VB and
the shielder-off set {Vi }. Then the shielder-off set {Vi } d-separates/m-separates
V A from VB .
Thus, in this sense shielder-off regions are d-separating.
Coming back to Definition 2 of local causality, the main question is when is a
LPT locally causal. We answer this question by the following
Proposition 1 Let the local von Neumann algebras of a LPT be atomic. Then Bell’s
local causality holds if the LPT obeys local primitive causality.

Proof If A is a projection and C is a minimal projection in an atomic von Neumann

algebra then C AC = r (C, A)C with r (C, A) ∈ {0, 1} in the case of Abelian and
r (C, A) ∈ [0, 1] ⊂ R in the case of non-Abelian algebras. Using the notation of
Definition 2 A becomes a projection in the atomic von Neumann algebra N (VC ) due
to local primitive causality. Thus if C ∈ N (VC ) is a minimal projection then
4 Bell’s Notion of Local Causality 35

φ(C ABC) φ(C AC B) φ(C B) φ(C AC) φ(C BC)

φC (AB) := = = r (C, A) =
φ(C) φ(C) φ(C) φ(C) φ(C)
=: φC (A)φC (B). (4.8)

Here we used that C B = BC due to commutativity in the case of a LCT and due to the
spacelike separation of VB and VC (ensured by requirement L 3Q ) and microcausality
in a LQT. 

By Proposition 1 our approach comes near to the position of Earman and Valente
(2014) who claim that local causality should be understood as local primitive causal-
ity. Here we see that in atomic LPTs local primitive causality implies, in fact, local
causality. [For a similar sufficient condition for local causality in stochastic LPTs
see Hofer-Szabó (2015b).]
In the case of LPTs with local primitive causality but with nonatomic von Neu-
mann algebras we do not know how to characterize the local manipulation on the
state described in Definition 2, therefore a similar proof cannot be applied. In the
case of LPTs without primitive causality the dynamics is not deterministic, hence
an initial state on a Cauchy surface algebra does not determine the state on the
whole quasilocal algebra A. States can be forced by a properly chosen state extension
procedure to show suitable causality properties. In this book we do not investigate
such state extensions. [For a simple causal stochastic Ising model see Hofer-Szabó
and Vecsernyés (2015)]. We stress the following: without a causal dynamics on the
observables (no local primitive causality) and without the notion of initial states
(no primitive causality) Bell’s local causality seems to be the only surviving local
causality requirement.
In the light of Proposition 1 the reader may ask how a local quantum theory can
be locally causal if local causality implies various Bell inequalities that are known
to be violated for certain sets of quantum correlations. We come back to this point
in Chap. 8.

References

J.S. Bell, Beables for quantum field theory, TH-2053-CERN, presented at the Sixth GIFT Seminar,
Jaca, 2–7 June (1975); reprinted in (Bell, 2004, 52–62)
J.S. Bell, Beables for quantum field theory, CERN-TH. 4035/84, (1984); reprinted in (Bell, 2004,
173–180)
J.S. Bell, EPR correlations and EPW distributions, in New Techniques and Ideas in Quantum
Measurement Theory (New York Academy of Sciences, 1986); reprinted in (Bell, 2004, 196–
200)
J.S. Bell, La nouvelle cuisine, in Between Science and Technology, ed. by J. Sarlemijn, P. Kroes
(Elsevier, 1990); reprinted in (Bell, 2004, 232–248)
J.S. Bell, Subject and object, in The Physicist’s Conception of Nature (D. Reidel, Dordrecht , 1973),
pp. 687–90; reprinted in (Bell, 2004, 40–44)
R. Clifton, The modal interpretation of algebraic quantum field theory. Phys. Lett. A 271(3), 167–
177 (2000)
36 4 Bell’s Notion of Local Causality

R. Clifton, H. Halvorson, Entanglement and open systems in algebraic quantum field theory. Stud.
Hist. Philos. Mod. Phys. 32(1), 1–31 (2001)
D. Dieks, The modal interpretation of algebraic quantum field theory, in Ontological Aspects of
Quantum Field Theory, ed. by M. Kuhlman, H. Lyre, A. Wayne (World Scientific, New Jersey,
2002), Chap. 11
J. Earman, G. Valente, Relativistic causality in algebraic quantum field theory. Int. Stud. Philos.
Sci. 28(1), 1–48 (2014)
C. Glymour, R. Scheines, P. Spirtes, Causation, Prediction, and Search (Cambridge: The MIT Press,
2000)
H. Halvorson, R. Clifton, Maximal beable subalgebras of quantum-mechanical observables. Int. J.
Theor. Phys. 38, 2441–84 (1999)
J. Henson, Comparing causality principles. Stud. Hist. Philos. Mod. Phys. 36, 519–543 (2005)
J. Henson, Non-separability does not relieve the problem of Bell’s theorem. Found. Phys. 43, 1008–
1038 (2013b)
J. Henson, Confounding causality principles: comment on Rédei and San Pedro’s “Distinguishing
causality principles”. Stud. Hist. Philos. Mod. Phys. 44, 17–19 (2013a)
G. Hofer-Szabó, Bell’s local causality is a d-separation criterion, in Springer Proceedings in Math-
ematics and Statistics (forthcoming) (2018)
G. Hofer-Szabó, Local causality and complete specification: a reply to Seevinck and Uffink, in
Recent Developments in the Philosophy of Science: EPSA13 Helsinki, ed. by U. Mäki, I. Votsis,
S. Ruphy, G. Schurz (Springer, Berlin, 2015a), pp. 209–226
G. Hofer-Szabó, Relating Bell’s local causality to the Causal Markov Condition. Found. Phys. 45(9),
1110–1136 (2015b)
G. Hofer-Szabó, P. Vecsernyés, On the concept of Bell’s local causality in local classical and
quantum theory. J. Math. Phys. 56, 032303 (2015)
T. Norsen, Local causality and completeness: Bell vs. Jarrett. Found. Phys. 39, 273 (2009)
T. Norsen, J.S. Bell’s concept of local causality. Am. J. Phys. 79, 12 (2011)
J. Pearl, Causality: Models, Reasoning, and Inference (Cambridge University Press, Cambridge,
2000)
M. Rédei, I. San Pedro, Distinguishing causality principles. Stud. Hist. Phil. Mod. Phys. 43, 84–89
(2012)
T.S. Richardson, P. Spirtes, Ancestral graph Markov models. Ann. Stat. 30, 962–1030 (2002)
L. Ruetsche, Interpreting Quantum Theories (Oxford University Press, Oxford, 2011)
K. Sadeghi, S. Lauritzen, Markov properties for mixed graphs. Bernoulli 20(2), 676–696 (2014)
M.P. Seevinck, J. Uffink, Not throwing our the baby with the bathwater: Bell’s condition of local
causality mathematically ‘sharp and clean’, in Explanation, Prediction, and Confirmation The
Philosophy of Science in a European Perspective, vol. 2, ed. by D. Dieks, W.J. Gonzalez, S.
Hartmann, Th. Uebel, M. Weber (2011), pp. 425–450
R. Werner, Local preparability of states and the split property in quantum field theory. Lett. Math.
Phys. 13, 325–329 (1987)
Chapter 5
The Common Cause Principle

