Deep-learning interatomic potential for iron at extreme conditions

Atomistic simulations play an important role in elucidating the physical properties of iron at extreme pressure and temperature conditions, which in turn provide crucial insights into the present state and thermal evolution of the earth's and planetary cores. However, simulations face challenges in retaining ab initio accuracy at the simulation size and time scales required to address some of the most important geophysical questions. We used deep-learning methods to develop interatomic models for iron covering pressures from 75--650 GPa and temperatures from 4000--7600 K. The models retain ab initio accuracy while being computationally cost effective. Rigorous validation tests attest their accuracy in large-scale simulations as well as in the presence of extended defects. The models pave the way to the determination of the thermodynamic and rheological properties of iron at extreme conditions with ab initio accuracy.