Infinitesimal Causality

A new mathematical framework places infinitesimal causality on a categorical footing, linking intervention geometry to the algebraic structure of Frobenius Markov categories, according to a paper posted to arXiv [1]. The work introduces Infinitesimal Causality (IDC), which models interventions not as discrete operations but as tangent deformations of the copy-and-discard structure that underpins classical information flow [1][2]. Two distinct Frobenius structures interact in the framework: the categorical Frobenius algebra that encodes copying, comparing, and discarding of classical variables, and a geometric Frobenius integrability condition requiring involutive closure of the intervention distribution [2]. Categorical causal sufficiency is defined as the compatibility of these two notions [2]. For structural causal models, the paper observes that infinitesimal causality is most naturally formulated in the slice of deterministic mechanisms over exogenous variables; visible stochastic kernels emerge only after a pushforward [2]. Interventions are treated as tangent vectors, and their Lie brackets measure whether the deformation preserves the classical information-flow structure [2]. Pearl's do-calculus serves as a guiding example. Ignoring irrelevant interventions corresponds to counit invariance, exchanging action and observation aligns with coproduct compatibility under pushforward, and independence conditions map to involutive bracket closure of the visible intervention distribution [2]. The approach draws on a broader lineage of infinitesimal causal structures. Earlier work by Segal formalized infinitesimal causal structures on manifolds, and subsequent research showed how Adjoint-invariant convex cones in symmetry algebras encode causality, horizons, and positive energy in quantum and classical theories with spacetime interpretations [3]. That formalism lifted causal structures to Hilbert spaces and classified physically inequivalent observables via the Dirac procedure on the complexified universal algebra [3]. In physics, causal structure remains a foundational concern. The chronology protection conjecture, proposed by Stephen Hawking, posits that laws beyond standard general relativity forbid closed timelike curves, even when the field equations permit them [5]. The conjecture is distinct from chronological censorship, where every closed timelike curve passes through an event horizon, hiding the causal violation from observers [5]. Lorentz transformations themselves, which encode the causal structure of special relativity, can be derived from principles ranging from Maxwell's equations to group theory [4]. While the IDC paper operates in the abstract setting of category theory, its treatment of interventions as deformations of information-flow structure echoes themes in machine learning, where diffusion models learn to reverse a noising process by modeling data as performing a random walk with drift through data space [6]. Those models, widely used in image and video generation, rely on Markov chains and stochastic differential equations trained via variational inference [6].