On Reichenbach's Common Cause Principle and Reichenbach's Notion of Common Cause

The British Society for the Philosophy of Science
On Reichenbach's Common Cause Principle and Reichenbach's Notion of Common Cause

Author(s): Gábor Hofer-Szabó, Miklós Rédei and László E. Szabó
Source: The British Journal for the Philosophy of Science, Vol. 50, No. 3 (Sep., 1999), pp.
377-399
Published by: Oxford University Press on behalf of The British Society for the
Philosophy of Science
Stable URL: http://www.jstor.org/stable/40072237
Accessed: 15-02-2018 17:13 UTC
REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/40072237?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
Oxford University Press, The British Society for the Philosophy of Science are
collaborating with JSTOR to digitize, preserve and extend access to The British Journal for
the Philosophy of Science
This content downloaded from 130.237.165.40 on Thu, 15 Feb 2018 17:13:29 UTC
All use subject to http://about.jstor.org/terms
Brit. J. Phil. Sci. 50 (1999), 377-399
On Reichenbach's Common Cause

Principle and Reichenbach's
Notion of Common Cause
Gabor Hofer-Szabo, Miklos Redei and
Laszlo E. Szabo
ABSTRACT
It is shown that, given any finite set of pairs of random events in a Boolea
which are correlated with respect to a fixed probability measure on the alg
algebra can be extended in such a way that the extension contains events th
regarded as common causes of the correlations in the sense of Reichenbach's
of common cause. It is shown, further, that, given any quantum probability
any set of commuting events in it which are correlated with respect to a fixed
state, the quantum probability space can be extended in such a way that the
contains common causes of all the selected correlations, where common caus
taken in the sense of Reichenbach's definition. It is argued that these re
strongly restrict the possible ways of disproving Reichenbach's Comm
Principle.
1 Informal statement of the problem

2 Reichenbach's notion of common cause
3 Common cause completeabilty: the classical case
4 Common cause completeability: the quantum case
5 Reichenbach's Common Cause Principle and common cause
completeability
6 Open questions
1 Informal statement of the problem

Reichenbach's Common Cause Principle and the mathematically explicit
notion of common cause formulated in terms of random events and their
probabilities goes back to Reichenbach's ([1956], Section 19) book. Both
the Common Cause Principle and the related concept of common cause have
been subjects of investigations in a number of works, especially in the papers
by Salmon ([1978], [1980], [1984]), by van Fraassen ([1977], [1982], [1989]),
© Oxford University Press 1999
378 Gdbor Hofer-Szabo, Miklos Redei and Ldszlo E. Szabo
by Suppes and Zanotti ([1970], [1981]), by Cartwright ([1987]) and by Spohn

([1991]). It seems that there is no general consensus as regards the status of the
Common Cause Principle and its relation to the notion of common cause. We
do not wish to evaluate here the relative merits of the different analyses of the
Principle and its relation to the common cause notion; nor do we aim at a
critical comparison of the different suggestions as to how best to specify a
technical notion of common cause. Rather, we take the notion of common
cause as it was formulated by Reichenbach himself, and we investigate the
following problem, raised first in Redei ([1998]). Let (S,fi) be a classical
probability space, i.e. S be a Boolean algebra of random events and /x be a
probability measure on 5. Assume that {(A,-,/?,-) | i £ /} is a set of pairs
of events in 5 that are statistically correlated with respect to /x:
l*(A ,-/?,■) > /x(A/)/x(/?,) (i E /), and assume, further, that 5 does not contain
events C,- that can be considered the common causes of the correlation between
A,- and Bi9 where common cause is taken in the sense of Reichenbach' s
definition ([1956], Section 19, recalled in Section 2 below). The problem is
whether the probability space (5, /x) can in principle be enlarged in such a way
that for each correlated pair (A,, Bt) there exists a common cause C, of the
correlation in the larger probability space (S', /jl). If the probability space
(S,/x) is such that for every pair (A,, Bt) in the set of correlated pairs
{(A,,/?,) | / E /} the space (S, ix) can be enlarged in the said manner, then
we say that (5, jx) is common cause completeable with respect to the given set
of correlations (see the Definition 5 in Section 3 for a precise formulation). We
shall prove that every classical probability space (5,/x) is common cause
completeable with respect to any finite set of correlations (Proposition 2 in
Section 3). That is to say, we show that given any finite set of correlations in a
classical event structure, one can always say that the correlations are due to
some common causes, possibly 'hidden' ones, i.e. ones that are not part of the
inital set S of events.
Reichenbach' s Common Cause Principle is the claim that if there is a

correlation between two events A and B and a direct causal connection between
the correlated events is excluded then there exists a common cause of the
correlation in Reichenbach' s sense. We interpret Proposition 2 as saying that
Reichenbach' s Common Cause Principle cannot be disproved by displaying
classical probability spaces that contain a finite number of correlated events
without containing a Reichenbachian common cause of the correlations - the
only justifiable conclusion one can draw is that the event structures in question
are common cause incomplete.
Statistical correlations also make sense in non-classical probability spaces
(£, 0), where L is a non-distributive, orthomodular lattice in the place of the
Boolean algebra S, and where 0 is a generalized probability measure ('state')
defined on £, taking the place of fi. Such non-classical probability spaces
On Reichenbach 's Common Cause Principle 379
emerge in non-relativistic quantum mechanics and in relativistic quantum field

theory. In these theories £ is the non-Boolean, orthomodular von Neumann
lattice T(M) of projections of a non-commutative von Neumann algebra M
determined by the observables of the quantum system. One of the difficulties in
connection with interpreting quantum theory is the alleged impossibility of the
existence of common causes of the correlations in those quantum event
structures. Note that it is not quite obvious what one means by a common
cause in a quantum event structure because Reichenbach' s original definition of
common cause was formulated in terms of events in a classical, Kolmogorovian
probability space, and the definition also makes essential use of classical
conditional probabilities, which do not make sense in general in non-
commutative (quantum) probability spaces. In fact, the definition of
common cause in the literature analysing the problem of common cause of
quantum correlations is quite different from the Reichenbachian one: typically
the definition is formulated in terms of hidden variables rather than in terms of
events, and it also makes (more or less tacitly) the extra assumption that the
hidden variables are common common causes (see below). In this paper we wish
to retain all the features of Reichenbach' s original definition while applying it to
quantum correlations, and we do this by explicitly requiring that the common
cause commutes with the events in the correlation. Having thus obtained a
definition of common cause in quantum event structures (see Definition 6 in
Section 4) we define (T(M),<t>) to be common cause completeable with
respect to a given set {(A,,/?,) | / E /} of pairs of (commuting) correlated
events in T(M) if the probability space (T(M), </>) can be enlarged in such a
way that each of the correlations has a common cause in the enlarged non-
classical probability space (T(M'), <t>') (see the Definition 9 in Section 4 for a
precise formulation). We shall prove that every (T(M),<I>) is also common
cause completeable with respect to the complete set of elements that are
correlated in a given quantum state <t> (Proposition 3 in Section 4). Proposition
3 allows us to conclude that Reichenbach' s Common Cause Principle cannot
be disproved even by finding quantum probability spaces that contain corre-
lated events without containing a Reichenbachian common cause of the
correlations; the only justifiable conclusion one can draw is that the quantum
event structures are common cause incomplete - just like in the classical case.
This conclusion seems to be in contradiction with the standard interpretation,
according to which quantum correlations cannot have a hidden common cause.
In Section 5 we localize the reason of this 'contradiction' . The essential point we
make is that the standard arguments showing the impossibility of the existence
of a common cause of quantum correlations assume that a common cause is
a common common cause, an assumption, we claim, which is not part of
Reichenbach' s notion of common cause. In Section 6 we formulate a couple
of open questions concerning the notion of common cause completeability.
2 Reichenbach's notion of common cause
Let (S, fi) be a classical probability space, where 5 is the Boolean algebra of
events and /x is the probability measure. If the joint probability fi(A AB)ofA
and B is greater than the product of the single probabilities, i.e. if
lx{A/\B)>ix{A)ii{B) (1)
then the events A and B are said to be (positively

