Deep Potential Molecular Dynamics Evaluation of the Structures and Thermophysical Properties of the LiF-BeF2 Molten Salt for a Molten Salt Reactor

Abstract

The molten salt reactor (MSR) is gaining increasing prominence in the nuclear power industry. The LiF-BeF2 molten salt system, utilized as a primary coolant and carrier salt in Generation IV nuclear reactors, poses considerable challenges for the experimental measurement of its performance owing to its highly corrosive characteristics and the acute toxicity of beryllium. The deep potential generator (DP-GEN) active learning framework features a fully automatic active learning closed loop, which greatly reduces the computational cost of density functional theory (DFT) calculations. In addition, the DP descriptor exhibits strong universality and is applicable to multi-ion molten salt systems. This work developed a machine learning potential (MLP) for the LiF-BeF2 system using the DP-GEN active learning framework. The potential function demonstrates high accuracy across a wide temperature range and enables a bridge between the structure and macroscopic thermal properties. The results show that the intensity of the first peak of the radial distribution function for Be–F is greater than that of Li–F, indicating that F– forms a stronger bond with Be2+. This made it difficult for F– to gain sufficient energy to escape the Be2+ coordination shell, thereby enabling the formation of a stable [BeF4]2– tetrahedral structure. The density, specific heat capacity, viscosity, and thermal conductivity obtained from the deep potential molecular dynamics simulation are in high agreement with the experimental data, with relative errors of 1.04%, 2.40%, 7.05%, and 1.08%, respectively.