(197o) Li Ion Diffusion in Solid Electrolyte Analyzed Using Deep Generative Models.

Molecular dynamics (MD) is a useful simulation technique to evaluate materials properties, and wide varieties of materials are analyzed using both classical and ab-initio MD. Recently, high performance computing system enables us to compute systems consisting of even over millions of atoms utilizing parallel computation, however, long time simulation is still a problem since time evolution must be solved sequentially.

Machine learning technique, especially generative model, can be a surrogate model for the above MD simulation. Endo et al., proposed a method called MD-GAN [1,2], which predicts time evolution of the (sub)system in an equilibrium state. We applied the method to predict the diffusivity of Li ions in solid state electrolyte. Set of trajectory data of the Li ions were obtained using ab-initio MD with SIESTA code [3]. The set of data were used as input data for MD-GAN, and predictive model was trained to predict long time behavior of the ions. The accuracy and effectiveness of the MD-GAN model are going to be discussed in this work.

1. Endo, Tomobe, and K. Yasuoka, Proceedings of the AAAI Conference on Artificial Intelligence, 32, 2192 (2018).
2. Kawada, et al., Chem. Inf. Model, 63, 76 (2023)
3. M. Soler et al., J. Phys. Condens. Matter, 14, 2745 (2002).