Electronic Structures of GeSe and GeSbTe Compounds Based on Machine Learning Empirical Pseudopotentials

The newly developed machine learning empirical pseudopotential method (ML-EPM) [Kim and Son, PhysRevB.109.045153 (2023)] overcomes poor transferability of traditional empirical pseudopotential method (EPM) by utilizing machine learning techniques while retaining its merit such as formal simplicity and less demanding resources. In this study, we perform electronic structure calculations by extending previous use of ML-EPM from binary to ternary compounds. With a training set of ab initio electronic structure calculations of various GeSe and GeSbTe compounds and their rotation-covariant descriptors, we successfully generate versatile and transferable empirical pseudopotentials of Ge, Se, Sb and Te, respectively. We demonstrate that, using the ML-EPM, computed electronic energy bands and wavefunctions of unlearned structures show good agreements with results from first-principles calculations. This agreement holds even for distinctive local atomic environments such as amorphous structures or more extended systems compared to those in training set, where computational costs are considerably decreased due to the absence of cumbersome self-consistency as well as reduction of plane wave basis.