Atomistic simulations for deep planetary interiors: Machine learning as a tool to break time and distance barriers

Machine learning molecular dynamics has opened up new areas for accurate simulations and predictions for the behavior of Earth and Planetary materials. New methods will be briefly reviewed and a number of current applications will be surveyed, ranging from determining multicomponent phase diagrams and melting relations under extreme conditions to melt and solid properties including thermal and electrical conductivity, viscosity, anelasticity, and glass transitions for oxides, silicates, and iron alloys under planetary core conditions.