Abstract In this chapter we turn to Reichenbach’s Common Cause Principle. The

principle is generalized to the LPT framework and the status of the various Common
Cause Principles in algebraic quantum field theory is investigated. Then we motivate
the noncommutative generalization of the principle and compare the Common Cause
Principle with Bell’s local causality.

Keywords Weak · Strong and Commutative · Noncommutative Common Cause

Principles

Local causality is closely related to Reichenbach’s (1956) Common Cause Principle.

The Common Cause Principle (CCP) states that if there is a correlation between two
events A and B and there is no direct causal (or logical) connection between the
correlated events, then there always exists a common cause C of the correlation.
Reichenbach’s original definition is formulated in a purely classical probabilistic
setting lacking any spatiotemporal considerations; however, it can readily be gener-
alized to the LPT framework. [For the steps of the generalization see Rédei (1997),
Rédei and Summers (2002, 2007), Hofer-Szabó and Vecsernyés (2012a, b, 2013a, b)
and Hofer-Szabó et al. (2013).] In this chapter we briefly overview the various CCPs
and relate them to Bell’s local causality.
Let {N (V ), V ∈ K} be a net representing a LPT. Let A ∈ N (V A ) and B ∈ N (VB )
be two events (projections) supported in spacelike separated regions V A , VB ∈ K
that are correlated in a locally normal and faithful state φ. The common cause of
the correlation is a set of events (projections) {Ck }k∈K that screen off the correlated
events from one another, and are localized in a region VC ∈ K in the causal past of
A and B. For the precise choice of this past one has (at least) three options. One can
localize the common cause either (i) in the union, J− (V A ) ∪ J− (VB ); or (ii) in the
intersection, J− (V A ) ∩ J− (VB ), of the causal past of the regions V A and VB ; or (iii)
more restrictively, in the spacetime region that lies in the intersection of causal pasts
of every point of V A ∪ VB , formally ∩x∈VA ∪VB J− (x); see (Rédei, Summers 2007).
38 5 The Common Cause Principle

We refer to the above three pasts in turn as the weak past, common past, and strong
past of A and B, respectively.
Now, we can define various CCPs in a LPT:
Definition 3 A LPT represented by a net {N (V ), V ∈ K} is said to satisfy the
(Weak/Strong) CCP, if for any pair A ∈ N (V A ) and B ∈ N (VB ) of projections
supported in spacelike separated regions V A , VB ∈ K and for every locally faithful
state φ establishing a correlation φ(AB) = φ(A)φ(B) between A and B, there
exists a nontrivial (see below) common cause that is a set of mutually orthogonal
projections {Ck }k∈K ⊂ N (VC ), VC ∈ K localized in the (weak/strong) common past
of V A and VB , which decompose the unit and satisfy

φCk (AB) = φCk (A)φCk (B), k ∈ K , (5.1)

where the state φCk is given by (3.8).

A common cause is called trivial if Ck ≤ X with X = A, A⊥ , B or B ⊥ for
all k ∈ K . If Ck commutes with both A and B for all k ∈ K , then we call it
a commuting common cause, otherwise a noncommuting one, and the appropriate
CCP a Commutative/Noncommutative CCP.
Trivial common causes provide solutions of (5.1) independently of the state φ. There-
fore they are considered as purely “kinematic” or “algebraic” solutions that are insen-
sitive to the actual physical environment provided by a particular state φ. If at least
one of the algebras N (V A ) and N (VB ) is finite dimensional, then even a more trivial
common cause can be given that is not sensitive to the given algebra elements A
and B either. Namely, any decomposition of the unit into minimal projections of the
corresponding finite dimensional algebra,1 that is, any maximal (atomic) decompo-
sition of the unit, provides a weak common cause solution of (5.1) irrespective of
the chosen events in N (V A ) and N (VB ), and irrespective of the correlating state φ
on them (Cavalcanti and Lal 2014). Therefore these trivial, maximal size solutions
reflect only the structure of the underlying finite-dimensional local algebras.
What is the status of these six different notions of the Common Cause Principle in
AQFT?
The question whether the Commutative Common Cause Principles are valid in
a Poincaré covariant local quantum theory in the von Neumann algebraic setting
was first raised by Rédei (1997). As an answer to this question, Rédei and Summers
(2002, 2007) have shown that the Commutative Weak CCP is valid in algebraic
quantum field theory with locally infinite degrees of freedom. Namely, in the von
Neumann setting they proved that in the case of type III local von Neumann algebras
for every locally normal and faithful state and for every superluminally correlated
pair of projections there exists a commuting common cause of size 2 in the weak

1 Of course the cardinality |K | of these (commuting or noncommuting) common causes is uniquely


determined by the finite-dimensional algebra: |K | = r nr if the finite-dimensional algebra is
isomorphic to the finite direct sum of full matrix algebras, ⊕r Mnr .
5 The Common Cause Principle 39

past of the correlated projections. They have also shown (Rédei and Summers 2002,
p. 352) that the localization
 of a common
 cause C < AB cannot be restricted
to J− (V A ) ∪ J− (VB ) \ J− (V A ) or J− (V A ) ∪ J− (VB ) \ J− (VB ) due to logical
independence of spacelike separated algebras.
Concerning the Commutative (Strong) CCP less is known. If one also admits
projections localized only in unbounded regions, then the Strong CCP is known
to be false: von Neumann algebras pertaining to complementary wedges contain
correlated projections but the strong past of such wedges is empty (see Summers
and Werner 1988 and Summers 1990). In spacetimes having horizons, for example,
those with the Robertson–Walker metric, the common past of spacelike separated
bounded regions can be empty, yet there exist states that provide correlations among
local algebras corresponding to these regions (Wald 1992).2 Hence, CCP is not
valid there. In the case of local algebras in Minkowski spaces there is no definite
statement. We are of the opinion that one cannot decide on the validity of the (Strong)
CCP without an explicit reference to the dynamics inasmuch as there is no bounded
region VC in the common past (hence neither in the strong past) for which isotony
would ensure that N (V A ∪ VB ) ⊂ N (VC

). But dynamics relates the local algebras

therefore N (V A ∪ VB ) ⊂ N (VC

+ t) = αt (N (VC

)) can be fulfilled for certain

VC ⊆ VC

⊂ J− (V A ) ∩ J− (VB ) and for certain time translation by t.

