Comparative studies of the structural, dynamic, and thermodynamic properties of molten salt FLiNaK using machine-learned neutral network and analytical forcefields

Ions in high temperature molten salts usually do not possess formal charges due to certain degrees
of covalent bonding. This also causes a non-spherically symmetric electronic structure, yielding
dipole or quadrupole moment. Although forcefields in the literature have included these features
in various ways, to what extent does the partial charges and atomic polarization affect the structures
and dynamics of molten salts remains unanswered. Using molten LiF-NaF-KF (FLiNaK) as a
model system, we investigated a few classical analytical forcefields and machine-learned neural
network forcefields (NNFF) via comparing numerous structural and dynamic properties with both
first-principles calculations and experimental data. The structures of molten FLiNaK are
dominated by partial charge effect while polarization on anions which does minimum influence
on structures dictates the overall dynamics. In general, these results might help us bridge the
knowledge gap in ionic liquids. In addition, NNFF captures the behavior of molten salts while
maintaining its computational efficiency, aiding the process of exploring different chemical
compositions in the interest of the development of Molten Salts Reactor (MSRs) and thermal
storage facilities.