Liquid Methanol from DFT and DFT/MM Molecular Dynamics Simulations

We present a comparative study of computational protocols for the description of liquid methanol from ab initio molecular dynamics simulations, in view of further applications directed at the modeling of chemical reactivity of organic and organometallic molecules in (explicit) methanol solution. We tested density functional theory molecular dynamics (DFT-MD) in its Car-Parrinello Molecular Dynamics (CPMD) and Quickstep/Born-Oppenheimer MD (CP2K) implementations, employing six popular density functionals with and without corrections for dispersion interactions (namely BLYP, BLYP-D2, BLYP-D3, BP86, BP86-D2, and B97-D2). Selected functionals were also tested within the two QM/MM frameworks implemented in CPMD and CP2K, considering one DFT molecule in a MM environment (described by the OPLS model of methanol). The accuracy of each of these methods at describing the bulk liquid phase under ambient conditions was evaluated by analyzing their ability to reproduce (i) the average structure of the liquid, (ii) the mean squared displacement of methanol molecules, (iii) the average molecular dipole moments, and (iv) the gas-to-liquid red-shift observed in their infrared spectra. We show that it is difficult to find a DFT functional that describes these four properties equally well within full DFT-MD simulations, despite a good overall performance of B97-D2. On the other hand, DFT/MM-MD provides a satisfactory description of the solvent-solute polarization effects with all functionals and thus represents a good alternative for the modeling of methanol solutions in the context of chemical reactivity in an explicit environment.