Machine learning approach to constructing tight binding models for solids with application to BiTeCl

Abstract

Finding a tight-binding (TB) model for a desired solid is always a challenge that is of great interest when, e.g., studying transport properties. A method is proposed to construct TB models for solids using machine learning (ML) techniques. The approach is based on the LCAO method in combination with Slater–Koster (SK) integrals, which are used to obtain optimal SK parameters. The lattice constant is used to generate training examples to construct a linear ML model. We successfully used this method to find a TB model for BiTeCl, where spin–orbit coupling plays an essential role in its topological behavior.