According to Reichenbach ([1956], Section 19
cause type explanation of a correlation like (1),
(common cause) that satisfies the conditions speci
Definition 1 C is a common cause of the corr

(independent) conditions hold:
li(AAB\Q = ii(A\Qii(B\Q (2)

M(A A B\CL) = ii{A\CL)ii{B\CL) (3)
li(A\C)>p(A\C1) (4)
fx(B\C)>fji(B\C1) (5)
where fi(X\Y) = /*C^ A Y)/fi(Y) denotes the conditional
condition Y, C1 denotes the complement of C and it is assum
probabilities fi(X), (X = A,B, C, C1) is equal to zero.
We shall occasionally refer to conditions (2)-(5) as 'Re

ditions'. Reichenbach proves the following
Theorem Conditions (2)-(5) imply (1); that is to say, if A

that they satisfy conditions (2)-(5), then there is a positiv
A and B in the sense of (1).
Some remarks and terminology:
(i) We emphasize that, from the point of view of the expla

the common cause of the correlation, each of the inde
(2)-(5) is equally important. For instance, the mere fa
satisfies (4) and (5) only, i.e. the fact that C is positi
relevant for both A and B, is not sufficient for C to
explanation of the correlation, since statistical relevan
is not sufficient to derive the correlation (1). This rema
addition to statistical relevance, we assume either fi(A
1, or C c A A B, since (3) can still fail, and again (1) c
(ii) Taking either A or B as C, the four conditions (2)-(5
means that Reichenbach's definition accomodates the case when there is
a direct causal link between the correlated events. To put this negatively:
as it stands, Reichenbach's definition does not distinguish between direct
causal influence between the correlated events and the correlation caused
by a common cause. Reichenbach' s definition also does not exclude the
following sort of common cause: C i= A (respectively C =£ B) but
(C < A or C > A) and fi(Q = /x(A) (6)
respectively
(C < B or C > B) and /*(C) = /i(fi) (7)

Such a C is a common cause in the sense of the Definition 1 but such a C
should not be regarded as a meaningful common cause because C is

identical with A (respectively B) up to an event of probability 0. If C is a
common cause such that none of (6) and (7) holds then we shall say that
C s a proper common cause, and in what follows, common cause
will always mean a proper common cause unless stated otherwise
explicitly.
(iii) It can happen that, in addition to being a probabilistic common cause, the
event C logically implies both A and B, i.e. C c A A B. If this is the case
then we call C a strong common cause. If C is a common cause such that
C<tA and C<£B then C is called a genuinely probabilistic common
cause.
(iv) A common cause C will be called deterministic if

M(A|C)=1=^|C)
M(A|C1)=0 = M(fi|C1)
Note that the notions of deterministic and genuinely probabilistic

cause are not negations of each other. There does not seem to e
straightforward relation between the notions of deterministic, genuine
abilistic and proper common causes, as we have defined them.
Next we wish to determine the restrictions imposed on the value
probabilities ji(C), fi(A\C), \x.(A\Cl), ii(B\C) and ii(B\CL) by the ass
that the correlation between A and B has C as a common cause. If we assume
that there exists a common cause C in (S,fi) of the given correlation
li(A AB)> ii(A)ii(E) then, using the theorem of total probability
/x(X) = ii(X\Y)rfY) + MXI^Xl " KH) X,YGS

we can write
fx(A) = ji(A|Om(O + ti{A\CL){\ - /i(C)) (8)

M(B) = rtfllOrtQ + M£|CX)(1 - iiiQ) (9)
M(A A B) = /x(A A fl|C)/x(C) + MA A B|C1)(1 - /i(C)) (10)
= /i(A|C)M*|C>(C) + MAIC-^BIC^Xl - MQ) (11)
((1 1) follows from (10) because of the screening off equations (2)-(3)). So the
assumption of a common cause of the correlation between A and B implies that

there exist real numbers
rC» rA|O rfl|O rA\CL » rB\CL
satisfying the following relations
0^rAlc,rB{C,rAlc±,rBlc± < 1 (12)

MA) = rAlcrc + rAlc±(l - rc) (13)
li(B) = rBlcrc + rfl|C±(l - rc) (14)
p(A AB) = rAlcrBlcrc + rAlc±rBlc±(l - rc) (15)
0 < rc < 1 (16)
r4|C> rA|C^ (17)
rB\c>rB\c^ (18)
Conversely, given a correlation ix
(S, pt), if there exists an element C
M(Q = rc (19)
MA|C) = rA|C (20)
MAIC1) = rAlc± (21)
M(B|Q = rB|C (22)
l»iB\CL) = rB|Cx (23)
and the numbers rc, rAjC, rB(C, îc1*^^1 satisfy the relation
then the element C is a common cause of the correlation in Reichenbach's
sense.
Proposition 1 Given any correlation pt(A AB)> fi(A)ii(B) in (

exists a non-empty two parameter family of numbers
rc(t, s)9 rA]C(t, s), rB{C(t, s), rA]C± (f, s), rB]C± (t, s)
that satisfy the relations (12)-(18).
Proof: Consider the system of 3 equations (13)- (15) with t = rAlc

as parameters. One can then express rc, rA\C± and rB\C± from eq
(15) as follows.
r C =
C =
M(A) - t , t ,x(AaB)-vl(A)s
r^ = -j-37- + , ' t = m-s (25>
KB)-s , it(AAB)-i*B)t »„
rB!C = -TIT- + , s = MA
Using the equations (24)-(26) it is easy to veri
parameters t, s within the bounds
the conditions (12)- (18) are satisfied.

