Session:

An X-ray Diffraction (XRD) analysis is widely used for determination of crystal structure. A measured XRD of the synthesized sample is identified by comparing it with the XRD patterns of the candidate crystal structures searched in database (DB). In addition, a Rietveld refinement is used to slightly modify the structure to reduce the gap the XRD patterns of the measured sample and the candidate structure. However, such a method strongly depends on the DB. If the measured XRD pattern is from an unknown structure, it is difficult to identify the correct crystal structure of the measured sample. Therefore, it is needed to develop an inverse design method to automatically create a crystal structure which reproduces the given XRD pattern.

In this study, we propose an inverse design scheme, named Evolv&Morph, to generate a crystal structure reproducing the target XRD pattern. It consists of Evolutionary algorithm [1] and crystal morphing [2]. This scheme tries to create enormous crystal structures with maximizing the similarity score between the XRD pattern of the created structure and the target. Evolutionary algorithm is an optimization method with generating various structures based on genetic operators such as random selection, crossover, and mutation. Crystal morphing is an interpolation method to generate intermediate structures between the input structures. It can also optimize the similarity score of XRD pattern with the support of Bayesian optimization. For sixteen different material systems, Evolv&Morph successfully created crystal structures with the XRD pattern almost the same as the given target XRD pattern.

This scheme is an automated crystal structure creation scheme, not relying on a database. Therefore, the suggested algorithm is applicapable to other material design application for specific properties. More detailed descriptions for Evolv&Morph can be found in Ref. [3].

References
1) A. R. Oganov, C. W. Glass, J. Chem. Phys. 124, 244704 (2006).
2) J. Oba, S. Kajita, Phys. Rev. Mater. 6, 023801 (2022).
3) J. Lee, J. Oba, N. Ohba, S. Kajita, submitted., preprint arXiv 2302.10464 (2023).

Co-Author(s): Junpei Oba (Toyota Central R&D Lab. Inc.), Nobuko Ohba (Toyota Central R&D Lab. Inc.), Seiji Kajita (Toyota Central R&D Lab. Inc.)

Joohwi Lee

Appearances