Presentation Information
[19p-C601-5]High-precision prediction of band gap by the application of a three-dimensional convolutional neural network to spatial charge distribution information
〇Masahiro Hayashi1, Ryosuke Oka1, Ryoki Toda1, Ryo Ishihara1, Hiroyuki Fujiwara1 (1.Gifu Univ.)
Keywords:
materials informatics,deep learning,band gap
The band gap (Eg) is an important indicator that determines the physical properties, and determining Eg is effective for exploring next-generation materials. Current first-principles calculations are highly accurate but computationally expensive, making them difficult for large-scale exploration. Predicting Eg through deep learning has been studied, but the accuracy is not yet practical. In this research, we focused on the fact that the physical properties of materials are determined by the charge distribution and successfully applied a 3D CNN to the three-dimensional charge distribution data of materials obtained from first-principles calculations, achieving high-precision Eg prediction.<o:p></o:p>