PFP: Universal Neural Network Potential for Material Discovery

28 Jun 2021  ·  So Takamoto, Chikashi Shinagawa, Daisuke Motoki, Kosuke Nakago, Wenwen Li, Iori Kurata, Taku Watanabe, Yoshihiro Yayama, Hiroki Iriguchi, Yusuke Asano, Tasuku Onodera, Takafumi Ishii, Takao Kudo, Hideki Ono, Ryohto Sawada, Ryuichiro Ishitani, Marc Ong, Taiki Yamaguchi, Toshiki Kataoka, Akihide Hayashi, Takeshi Ibuka ·

Computational material discovery is under intense research because of its power to explore the vast space of chemical systems. Neural network potentials (NNPs) and their corresponding datasets have been shown to be particularly effective in performing atomistic simulations for such purposes. However, existing NNPs are generally designed for rather narrow target materials and domains, making them unsuitable for broader applications in material discovery. To overcome this issue, we have developed a universal NNP named PFP capable of handling any combination of 45 elements under a wide range of conditions. Particular emphasis is placed on the generation of training datasets, which include a diverse set of virtual structures to attain the universality of the NNP. In addition, the NNP architecture has been designed to be able to train this diverse dataset. We demonstrated the applicability of PFP, in selected domains: lithium diffusion in LiFeSO${}_4$F, molecular adsorption in metal-organic frameworks, order-disorder transition of Cu-Au alloys, and material discovery for a Fischer-Tropsch catalyst. These examples showcase the power of PFP, and this technology provides a highly useful tool for material discovery.

PDF Abstract


Materials Science Computational Physics