Hi Lammps Users,
I have battled with this problem for a while and I was wondering if anyone can help me. I am trying to simulate graphene to calculate its thermal-conductivity. I am currently using the direct-thermostatting method and getting very small values ~23W/mK.
If anyone has any suggestions they would be greatly appreciated.
ter.stat (3.06 KB)
I gave a quick look at your input (try simplifying it as much as possible when requesting help on a mailing list), it looks like you’re using a rather small system (not sure since you haven’t attached the structure, but looking at thermostat regions… ). When using NEMD, size effects can be strong in systems where you have many propagating modes actively contributing to the thermal conductivity (silicon, etc). That’s especially true in graphene where size effects are probably the strongest - some people have even suggested a log divergence of the thermal conductivity wrt sample size. Check the literature for that. I remember a paper where the authors performed NEMD with system length > 1µm : K wasn’t converged yet.
You may want to perform extrapolation at infinite size (several simulations with different sample sizes in the transport direction), plot the inverse of the thermal conductivity = f(inverse of the system size), and try to extrapolate at 1/L = 0. For many systems (but not all of them), 1/K scales linearly with 1/L.
You can also compare with equilibrium molecular dynamics. Here for graphene and a short-range potential like Tersoff, the heat flux correlation function is typically converged for systems >~ 5000-6000 atoms.
Also, you are performing NPT then equilibration at constant V then sampling. Between the NPT and the constant V equilibration, you should compute the equilibrium lx and ly values. And 100ps is maybe a bit short of a thermalization run.