[9a-N304-1~11] 18.1 Materials Informatics

9:30 AM - 9:45 AM JST(12:30 AM - 12:45 AM UTC)Presentation by Applicant for JSAP Young Scientists Presentation Award

[9a-N304-3] Prediction of Dielectric Constants of CaTiO₃ Using Δ-Learning Model with HSE

〇(M2)Koki Yoshimochi¹, Alex Kutana¹, Ryosuke Jinnouchi¹, Ryouji Asahi¹ (1.Nagoya Univ.)

Keywords:

machine learning potential,graph neural network,dielectric constants

In this study, we aimed to construct a dielectric constant prediction model with HSE-level accuracy by applying Δ-learning to a rotation-equivariant graph neural network model trained with PBEsol. In this Δ-learning approach, the model was trained to learn the difference between PBEsol and HSE. We validated the approach using CaTiO₃ as a target material, and found that the prediction error for the difference Δ was sufficiently small. Using the constructed HSE-level model, we evaluated the physical properties and analyzed the contributions of the Born effective charge and structure to the temperature dependence of the dielectric constant.