Development of Neural Network Potentials for Studying Chemical Behaviors of La3+/Nd3+ Ions in Molten LiCl–KCl–CsCl in Combination with Raman Spectroscopy

Abstract

The chemistry of molten salts has attracted great research interest owing to their wide applications in diverse fields. In the pyrochemical reprocessing of spent nuclear fuel or molten salt nuclear reactors, lanthanide elements as the principal fission products bring about changes in the composition and properties of molten salts. Herein, we report a comprehensive study on the coordination chemistry of the representative trivalent lanthanide ions (La³⁺/Nd³⁺) in LiCl–KCl–CsCl using a multiscale strategy combining Raman spectroscopy, deep learning, and large-scale molecular dynamics (MD) simulations. The neural network potential (NNP)-based MD and Raman spectroscopy studies revealed that La³⁺/Nd³⁺ ions prefer to form persistent octahedron complexes with the six-coordinated species as the dominant species at high temperatures. Compared to LaCl₆^3–, NdCl₆^3– shows higher stability with obviously longer lifetimes in LiCl–KCl–CsCl, as confirmed by the observed stronger interaction of Nd³⁺–Cl^– pairs. The total and partial structure factors further indicated the formation of a more stable network structure in LiCl–KCl–CsCl containing NdCl₃. Besides, the temperature exerts a larger influence on the local structures of the La³⁺ species compared to the Nd³⁺ analogues. According to the potential mean force calculations, the bond dissociation energies follow the order Ln–Cl > Li–Cl > K–Cl > Cs–Cl in LiCl–KCl–CsCl–LnCl₃. The NNP-based large-scale MD simulations have been verified to be an efficient and powerful way in molten salt chemistry research.