Thermostat

From OxDNA

The best thermostat implemented in the simulation code (john thermostat) is a simple thermostat that emulates Brownian dynamics. The system is evolved integrating Newton's equations of motion ('NVE' ensemble) for a given (small) number of steps. Then the velocity and momentum of each particle are refreshed, with a given fixed probability. The new velocities and momenta are chosen according to the Maxwell distribution of the temperature at which the simulation is run. This approximates a Brownian dynamics on time scales much longer than the refresh interval.

  • dt time steps of the integration
  • newtonian steps number of steps between refresh attempts
  • pt the probability with which each particle gets its velocity and momentum refreshed at each attempt.
  • diff_coeff the overall monomer diffusion coefficient resulting from the thermostat. The code internally sets pt to get this value. Specifying pt will override this, regardless which comes first in the input file.

The algorithm works as follows: the system is evolved for a number of steps equal to newtonian_steps according to Newton's equations of motion. Then for each particle a random number is extracted; if it is larger than the value for pt (either set explicitly or derived from diff_coeff) the particle is left untouched. If the random number extracted is lower than the value of pt, each of the components of the the velocity and angular momentum of the particle are refreshed according to the Maxwell distribution dictated by the value of the temperature.

A completely Brownian dynamics (on the time scale set by dt) can be obtained setting pt = 1 and newtonian_steps = 0. Of course, this makes little sense.

Increasing the value of pt makes the dynamics more damped, so that overall diffusion is slower but local motion is somewhat better explored. We found that a good thermostat setting to study diffusion-limited events is to set diff_coeff = 2.5 and newtonian_steps = 103. If one is not limited by diffusion, internal relaxation can be sped up by lowering the value of diff_coeff by a factor 2 or 4.