Reichenbach's Principle of the Common Cause

• The Principle of the Common Cause (PCC) is a causal explanation framework that asserts positively correlated events must have a common antecedent cause, as formalized by Reichenbach.
• PCC extends into diverse formulations—including probabilistic, algebraic, and quantum models—addressing screening-off conditions and notions of common cause completeness.
• Modern applications of PCC tackle issues in latent confounder identification, multipartite coordination, and challenges in Bell scenarios, bridging classical and quantum theories.

The Principle of the Common Cause (PCC) is Reichenbach’s claim that if two events are correlated and do not stand in a direct causal relation, then the correlation should be explained by a common cause in their past. In contemporary work, PCC is treated not as a single fixed doctrine but as a family of probabilistic, algebraic, spacetime, and quantum-causal constraints. Its standard Reichenbachian form is a screening-off principle; stronger versions require that every positive correlation in a model have an internal common cause; weaker or revised versions alter screening-off, localization, or even background causal structure. Recent work has sharpened PCC in countable Boolean algebras, relativistic causal frameworks, Bell scenarios, latent-variable inference, and multipartite coordination tasks (Burešová, 24 Jan 2025, Rédei et al., 2012, Cavalcanti et al., 2013).

1. Reichenbachian core and standard probabilistic form

In its standard probabilistic form, PCC begins with positive correlation. If $A$ and $B$ are events in a probability space, then the relevant condition is

$P(A \wedge B) > P(A)P(B).$

Equivalently, in Boolean-algebraic language,

$\mathrm{COV}_p(a,b)=p(a\wedge b)-p(a)p(b)>0.$

The search for a common cause is restricted to positively correlated pairs, because a Reichenbachian common cause entails positive covariance (Burešová, 24 Jan 2025).

A Reichenbachian common cause $C$ of $A$ and $B$ satisfies two screening-off conditions and two probability-raising conditions: $P(A\wedge B\mid C)=P(A\mid C)P(B\mid C),$

$$P(A\wedge B\mid \neg C)=P(A\mid \neg C)P(B\mid \neg C),$$

$$P(A\mid C)>P(A\mid \neg C), \qquad P(B\mid C)>P(B\mid \neg C).$$

In Reichenbach’s conjunctive-fork model, these conditions explain the unconditional correlation as arising from mixing over $B$0 and $B$1. The same pattern appears in Boolean algebras, general orthomodular lattices, and quantum probability spaces restricted to compatible projections (Mazzola et al., 2017, Kitajima et al., 2015).

Two distinctions are fundamental. First, PCC is an explanatory principle, not merely a probabilistic identity: correlations are taken to call for direct causation or common causes. Second, the screening-off equations are only one way to explicate that explanatory demand. Later work therefore separates “PCC proper” from particular probabilistic realizations such as factorization conditions in Bell scenarios (Cavalcanti et al., 2013).

2. From single causes to systems, deviations, and completeness

Reichenbach’s original model uses a single binary cause. One major generalization is the Reichenbachian common cause system (RCCS), a finite partition $B$2 such that each cell screens off the correlation and the conditional probabilities of the effects vary together across cells. In this framework, the observed correlation is explained by a many-valued hidden cause rather than a single event. Mazzola and Evans showed that, for any positively correlated pair of events in a classical probability space and any finite $B$3, there exists an extension of the original space containing an RCCS of size $B$4, correcting a logical defect in an earlier proof by Hofer‑Szabó and Rédei (Mazzola et al., 2017).

A further extension replaces “unexpected correlation” by “positive deviation from expected correlation.” Let

$B$5

On this reading, a common cause need not restore independence; it may restore the expected correlation $B$6. Mazzola’s generalized common-cause models therefore require

$B$7

and analogous partition-based versions for systems of causes. Two system models are developed: generalized Hofer‑Szabó–Rédei systems and generalized Mazzola systems. For both, existence in an extension of the probability space is characterized by the same condition,

$B$8

which marks a precise limit on how far the extended PCC can be pushed in classical probability (Mazzola, 2017).

A stronger strengthening of PCC is common cause completeness (CCC). A probabilistic structure $B$9 is common cause complete if every pair of positively correlated events in $P(A \wedge B) > P(A)P(B).$0 has a Reichenbachian common cause within the same Boolean algebra. This is a model-internal realization of PCC: $P(A \wedge B) > P(A)P(B).$1 If no positive correlations exist, the algebra is only trivially common cause complete (Burešová, 24 Jan 2025).

