Optimal Bias Potentials via Ergodic Optimal Control and Generator Learning

Data and code accompanying the study on Optimal Bias Potentials via Ergodic Optimal Control and Generator Learning.

The archive contains the Python implementation and numerical data used to reproduce the results presented in the associated manuscript. The benchmark systems include the double-well, lemon-slice, and three-hole potentials, together with an application to alanine dipeptide.

The deposited material includes implementations of the gEDMD eigen-pair approximation using random Fourier features (RFF) and B-spline basis functions, the neural Cole–Hopf (NCH) approximation, temperature-reweighting calculations, and closed-loop first-passage-time simulations. Numerical results and data used for the figures are provided for the benchmark systems and the alanine-dipeptide application.

The archive is organized into code.zip, containing the numerical methods, simulation and plotting scripts, and parameter files; Data.zip, containing the numerical data and saved results; and Fig.zip, containing the figures used in the manuscript. See README.md for the directory structure and instructions for reproducing the numerical results and figures.