The connection between measure-theoretic optimal transport and dissipative nonequilibrium dynamics provides a framework for quantifying nonequilibrium control costs, leading to a geometric interpretation of control and a collection of thermodynamic speed limits. However, realizing such optimal control strategies is challenging both in laboratory driving experiments or numerical simulations. In this work, we examine the validity of thermodynamic speed limits in non-optimal driving scenarios and develop a strategy for optimizing minimally dissipative protocols based on optimal transport flow matching. This generative machine learning technique enables us to learn control forces that closely approximate theoretically optimal trajectories while remaining computationally tractable. We further demonstrate that these combined geometric and machine learning approaches extend to a broader class of control paradigms, offering opportunities for designing minimally dissipative logical operations. Notably, our protocols saturate the thermodynamical Landauer limit associated with the energy cost of bit erasure, establishing a practical foundation for energy-efficient implementation of complex, multi-bit computational operations.
Biography:
Jérémie Klinger has been a Postdoctoral Researcher in the Rotskoff Group at Stanford University since 2023, following the completion of his PhD in Theoretical Physics at LPTMC (Laboratoire de Physique Théorique de la Matière Condensée) in Paris, France, from 2020 to 2023. Prior to his doctoral studies, he graduated from École Polytechnique and École Normale Supérieure in 2020 with joint degrees in Physics and Applied Mathematics. Jérémie's research focuses on uncovering universal behaviors in stochastic models of physicochemical systems. His work particularly emphasizes the elucidation of general transport properties, ranging from the effects of transport medium structure on first-passage observables to system-agnostic energetic laws governing stochastic motion. Recently, Jérémie has expanded his research to incorporate machine learning approaches, investigating novel control opportunities in fluctuating systems.
Location: Cory 373