Generative diffusion model for surface structure discovery

The generative diffusion method is adapted for use in connection with surface structure determination. The authors introduce here the concepts required to train a diffusion model on structures from small-cell density functional theory calculations. Using the diffusion model to generate large-cell surface structures, the authors demonstrate that highly plausible structures result. The efficiency of generating low-energy structures is compared to that of the random structure search method. Finally, using the diffusion method, a structure for a Ag-based surface-oxide grain boundary is proposed.