Title:The Topology and Geometry of Causality

Abstract:We provide a unified operational framework for the study of causality, non-locality and contextuality, in a fully device-independent and theory-independent setting. Our investigation proceeds from two complementary fronts: a topological one, using tools from sheaf theory, and a geometric one, based on polytopes and linear programming. From the topological perspective, we understand experimental outcome probabilities as bundles of compatible contextual data over certain topological spaces, encoding causality constraints. From the geometric perspective, we understand the same experimental outcome probabilities as points in high-dimensional causal polytopes, which we explicitly construct and fully characterise.
Our work is a significant extension of both the established Abramsky-Brandenburger framework for contextuality and the current body of work on indefinite causality. We provide definitions of causal fraction and causal separability for empirical models relative to a broad class of causal constraints: this allows us to construct and characterise novel examples which explicitly connect causal inseparability to non-locality and contextuality. In particular, we clearly demonstrate the existence of "causal contextuality", a phenomenon where causal structure is explicitly correlated to the classical inputs and outputs of local instruments, so that contextuality of the associated empirical model directly implies causal inseparability.