Pressure dependence of liquid iron viscosity from machine-learning molecular dynamics

TL;DR

This study uses machine-learning-enhanced molecular dynamics to accurately determine the viscosity of liquid iron under Earth's core conditions, providing a detailed pressure-temperature viscosity map for geophysical applications.

Contribution

The paper introduces a machine-learning potential for liquid iron, enabling precise viscosity calculations across core conditions with reduced uncertainty.

Findings

Viscosity of liquid iron is around 10s mPa.s under core conditions.

The Einstein-Stokes relation does not hold at outer core conditions.

Provides a comprehensive viscosity map for Earth's outer core.

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 Einstein-Stokes relation is not accurate for outer core conditions. The viscosity is on the order of 10s \si{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.