Deep-learning-based Molecular-Dynamics Simulations of Iron in Planetary Core Conditions

Developments of machine-learning techniques in recent years have enabled large-scale simulations of a variety of systems with high accuracy, while applications of such approaches to the study of materials at extreme conditions are relatively few. In this presentation, we report on the construction and application of deep-learning potentials for molecular-dynamics (MD) simulations of iron at planetary-core conditions (multi-megabar in pressures and several thousand Kelvin in temperature). The simulations show cooperative diffusion along <111> directions of body-centered cubic iron, a special state of iron near melting that has many geophysical implications, as clarified lately through ab initio MD simulations.[1] Furthermore, we will discuss findings about melting and solidification of iron through equilibrium and nonequilibrium simulations, which shed light on the high-pressure phase diagram of iron and its dependence on the compression technique. Both are important but open questions in high-energy-density sciences.

[1] M. Ghosh, et al., “Cooperative Diffusion in Body-Centered-Cubic Iron at Earth and Super-Earth’s Inner Core Conditions,” submitted to Nature Communications.