[17p-A401-1~15] 23.1 Joint Session N "Informatics"

△ [17p-A401-4] Development of a machine learning model to predict various bonding species formed on graphene

〇(M1)Kento Nishio¹, Kiyou Shibata^1,2, Mizoguchi Teruyasu^1,2 (1.Sch. Eng., the Univ. of Tokyo, 2.IIS, the Univ. of Tokyo)

Keywords:materials infomatics, graphene, graph convolutional neural network

In this study, we attempted to construct a model that can simultaneously predict the physical properties and adsorption site dependence of systems in which a wide variety of elemental species with different bonding properties are adsorbed on graphene. Specifically, structures of atoms up to period 4 of the periodic table adsorbed on the graphene surface were input to a graph convolutional neural network, and the formation energy of the system was regressed. As a result, we succeeded in constructing a model that can predict formation energies with higher accuracy than recently reported models. This model, which can be applied to various elemental species, is very important for applications such as catalyst search.