[22p-52A-1~17] 23.1 Joint Session N "Informatics"

3:30 PM - 3:45 PM JST(6:30 AM - 6:45 AM UTC)

[22p-52A-10] Development of a Graph Neural Network Considering Anisotropy: Application to Prediction of Anisotropic ELNES/XANES

〇Kiyou Shibata¹, Teruyasu Mizoguchi¹ (1.The Univ. of Tokyo)

Keywords:

materials informatics,core loss spectrum,machine learning

Recently, graph neural networks (GNNs) have been proposed that use the graph structure constructed from atomic configurations as explanatory variables, but most GNNs use only the structure graph as explanatory variables and cannot handle orientation dependence such as external fields on the structure. In this study, we propose a GNN that can predict the shape of spectra considering anisotropy by taking the orientation of transition dipole moments as explanatory variables in addition to structural information, and apply it to a database of carbon K-edge spectra of organic molecules to verify its prediction performance.