Thermodynamics and Phase Behavior of Alkali Metal Mixture Using Ab-initio-based Machine Learning Interatomic Potentials*

Alkali metal mixtures have characteristic composition dependence of melting points. It is experimentally reported that they have lower melting points than pure alkali metals and have lowest melting points at the different concentrations depending on the combination of alkali metals.

We succeeded in qualitatively reproducing the phase diagram of Rb-Na mixture by theoretical calculations as well as revealing the parameters that determine the composition dependence of the phase diagram. In order to numerically estimate a melting temperature, Helmholtz free energies were calculated by the thermodynamic integration method, which is statistically reliable. Although ab-initio calculation is one of the powerful tools to study physical phenomena, it is unrealistic for directly applying to thermodynamic integration because of its cost problem. We constructed interatomic potentials based on ab-initio results by training with artificial neural networks. The constructed potential enables us to perform long enough simulation to apply thermodynamic integration while retaining the ab-initio accuracy.

In the presentation, we will also discuss the thermodynamic quantities obtained by our simulation.