Coming back to the proof of Rédei and Summers, their statement is based on the
crucial premise that the algebras in question are von Neumann algebras of type III.
Although these algebras arise in a natural way in the context of Poincaré covariant
theories, other local quantum theories apply von Neumann algebras of other types.
For example, theories with locally finite degrees of freedom are based on finite-
dimensional (type I) local von Neumann algebras. This raises the question of whether
the commutative weak CCP is valid in these local quantum theories. To address the
problem Hofer-Szabó and Vecsernyés (2012a) have chosen the local quantum Ising
model (see Müller and Vecsernyés) having locally finite degrees of freedom. It turned
out that the Commutative Weak CCP is not valid in the local quantum Ising model
and it cannot be valid either in theories with locally finite degrees of freedom in
general.
But why should we require commutativity between the common cause and its
effects at all? Commutativity has a well-defined role in any quantum theory: observ-
ables should commute to be simultaneously measurable. In AQFT commutativity of
observables with spacelike separated supports is an axiom. To put it simply, com-
mutativity can be required for events that can happen “at the same time.” But cause
and effect are typically not these sorts of events. If one considers ordinary quantum
mechanics, one easily sees that observables do not commute even with their own
time translates in general. For example, the time translate x(t) := U (t)−1 xU (t) of
the position operator x of the harmonic oscillator in quantum mechanics does not
commute with x ≡ x(0) for generic t, because in the ground state vector ψ0 we have

2 We thank David Malament for calling our attention to this point and Wald’s paper.
40 5 The Common Cause Principle

  −i sin (ωt)

x, x(t) ψ0 = ψ0 ≡ 0. (5.2)
mω

Thus, if an observable A is not a conserved quantity, that is A(t) = A, then the

commutator [A, A(t)] = 0 in general. Therefore why should the commutators [A, C]
and [B, C] vanish for the events A, B and for their common cause C supported in
their (weak/common/strong) past? We think that commuting common causes are
only unnecessary reminiscences of their classical formulation. Due to the time delay
between the correlated events and the common cause, it is also an unreasonable
assumption.
Abandoning commutativity in the definition of the common cause is therefore a
natural move. To our knowledge the first who raised the possibility of the noncom-
muting common causes were Clifton and Ruetsche (1999) criticizing Rédei (1997)
who required the common cause to be commuting. They say: “[requiring commuta-
tivity] bars from candidacy to the post of common cause the vast majority of events
in the common past of events problematically correlated” (p. 165). And indeed, the
benefit of allowing noncommuting common causes is that the noncommutative ver-
sion of the result of Rédei and Summers can be regained: as shown in (Hofer-Szabó
and Vecsernyés 2013a, b), by allowing common causes that do not commute with the
correlated events, the Weak CCP can be proven in local UHF-type quantum theories.
We close this chapter by comparing CCPs with Bell’s local causality. First, note that
the core mathematical requirement of both principles is the screening-off conditions
(4.4) or equivalently (5.1). However, the localization of the screener-off events is
different: they are localized symmetrically in the weak/common/strong past in the
case of the CCPs and asymmetrically in the causal past of one of the correlating
events in the case of local causality. Moreover, the subjects of these conditions are
also very different: In the first case the screening-off should hold for all pairs of
algebra elements supported in the spacelike regions V A , VB ∈ K. On the contrary,
different common causes are not only allowed for different triples (A, B, φ) but also
a nontrivial dependence is expected on physical grounds. Second, in the case of local
causality the screening-off condition (4.4) is required for every atomic event (satis-
fying certain localization conditions). On the contrary, in the case of the CCP the
screener-offs are typically not atomic events, because finding a common cause for a
correlation does not mean that we have arrived at the most detailed physical descrip-
tion of the situation. It simply means that at this level of description correlations
can be causally accounted for. [For an opposite view see Uffink (1999) and Henson
(2005).] In the case of local causality, however, it is necessary for the screener-off
events to be atomic (provided the local algebras are atomic), because they express
the Cauchy data on the shielding-off region blocking any causal information from
the past. Finally, both the CCPs and Bell’s local causality imply Bell’s inequalities.
But this is the topic for Chap. 7.
References 41

References

E.G. Cavalcanti, R. Lal, On modifications of Reichenbach’s principle of common cause in light of

Bell’s theorems. J. Phys. A 47, 42 (2014)
R. Clifton, L. Ruetsche, Changing the subject: Redei on causal dependence and screening off in
relativistic quantum field theory. Philos. Sci. 66, S156–S169 (1999)
J. Henson, Comparing causality principles. Stud. Hist. Phil. Mod. Phys. 36, 519–543 (2005)
G. Hofer-Szabó, P. Vecsernyés, Reichenbach’s common cause principle in AQFT with locally finite
degrees of freedom. Found. Phys. 42, 241–255 (2012a)
G. Hofer-Szabó, P. Vecsernyés, Noncommuting local common causes for correlations violating the
Clauser-Horne inequality. J. Math. Phys. 53, 12230 (2012b)
G. Hofer-Szabó, P. Vecsernyés, Noncommutative common cause principles in AQFT. J. Math. Phys.
54, 042301 (2013a)
G. Hofer-Szabó, P. Vecsernyés, Bell inequality and common causal explanation in algebraic quantum
field theory. Stud. Hist. Phil. Mod. Phys. 44(4), 404–416 (2013b)
G. Hofer-Szabó, M. Rédei, L.E. Szabó, The Principle of the Common Cause (Cambridge University
Press, Cambridge, 2013)
V.F. Müller, P. Vecsernyés, ”The phase structure of G-spin models”, to be published
M. Rédei, Reichenbach’s common cause principle and quantum field theory. Found. Phys. 27,
1309–1321 (1997)
M. Rédei, J.S. Summers, Local primitive causality and the Common Cause Principle in quantum
field theory. Found. Phys. 32, 335–355 (2002)
M. Rédei, S.J. Summers, Remarks on causality in relativistic quantum field theory. Int. J. Theor.
Phys. 46, 2053–2062 (2007)
H. Reichenbach, The Direction of Time (University of California Press, Los Angeles, 1956)
S.J. Summers, R. Werner, Maximal violation of Bell’s inequalities for algebras of observables in
tangent spacetime regions. Ann. Inst. Henri Poincaré Phys. Théor. 49, 215–243 (1988)
S.J. Summers, On the independence of local algebras in quantum field theory. Rev. Math. Phys. 2,
201–247 (1990)
J. Uffink, The principle of the common cause the Bernstein Paradox. Philos. Sci. 512–525 (1999)
R.M. Wald, Correlations beyond the horizon. Gen. Rel. Gravit. 24, 1111–1116 (1992)
Chapter 6
Bell’s Inequalities

Abstract This Chapter collects the most important concepts and some of the rep-
resentative propositions concerning Bell’s inequalities in the general C ∗ -algebraic
setting and in the special LPT framework.

