Effect of the Berendsen thermostat on dynamical properties of water

The effect of the Berendsen thermostat on the dynamical properties of bulk SPC/E water is tested by generating power spectra associated with fluctuations in various observables. The Berendsen thermostat is found to be very effective in preserving temporal correlations in fluctuations of tagged particle quantities over a very wide range of frequencies. Even correlations in fluctuations of global properties, such as the total potential energy, are well-preserved for time periods shorter than the thermostat time constant. Deviations in dynamical behaviour from the microcanonical limit do not, however, always decrease smoothly with increasing values of the thermostat time constant but may be somewhat larger for some intermediate values of τ B , specially in the supercooled regime, which are similar to time scales for slow relaxation processes in bulk water. 2 The ideal ensemble to extract dynamical information from a molecular dynamics simulation is the microcanonical (NVE) ensemble [1, 2]. Since the total energy is conserved in this ensemble, the Newtonian equations of motion can be assumed to represent the natural evolution of the system, subject to the accuracy of using classical mechanics to describe the atomic dynamics. The constant energy (E) and volume (V) conditions do not, however, correspond to the most common experimental conditions and therefore it is often desirable to implement MD simulations in the canonical (NVT) and isothermal-isobaric (NPT) ensembles. An additional reason for preferring the NVT ensemble is that when long run lengths are required, as in the case of studies of slow dynamics in liquids or glasses, there may be significant energy drift in the NVE ensemble. Two types of approaches have been developed to adapt MD simulations to the canonical ensemble: (i) extended Lagrangian methods, such as Nose-Hoover thermostats and (ii) resampling or rescaling of velocities, as in the case of the Andersen and Berendsen thermostats. The extended Lagrangian methods will generate the true canonical distribution of velocities. The rescaling approaches will only ensure that the average kinetic energy of the system corresponds to the expected value at the desired temperature but have the advantage that they can be combined very simply with the Verlet algorithm. In both approaches, the degree of perturbation of the real time evolution of the system can be adjusted by manipulating various thermostat parameters. The Berendsen thermostat represents a proportional scaling of the velocities per time step in the algorithm with the scaling factor being given 3 by λ …