Stein Kernelized Molecular Dynamics for Active Learning of Interatomic Potentials

In three linesSKMD (Stein Kernelized Molecular Dynamics) is an enhanced sampling method for active learning of ML-based interatomic potentials. It uses interacting particle dynamics and preserves the Boltzmann distribution, balancing exploration with convergence to high-probability energy landscape regions. Demonstrated on Müller-Brown potential and MACE potential for alanine dipeptide.

Summary generated by Claude — human-verified