Keywords Separable states · Clauser-Horne inequality

There is a notion that is even more tightly linked to Bell’s name than local causality,
namely Bell’s inequalities. In this chapter we collect the most important concepts
and some of the representative propositions concerning Bell’s inequality in LPTs
[see Summers (1990) and Halvorson (2007)]. We start with the general C ∗ -algebraic
setting and then go over to the special algebraic quantum field-theoretical formula-
tion.
In the general C ∗ -algebraic setting Bell’s inequality is treated in the following way.
Let A and B be two mutually commuting unital C ∗ -subalgebras of a C ∗ -algebra C.
A Bell operator R for the pair (A, B) is an element of the set:

1 
B(A, B) := X 1 (Y1 + Y2 ) + X 2 (Y1 − Y2 )  X i = X i∗ ∈ A; Yi = Yi∗ ∈ B;
2

− 1  X i , Yi  1

where 1 is the unit element of C. For any Bell operator R the following can be proven.
√
Theorem 1 For any state φ : C → C, one has |φ(R)|  2.

Theorem 2 For separable states (i.e., for convex combinations of product states)
|φ(R)|  1.
44 6 Bell’s Inequalities

The Bell correlation coefficient of a state φ is defined as

  
β(φ, A, B) := sup |φ(R)|  R ∈ B(A, B)

√ is said to be violated if β(φ, A, B) > 1, and maximally violated

and Bell’s inequality
if β(φ, A, B) = 2. An important result of Bacciagaluppi (1994) is:

Theorem 3 If A and B are C ∗ -algebras, then there are some states violating Bell’s
inequality for A ⊗ B iff both A and B are non-abelian.

Going over to von Neumann algebras Landau (1987) has shown that the maximal
violation of Bell’s inequality is generic in the following sense.
Theorem 4 Let N1 and N2 be von Neumann algebras, and suppose that N1 is
abelian and N1 ⊆ N2 (N  being the commutant of N ). Then for any state
β(φ, N1 , N2 )  1. On the other hand, if both N1 and N2 are non-abelian von
Neumann algebras such that N1 ⊆ N2 , and if (N1 , N2 ) √
satisfies the Schlieder prop-
erty, then there is a state φ for which β(φ, N1 , N2 ) = 2.
1

Adding further constraints on the von Neumann algebras one obtains other important
results such as the following two.

Theorem 5 If N1 and N2 are properly infinite2 von Neumann algebras on the Hilbert
space H such that N1 ⊆ N2 , and (N1 , N2 ) satisfies the Schlieder property, then there
is a dense set of vectors in H inducing states that violate Bell’s inequality across
(N1 , N2 ) (Clifton and Halvorson 2001).

Theorem 6 Let H be a separable Hilbert space and let R be a von Neumann factor
of type III1 acting on H. Then every normal state φ of B(H) maximally violates
Bell’s inequality across (R, R ) (Summers and Werner 1988).

Type III factors featuring in Theorems 5 to 6 are the typical local von Neumann
algebras in AQFT with locally infinite degrees of freedom.
The Bell inequality typically used in AQFT is of the form:
  
φ X 1 (Y1 + Y2 ) + X 1 (Y1 − Y2 )   2, (6.1)

where X m ∈ N (V A ) and Yn ∈ N (VB ) are self-adjoint contractions (i.e., −1 

X m , Yn  1 for m, n = 1, 2) supported in spatially separated spacetime regions
V A and VB , respectively. This type of Bell’s inequality is usually referred to as the
Clauser–Horne–Shimony–Holte (CHSH) inequality (Clauser et al. 1969).

1 The commuting pair (A, B ) of C ∗ -subalgebras in C obeys the Schlieder property, if 0

= A ∈ A and
0 = B ∈ B, then AB = 0. Because in the case of von Neumann algebras A and B can be required
to be projections, the Schlieder property is the analogue of logical independence in classical logic.
2 The center contains no finite projections.
6 Bell’s Inequalities 45

Sometimes another Bell-type inequality is used in the EPR-Bell literature instead

of (6.1): the Clauser–Horne (CH) inequality (Clauser and Horne 1974) defined in
the following way.

− 1  φ(A1 B1 + A1 B2 + A2 B1 − A2 B2 − A1 − B1 )  0, (6.2)

where Am and Bn are projections located in N (V A ) and N (VB ), respectively. It is

easy to see, however, that the two inequalities are equivalent: in a given state φ the
set {(Am , Bn ); m, n = 1, 2} violates the CH inequality (6.2) if and only if the set
{(X m , Yn ); m, n = 1, 2} of self-adjoint contractions given by

X m := 2 Am − 1 (6.3)
Yn := 2Bn − 1 (6.4)

violates the CHSH inequality (6.1). Therefore, the two types of Bell’s inequalities
can be used ad libitum.

References

G. Bacciagaluppi, Separation theorems and Bell inequalities in algebraic QM, in Symposium on

the Foundations of Modern Physics 1993: Quantum Measurement, Irreversibility and Physics of
Information, (World Scientific, 1994) pp. 29–37
J.F. Clauser, M.A. Horne, A. Shimony and R.A. Holt, Proposed experiment to test local hidden-
variable theories, Phys. Rev. Lett. 23, 880–884 (1969)
J.F. Clauser and M.A. Horne, Experimental consequences of objective local theories, Phys. Rev. D.
10, 526–535 (1974)
R. Clifton and H. Halvorson, Entanglement and open systems in algebraic quantum field theory,
Stud. Hist. Phil. Mod. Phys. 32 (1), 1–31 (2001)
H. Halvorson, Algebraic quantum field theory, in Philos. Phys., vol. I, ed. by J. Butterfield, J. Earman
(Elsevier, Amsterdam, 2007), pp. 731–922
L. Landau, On the violation of Bell’s inequality in quantum theory, Phys. Lett. A. 120, 54–56 (1987)
S.J. Summers, R. Werner, Maximal violation of Bell’s inequalities for algebras of observables in
tangent spacetime regions. Ann. Inst. Henri Poincaré Phys. Théor. 49, 215–243 (1988)
S.J. Summers, On the independence of local algebras in quantum field theory. Rev. Math. Phys. 2,
201–247 (1990)
Chapter 7
The EPR Scenario

Abstract In this Chapter we outline the Einstein–Podolsky–Rosen (EPR) scenario

and show that the violation of Bell’s inequalities does not block the implementation
of the EPR situation in a locally causal LQT, neither it excludes a noncommuting
common causal explanation for the EPR correlations.

