Creation of crystal structure reproducing X-ray diffraction pattern without using database

When a sample's X-ray diffraction pattern (XRD) is measured, the corresponding crystal structure is usually determined by searching for similar XRD patterns in the database. However, if a similar XRD pattern is not found, it is tremendously laborious to identify the crystal structure even for experts. This case commonly happens when researchers develop novel and complex materials. In this study, we propose a crystal structure creation scheme that reproduces a given XRD pattern. We employed a combinatorial inverse design method using an evolutionary algorithm and crystal morphing (Evolv&Morph) supported by Bayesian optimization, which maximizes the similarity of the XRD patterns between target one and those of the created crystal structures. For sixteen different crystal structure systems with twelve simulated and four powder target XRD patterns, Evolv&Morph successfully created crystal structures with the same XRD pattern as the target (cosine similarity 99% for the simulated ones and >96% the experimentally measured ones). Furthermore, the present method has merits in that it is an automated crystal structure creation scheme, not dependent on a database. We believe that Evolv&Morph can be applied not only to determine crystal structures but also to design materials for specific properties.