[18p-A24-1~13] 23.1 Joint Session N "Informatics"

2:15 PM - 2:30 PM JST(5:15 AM - 5:30 AM UTC)

[18p-A24-4] Material Design Method Using Gradient-Based Inverse Problem Solving to Achieve Desired Crystal Structures and Properties

〇Akihiro Fujii¹, Augustin Lu¹, Yoshitaka Ushiku², Satoshi Watanabe¹ (1.Tokyo Univ., 2.OMRON SINIC X Corp.)

Keywords:

material design,inverse problem,deep learning

In this study, we propose a material design method that optimizes crystal structures towards target property values using a property prediction deep learning model and its gradients. Unlike generative models, this approach allows the optimization of crystal structures while imposing various conditions, such as electrical neutrality and atomic configuration. By employing a pre-trained bandgap prediction model, we optimize perovskite materials while maintaining electrical neutrality and the atomic configuration of the initial structure, and propose candidate materials with the desired bandgap values.