Research

Accelerated Langevin Dynamics

A neural network importance function biases Langevin dynamics toward rare transitions, while path reweighting and a branching random walk recover unbiased rates and reaction-channel probabilities — matching Kramers theory on a two-channel potential and the LJ7 benchmark.

Accelerated Markov Chain Monte Carlo Simulation

A neural network learns the bias potential in log-space, and simulated annealing bootstraps training from high temperature down to the rare-event regime, making importance-sampled Markov chains tractable in high dimensions.

Agent-Guided Variance Reduction for Protein Kinetics

An agentic search driven by Claude Code selects the weight-control scheme that tames path-weight degeneracy, recovering the alanine dipeptide isomerization rate to within 3–6% of brute-force MD.