Comparison of model potentials for molecular-dynamics simulations of silica

Structural, thermomechanical, and dynamic properties of pure silica SiO2 are calculated with three different model potentials, namely, the potential suggested by van Beest, Kramer, and van Santen (BKS) [Phys. Rev. Lett. 64, 1955 (1990)], the fluctuating-charge potential with a Morse stretch term for the short-range interactions proposed by Demiralp, Cagin, and Goddard (DCG)[Phys. Rev. Lett. 82, 1708 (1999)], and a polarizable force field proposed by Tangney and Scandolo (TS) [J. Chem. Phys. 117, 8898 (2002)]. The DCG potential had to be modified due to flaws in the original treatment. While BKS reproduces many thermomechanical properties of different polymorphs rather accurately, it also shows qualitatively wrong trends concerning the phononic density of states, an absence of the experimentally observed anomaly in the c/a ratio at the quartz alpha-beta transition, pathological instabilities in the beta-cristobalite phase, and a vastly overestimated transition pressure for the stishovite I --> II transition. These shortcomings are only partially remedied by the modified DCG potential but greatly improved by the TS potential. DCG and TS both reproduce a pressure-induced transition from alpha-quartz to quartz II, predicted theoretically based on the BKS potential.