As the above proposition shows, the Reichenbach conditions allow, in principle,
for a number of different common causes, each characterized probabilistically by
the five real numbers rc, rA(C, rB\C, rA\C± , rB\C± satisfying the relations (12)- (18).
Given a correlation n(A AB)> n(A)ii(B), we call a set of five real numbers rc,
rA|c> rB|c» rA\c1'> rB\cL admissible if they satisfy conditions (12)- (18).
Definition 2 A common cause C of a correlation pt(A AB)> fi(A)fi(B) is said

to have (be of) the type (rc, rA(C, rfi)C, rA\C± , rB\C± ) if these numbers are equal to
the probabilities indicated by the indices, i.e. if the equations (19)-(23) hold.
3 Common cause completeability: the classical case

Given a statistically correlated pair of events A, B in a probability space (5, ^),
a proper common cause C in the sense of Reichenbach' s definition does
not necessarily exist in S. (For instance the set of events might contain only
I,A9B and their orthogonal complements and hence be too small to contain a
proper common cause.) If this is the case, then we call (5, ix) common cause
incomplete. The existence of common cause incomplete probability spaces
leads to the question of whether such probability spaces can be enlarged so that
the larger probability space contains a proper common cause of the given
correlation. What is meant by 'enlargement' here is contained in Definition 3
below. Before we give this definition recall that the map h: Sx - > S2 between
two Boolean algebras 5! and S2 is a Boolean algebra homomorphism if it
preserves all lattice operations (including orthocomplementation). A Boolean
algebra homomorphism h is an embedding ifX^Y implies h(X) =t h(Y).
Definition 3 The probability space (S', //) is called an extension of(S, pt) if
there exists a Boolean algebra embedding h of S into Sf such that
ti(X) = n'(h(X)) for all XGS (29)

This definition, and in particular the condition (29), implies that if (S',
extension of (5, /a) (with respect to the embedding h), then every single
tion fi(A AB)> fi(A)fi(B) in (S, /x) is carried over intact by h into the co
ti'(h(A)Ah(B))= li\h(AAB)) = tjL(AAB) > /i(A)/i(B) = ii\h(A))

Hence, it makes sense to ask whether a correlation in (S, fi) has a Reiche
common cause in the extension (5', p). So we stipulate
Definition 4 We say that (S', //) is a type (rc, rA\C, rB\C, rA\C± , rB\C± ) common
cause completion of (5, /x) with respect to the correlated events A, B if(S\ /i)
is an extension of (5,pt), and there exists a Reichenbachian common
cause CElS' of type (rc,rA\C,rB\C,rA\C±,rB\C±) of the correlation
Hf(h(A) A h(B)) > ti'(h(A))ti'(h(B))-
Definition 5 Let (S, pi) be a probability space and { (Ah Bt) | i G / } be a set of
pairs of correlated events in S. We say that (S, fi) is common cause complete-
able with respect to the set {(A,, B() \ i G /} of correlated events if given any
set of admissible numbers (f*c>rA|C>rfl|C>rA|c1>r#|c1) for everJ i^-L there
exists a probability space (S\fi) such that for every i G / the space (S\fi)
is a type-(rlc,rlA\C,rB\C,rA\C±,rB\C±) common cause extension of(S>n) with
respect to the correlated events Ah Bt.
Proposition 2 Every classical probability space (5, /x) is common cause

completeable with respect to any finite set of correlated events.
The proof of this statement proceeds by induction on the index /. One shows
first that given the single pair (AuBi) = (A,B) of correlated events in (S,n)
and any admissible numbers (rc,rA\C,rB\C,rA\C±,rB\C±) there exists a type-
(rc> rA|c» rB\c* rA\cL » rB\cL ) common cause completion of (S, fi) with respect to
(A,B). We prove this statement in two steps. In Step 1 we construct an
extension (S',ii) of (S,/x). In Step 2 we show that given any admissible
numbers, the probability measure // can be chosen in such a way that there
exists a proper common cause in S' that has the type specified by the admissible
numbers. Finally, we shall argue that if (Sn~l 9 fin~l) is a common cause
completion of (S, fi) with respect to the set of n - 1 correlations between A,
and Bt(i = 1, ... n - 1), then there exists a common cause completion of (S, fi)
with respect to the n correlations. For the details see the Appendix.
4 Common cause completeability: the quantum case

Statistical correlations also make sense in non-classical probability structures
(£, <t>), where L is a non-distributive lattice of events and </> is a generalized
probability measure ('state') on L. Such non-classical probability spaces arise
in quantum theory, where L is the non-distributive, orthomodular lattice of
projections T(M) of a non-commutative von Neumann algebra M deter-
mined by the set of observables of a quantum system (for the operator algebraic
notions used here without definition see e.g. Kadison and Ringrose ([1983],
[1986]). A map <j>: ?{M) - ► [0, 1] on such an event structure is called a state if
it is additive on orthogonal projections in the following sense:
<KW = 5><P<) pi±pJ ii=J <3°)
A positive, linear functional on a von Neumann algebra M is called normal

state if its restriction to the lattice of projections is a state (i.e. it is additive) in
the above sense. The restriction of a normal state to a Boolean sublattice of
TiM) is a classical probability measure, so normal states are the analogues of

classical probability measures. If M acts on the Hilbert space 3~C, then a
normal state is always of the form </>(X) = Tr{ WX) with some density matrix
W. We call a pair (?(JVf),0) with a normal state 0 a quantum probability
space. Two commuting events A, B in a quantum probability space (T(M), <j>)
are called (positively) correlated if
0(A A B) >(f>(A)<t>(B) (31)
Given a correlation in a quantum probability space, we m

is a Reichenbachian common cause in T(M) of the corr
enbachian common cause we mean a C E P( JVf) which
and B and satisfies the Reichenbachian conditions (2)-
stipulate the following
Definition 6 The event C E T(M) is a common cause of

between two commuting events A, B in a quantum prob
if
1. C commutes with both A and B;

2. the following four conditions (analogous to (2)-(5)) are satisfied
<t>(A A B A C) _ <t>(A A C)<t>{B A C)

0(Q " </>(C) 0(Q
0(A A £ A C1) _ 0(A A Cx)0(B A C1)
(KC1) ~ ^(C1) <t>{CL)
0(A A C) cKAAC1)
4>(C) > tfKC1)
0(5 A C) <j>(BhCL)
0(O > cKC1)
Similarly to the classical case, a common cause C is called proper i/if differs
from both A and B by more than an event of ^-measure zero.
Having this definition, we can define the type of the common cause in a
quantum probability space in exactly the same way as in the classical case, and
we can also speak of admissible numbers etc. Just like a classical probability
space, a quantum probability space {T{M\<t>) may contain a correlation
without containing a proper common cause of the correlation in the sense of
Definition 6. If this is the case, then we call the quantum probability space
common cause incomplete, and we may ask if the quantum probability space
can be enlarged so that the enlarged space contains a proper common cause.
What is meant by 'enlargement' is specified in the next definition, which is

completely analogous to Definition 3.
Definition 7 The quantum probability space (T(M'), <t>f) is an extension of the

quantum probability space (T(M), <j>) if there exists an embedding h ofT(M)
into T(M') such that
<t>'(h(X)) = 4>(X) for all X E T(M)

