Crystal structure prediction by combining graph network and Bayesian optimization

Abstract

We developed a density functional theory (DFT)-free approach for crystal structure prediction, in which a graph network (GN) is adopted to establish a correlation model between the crystal structure and formation enthalpies, and Bayesian optimization (BO) is used to accelerate the search for crystal structure with optimal formation enthalpy. The approach of combining GN and BO for crystal structure searching (GN-BOSS) can predict crystal structures at given chemical compositions with and without additional constraints on cell shapes and lattice symmetries. The applicability and efficiency of the GN-BOSS approach is then verified by solving the classical Ph-vV challenge. The approach can accurately predict the crystal structures with a computational cost that is three orders of magnitude less than that required for DFT-based approaches. The GN-BOSS approach may open new avenues for data-driven crystal structural predictions without using expensive DFT calculations.