Keywords EPR scenario · Common causal explanation · Clauser-Horne

inequality

Let us start with a succint description of the EPR scenario. Consider a pair of spin- 21
particles prepared in the singlet state φs (see Fig. 7.1). Measure the spin of the left
particle in directions am (m = 1, 2) and the spin of the right particle in directions bn
(n = 1, 2). Quantum mechanics yields the following probabilities for the outcomes

1
φs (Am ) = (7.1)
2
1
φs (Bn ) = (7.2)
2  
1 θam bn
φs (Am Bn ) = sin2 (7.3)
2 2

where Am and Bn denote those projections of the tensor product matrix algebra
M2 (C) ⊗ M2 (C) that are spectral projections with eigenvalue + 21 of the spin oper-
ators associated with directions am and bn , respectively, and θam bn denotes the angle
between directions am and bn .
In Chap. 8 we show that the EPR scenario can be represented in the LPT framework
where all the projections are localized in a well-defined spacetime region and there
is a state on the LPT which yields the above probabilities [see Hofer-Szabó and
Vecsernyés (2013)]. From this fact, however, it follows neither that this LPT will
48 7 The EPR Scenario

Fig. 7.1 EPR scenario for spin- 21 particles

be locally causal nor that there will be a common causal explanation for the EPR
correlations in this LPT. Hence, one can raise the following questions.
Question 1: Can the EPR scenario be implemented in a LPT that is locally causal?
Question 2: Can the EPR scenario be implemented in a LPT such that there exists
a (weak/strong) common causal explanation of the EPR scenario?
The standard answer to both questions is no. The brief argument is this: both lo-
cal causality and also the common causal explanation imply Bell’s inequalities that
are violated for certain measurement settings in the EPR scenario. Hence the imple-
mentability of the EPR scenario in a LPT that is locally causal or provides a common
causal explanation for the correlations is impossible.
However, as we shortly show, the above reasoning crucially hinges on the assump-
tion of commutativity; that is, the common cause accounting for the correlations is
commuting and the LPT implementing the correlations is classical, that is, a LCT.
It turns out that in the general noncommutative case Bell’s inequalities cannot be
derived, hence, their violation blocks neither a noncommutative common causal ex-
planation nor the implementation of the EPR scenario in a noncommutative LPT,
that is, into a LQT.
The two reasonings relating local causality and the CCPs to Bell’s inequalities are
parallel. Hence, we consider here only the first. The following proposition shows that
a locally causal LPT does not necessarily imply Bell’s inequalities. For an analogous
proposition relating the common causal explanation to Bell’s inequalities see (Hofer-
Szabó and Vecsernyés 2013; Proposition 2).
Proposition 2 Let {N (V ), V ∈ K} be a locally causal LPT with atomic (type I)
local von Neumann algebras. Let A1 , A2 ∈ A(V A ) and B1 , B2 ∈ A(VB ) be four pro-
jections localized in spacelike separated spacetime regions V A and VB , respectively,
that are correlated pairwise in the locally faithful state φ:

φ(Am Bn ) = φ(Am ) φ(Bn ) (7.4)

for any m, n = 1, 2. Let VC ∈ K be a region satisfying requirements L 1 , L 2 , and

L 3Q /L C3 in Definition 2 of local causality and let {Ck }k∈K ⊂ N (VC ) be a maximal
7 The EPR Scenario 49

partition of the unit containing mutually orthogonal atomic projections. Then the
Clauser–Horne inequality

− 1  (φ ◦ T{Ck } )(A1 B1 + A1 B2 + A2 B1 − A2 B2 − A1 − B1 )  0. (7.5)

holds for the state φ ◦ T{Ck } . If {Ck } commutes with A1 , A2 , B1 , and B2 , then the
Clauser–Horne inequality holds for the original state φ:

− 1  φ(A1 B1 + A1 B2 + A2 B1 − A2 B2 − A1 − B1 )  0. (7.6)

Proof It is an elementary fact of arithmetic that for any α, α  , β, β  ∈ [0, 1] the

number

αβ + αβ  + α  β − α  β  − α − β (7.7)

lies in the interval [−1, 0]. Now let α, α  , β, β  be the conditional probabilities:

α := φCk (Am ) (7.8)


α := φCk (Am  ) (7.9)
β := φCk (Bn ) (7.10)
β  := φCk (Bn  ) (7.11)

Plugging (7.8)–(7.11) into the (7.7) and using that the atomic partition {Ck }k∈K
screens off all correlations, that is,

φCk (Am Bn ) = φCk (Am ) φCk (Bn ) (7.12)

we get

− 1  φCk (Am Bn ) + φCk (Am Bn  ) + φCk (Am  Bn )

−φCk (Am  Bn  ) − φCk (Am ) − φCk (Bn )  0. (7.13)

Multiplying the above inequality by φ(Ck ), using (3.8), that is,

(φ ◦ T{Ck } )(XCk )
φCk (X ) = (7.14)
φ(Ck )

and summing up for the index k one obtains (7.5). If {Ck }k∈K is a commuting joint
common cause, then T{Ck } drops out from the above expression. Therefore (7.5)
becomes identical to (7.6), which completes the proof. 