By an embedding is meant here a lattice homomorphism that preserves all
lattice operations (including the orthocomplementation) and such that X =£ Y
implies h(X) # h(Y).
Definition 8 We say that (T{M\<t>') is a type-(rc,rAlc,rBlc,rAlc±9rBlc±)

common cause completion of (T(M), <t>) with respect to the correlated events
A, B if(?(M\<t>) is an extension of (T{M\<t>\ and there exists a Reich-
enbachian common cause C E T(Mf) of type (rc, rA\C, rB\C, rA\C±y rB\C±) of
the correlation <t>'(h(A) A h(B)) > <f>' \h(A))<t>' \h(B)).
We can now give the definition of common cause competeability in the

quantum case.
Definition 9 Let (T(M),<t>) be a quantum probability space and

{(AhBj) | / E /} be a set of pairs of correlated events in T(M). We say that
(T(M),<I>) is common cause completeable with respect to the set
{(Ai,Bj) | / E /} of correlated events if given any set of admissible numbers
(rc> rA\c> rfi|c» rA\cL>rB\cL}for every i €= 1> there exists a quantum probability
space (?(M'\ <£') such that for every i E / the space (T(M\ <j>f) is a type-
(rc> rA|c» rB|c> r\\cL » rB\cL ) common cause extension of(T(M), <t>) with respect
to the correlated pair {AhBt).
Proposition 3 Every quantum probability space (T(M), <f>) is common cause

completeable with respect to the set of pairs of events that are correlated in the
state <t>.
The proof of this statement is divided into two parts. In the first part we
construct an extension of the quantum probability space (T{M),<j>)\ this is
done in two steps. In Step 1 the quantum probability space (T(M),<t>) is
embedded into the quantum probability space (T(3~C 0 3^),<j>2) with a sui-
table state <t>2 extending </>, where 3~C is the Hilbert space on which the von
Neumann algebra M. acts. In Step 2 this latter quantum probability space is
embedded into (T(3-C), </>'), where 3^' is a Hilbert space constructed expli-
citly. We show in the second part of the proof that for any correlated pair (A, B)
in (T(M), <t>) and for any admissible set of numbers there exists in (T(3-C), <t>)
a Reichenbachian common cause of type defined by the admissible numbers.
For the details see the Appendix.
5 Reichenbach's Common Cause Principle and common

cause completeability
Reichenbach 's Common Cause Principle is a non-trivial metaphysical claim
about the causal structure of the physical world: if a direct causal influence
between the probabilistically correlated events A and B does not exist, then
there exists a (proper) common cause of the correlation (in Reichenbach's
sense). There exist both classical and quantum probability spaces that contain
correlations without containing proper common causes of the correlations;
hence, if one wants to maintain Reichenbach's Common Cause Principle, one
must be able to claim that there exist 'hidden' events ('hidden' in the sense of not
being accounted for in the given event structure which is thus common cause
incomplete) that can be interpreted as the common causes of the correlations. If
such 'hidden' common cause events exist, then there must exist an extension of the
original probability space, an extension that accommodates the common causes.
Propositions 2 and 3 tell us that such extensions are always possible. In other
words, Propositions 2 and 3 show that a necessary condition for a common
cause explanation of correlations in both classical and quantum event struc-
tures can always be satisfied. To put this negatively: one cannot disprove
Reichenbach's Common Cause Principle by proving that the necessary con-
dition (common cause completeability) for its validity cannot be satisfied.
It is generally accepted that the Reichenbachian conditions (2)-(5) are just
necessary conditions. If an event C must satisfy also some Supplementary
Conditions (in addition to the Reichenbachian conditions) to qualify as a
common cause, then a disproof of Reichenbach's Common Cause Principle
requires establishing that there exists no event whatsoever that satisfies both
the Reichenbachian conditions and the Supplementary Conditions. It goes
without saying that such a disproof first requires the specification of the
Supplementary Conditions. Propositions 2 and 3 impose strong restrictions
on the possible mathematical specifications of the Supplementary Conditions:
these conditions cannot be formulated in terms of the probabilities p(C),
p(A\C), p(B\C), p{A\CL) and p{B\CL). This is because the assumptions
in Propositions 2 and 3 contain no restrictions whatsoever on these proba-
bilities - beyond the Reichenbach conditions. Therefore, the hypothetical
Supplementary Conditions would have to be specified entirely in terms of
the algebraic/logical structure of events.
The conclusion that probability spaces are typically common cause
completeable seems to contradict the received view concerning correlations
predicted by quantum mechanics and quantum field theory. According to the
standard interpretation an explanation of the quantum correlations by assuming a
direct causal influence between the correlated quantum events is excluded by the
theory of relativity, and furthermore there cannot exist Reichenbachian
common causes of the correlations because, as the argument goes, the assump-
tion of a common cause of correlations in Reichenbach's sense implies Bell's
inequality (this view is present in van Fraassen ([1989]), Skyrms ([1984]),
Butterfield ([1989]) and Spohn ([1991]) and it has made its way into textbooks
already (see Salmon et al. ([1992])).
But there is no contradiction here at all because the present paper's analysis
differs from the standard interpretation. To see where the differences are, let us
recall using the present paper's terminology and notation the standard argu-
ment in favour of the claim 'the existence of Reichenbachian common causes
of correlations implies Bell's inequality'. Consider four pairs
(Al,B2yAAuB2);(A2,Bl);(A2,B2)
of commuting events in T(M) that are correlated in the state <j>. Assume that
there exists a common common cause C of the four correlations; i.e. assume
that there exists a single C E T(M) that is a common cause (in the sense of the
Definition 9) of all four correlated pairs. Using the notation <t>(X\Y) = ^p^
for commuting XJE T(M) we can write then
<t>(AiABj\C) = <t>(Ai\C)<t>(Bj\C)
<t>{Ai/KBj\CL) = <t>{Ai\CL)<t>{Bj\CL) U=l>2 (32)

