Recently, I have been investigating the thermal conductivity methods that LAMMPS offers. From what I have found, there are the Muller-Plathe and Green-Kubo methods which can be used. I am interested in determining the thermal conductivity in Silicon crystal, which I have seen on the lammps mailing list, is not a trivial task.

Thus far, I have only implemented the Muller-Plathe method, as it does not appear to require nearly the same computational time as Green-Kubo. However, I first noticed that the thermal conductivities were fairly small. After reading some more, I found a LAMMPS workshop ppt that talked about the effect of the system size on the results, which sure enough, increases the thermal conductivity. Then they showed how one can extrapolate to infinite length in an attempt to calculate thermal conductivity in bulk materials.

Can anyone offer advice on the validity of this method? I also notice that SW potential gives higher thermal conductivities than tersoff (guessing that is just due to differences in the force field, as no force field is perfect).

Another question: How linear can I really expect the temperature gradient to be? I attached a figure that shows the temperature gradient for a 4x4x96 (lattice units) simulation at 500 K. Unfortunately, despite following the exact method as listed in the PPT, the thermal conductivity I get is about 12 W/mK as opposed to 19.5 W/mK listed at the workshop ppt website. My N value in the fix thermal_conductivity is 50 (looks better with 50 than 100)

(http://lammps.sandia.gov/workshops/Feb10/Dave_Rigby/MD_LammpsWorkshop_Feb2010.pdf)

I am not sure if this could be an artifact of the temperature gradient. Is this best way to measure the gradient to take the point at N/2 and N and find the slope? Clearly, the results are not linear. I only ran the simulation for thermal conductivity for 200 ps (last 100 ps is where I collected data). I know this is not a very long time.

Basically, I am just looking for people who have investigated this sort of thing before and any advice they have. I want to reproduce the thermal conductivity of silicon with tersoff FF, but at 300K, the thermal conductivity that I find is ~3 W/mK with Tersoff and 5x5x200 simulation.

The version of LAMMPS I am using is fairly recent (30 July 2016), so I imagine I am on the right track with everything (not getting any error messages), just finding the thermal conductivity to be much lower than I was expecting.

Thanks,

Jackson