Efficient and Accurate Car-Parrinello-like Approach to Born-Oppenheimer Molecular Dynamics

We present a new method which combines Car-Parrinello and Born-Oppenheimer molecular dynamics in order to accelerate density functional theory based ab initio simulations. Depending on the system a gain in efficiency of 1 to 2 orders of magnitude has been observed, which allows ab initio molecular dynamics of much larger time and length scales than previously thought feasible. It will be demonstrated that the dynamics is correctly reproduced and that high accuracy can be maintained throughout for systems ranging from insulators to semiconductors and even to metals in condensed phases. This development considerably extends the scope of ab initio simulations.

Figure 1

Deviations from the BO surface of liquid ${\mathrm{SiO}}_2$ with respect to total energies (upper panel) and mean force deviations (lower panel). The deviation in the energies corresponds to a constant shift of $4.16\times {10}^{-4}$ Hartree per atom for one corrector step and $3.5\times {10}^{-5}$ Hartree per atom for two corrector steps. The average mean force deviation is unbiased.

Figure 2

Partial pair-correlation functions g(r) of liquid Si (upper left panel) and liquid ${\mathrm{SiO}}_2$ at 3000 K and 3500 K, respectively, using a DZVP Gaussian basis set.

Figure 3

The kinetic energy distribution calculated from a 1 ns trajectory with a time step of 3.25 fs of metallic liquid ${\mathrm{Si}}_{64}$ using a DZVP Gaussian basis set and a density cutoff of 100 Ry (left panel). Velocity autocorrelation function (upper right) and its Fourier transform (lower right) of 32 Water at 325 K using a TZV2P Gaussian basis set, a density cutoff of 280 Ry and the BLYP exchange-correlation functional. The Langevin friction coefficients are ${\gamma }_L=0$ and ${\gamma }_D\sim {10}^{-8}{\mathrm{fs}}^{-1}$.