Kolmogorov-Chapman Maps: Trajectory-Free Learning for Simulating Molecular Dynamics
KC-Maps learned peptide dynamics without long training trajectories and used fewer GPU-hours than classical molecular dynamics in the authors’ tests.
The method trains on marginal samples—states sampled without linked trajectories—plus known drift and diffusion coefficients. Given a starting state and time, it predicts the distribution of states at a requested later time. Its architecture enforces boundary conditions, while training uses the Kolmogorov–Chapman equation to keep transitions consistent across time intervals. The abstract reports a computational advantage but gives no numerical speedup.
ArXiv · AI/CL/LG's note
The method trains on marginal samples—states sampled without linked trajectories—plus known drift and diffusion coefficients. Given a starting state and time, it predicts the distribution of states at a requested later time. Its architecture enforces boundary conditions, while training uses the Kolmogorov–Chapman equation to keep transitions consistent across time intervals. The abstract reports a computational advantage but gives no numerical speedup.
ArXiv · AI/CL/LG's note
score 5