An Accurate Machine-Learned Potential for Krypton under Extreme Conditions.

We have developed two machine-learned pair potentials for krypton based on CCSD(T) quantum chemical calculations on two and three atom clusters. Through extensive testing with molecular dynamics, we find both potentials give good agreement with the experimental equation of state, melting point, and neutron scattering data for the fluid. Compared with the most widely used Lennard-Jones model, our potentials produced similar results in low-pressure melting and equation of state. However, extending the regime to higher pressures of ≤30 GPa showed a remarkable divergence of the Lennard-Jones model from the experimental (solid) equation of state. Our potential showed extremely good agreement, despite having no solid phases in the training set.