1. analysis of ks eigenfunctions using a cnn in simulations of the mit in doped semiconductors

Keith Slevin & Tomi Ohtsuki

Osaka University & Sophia University, Japan

28 June 2022 Tue 3 pm

                                      IBS Center for Theoretical Physics of Complex Systems (PCS), Administrative Office (B349), Theory Wing, 3rd floor

                                      Expo-ro 55, Yuseong-gu, Daejeon, South Korea, 34126 Tel: +82-42-878-8633                     

Machine learning has recently been applied to many problems in condensed matter physics. A common point of many proposals is to save computational cost by training the machine with data from a simple example and then using the machine to make predictions for a more complicated example. Convolutional neural networks (CNN) have proved to work well for assessing eigenfunctions in disordered systems [1]. Here we apply a CNN to assess Kohn–Sham (KS) eigenfunctions obtained in density functional theory simulations of the metal–insulator transition in a doped semiconductor. We find that a CNN trained using data from a simulation of a doped

semiconductor that neglects electron spin successfully predicts the critical concentration when presented with eigenfunctions from simulations that include spin [2].

[1] T. Ohtsuki, T. Mano, Drawing Phase Diagrams of Random Quantum Systems by Deep Learning the Wave Functions, J. Phys. Soc. Jpn. 89, 022001 (2020).

[2] Y. Harashima, T. Mano, K. Slevin, T. Ohtsuki, Analysis of Kohn–Sham Eigenfunctions Using a Convolutional Neural Network in Simulations of the Metal-Insulator Transition in Doped Semiconductors, J. Phys. Soc. Jpn. 90, 094001 (2021).