The moral of Proposition 2 is the following. Bell’s inequalities can be derived in a

locally causal LPT only for a modified state φ ◦ T{Ck } in general. It can be derived for
the original state φ if the set of atomic projections {Ck } localized in VC commutes
50 7 The EPR Scenario

with A1 , A2 , B1 , and B2 . Clearly, if the LPT is classical, the elements taken from any
local algebra will commute, therefore Bell’s inequalities hold for the original state
φ in LCTs. However, going over to locally causal LQTs, commutation of {Ck } with
the correlated events is not guaranteed. If VC is spatially separated from VB (ensured
by requirement L 3Q but not L C3 ), then {Ck } will commute with B1 and B2 due to
microcausality; hence (4.4) will be satisfied, even if the B1 and B2 do not commute.
However, in the case of local primitive causality one cannot pick a maximal partition
of unit {Ck } in N (VC ) (which is needed for the states φCk to be pure on N (VC ))
such that {Ck } commutes also with projections A1 and A2 , if [A1 , A2 ] = 0. Namely,
N (V A ) ⊂ N (VC ) = N (VC ) due to isotony and local primitive causality, and the
image T{Ck } (N (VC )) is a maximal abelian subalgebra of N (VC ) containing exactly
those elements that commute with {Ck }. Hence, in order to commute with {Ck },
both A1 and A2 should be contained in T{Ck } (N (VC )), which cannot be the case, if
[A1 , A2 ] = 0.
The conclusion is that in the case of noncommuting projections A1 and A2 the
theorem of total probability, k φ(Ck Am Ck ) = φ(Am ), will not hold for the original
state1 φ at least for one of the projections A1 and A2 . This fact blocks the derivation
of Bell’s inequalities for the original state φ. (For the details see (Hofer-Szabó and
Vecsernyés 2013, p. 410.)) In short, Bell’s inequalities can be derived in a locally
primitive causal LQT with atomic von Neumann algebras, hence in a locally causal
LQT, if the projections supported on one of the regions do commute. However, this
is exactly what one expects relying on the first part of Theorem 4.
Coming back to Question 1 raised above, namely how a local quantum theory can
be locally causal in the face of Bell’s inequalities, we have arrived at the following
answer. Bell’s inequalities can be derived from local causality if the “beables” of the
local theory are represented by commutative local algebras. This fact is completely
analogous with the relation shown in (Hofer-Szabó and Vecsernyés 2013): Bell’s
inequalities can be derived from a (joint, non-conspiratorial, local) common cause
if it is a commuting common cause. Thus, contrary to common wisdom, the answer
to both Questions 1 and 2 is a qualified yes. The violation of Bell’s inequalities does
not block the implementation of the EPR situation in a LPT that is locally causal
or provides a common causal explanation for the EPR correlations. Both common
causal explanation and local causality are more general notions than what is captured
by Bell’s inequalities.

Reference

G. Hofer-Szabó, P. Vecsernyés, Bell inequality and common causal explanation in algebraic quantum
field theory. Stud. Hist. Phil. Mod. Phys. 44(4), 404–416 (2013)

holds only for the state φ{Ck } for which φ{Ck } (Am ) = φ(Am ) at least for one of the projections
1 It

A1 and A2 .
Chapter 8
A Noncommutative Locally Causal Model
for the EPR Scenario

Abstract In this Chapter we explicitly construct a locally causal LQT for the EPR
scenario. The model is the 1+1 dimensional local quantum Ising model (Müller and
Vecsernyés). We show that the model is not only locally causal in Bell’s sense but
also able to implement four pairs of events correlating in the singlet state.

Keywords Discretized Minkowski spacetime · Quantum Ising model · Singlet

state

In the previous chapter it was shown that the violation of Bell’s inequalities does
not prohibit the implementation of the EPR situation in a locally causal LPT. Here
we explicitly construct such a model. Consider a discretized version of the two-
dimensional Minkowski spacetime M2 composed of minimal double cones Om (t, i)
of unit diameter with their center in (t, i) for t, i ∈ Z or t, i ∈ Z + 1/2. The set
{Oim , i ∈ 21 Z} of such minimal double cones with t = 0, −1/2 defines a “thick-
ened” Cauchy surface in this spacetime (see Fig. 8.1). The double cone Oi,m j stuck
to this Cauchy surface is defined to be the smallest double cone containing both
Oim and Omj : Oi,m j := Oim ∨ Omj . Similarly, let Om (t, i; s, j) := Om (t, i) ∨ Om (s, j).
The directed partially ordered set (with respect to inclusion) of such double cones
is denoted by Km . The directed subset of Km whose elements are stuck to a Cauchy
surface is denoted by KCm S . Obviously, KCm S is invariant with respect to integer space
translations and Km is invariant with respect to integer space and time translations.
The net of local algebras is defined as follows. The “one-point” observable
algebras associated with the minimal double cones Oim , i ∈ 21 Z are defined to be
A(Oim )  M1 (C) ⊕ M1 (C). For the unitary self-adjoint generators Ui ∈ A(Oim ) of
these local algebras one demands the commutation and anticommutation relations:

−U j Ui , if |i − j| = 21 ,
Ui U j = (8.1)
U j Ui , otherwise.
52 8 A Noncommutative Locally Causal Model for the EPR Scenario

Fig. 8.1 A thickened

Cauchy surface in the two
dimensional Minkowski
space M2

m m m m m
O−2 O−1 O0 O1 O2
m m m m
O−3/2 O−1/2 O1/2 O3/2

The local algebras A(Oi, j ), Oi, j ∈ KCm S are linearly spanned by the monoms

ki+ 1 k j− 1 k
Uiki Ui+ 12 . . . U j− 12 U j j (8.2)
2 2

where ki , ki+ 21 . . . k j− 21 , k j ∈ {0, 1}.1

Because the local algebras A(Oi,i− 21 +n ), i ∈ 21 Z for n ∈ N are isomorphic to
the full matrix algebra M2n (C), the quasilocal observable algebra A is a uniformly
hyperfinite (UHF) C ∗ -algebra and consequently there exists a unique (nondegener-
ate) normalized trace Tr : A → C on it. We note that all nontrivial monoms in (8.2)
have zero trace.
In order to extend the “Cauchy surface net” {A(O), O ∈ KCm S } to the net {A(O),
O ∈ Km } one has to classify the integer-valued time evolutions given by a group
{βt , t ∈ Z} of automorphisms of A (the quasilocal algebra of a Cauchy surface net)
that are covariant and causal, and commute with integer space translation automor-
phisms of A. The classification was given in (Müller and Vecsernyés) and it was also
shown that the extended net satisfies isotony, microcausality, algebraic Haag duality,
Z × Z covariance with respect to integer time and space translations, and primitive
causality:

A(V ) = A(V  ), (8.3)

where V is a finite connected piece of a thickened Cauchy surface (composed of

minimal double cones from KCm S ). The double spacelike complement of V is denoted
by V  , which is the smallest double cone in Km containing V .
Because {A(O), O ∈ Km } is an atomic LPT obeying local primitive causality,
Proposition 1 in Chap. 4 implies that the net {A(O), O ∈ Km } is locally causal.

