AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-20
Generative marginal learning for finite-size molecular simulation with application to stress calculation
Molecular simulations of small periodic systems suffer from finite-size effects when estimating local observables. The authors propose generative marginal learning (GML), which trains a conditional diffusion model on large-system simulations to generate configurations of a retained region without simulating its surroundings. Applied to the spatially averaged Irving–Kirkwood stress near a vacancy defect, GML substantially improved pair structure and local stress accuracy over small periodic MD simulations.