Comparing the efficiencies of stochastic isothermal molecular dynamics methods (Q634158)

The authors put forward a quantitative criterion, which they call ``efficiency'', in order to determine how well a given thermostatting method can drive the dynamical system to the target temperature with small perturbations. In the study of time-dependent properties, such as autocorrelation functions or diffusion constants, by means of molecular dynamics, the temperature rather than the energy is fixed, and one needs a thermostat to rapidly drive the system to the Gibbs canonical distribution at a prescribed temperature. However, the thermostat induces cumulative deviations from the Hamiltonian dynamics during the time necessary to bring the system to equilibrium at the thermostat temperature. As a figure of merit for the trade-off between these two competing effects, the authors introduce the ``efficiency'' defined as the reciprocal of the amount of error (in the velocity autocorrelation function) that accumulates during the time that the system is brought to equilibrium at the thermostat temperature. The authors provide analytical estimates for this quantity for the case of several thermostats, namely the Langevin thermostat, the Nosé-Hoover-Langevin thermostat and the recently introduced ``stochastic velocity rescaling'' thermostat. The analytical work shows that the Nosé-Hoover-Langevin thermostat and the stochastic velocity rescaling thermostat are more efficient by a factor of order \(n\) with respect to the Langevin thermostat for systems with \(n\) degrees of freedom. The analytical result is also supported by numerical experiments.