1 Fora detailed Hopf algebraic description of the local quantum spin models see Szlachányi and
Vecsernyés (1993), Nill and Szlachányi (1997), Müller and Vecsernyés.
8 A Noncommutative Locally Causal Model for the EPR Scenario 53

Next we show that the EPR scenario can be implemented in our 1+1-dimensional
Ising model. Consider the following group of causal time translation automorphisms
of the net {A(O), O ∈ KCm S } given by its generator β ≡ β1 .

1
β(Ux ) = Ux− 21 Ux Ux+ 21 , x ∈ Z + . (8.4)
2
(In our following example we need not even specify the particular choice for
β(Ux ), x ∈ Z from the allowed ones.) Let us consider the double cones O A :=
Om (0, −1) ∪ Om ( 21 , − 21 ), O B := Om ( 21 , 21 ) ∪ Om (0, 1), and the “two-point” alge-
bras A(O A ), A(O B ) pertaining to them. (See Fig. 8.2.) A linear basis of the algebra
A(O A ) is given by the monoms

1, U−1 , β(U− 21 ) ≡ U−1 U− 21 U0 , iU−1 β(U− 21 ) ≡ iU− 21 U0 (8.5)

(where i in the fourth monom is the imaginary unit). They satisfy the same commu-
tation relations as the Pauli matrices σ0 = 1, σx , σ y , and σz in M2 (C).
Therefore, introducing the notation

U := (U−1 , U−1 U− 21 U0 , iU− 21 U0 ), (8.6)

any minimal projection in A(O A ) can be parametrized as

1
A(a) := (1 + aU) (8.7)
2

where a = (a1 , a2 , a3 ) is a unit vector in R3 . In the same vein, any minimal projection
in A(O B ) can be parametrized as

1
B(b) := (1 + bV) , (8.8)
2

Fig. 8.2 Projections in

A(O A ) and A(O B )

Am Bn
54 8 A Noncommutative Locally Causal Model for the EPR Scenario

where

V := (U1 , −U0 U 21 U1 , iU0 U 21 ) (8.9)

is the triple composed of the generators of A(O B ) and b = (b1 , b2 , b3 ) is a unit vector
in R3 . The projections A(a) and B(b) can be interpreted as the event localized in
A(O A ) and A(O B ), respectively, pertaining to the generalized spin measurement in
direction a and b, respectively.
Now, consider two projections Am := A(am ); m = 1, 2 localized in O A , and two
other projections Bn := B(bn ); n = 1, 2 localized in the spacelike separated double
cone O B . A faithful state φ : A → C on our system can be given in terms of an
invertible density matrix ρ ∈ A: φ( · ) = T r (ρ · ). We choose a one-parameter family
of such states
 
ρ(λ) := 1 + λ U−1 U− 21 U 21 U1 − U−1 U1 + U− 21 U 21 , λ ∈ [0, 1). (8.10)

For λ = 1 the state defined by (8.10) is not faithful on A (because the corresponding ρ
is not invertible) but leads to the usual singlet state on A(O A ) ∨ A(O B )  M2 (C) ⊗
M2 (C). It is easy to see that the correlation between Am and Bn in the state (8.10)
will be:
λ  m n
corr (Am , Bn ) := φ(Am Bn ) − φ(Am ) φ(Bn ) = − a ,b (8.11)
4

where  , is the scalar product in R3 . In other words, Am and Bn will correlate

whenever am and bn are not orthogonal. Now, if am and bn are chosen as

a1 = (0, 1, 0) (8.12)
a2 = (1, 0, 0) (8.13)
1
b1 = √ (1, 1, 0) (8.14)
2
1
b2 = √ (−1, 1, 0) (8.15)
2

the CH inequality (6.2) will be violated at the lower bound because


φ(A1 B1 + A1 B2 + A2 B1 − A2 B2 − A1 − B1
√
1 λ  1 1   1 2   2 1   2 2  1+λ 2
=− − a ,b + a ,b + a ,b − a ,b = − , (8.16)
2 4 2
8 A Noncommutative Locally Causal Model for the EPR Scenario 55

which is smaller than −1 if λ > √1 .

2
Or, equivalently, the CHSH inequality (6.1)
where

X m := 2 Am − 1 (8.17)
Yn := 2Bn − 1 (8.18)

will be violated for the above setting because

φ(X 1 (Y1 + Y2 ) + X 1 (Y1 − Y2 ))

    √
= −λ a1 , b1 + b2 + a2 , b1 − b2 = −λ2 2 (8.19)

is smaller than −2 if λ > √12 . Both the CH and the CHSH inequality are maximally
violated for the singlet state, that is, if λ = 1.
To sum up, we have constructed a locally causal LPT in which the EPR scenario
can be implemented. That is, we localized two pairs of noncommuting projections
in two spatially separated regions such that the projections represented different spin
outcomes in certain directions. The singlet state of the model provided just the EPR
statistics of these outcomes and the CH inequalities were violated. In short, we have
provided a locally causal LPT for the EPR violating Bell’s inequalities.
Note again that the above model does not stand in contradiction to the common
wisdom, namely that the violation of Bell’s inequalities excludes a local realistic
model for the EPR. The net we have provided is a LQT, a local quantum theory
and not a LCT, a local classical theory. All we wanted to show that we can preserve
local causality at the price of abandoning classicality. Noncommutative locally causal
models are not excluded by the violation of Bell’s inequalities.
Of course, it is another and highly nontrivial philosophical question as to what is
the correct interpretation of a noncommutative model. Noncommuting projections
are as a standard interpreted operationally: they represent the outcomes of incom-
mensurable measurements. In order to provide a locally causal explanation of the
EPR scenario in the Bellian sense one should interpret these noncommuting observ-
ables as beables representing some independently existing properties of the system.
Our book leaves open the question as to whether such a noncommutative ontology
can be consistently developed.
The Ising model sketched above also provides a common causal explanation for
the EPR correlations. In Hofer-Szabó and Vecsernyés (2012b, Proposition 1) we
have proven that the above four correlations violating the CH inequality (6.2) have a
strong common cause localized in the region shown in Fig. 8.3. This result positively
answers Question 2 raised in the previous chapter as to whether there is a LPT
providing a common causal explanation for the EPR correlations violating Bell’s
inequalities.
56 8 A Noncommutative Locally Causal Model for the EPR Scenario

Fig. 8.3 Localization of a

common cause for the
correlations {(Am , Bn )}

Am Bn
C

References

G. Hofer-Szabó, P. Vecsernyés, Noncommuting local common causes for correlations violating the
Clauser-Horne inequality. J. Math. Phys. 53, 12230 (2012b)
V.F. Müller, P. Vecsernyés, The phase structure of G-spin models, to be published
F. Nill, K. Szlachányi, Quantum chains of Hopf algebras with quantum double cosymmetry. Com-
mun. Math. Phys. 187, 159–200 (1997)
K. Szlachányi, P. Vecsernyés, Quantum symmetry and braid group statistics in G-spin models.
Commun. Math. Phys. 156, 127–168 (1993)
Chapter 9
Summary and Outlook

Abstract In this Chapter we summarize the main results of this book and inves-
tigate their philosophical consequences. We argue that embracing noncommuting
“beables” in our ontology significantly extends our explanatory sources in account-
ing for correlations. We also examine what price we need to pay for abandoning
classicality in order to preserve local causality.

