Molecular dynamics calculations: Machine learning

Molecular dynamics calculations have been an essential tool to model the structural and dynamic properties of atomic systems. Apart from providing the experimental tool set for statistical mechanics, molecular dynamics simulations are routinely used to understand, interpret and predict microscopic processes in realistic materials. Machine learning techniques have helped to accelerate molecular dynamics in several ways: surrogate models, trajectory analysis, enhanced sampling, to name a few. This entry focusses largely on machine learning interatomic potentials (MLIPs), presenting their development grounded on ab initio methods and analytical potentials. The basic ingredients of the methodology is discussed through examples, and some concrete implementations are mentioned. Applications highlighting the new possibilities offered by MLIPs are shown, as well as some remarks on future directions.