Lattice dynamics modeling of thermal transport in solids using machine-learned atomic cluster expansion potentials: A tutorial

Abstract

Lattice dynamics (LD) plays a crucial role in investigating thermal transport in terms of not only underlying physics but also novel properties and phenomena. Recently, machine learning interatomic potentials (MLIPs) have emerged as powerful tools in computational physics and chemistry, showing great potential in providing reliable predictions of thermal transport properties with high efficiency. This tutorial provides a comprehensive guideline for MLIPs’ development and how they are used for the computational modeling of thermal transport. Using atomic cluster expansion (ACE) as the paradigmatic potential, we introduce the essential fundamentals of MLIPs, including data construction, model training, and hyperparameter optimization. With the developed ACE potentials, we further showcase their applications in the LD modeling of thermal transport for crystalline silicon and amorphous carbon. The corresponding code implementations for MLIP applications in calculating thermal conductivity are also provided for beginners to follow.