Pressure Dependence of Liquid Iron Viscosity From Machine‐Learning Molecular Dynamics

Abstract We have developed a machine‐learning potential that accurately models the behavior of iron under the conditions of Earth's core. By performing numerous nanosecond scale equilibrium molecular dynamics simulations, the viscosities of liquid iron for the whole outer core conditions are obtained with much less uncertainty. We find that the Stokes‐Einstein relation is not accurate for outer core conditions. The viscosity is on the order of tens of mPa s, in agreement with previous first‐principles results. We present a viscosity map as a function of pressure and temperature for liquid iron useful for geophysical modeling.