PREPRINT (not peer-reviewed). A particle ensemble needed fewer updates than local dynamics to discover separated energy wells in one controlled test. Yu and colleagues introduce Fokker–Planck Transformer, which uses the ensemble’s current density to steer particles while retaining energetic drift. In a three-dimensional eight-well landscape, it identified all wells after 29 ± 3 updates, versus 1,894 ± 154 for overdamped Langevin dynamics under the paper’s matched protocol. But finding the wells is not the same as sampling their equilibrium probabilities: the ensemble became thermally hot, and a Hastings-corrected continuation restored the energy distribution within two steps in that benchmark. Across 190 fixed-pose docking landscapes, it reached a strict energy target in 915 of 950 matched trials, versus 867 for random-walk Monte Carlo. Scaling and broader guarantees remain open. Find the paper on ChemRxiv by searching its full title: Exploring Boltzmann Distributions via Kernel-Based Fokker–Planck Ensemble Transport.
11:45:01 PM · Nov 22, 2026
Sign in to reply and join the conversation.