0(Af- A Bj) = <t>{At A Bj\C)<t>(C) + 0(Af- A ^•|C1)0(C1)
= 0(A,|CMB,|Q0(O + ÎC1)^ C1)*^1)
The elementary inequality for numbers ah bj E [0, 1] (ij =1,2)
\a(bi + atbj + ajbi - afyl < 0 (33)

implies
\<t>{Ax\C)<t>(Bx\C) + <f>{Ax\C)<t>(B2\C) + <t>(A2\C)<t>(Bx\C)
-<KA2\C)<t>(B2\C)\<0 (34)
Multiplying (34) by <j>(C) and by ^(C1), adding

and using (32) we obtain
\<t>(Ax A Bx) + <t>{Ax A B2) + <t>(A2 A
The inequality (35) is known as the Clauser-Ho

not to hold for every quantum state <f> that p
projections (see e.g. Summers ([1990a], [199
The crucial assumption in the above derivatio
C is a common cause for all four correlated p
common cause, shared by the different correla
Bell's inequality cannot be derived. But there d
reason why common causes should also be com
of quantum or of any other sort of correlati
Reichenbach' s notion of common cause there is nothing that would justify such
an assumption; hence if such an assumption is made, it needs extra support. It
should be mentioned that while the impossibility of (non-probabilistic) common
common causes of the (non-probabilistic) GHZ correlations has been proved in
the paper Belnap and Szabo ([1996]), it remains open in that paper whether
non-common common causes of the GHZ correlations exist.
6 Open questions
To decide whether a particular event structure is common cause incomplete
does not seem to be a trivial matter. In a previous paper the problem was raised
whether the event structure defined by (algebraic relativistic) quantum field
theory is common cause incomplete, and this problem is still open (see Redei
[1997], [1998], for a precise formulation of the question). It is even conceivable
that the explicitly formulated axioms that define algebraic quantum field
theory - and thereby the set of all events - are not strong enough to decide
the issue. Such undecideability would be especially interesting.
It also is an open mathematical question whether one can have common
cause closed probability spaces, where a probability space is said to be
'common cause closed' iff for every pair of correlated events there exists in
that probability space a proper common cause of the correlation in Reich-
enbach's sense. It is important here that common cause means a proper
common cause. This qualification on non-triviality is necessary; for it is not
difficult to show (using standard tensor product procedures) that every quan-
tum probability space can be enlarged in such a way that the enlarged quantum
probability space is common cause closed in the improper, formal sense that
for every correlated pair (A, E) there exists at least one C < A, C ± A such that
<j){C) = <j>(A) and that such a C satisfies the Reichenbach conditions.
Whether or not common cause closed probability spaces exist, it is not
reasonable to expect a probabilistic physical theory to be common cause closed.
This is because one does not expect to have a proper common cause explanation
of probabilistic correlations that arise as a consequence of a direct physical
influence between the correlated events, or which are due to some logical
relations between the correlated events. One would want to have a common
cause explanation of correlations only between events that are neither directly
causally related, nor do they stand in a straightforward 'logical consequence
relation' to each other. Thus a precise notion of causal (in)dependence, different
from the notion of the standard probabilistic independence (correlation), is
needed. Perhaps the notion of 'logical independence' (see the References;
Redei ([1995a], [1995b], [1998])) can be useful here. Two orthocomplemented
sub-lattices Lx and £2 of an orthomodular lattice L are called logically indepen-
dents A A B # 0 for any A G L\ and B G £2. This is an independence condition
that obtains between spacelike separated local systems in the sense of (algebraic)
quantum field theory; so this logical independence condition can be viewed as
a lattice theoretic formulation of 'separatedness' of certain events. It seems
reasonable then to expect a probabilistic physical theory (£, fi) to be common
cause closed with respect to the correlated elements in every two, logically
independent, commuting sublattices Lx , £2. It is not known if this is possible.
As we have argued at the end of Section 5, quantum correlations cannot have
a common common cause in general. This raises the question of whether
classical correlations exist that cannot have a common common cause. Note
that Bell's inequality is never violated in classical probability theory, so one

cannot obtain an answer to this question by referring to the violation of Bell's
inequality. In this sense the behaviour of correlations with respect to Bell's
inequality is not a good indicator of (non)existence of common common
causes. (That Bell's inequality is not a good indicator of (non)existence of
non-common common causes is clear from Propositions 2 and 3, since these
propositions show that common causes can be constructed both in the classical
and in the quantum case, whereas Bell's inequality is always satisfied in the
classical case and it does not hold in the quantum case.) To return to the
question of this paragraph, the answer is yes. That is, there exist classical
probability spaces containing different pairs of events that are correlated with
respect to a fixed probability measure and which cannot have a common
common cause. (The proof of this assertion will be published elsewhere.)
Acknowledgements
Work supported by AKP, by OTKA (contract numbers: T025841 andF023447)
and by the Dibner Institute MIT, where M. Redei was staying during the 1 997/98
academic year as a Resident Fellow. We also wish to thank the audiences -
especially N. Belnap, R. Clifton and W. Salmon - of two seminars held in
October 1997 at the Center for Philosophy of Science, Pittsburgh University.
Department of Philosophy
Technical University of Budapest
Hungary
gszabo @ hps. elte. hu
Department of History and Philosophy of Science
Eotvos University
Budapest, Hungary
redei @ ludens. elte. hu
Department of Theoretical Physics

Department of History and Philosophy of Science
Eotvos University
Budapest, Hungary
szabol @ caesar. elte. hu
Appendix
Proof of Proposition 2
Step 1 By Stone's theorem we may assume without loss of generality that 5 is a
field of subsets of a set 12. Let Q{ and Q2 be two identical copies of Q,
distinguishable by the indices 1 and 2, and let S{ and S2 be the corresponding
two copies of S:
fy = {(*,/) | jcGQ} i = l,2

S, = {{(*,/) | jcGX} | XGS] i=l,2
Let hi (i = 1, 2) denote the Boolean algebra isomorphisms betw

(i = l,2):
S3X^hi(X)={(x,i) | xGX] i = l,2
Furthermore, let S' be the set of subsets of 12] u Q2 hav

/i1(X)u/i2(r),i.e.
s'-(»ifflu*2(n i xjg5)
We claim that S' is a Boolean algebra of subsets of Qj u Q2 with respec
usual set theoretical operations u, n, 1 and that the map h defined b
h(X) = hx(X)u h2(X) X G 5 (36)

is an embedding of S into Sf. To see that S' is a Boolean algebra o
to show that Sf is closed with respect to the set theoretical opera
meet and complement, and this is a straightforward consequence
that 5, itself being a Boolean algebra with respect to the set
operations, is closed with respect to these operations. Checking
morphism properties of h is a routine task.
We now define a measure /z' on Sf that has the property
(/ = 1,2,3,4) be four arbitrary real numbers in the interval [0,1
define a yt! measure on S ' by
ti{hx(X) u h2(Y)) = rlfx(X n(An B)) + r2fi(X n(An B1))

