Recent advances in machine learning interatomic potentials for cross-scale computational simulation of materials

Abstract

In recent years, machine learning interatomic potentials (ML-IPs) have attracted extensive attention in materials science, chemistry, biology, and various other fields, particularly for achieving higher precision and efficiency in conducting large-scale atomic simulations. This review, situated in the ML-IP applications in cross-scale computational models of materials, offers a comprehensive overview of structure sampling, structure descriptors, and fitting methodologies for ML-IPs. These methodologies empower ML-IPs to depict the dynamics and thermodynamics of molecules and crystals with remarkable accuracy and efficiency. More efficient and advanced techniques from interdisciplinary research field play an important role in opening a wide spectrum of applications spanning diverse temporal and spatial dimensions. Therefore, ML-IP method renders the stage for future research and innovation promising revolutionary opportunities across multiple domains.