The origin of layer structure artifacts in simulations of liquid water

A recent paper (Yonetani, Chem. Phys. Lett. 2005, 406, 49-53) shows that in computer simulations of TIP3P water (Jorgensen et al. J. Chem. Phys. 1983, 79, 926-935) a strange layer formation can occur when a long cutoff is used. This result is counterintuitive because, in principle, increasing the cutoff should give more accurate results. Here we test this finding for different water models and try to explain why layer formation occurs. In doing so we find that under certain conditions, layer formation coincides with a sharp density increase to 1050 g/L, while simultaneously a pressure of 600 bar develops and water diffusion becomes anisotropic. This leads us to conclude that a group-based cutoff (of at least 1.4 nm) stabilizes an anomalous phase with most water models. In some cases the ordering is strengthened further by periodicity in the simulation cell, but periodicity effects can even be observed with a short cutoff (0.9 nm) and a relatively large box of 4 nm. Water models that have a relatively large quadrupole moment, more in accord with the experimental gas-phase values, in particular TIP4P (Jorgensen et al. J. Chem. Phys. 1983, 79, 926-935), are much less affected by the problem, because the dipole-dipole interaction is quenched at long distance. A comparison of different cutoff treatments, namely truncation, reaction field, particle mesh Ewald (PME), and switch and shift functions, for the simulation of water shows that only PME and shift functions yield realistic dipole-dipole interactions at long distance. The impact for biomolecular simulations is discussed.