Machine Learning Interatomic Potentials for Heterogeneous Catalysis

Abstract

Atomistic modeling can provide valuable insights into the design of novel heterogeneous catalysts as needed nowadays in the areas of, e. g., chemistry, materials science, and biology. Classical force fields and ab initio calculations have been widely adopted in molecular simulations. However, these methods usually suffer from the drawbacks of either low accuracy or high cost. Recently, the development of machine learning interatomic potentials (MLIPs) has become more and more popular as they can tackle the problems in question and can deliver rather accurate results at significantly lower computational cost. In this review, the atomistic modeling of catalytic systems with the aid of MLIPs is discussed, showcasing recently developed MLIP models and selected applications for the modeling of heterogeneous catalytic systems. We also highlight the best practices and challenges for MLIPs and give an outlook for future works on MLIPs in the field of heterogeneous catalysis.

Keywords: MLIPs; computational chemistry; heterogeneous catalysis; molecular dynamics.