3. Algebraic realizations in classical and quantum probability

The algebraic status of PCC depends sharply on the structure of the event space. In Boolean algebras, earlier sufficient conditions for CCC required $P(A \wedge B) > P(A)P(B).$2-completeness, the Darboux property, and non-atomicity. A 2025 construction shows these conditions are not necessary: there exists a countable, non-trivially common-cause-complete Boolean algebra, namely the algebra of finite disjoint unions of left-open, right-closed subintervals of $P(A \wedge B) > P(A)P(B).$3 with rational endpoints, equipped with the restriction of Lebesgue measure. This construction shows that a very small event structure can still satisfy the strong Reichenbachian demand that every positive correlation have an internal common cause (Burešová, 24 Jan 2025).

In quantum probability spaces, common-cause existence can be asked for commuting projections in the projection lattice of a von Neumann algebra. Gyenis and Rédei prove that such a quantum probability space is common cause closed iff it has at most one measure-theoretic atom. Since local algebras in algebraic quantum field theory are type III and hence atomless, this result supports common-cause closedness in those settings; by contrast, finite-dimensional quantum systems with many atoms are not common-cause closed (Kitajima et al., 2015).

Recent work on identical particles exposes a distinct obstruction. For permutation-symmetric states and observables, one may demand that the common cause also be permutation symmetric. Depending on how joint probabilities are extracted from symmetric measurement data, two outcomes emerge. Either symmetric common causes need not exist, which suggests hidden distinguishability at the level of the screening variable, or symmetric screeners exist only in a trivial sense, because the “cause” collapses into re-labeled versions of the events it is supposed to explain. This shows that even with commutative measurements, the quantum theory of identical particles places nontrivial constraints on Reichenbachian explanation (Hovhannisyan et al., 19 Jun 2026).

4. Spacetime localization, screening-off, and relativistic QFT

In relativistic contexts PCC must be localized in spacetime. Henson’s framework introduces a least domain of decidability map $P(A \wedge B) > P(A)P(B).$4, full specifications of spacetime regions, and two past regions for spacelike separated domains $P(A \wedge B) > P(A)P(B).$5: the mutual past

$P(A \wedge B) > P(A)P(B).$6

and the truncated joint past

$P(A \wedge B) > P(A)P(B).$7

This yields screening-off principles SO1 and SO2, which require that all full specifications of $P(A \wedge B) > P(A)P(B).$8 or $P(A \wedge B) > P(A)P(B).$9 screen off correlations between spacelike-separated events (Rédei et al., 2012).

Redei and San Pedro distinguish finite and infinite versions of SO1 and SO2, depending on whether the spacetime regions are causally finite. Their analysis argues that the four principles are non-equivalent. In particular, the step from SO2 to SO1 in Henson’s equivalence proof relies on causally infinite regions, which are physically dubious and unavailable in the finite case (Rédei et al., 2012).

In algebraic quantum field theory, existing results support only Finite SO2: under local primitive causality, correlations between projections in spacelike separated double cones admit common causes in the truncated joint past. There is no corresponding general theorem for Infinite SO2 or for either version of SO1. This sharpens PCC in relativistic quantum theory: a plausible “weak PCC” is finite and truncated-joint-past based, rather than a universal mutual-past principle (Rédei et al., 2012).

5. Bell’s theorem, screening-off, and revised causal readings

Bell scenarios put classical PCC under pressure because the standard package

$\mathrm{COV}_p(a,b)=p(a\wedge b)-p(a)p(b)>0.$0

entails Bell inequalities, while quantum experiments violate them. Cavalcanti and Lal therefore distinguish PCC proper from Factorization of Probabilities (FP). PCC proper is the explanatory demand that correlations require direct causes or common causes; FP is the specific screening-off condition

$\mathrm{COV}_p(a,b)=p(a\wedge b)-p(a)p(b)>0.$1

This decomposition permits different reactions to Bell’s theorem: abandon PCC, relax relativistic causal structure, modify FP, or alter the role of total probability for hidden causes (Cavalcanti et al., 2013).

One attempted repair is the use of non-commutative common causes in algebraic quantum field theory. Cavalcanti and Lal argue that the Hofer‑Szabó–Vecsernyés definition becomes trivial: any orthonormal product basis can serve as a common-cause system for the correlations of any quantum state, including Bell-violating states. Because the resulting “causes” do not recover the observed correlations via a law-of-total-probability decomposition, they fail to preserve the explanatory role of PCC (Cavalcanti et al., 2013).

A different response treats Bell correlations as a familiar exception to PCC rather than as a refutation of causal locality. Price argues that EPR–Bell correlations can be regarded as selection artefacts, akin to collider bias: correlations that arise or change under preselection or postselection are well-known exceptions to screening-off principles. On that view, failure of Bell factorizability need not imply violation of intuitive locality; it may instead signal the ubiquity of selection effects in quantum phenomena (Price, 19 Feb 2026).

More radical revisions alter the background causal order itself. One proposal is to replace a common cause in the common past by a common future interaction enabled by closed timelike curves: to-be-correlated pieces of classical information meet locally in the future, the correlation is established there, and outputs are propagated back along the curves. The proposal is presented as a way of preserving a Reichenbach-style demand for mechanism while abandoning a globally well-founded causal structure (Baumeler et al., 2017).

