Comparative studies of the structural, dynamic, and thermodynamic properties of molten FLiNaK using machine-learned neural network and analytical forcefields
Molten salts and their properties have drawn increasing attention from nuclear communities due to their great potential of implementing as the primary coolant or the fuel in the Molten Salt Reactor (MSR). However, existing data of molten salts show large discrepancies while atomic scale computation at first-principles level fails to reach the timescale for probing bulk properties. In this study, a Machine-Learned (ML) Neutral Network (NN) model trained on first-principles calculation is used to predict properties of molten LiF-NaF-KF (FLiNaK). The structures of FLiNaK are shown to be related to charge screening effect via comparing with a reparametrized Tosi-Fumi (TF) model (rTF) while the overall dynamics near the melting point might result from the polarization of the anions. In general, NN model is capable of capturing properties of ionic liquids and extending the computational timescale which is not reachable by first-principles calculation.
Bio: Shao-Chun Lee is currently a graduate student with his advisor, Professor Y Z, at Nuclear, Plasma and Radiological Engineering Department at University of Illinois Urbana-Champaign. He is studying physical and thermodynamic properties of various materials using molecular dynamics methods at both classical and first-principles level. Materials he studies include molten fluoride and chloride salts and electrolytes for Li-ion batteries. His work also includes developing machined-learning based forcefields from first-principles calculations. He received his B.S. (Engineering and System Science), M.S. (Nuclear Engineering and Science) from National Tsing Hua University, Hsinchu, Taiwan.