+ r3fx(X n (A1 n B)) + rAtx(X n (A1 n BL))
+ (1 - rO^r n (A nB)) + (1 - ^(ynf^nB1)
+ (1 - r3)^(y n (A1 n5)) + (1 - r4)/*(y n (A1 n f
Since AnÂn^^^fi and A x n B1 are disjoint and their un

follows that
//(/*! (X) u *2(X)) = //(/*(X)) = /x(X) X G 5

Hence (5 V) is indeed an extension of the original probability sp
Step 2 Choose any value of the parameters t , 5 within the bounds
(27)-(28), and consider the corresponding numbers rA(C = t, rB\C = s and

rc, rA|Ci,rfi|C±, the latter ones defined by (24)-(26). We claim that the prob-
ability space (S', fi) constructed in Step 1 is a common cause extension of (S, /x)
of type (rc, rA)C, rB\C, rA|C± , rfi|C± ) with respect to the correlation between A and
B, if the numbers r, (i = 1, 2, 3, 4) defining // by the formula (37) are given by
_ rCrA\CrB\C
ri ~ _ fi(A A 5)
_ rCrA\c(\ ~ rfl|c)
r2"/i(A)-/i(AAB) _
r3 ti(B)-ti(AAB)
_ rc(l ~ rA|C ~ rB\C + rA|Crfl|c)
r4" _ /x(A1AB1)
To show that (S', //) is a common cause completion of (5, fi) one only has to
display a proper common cause C in Sf of the correlation. We claim that
C = /*i(0) u /*2(0) is a proper common cause. Clearly, C is a proper common
cause if it is a common cause. To see that C is a common cause indeed one can
check by explicit calculation that the following hold
M/(/i1(Q)u/i2(0)) = rc (37)
*i'((*i(A) u MAMhAQ) u *2(0)) = rAlc (38)
/((AiW u ^(BMKfcÔ) u *2(0)) = rB|C (39)
M((»i(A) u ^(A))!!*^!]) u ^(O)]1) = rMC, (40)
^((Ai W u *2(*))|[*i(Q) u ^(B)]1) = rB|cx (41)
Since the numbers r^^, rB\C, rc, rA\C± , rB(Ci were chosen so that they satisfy the
conditions (12)- (18), C is indeed a common cause.
Assume now that there exists a common cause completion (Sn~l,iin~l) of
(5, n) that contains a common cause C,- of each correlation /*(AZ A /?,-) >
li(Ai)fA(Bi) (i= 1, ... n - 1). Consider the correlation between An and #„. By
repeating the two steps (Step 1-Step 2) one can construct a common cause
completion (5n,/xw) of (Sn~\iin~l) that contains a common cause Cn of the
correlation between An and Bn. To complete the induction one only has to see
that (5n, ixn) also contains common causes of each of the correlations between
hn(Aj),hn(Bi) (i = 1, . . .n - 1), where hn is the Boolean algebra embedding
of Sn~l into Sn. But /in(Q) (i = 1,2...) are clearly common causes of
the correlations between hn(Ai),hn(Bi) (i = l,...n- 1) because hn is a
homomorphism preserving ^n_i.
Proof of Proposition 3
Step 1 We may assume without loss of generality that M is acting on a Hilbert
space 3~C . Let 3~C © 3~C be the direct sum of 3i with itself and consider the
map h2 defined by
T(M) 3X^> h2(X) = X © X G ?( Jf © 5f )
Then h2 is an embedding of T(M) as an orthomodular lattice into the

orthomodular lattice T{3< © H ). Let 02 be a state defined on T( J{ © Jf )
by the density matrix ^ W © ^ W, where W is the density matrix belonging to <t>.
Clearly, 02 nas tne property
</>2(/i2(X)) = <t>{X) X G 2>( JH) (42)
So the probability space (!P( !hC © J/*), 02) is an extension of (T(M),<t>

Since every density matrix is a convex combination of (possibly counta
infinite number of) one dimensional projections, there exist vect
\l/k G ( 3~C © 3{) and non-negative numbers X* (k = 1 . . .) such t
£rX*=land
iweJw
z z it z z = f^W4 (43)
(Here, and in what follows, P% denote
subspace spanned by the Hilbert space
Step 2 Let Jf be the set of functions

such that .sqpJIgMlk < °°, where U\\2
set Jfr is a complex linear spac
((*i£i + K2g2)(n) = K{gx(n) + K2g2(n
becomes a Hilbert space with the scala
oo
(Si,S2>' = $>*tei(*XS2(*)>2
where (, )2 is the scalar product in 3~C © 3~C and the number

(43). The map ti\ 3(3< © J/") - S( Jf ') defined by
(h'(Q))g(n) = Q(g(n)) n G N, g G 5f ' (45)
is an algebra homomorphism from the algebra 2?( 3{ © Jf ) of all bounded

operators on 3~C © 3f into the algebra H(J-C) of all bounded operators
on 3-f; furthermore, ti(Qx) i= h'(Q2) if Qx # Q2, and routine reasoning
shows that if Qn G 2?( Jf © 3~C ) is a sequence of operators such that for
some Q G S( Jf © 3<) we have &,£ - Q% for all £ G Jf © Jf , then
(h'(Qn))g - ► ^(0^ for all g G ^T'. This means that // is continuous in the
respective strong operator topologies in 2?( 3~C © 3~C ) and 3(3-C). It follows
that if A, B are two projections in T( 34 0 3~C ), then
h\A AB) = h\s - \\m{AB)n) = s- \im(h\AB)n) = s - \im(h' (A)ti \B))n

= h\A) A ti(B)
It follows then that the restriction of h' to ?( J{ 0 Jf ) is an embedding of

?( J{ 0 Jf) as an orthomodular lattice into the orthomodular lattice T(3-C').
Let £/ (/ = 1, . . . dim(J{ 0 Jf )) be an orthonormal basis in Jf 0 5i. Then
the elements gkl E 3~C ' defined by
(dnk^=^ x^ ifX,#O
to(n) = { V x^
I 0 if X* = 0
form an orthonormal basis in 3~C . (bnk denotes the Kronecker symbol.) Th
linear operator W' on 3~C defined by
(WgXn) = KP+m8(n) ^N,gGjf

is easily seen to be a density matrix, hence it defines a state <f> on T(3~Cf). Fo
A G T( J{ 0 Jf ) we have
oo dim(3<®3{)
<t>\h\A)) = Tr{W'ti{A)) = Y, Yl (S*" W'tiW8ki)'

k=\ l=\
oo dim(J{®J{) oo
= E E E x« ^w> [w'h'M8ki}
A:=l /=1 «=1
oo dim(${®3<) oo j .
jt=l /=1 n=\ VKk VKk
oo dim(J{@J{)
= E E ^«/-^/*/>2
it=i /=i
dim(J{®J{) oo
= E <*/• [ E x^p^] */>2 = Tr( WA) = <ma)

1=1 k=\
So (T(3-f' ),<!>') is an extensio

(?(J/"0 Jf),02). It follows that (T
ability space (T( M\ </>), where th
h'oh2.
We now claim that for any given pair of events (A, B) in T(M) that are
correlated with respect to 0 and for any given set of admissible numbers
(rc»rA|c»rmc.rA|c1»rB|c-L) the probability space (T(J-f'),<t>') constructed
above contains a proper Reichenbachian common cause C of type
(rc»rA|c»rB|c»r4|c-L»rfi|c-L) °f tne correlation. Indeed, given the numbers
(rc»rA|c»rB|c»r4|c1»rfi|c-L)» me event C defined below by (46;) is a common
cause of type (rc, rA]C, rB{C, rA{C± , rB|Ci ).
On Reichenbach's Common Cause Principle 395
C:=PaVPpVPyVPd (46)
where Pa, P@, Py and Pd are projections in T(3-C) de
(Pag)(k) = Pakg(k)
(Pyg)(k) = Pykg(k) kGK.gE

