Theory-independent limits on correlations from generalised Bayesian networks

Abstract: Bayesian networks provide a powerful tool for reasoning about probabilistic causation, used in many areas of science. They are, however, intrinsically classical. In particular, Bayesian networks naturally yield the Bell inequalities. Inspired by this connection, we generalise the formalism of classical Bayesian networks in order to investigate non-classical correlations in arbitrary causal structures. Our framework of `generalised Bayesian networks' replaces latent variables with the resources of any generalised probabilistic theory, most importantly quantum theory, but also, for example, Popescu-Rohrlich boxes. We obtain three main sets of results. Firstly, we prove that all of the observable conditional independences required by the classical theory also hold in our generalisation; to obtain this, we extend the classical $d$-separation theorem to our setting. Secondly, we find that the theory-independent constraints on probabilities can go beyond these conditional independences. For example we find that no probabilistic theory predicts perfect correlation between three parties using only bipartite common causes. Finally, we begin a classification of those causal structures, such as the Bell scenario, that may yield a separation between classical, quantum and general-probabilistic correlations.

An Examination of Generalised Bayesian Networks for Non-Classical Correlations

The paper "Theory-independent limits on correlations from generalised Bayesian networks" by Henson, Lal, and Pusey offers a comprehensive exploration into the extension of Bayesian network formalism beyond classical contexts. This work innovatively addresses the limitations of classical Bayesian networks by integrating resources from quantum and generalized probabilistic theories. This extension has profound implications for understanding non-classical correlations in complex causal structures.

Overview of Generalised Bayesian Networks

Bayesian networks, as formalised structures composed of directed acyclic graphs (DAGs), have long been instrumental in modelling and interpreting causal relationships among random variables. However, their traditional framework is inherently classical, thus imposing constraints like the Bell inequalities. To surpass these limitations, the authors introduce 'generalised Bayesian networks', a novel construct that supplants the standard latent variables of Bayesian networks with resources intrinsic to any generalized probabilistic theory (GPT), such as quantum correlations and Popescu-Rohrlich boxes.

In this framework, observed nodes are associated with classical outcomes, while unobserved nodes facilitate the deployment of resources, mimicking complex causal interactions found in quantum systems. Definitions and conditions are rigorously established to maintain the integrity of causal dependencies across these non-traditional nodes.

Key Results and Theoretical Contributions

The paper delineates three primary contributions of theoretical and practical significance:

1. Extension of the $d$-Separation Theorem: The authors prove that the observable conditional independences outlined by the classical $d$-separation theorem are preserved within the generalised framework. This result underscores their approach's soundness and applicability, allowing researchers to infer conditional independences in these broader settings without revisiting the foundational principles of Bayesian networks.

2. Theory-Independent Constraints: Beyond conditional independences, the paper illuminates specific theory-independent constraints on probabilities. For instance, they demonstrate that no probabilistic model can yield perfect correlation between three parties relying strictly on bipartite common causes, irrespective of the probabilistic theory in use.

3. Causal Structure Classification: The work embarks on a preliminary classification of causal structures that reveal separations between classical, quantum, and general probabilistic correlations. This aspect is particularly promising for quantum foundations and provides a scaffold for identifying causal structures that exhibit distinct probabilistic discrepancies across theories.

Implications for Quantum and Beyond

The implications of Henson et al.'s work extend both theoretically and practically. By offering a framework that envelops classical, quantum, and even more general correlational resources, this research propels the understanding of quantum non-classicality beyond traditional Bell scenario examinations. It establishes a foundation for exploring limits on correlations dictated solely by causal structure, divorced from specific theory dependences. This exploration opens new avenues in the pursuit of characterizing the uniqueness of quantum correlations amongst potential GPTs.

Speculations on Future Developments

The generalised Bayesian network framework invites further exploration into more nuanced classifications of causal scenarios where quantum or post-quantum theories manifest advantages over classical interpretations. Future work may focus on expanding the realm of generalised probabilistic theories and exploring their theoretical boundaries within causal networks. This could bridge persistent gaps in device-independent quantum information processing and offer refined tools for characterizing correlations not achievable by classical causal structures.

Through robust theoretical groundwork and generalisation, this paper paves the way for advancing causal inference methodologies suited for a quantum-driven era, with implications destined to resonate across quantum foundations, information theory, and beyond.