KC-Maps learns SDE transition kernels without trajectory data
Original title: Kolmogorov-Chapman Maps: Trajectory-Free Learning for Simulating Molecular Dynamics
The method uses marginal samples plus known drift and diffusion, and beats classical molecular dynamics on peptide systems by GPU-hours.
Researchers introduce Kolmogorov-Chapman Maps, a neural method that learns the transition kernel of a stochastic differential equation without long simulated trajectories. KC-Maps uses trajectory-free marginal samples and known drift and diffusion coefficients, building boundary conditions into the network and training against the Kolmogorov-Chapman equation. On peptide molecular systems, the authors report it outperforms classical molecular dynamics in GPU-hours.
- Training needs only marginal samples and known drift and diffusion coefficients.
- The network encodes boundary conditions and a Kolmogorov-Chapman training objective.
- Authors report better GPU-hour efficiency than classical molecular dynamics on peptides.
From the original article
1 min read at arxiv.orgStochastic differential equations (SDEs) describe stochastic processes across a wide range of scientific domains, including physics, biology, and finance. Although individual numerical integration steps are inexpensive, observing meaningful events or transitions can require millions of sequential steps, resulting in substantial wall-clock simulation time and poor hardware utilization.
By Majdi Hassan · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.