6. Methodological, inferential, and algorithmic uses of PCC

PCC is also used as an inferential template. In the analysis of Simpson’s paradox, Hovhannisyan and Allahverdyan assume a latent common cause $\mathrm{COV}_p(a,b)=p(a\wedge b)-p(a)p(b)>0.$2 satisfying

$\mathrm{COV}_p(a,b)=p(a\wedge b)-p(a)p(b)>0.$3

For the minimal binary setting—binary $\mathrm{COV}_p(a,b)=p(a\wedge b)-p(a)p(b)>0.$4, $\mathrm{COV}_p(a,b)=p(a\wedge b)-p(a)p(b)>0.$5, $\mathrm{COV}_p(a,b)=p(a\wedge b)-p(a)p(b)>0.$6, and binary $\mathrm{COV}_p(a,b)=p(a\wedge b)-p(a)p(b)>0.$7—they prove that every such $\mathrm{COV}_p(a,b)=p(a\wedge b)-p(a)p(b)>0.$8 predicts the same direction of association between $\mathrm{COV}_p(a,b)=p(a\wedge b)-p(a)p(b)>0.$9 and $C$0 as conditioning on $C$1, not as marginalizing over $C$2. In that minimal case, PCC selects the stratified rather than the marginal option (Hovhannisyan et al., 2024).

When the common cause is known to exist but is unobserved, the problem becomes one of latent-confounder identification. Allahverdyan and Danageozian propose a generalized maximum likelihood method for selecting a “most likely” $C$3 consistent with

$C$4

Near $C$5, the objective is equivalent to maximizing the joint entropy $C$6 under the PCC constraints, linking the method closely to maximum entropy. In the binary symmetric case, the optimal conditional probabilities exhibit non-analytic behavior reminiscent of a second-order phase transition when the observed distribution passes from correlation to anti-correlation (Hovhannisyan et al., 2023).

PCC has also been translated into machine-learning language. In a 2025 study, the joint probability matrix $C$7 of pixels $C$8 and images $C$9 is modeled as

$A$0

with $A$1 interpreted as the value of a common cause variable $A$2. On this reading, nonnegative matrix factorization is an approximate implementation of PCC. The paper further proposes a PCC-based predictability criterion for the effective rank of NMF and reports that the corresponding basis images are stable against weak noise and against seeds of local optimization (Khalafyan et al., 3 Sep 2025).

Not all operational uses of PCC succeed. In probabilistic cellular automata, several PCC-inspired axioms—non-signaling, non-correlation, screening-off, and screening-off-completable $A$3-causality—fail to characterize exactly the class of local probabilistic dynamics. Direct screening-off is too strong, while weakened PCC-style conditions remain too weak. The conclusion is that probabilistic locality requires an explicit operational localizability condition rather than an abstract common-cause principle alone (Arrighi et al., 2011).

7. Multipartite coordination as a new causal principle

Recent work proposes coordination as a genuinely multipartite extension of PCC. The principle states that perfect coordination, in the form of agreement on a uniformly random output among $A$4 parties, is possible only if they share a common cause. For four parties, this means that the distribution

$A$5

cannot be generated in a network lacking a common ancestor of all four parties (Centeno et al., 4 May 2026).

Quantum theory satisfies this coordination principle. The proof uses quantum inflation: assuming perfect coordination in a network without a global common cause, one constructs an inflated network in which certain adjacent pairs reproduce the original perfect coordination while another pair becomes causally independent. A projector argument then forces that independent pair to be perfectly coordinated as well, contradicting independence (Centeno et al., 4 May 2026).

The principle is strictly stronger than standard no-signaling plus independence. An explicit operational probabilistic theory is constructed that obeys both principles while still allowing perfect coordination without a common cause. This shows that multipartite PCC does not follow from the usual pairwise causal constraints (Centeno et al., 4 May 2026).

The same program yields Bell-like inequalities certifying the presence of a common cause and, in a genuinely quantum coordination task, the presence of a quantum common cause. In particular, the preparation of a multipartite GHZ state cannot be explained by networks with only classical common causes shared among all parties; a quantum common cause is required, and experimentally testable Bell-like inequalities witness that requirement (Centeno et al., 4 May 2026).

Taken together, these developments suggest that PCC is best understood not as a single immutable doctrine but as a hierarchy of causal principles. At one end lie Reichenbach’s original screening-off conditions for pairs of events; at the other lie common-cause completeness, relativistic localization principles, selection-bias exceptions, quantum generalizations, and multipartite coordination constraints. The contemporary literature treats these not as mutually exclusive replacements but as technically distinct ways of asking when, where, and in what form correlations demand causal explanation.