I am trying to compute thermal conductivity of bulk silicon (using Stillinger Weber potential) at 300K via LAMMPS using the reverse NEMD approach (fix/thermal conductivity command). I ran a few simulations and notice that the thermal conductivity is almost a factor of 100 lower than what I would expect (i.e. around 1 W/mK as opposed to 100 W/mK). I am wondering if anyone has any experience with these simulations in terms of getting good match between experimental thermal conductivity and LAMMPS predictions (for silicon). Specifically:
1) The kinetic energy is exchanged every certain number of timesteps, and temperature gradient is also computed at a certain frequency. I wonder if any LAMMPS users have identified an optimum combination for these parameters.
2) Is there a total time duration for which these simulations need to be run?
I have currently specified the energy exchange to take place every 50 timesteps, and my timestep size is 0.001 psec. In addition, I am computing the temperature gradient every 1000 timesteps.