Séminaire / Seminar GReCO |
| « From Multiphysics Complexity to Causal Histories: Path-Space Monte Carlo Beyond Linear Transport » |
|
Daniel Yaacoub |
|
Many physical systems are described by coupled fields. Their complexity arises from the combination of distinct physical processes, nested phenomenologies, intricate geometries and morphologies, or widely separated spatial and temporal scales. The EDStar community addresses this challenge by reformulating Eulerian models in path space and by estimating observables through Monte Carlo sampling, building on the Feynman-Kac perspective and developments at the interface between physics and computer graphics. This probabilistic viewpoint accommodates linearly coupled processes within a common propagation framework while reducing reliance on globally resolved meshes and making complex geometries computationally tractable.
However, the path-space perspective offers more than an alternative computational strategy: it provides a renewed physical reading of these systems by making explicit the causal paths through which sources and interactions contribute to an observable. In linear transport, these causal paths are embodied by light rays, fluid streamlines, or Brownian trajectories. Beyond linear transport, nonlinear mechanisms including molecular collisions, growth–fragmentation dynamics, and self-consistent or field-mediated interactions call for different path-space constructions, often involving branching stochastic processes. Despite their differences, these mechanisms raise a common physical question: how do sources and local interactions contribute to an observable? In self-consistent transport, where the field driving motion depends on the state it transports, these dependencies become implicit in the effective description. Making them explicit can recover causal maps connecting physical sources to observables and reveal many distinctions: different geometries may produce nearly identical observable fields while organizing their causal contributions differently. Path-space methods thus offer a common perspective for investigating physical causality across linear and nonlinear systems. Monte Carlo sampling makes this perspective computationally accessible in complex systems, connecting a renewed interpretation of physical phenomena with practical approaches to multiphysics and multiscale problems. |
|
lundi 26 octobre 2026 - 11:00 Salle des séminaires Évry Schatzman Institut d'Astrophysique de Paris |
| Pages web du séminaire / Seminar's webpage |