Developing interaction potentials from first principles

Abstract

Interaction potentials for the double-perovskite cryolite, Na3AlF6, have been developed for use in classical Molecular Dynamics (MD) simulations using a method whereby ionic configurations are generated with empirical pair potentials, the multipoles and forces on the ions and the stress tensor of the cell are extracted from ab initio single-point DFT calculations, and then the multipoles, forces and stresses from the MD simulations are ‘fit’ to the ab initio quantities in a series of steps in which the potential parameters are optimized, for models of varying complexity. Previously, interaction potentials have been developed empirically by tuning the parameters to reproduce experimentally-derived properties such as structure factors and densities, and so the testing and development of the newer method is necessary in order to standardize a way of obtaining potentials from first principle considerations. A fitted potential was then used to characterize the ion dynamics in crystalline cryolite: a monoclinic to orthorhombic phase transition and the low-temperature-phase tilt-domain structure of the AlF3− 6 , the dominant structural features, are reproduced. The motional processes, which have been studied indirectly in NMR, conductivity and diffraction experiments, include oscillation of the AlF3− 6 and sodium ion diffusion - it has been suggested that these occur at a remarkably fast rate. The nature of the AlF3− 6 oscillatory motion is studied in more depth than accessible to experiment, and its connection with diffusion is investigated. Given the intrinsically defective nature of cryolite and the absence of diffusion in the initial simulations, defects are introduced to observe their effect on the dynamics: they are shown to be necessary for diffusion. This work has been written up in an article accepted for publication in the Journal of Physical Chemistry. The ab initio potentials developed as above involve representing a system with formally charged monatomic ions. We extended the scope of the method significantly with technical developments to allow for the inclusion of molecular ions, such as the hydroxide ion, the sulphate ion or the uranyl ion, where the intraionic bonding has significant covalent character. The appropriate modifications of the MD code were made and a modified force-fitting procedure was developed. The new method was applied to Mg(OH)2 which is an important mineral (brucite) and to the melts of uranyl chloride which are of interest in nuclear waste reprocessing. Although we found good potentials were harder to obtain for these compounds, we found this arose from their layered structure rather than the molecular nature of the ions, and that our method could achieve a level of success approaching that used in the cryolite work on further iterations of the fitting process.