Choice of time step in Molecular Dynamics simulations
In Molecular Dynamics (MD) simulation of atomic structures it is important to use as large a time step as possible in order to sample phase space rapidly and save on computer expense. Technically, the effect of the resulting algorithm errors in the trajectories of the particles is important. In previous researches, empirical/computational investigations into this subject is reported. Fincham [1] compared outputs of several MD simulations differing only in the used time step. This research outputs indicated much larger time steps than usual can be employed without producing significant errors in observed thermodynamic, structural or dynamic properties.
Furthermore, previous researches revealed the time steps with larger value can be used in common MD simulations than has hitherto been thought possible. In conventional constant energy MD results with a large time step are probably reliable as long as there is no appreciable temperature drift; this is better rule-of-thumb than one based on an arbitrary requirement on the total energy fluctuations. The maximum time step acceptable under the stated criteria for constant energy MD is probably the best guide to the time step choice for constant temperature simulations, if very accurate thermodynamic results are required. Otherwise some further increase is possible. Contrary to what is often supposed, thermodynamic quantities are more susceptible to algorithm errors than single-particle dynamical quantities. This is because of their sensitivity to small structural changes. Numerically, with constrained temperature dynamics one can get a very good velocity autocorrelation function out to 1.2 ps with the huge time step of 60 fs.
Reference
[1] Fincham, D. (1986). Choice of timestep in molecular dynamics simulation. Computer Physics Communications, 40(2-3), 263–269. doi:10.1016/0010-4655(86)90113-x.