Assessment of the pairwise additive approximation and evaluation of many-body terms for water clusters.

We have tested the ability of four commonly used density functionals (three of which are semilocal and one of which is nonlocal) to outperform accurate pairwise additive approximations in the prediction of binding energies for a series of water clusters ranging in size from dimer to pentamer. Comparison to results obtained with the Weizmann-1 (W1) level of wave function theory shows that while all density functionals are capable of outperforming the accurate pairwise data, the choice of basis set used is crucial to the performance of the method, and if a poor choice of basis set is made the errors obtained with density functional theory (DFT) can be larger than those obtained with the simple pairwise approximation. We have also compared the binding energies and many-body terms determined with DFT to those obtained with W1, and have found that all semilocal functionals have significant errors in the many-body components of the full interactions energy. Despite this limitation, however, we have found that, of the four functionals tested, PBE1W/MG3S is the most accurate for predicting the binding energies of the clusters.