Applications of machine-learning interatomic potentials for modeling ceramics, glass, and electrolytes: A review

The emergence of artificial intelligence has provided efficient methodologies to pursue innovative findings in material science. Over the past two decades, machine-learning potential (MLP) has emerged as an alternative technology to density functional theory (DFT) and classical molecular dynamics (CMD) simulations for computational modeling of materials and estimation of their properties. The MLP offers more efficient computation compared to DFT, while providing higher accuracy compared to CMD. This enables us to conduct more realistic simulations using models with more atoms and for longer simulation times. Indeed, the number of research studies utilizing MLPs has significantly increased since 2015, covering a broad range of materials and their structures, ranging from simple to complex, as well as various chemical and physical phenomena. As a result, there are high expectations for further applications of MLPs in the field of material science and industrial development. This review aims to summarize the applications, particularly in ceramics and glass science, and fundamental theories of MLPs to facilitate future progress and utilization. Finally, we provide a summary and discuss perspectives on the next challenges in the development and application of MLPs.