(P8g)(k) = P6kg(k)
where Pak, Ppk, Pyk and P&k are projections in J{ 0 3
_ f 0 if (\l/k,(A AB1@A AB1

<Xk~ _ \ cos <4ol\ + sinoj?^ if Ofo,(A A 51 0
where
cos «4 = rA|z(l - rw)rz
^fe(AAêAA^)
{«Wk)2 = \
( 0 if (fc(AAB©AAB)W2=O
* = \ cos Jk 01 + sin uPk&i if Uk, (AAB®AA
where
2 3 (fc(AAfieAAB)^)2
cos 2 Jk=rMZrB,zrc 3
^(Â^^BÂW*
Pl±pj £A AB©A AB
J 0 if (iMA1 A B0 A1 AB)W2 =
lk ~ \ cos o)ykyl + sin wJy? if {ik, (A1 AB8A1 A B)^>2 #
where
cos «J = re|Z(l - rA|Z)rz
7^±7^GA1 AB©AX AB
<7i7*2)'=l
f 0 if (iM^1 AB1®A1 AB1\Pk)2) = 0

k~ jcosw^l + sincoj^* if (^(A1 A B1 ® A1 AB^fohÔ
where
cos2 4 = rz{\ - rAlz - rBlz + rA|Zrfi|Z)^^
6^15* E A1 AB1eA1 AB1

<6i5^>2= 1
Since «i, ft[, 7^ and 6{ are unit vectors, ak, /3h yk, dk also are unit vectors in
3-C 0 3{. The element C commutes with h{A) and h(B) because
Pak KAAB^QAAB1
Ppk<AAB@AAB
Pyk <A1 A50A1 A B
Pdk<AL AfiÂ1 AB1
So to show that C is indeed a common cause of the said type we just have to
show that the following hold
4>\C) = rc (47)
t'jh'jA) A C)
4>'iC) = ^ (48)
<t>'(ti(B) A Q
*<£) = rB|C (49)
<t>'(h'(A) A C1)
4,'^) =r«cx (50)
(t>'(h'(B)AC±)
*<£>■) =r°^ (51)
We show (48)-(51) by showing first that the following hold
<j>'(ti(A) AC) = rMCrc (52)

<t>'(h'(B) AC) = rB\Crc (53)
<l>'(h'(A)ACL) = rA\C,rc, (54)
<t>'(ti(B)AC1) = rB{C,rc, (55)
which imply (48)-(51) if 0'(C) = rc-
We compute <t>'(ti(A) A C) first.
<t>'(h(A) A C) = 4>'((ti(A) A (Pa V Pff)) = <t>'(Pa V Pp) = <t>'(Pa) + *'(/>„)
where we have used the fact that Pa, Pp, Py and Pb are pairwise orthogonal
projections. We can compute <t>\Pa) as follows
oo Jim(JfeJf)
<t>'(Pa) = Tr(WfPa) = J2 E <*«• Wfp«8ki)f

k=\ l=\
oo dim(J{@2{) oo . ^
k=\ l=\ n=\ VKk V*k
oo dim(J{@3{)
= E E ^(f/, ^4^S/>2 = Tr2(P+kP

k=\ l=\
oo oo
= J2
k=l k=\
OO
= EMc°s«?Mfc,aJ) + sino)?(^^>]2 (56)

The second term in (56) is equal to zero because ccxkLct\ by definition, and in
view of the definition of a\ and ol\ we can also write
0 = (ai,c^)2 = ([AAB±®AA Bâjh

= (^ [AAB1 ®AABL]*l)2 = (iha2k)2
Since we have
<**,aj!)2 = l
it follows that
^(p«) = Ex* cos2 w*
= ^|c(l-^|cVc^(A^1)EX^^(AAgl@AAjgl)^)2
= ^|c(l - ^|cyc<^)(AAgi)</>(Âg-L) = rA)C(l - rB)C)rc

In a completely analogous way one obtains
<l>'(Pfi) = rA\crB\crc
And so
<t>'(Pa) + 0r(^) = rMCrc

which is (52). A similar derivation shows that
4>\h(B) A C) = rBlcrc
which is (53). Since C commutes with both h(A) and h(E) we can write
4>'WA) A C1) = <t>'(h(A)) - <j>\h{A) A C) = 0(A) - rA]Crc = rA|C.rcx

*'(*(«) A C1) = </>'(/*(£)) - <t>'(h(B) A C) = 0(5) - rfl|Crc = rB^rc,
which establishes (54) and (55). One can compute <t>{Py) and <t>'{Pb) exactly
the same way as <t>(Pct) and <t>(Pp), and one obtains
0'(C) = <t>\Pa VPpVPyV P6) = <t>'(Pa) + <t>\P^) + <t>'(Py) + <t>\Pb)

= rA|c(l - îcVc + rAicîc^c + rB|c(l - rMC)rc
+ rc(l - îc - îc + rîcîc) = rc
which shows (47), which, together with (52)-(55) proves (47)-(51).
References
Belnap, N. and Szabo, L. E. [1996]: 'Branching Space-time Analysis of the GHZ
Theorem', Foundations of Physics , 26 ■, pp. 989-1002.
Butterfield, J. [1989]: 'A Space-time Approach to the Bell Inequality', in J. Cushing and
E. McMullin (eds), Philosophical Consequences of Quantum Theory, Notre Dame:
University of Notre Dame Press, pp. 1 14-44.
Cartwright, N. [1987]: 'How to Tell a Common Cause: Generalization of the Conjunctive
Fork Criterion', in J. H. Fetzer (ed.), Probability and Causality, Boston: Reidel Pub.
Co., pp. 181-88.
Kadison, R. V. and Ringrose J. R [1983]: Fundamentals of the Theory of Operator
Algebras. Vol. I. Elementary Theory, New York: Academic Press Inc.
Kadison, R. V. and Ringrose, J. R [1986]: Fundamentals of the Theory of Operator
Algebras. Vol. II. Advanced Theory, New York: Academic Press Inc.
Redei, M. [1995a]: 'Logical Independence in Quantum Logic', Foundations of Physics ,
25, pp. 41 1-22.
Redei, M. [1995b]: 'Logically Independent von Neumann Lattices', International
Journal of Theoretical Physics , 34, pp. 1711-18.
Redei, M. [1997]: 'Reichenbach's Common Cause Principle and Quantum Field
Theory', Foundations of Physics, 27, pp. 1309-21.
Redei, M. [1998]: Quantum Logic in Algebraic Approach, Dordrecht: Kluwer Aca-
demic Publishers.
Reichenbach, H. [1956]: The Direction of Time, Los Angeles: University of California