Keywords Noncommuting beables · Noncommutative ontology

In this book we aimed to implement Bell’s notion of local causality in a framework,

called local physical theory, that integrates probabilistic and spatiotemporal concepts
in a common conceptual scheme. After giving a brief overview of the various locality
and causality principles we provided a clear-cut definition of local causality. This
definition transcended Bell’s original intuition in the sense that it also incorporated
noncommutative “beables.” Having formulated local causality we gave sufficient
conditions for a local physical theory to be locally causal: a theory will be locally
causal if local primitive causality holds and the local von Neumann algebras are
atomic (Proposition 1). Then we compared Bell’s local causality with the various
Common Cause Principles, overviewed the main concepts and theorems concerning
Bell’s inequalities in a local physical theory, and briefly outlined the EPR scenario.
We found a nice parallelism here: Bell’s inequalities cannot be derived either from
local causality or from a common cause unless the local physical theory is classical or
the common cause is commuting, respectively (Proposition 2). Finally, we explicitly
constructed a simple local quantum theory implementing the EPR scenario that was
locally causal and provided a common causal explanation for the EPR correlations.
All our results point in a common direction, namely towards noncommutativity.
It turned out that the violation of Bell’s inequalities does not exclude either a locally
causal local physical theory or a local physical theory providing a common causal
explanation for the correlations. Therefore Bell’s inequalities in the nonclassical case
do not play the same role as in the classical one. In the classical case there was a direct
logical link between the local causality/common causal explanation and the validity
58 9 Summary and Outlook

of Bell’s inequalities. Thus the violation of Bell’s inequalities excludes this and only
this subset of LPTs as locally causal models/possible common causal explanations.
To put it differently, taking seriously the noncommutative type of description where
events are represented by not necessarily commuting projections, one can provide
a locally causal account/common causal explanation in a much wider range than
simply sticking to commutative events.
However, it is one thing to construct a noncommutative locally causal mathemat-
ical model for the EPR and another thing to provide a reasonable physical interpre-
tation for the model. What does it mean, for example, that a certain correlation has
a noncommuting common cause? If the common cause {Ck } is noncommuting, then
the theorem of total probability φ(X ) = φ{Ck } (X ) does not hold for X = A, B, and
AB. This fact is a straight consequence of noncommutativity: specifying a state on
a maximal commutative subalgebra does not specify it on the full noncommutative
algebra. The violation of the theorem of total probability means that the original
probability of the correlated events cannot be reconstructed from the probabilities
conditioned on the common cause in the sense of

φ(X ) = φCk (X )φ(Ck )
k

Thus the common causal explanation cannot be regarded as the description of the
same physical situation at a finer level which is the traditional understanding of the
common causal explanation.
Similarly, in a locally causal LQT the atomic partitions in a shielder-off region
screen off the correlation, but if the partition does not commute with the correlated
events then the probability of these latter cannot be reconstructed from those of
the former. Thus the screener-off partitions cannot be regarded as a more detailed
description of the correlations in question.
One can react to this fact in different ways. One reaction is to say that the violation
of the theorem of total probability completely ruins the program of going noncom-
mutative and so to preserve local causality and the common causal explanation [see
Cavalcanti and Lal (2014)]. Another reaction would be, however, the following.
Observe that the definition of the common cause does not contain the requirement
(which our classically informed intuition would dictate) that the conditional proba-
bilities, when added up, should give back the unconditional probabilities. To put it
in a more formal way: the theorem of total probability is not part of the definition
of the common cause. The defining property of the common cause is simply the
screening-off. That is, common causes might not be measured without the distor-
tion of the statistics of the original correlated events. But this fact is ubiquitous for
noncommuting observables in quantum mechanics. If we tolerate this fact in general,
then why not tolerate it for common causes? As we have seen, allowing noncommut-
ing common causes can help us maintain Bell’s original intuition concerning local
causality.
An analogy might help here. Reichenbach’s original definition of the common
cause was somewhat different from the one used in this book. It contained the
9 Summary and Outlook 59

screener-off conditions but also contained some extra requirements. It seems that the
prime reason for Reichenbach to add these extra requirements [the so-called positive
statistical relevancy condition, see Reichenbach (1956, p. 159)] to the screener-off
conditions was that these conditions together logically implied the correlation. In this
sense the Reichenbachian common cause provided a Hempelian explanation for the
correlation: if the common cause were present, the correlation logically followed.
However, as time passed it turned out that these extra conditions are just unnecessary;
they do not form part of the intuition of what a common cause is. Therefore these
extra conditions have been completely dropped from the definition of the common
cause in the literature. [For an attempt to define the notion of the common cause such
that it preserves this deductive relation see Hofer-Szabó and Rédei (2004, 2006).]
We do not claim that abandoning commutativity in the definition of the common
cause or local causality is just as unproblematic as abandoning positive statistical
relevancy in Reichenbach’s definition of the common cause. But we do claim that
embracing noncommuting events substantially extends our explanatory sources in
accounting for correlations in a locally causal manner. We are fully aware that adopt-
ing noncommuting beables in our ontology is a high price to pay for preserving local
causality. This book is an invitation for philosophers to explore the prospects of such
a noncommutative ontology.

References

E.G. Cavalcanti, R. Lal, On modifications of Reichenbach’s principle of common cause in light of

Bell’s theorems. J. Phys. A 47, 42 (2014)
G. Hofer-Szabó, G., M. Rédei, Reichenbachian common cause systems. Int. J. Theor. Phys. 34,
1819–1826 (2004)
G. Hofer-Szabó, G. M. Rédei, Reichenbachian common cause systems of arbitrary finite size exist.
Found. Phys. 35, 745–746 (2006)
H. Reichenbach, The Direction of Time (University of California Press, Los Angeles, 1956)