Preprint
Jul 2026
A Distributional Framework for Generative Modeling of Molecular Crystals
This work develops a canonical parameterization, and trains energy-based generative flow networks to approximate the Boltzmann distribution over crystal structures for target molecules and space groups, and sample and analyze distributions of crystals for two molecules.
Michael Kilgour, A. Dong, M. Tuckerman et al.
· 1 citation