Press.
Salmon W. C. [1978]: 'Why ask "Why?"?', in Proceedings and Addresses of the

American Philosophical Association, 51, pp. 683-705.
Salmon, W. C. [1980]: 'Probabilistic Causality', Pacific Philosophical Quarterly, 61,
pp. 50-74.
Salmon, W. C. [1984]: Scientific Explanation and the Causal Structure of the World,
Princeton: Princeton University Press.
Salmon, M. H. et al. [1992]: Introduction to the Philosophy of Science, Englewood

Cliffs: Prentice Hall Inc.
Skyrms, B. [1984]: 'EPR: Lessons for Metaphysics', Midwest Studies in Philosophy, 9,

pp. 245-55.
Spohn, W. [1991]: 'On Reichenbach' s Principle of the Common Cause', in W. Salmon
and G. Wolters (eds), Logic, Language and the Structure of Scientific Theories,
Pittsburgh: University of Pittsburgh Press, pp. 211-35.
Summers, S. J. [1990a]: 'Bell's Inequalities and Quantum Field Theory', in Quantum
Probability and Applications V. Lecture Notes in Mathematics # 1441, Berlin-
Heidelberg-New York: Springer, pp. 393-413.
Summers, S. J. [1990b]: 'On the Independence of Local Algebras in Quantum Field
Theory', Reviews in Mathematical Physics , 2, pp. 201-47.
Suppes, P. [1990]: 'Probabilistic Causality in Quantum Mechanics', Journal of Statis-
tical Planning and Inference, 25, pp. 293-302.
Suppes, P. and Zanotti, M. [1970]: A Probabilistic Theory of Causality, Amsterdam:
North-Holland.
Suppes, P. [1981]: 'When are Probabilistic Explanations Possible?', Synthese, 48,

pp. 191-9.
Van Fraassen, B. C. [1977]: 'The Pragmatics of Explanation', American Philosophical
Quarterly, 14, pp. 143-50.
Van Frassen, B. C. [1982]: 'Rational Belief and the Common Cause Principle', in
R. McLaughlin {ed.\ What? Where? When? Why?, Boston: D. Reidel Pub. Co., pp.
193-209.
Van Fraassen, B. C. [1989]: 'The Charybdis of Realism: Epistemological Implications

of Bell's Inequality', in J. Cushing and E. McMullin {eds.), Philosophical Conse-
quences of Quantum Theory, Notre Dame: University of Notre Dame Press, pp. 97-
113.

On Reichenbach's Common Cause Principle and Reichenbach's Notion of Common Cause - Gábor Hofer-Szabó, Miklós Rédei and László E. Szabó

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

On Reichenbach's Common Cause Principle and Reichenbach's Notion of Common Cause - Gábor Hofer-Szabó, Miklós Rédei and László E. Szabó

Uploaded by

Copyright:

Available Formats

The British Society for the Philosophy of Science

On Reichenbach's Common Cause

1 Informal statement of the problem

1 Informal statement of the problem

© Oxford University Press 1999

by Suppes and Zanotti ([1970], [1981]), by Cartwright ([1987]) and by Spohn

Reichenbach' s Common Cause Principle is the claim that if there is a

emerge in non-relativistic quantum mechanics and in relativistic quantum field

2 Reichenbach's notion of common cause

then the events A and B are said to be (positively

Definition 1 C is a common cause of the corr

li(AAB\Q = ii(A\Qii(B\Q (2)

We shall occasionally refer to conditions (2)-(5) as 'Re

Theorem Conditions (2)-(5) imply (1); that is to say, if A

Some remarks and terminology:

(i) We emphasize that, from the point of view of the expla

(C < A or C > A) and fi(Q = /x(A) (6)

(C < B or C > B) and /*(C) = /i(fi) (7)

should not be regarded as a meaningful common cause because C is

(iv) A common cause C will be called deterministic if

Note that the notions of deterministic and genuinely probabilistic

/x(X) = ii(X\Y)rfY) + MXI^Xl " KH) X,YGS

fx(A) = ji(A|Om(O + ti{A\CL){\ - /i(C)) (8)

assumption of a common cause of the correlation between A and B implies that

rC» rA|O rfl|O rA\CL » rB\CL

satisfying the following relations

0^rAlc,rB{C,rAlc±,rBlc± < 1 (12)

Proposition 1 Given any correlation pt(A AB)> fi(A)ii(B) in (

that satisfy the relations (12)-(18).

Proof: Consider the system of 3 equations (13)- (15) with t = rAlc

parameters t, s within the bounds

the conditions (12)- (18) are satisfied.

Definition 2 A common cause C of a correlation pt(A AB)> fi(A)fi(B) is said

3 Common cause completeability: the classical case

ti(X) = n'(h(X)) for all XGS (29)

ti'(h(A)Ah(B))= li\h(AAB)) = tjL(AAB) > /i(A)/i(B) = ii\h(A))

Proposition 2 Every classical probability space (5, /x) is common cause

4 Common cause completeability: the quantum case

<KW = 5><P<) pi±pJ ii=J <3°)

A positive, linear functional on a von Neumann algebra M is called normal

TiM) is a classical probability measure, so normal states are the analogues of

0(A A B) >(f>(A)<t>(B) (31)

Given a correlation in a quantum probability space, we m

Definition 6 The event C E T(M) is a common cause of

1. C commutes with both A and B;

<t>(A A B A C) _ <t>(A A C)<t>{B A C)

What is meant by 'enlargement' is specified in the next definition, which is

Definition 7 The quantum probability space (T(M'), <t>f) is an extension of the

<t>'(h(X)) = 4>(X) for all X E T(M)

Definition 8 We say that (T{M\<t>') is a type-(rc,rAlc,rBlc,rAlc±9rBlc±)

We can now give the definition of common cause competeability in the

Definition 9 Let (T(M),<t>) be a quantum probability space and

Proposition 3 Every quantum probability space (T(M), <f>) is common cause

5 Reichenbach's Common Cause Principle and common

of correlations implies Bell's inequality'. Consider four pairs

<t>{Ai/KBj\CL) = <t>{Ai\CL)<t>{Bj\CL) U=l>2 (32)

\a(bi + atbj + ajbi - afyl < 0 (33)

Multiplying (34) by <j>(C) and by ^(C1), adding

\<t>(Ax A Bx) + <t>{Ax A B2) + <t>(A2 A

The inequality (35) is known as the Clauser-Ho

that Bell's inequality is never violated in classical probability theory, so one

Department of Theoretical Physics

//(/! (X) u 2(X)) = //(/*(X)) = /x(X) X G 5

= E </• [ E x^p^] />2 = Tr( WA) = <ma)