Comparison of five empirical potential models for aluminosilicate systems: Albite and anorthite as test cases

In this work we have carried out a systematic investigation of the performance of different force fields in correctly reproducing the structure, densities as a function of temperature and the properties (Young's and Bulk moduli and the coefficient of thermal expansion) of both albite and anorthite crystals and glasses. After this first step of comparison, we selected the potential which gave the better overall performances (the Bertani-Menziani-Pedone potential) and investigated the effect of the cooling rate on structure and properties of albite and anorthite glasses and verified the predicting power of MD simulations for the simulation of the coefficient of thermal expansion (CTE) for a series of several aluminosilicate glasses whose compositions go from the albite-like one to the anorthite-like one. The tested interatomic potentials provide reliable short-range order around the constituting cations, but some differences are detected in the medium-range order described by the T-O-T angles (T= network formers cations, Si and Al), the Al/Si distributions and Qn distributions. In general, all the potentials overestimate the CTE because of the high quenching rate used but we have shown that extrapolated properties at room temperature provide a more reliable comparison with experiments. The CTE trend of the albite-anorthite plagioclase glass series is well reproduced and associated with the substitution of Na with Ca ions rather than the Al/Si one. The error associated with the computed CTE values decreases with box dimension and we found that simulation boxes containing 50k atoms are necessary to reliably estimate this property.