Previous research has indicated the possible existence of a liquid-liquid critical point (LLCP) in models of silica at high pressure. To clarify this interesting question we run extended molecular dynamics simulations of two different silica models (
WAC and BKS) and perform a detailed analysis of the liquid at temperatures much lower than those previously simulated. We find no LLCP in either model within the accessible temperature range, although it is closely approached in the case of the WAC potential near 4000 K and 5 GPa. Comparing our results with those obtained for other tetrahedral liquids, and relating the average Si-O-Si bond angle and liquid density at the model glass temperature to those of the ice-like beta-cristobalite structure, we conclude that the absence of a critical point can be attributed to insufficient stiffness in the bond angle. We hypothesize that a modification of the potential function to mildly favor larger average bond angles will generate a LLCP in a temperature range that is accessible to simulation. The tendency to crystallize in these models is extremely weak in the pressure range studied, although this tendency will undoubtedly increase with increasing stiffness.
The liquid-liquid critical point scenario of water hypothesizes the existence of two metastable liquid phases---low-density liquid (LDL) and high-density liquid (HDL)---deep within the supercooled region. The hypothesis originates from computer simul
ations of the ST2 water model, but the stability of the LDL phase with respect to the crystal is still being debated. We simulate supercooled ST2 water at constant pressure, constant temperature and constant number of molecules N for N<=729 and times up to 1000 ns. We observe clear differences between the two liquids, both structural and dynamical. Using several methods, including finite-size scaling, we confirm the presence of a liquid-liquid phase transition ending in a critical point. We find that the LDL is stable with respect to the crystal in 98% of our runs (we perform 372 runs for LDL or LDL-like states), and in 100% of our runs for the two largest system sizes (N=512 and 729, for which we perform 136 runs for LDL or LDL-like states). In all these runs tiny crystallites grow and then melt within 1000 ns. Only for N<=343 we observe six events (over 236 runs for LDL or LDL-like states) of spontaneous crystallization after crystallites reach an estimated critical size of about 70+/-10 molecules.
