[lammps-users] Thermal conductivity


Does any one have a recommendation for calculating thermal conductivity of a solid with LAMMPS using the Tersoff potential. I am interested in using the Green-Kubo method, but is this compatible with the Tersoff potential? Looking at the mailing list archives it seems like this might be incompatible.

Alternatively, is it possible to use the Muller-Plathe method for my system given that there are multiple atom types? (I am concerned by the “the masses of all exchanged atom pairs should be the same to use this fix” in the manual.)

Another alternative would be to use the Einstein relation, which is equivalent to Green-Kubo. Has any one used this and is it compatible with LAMMPS?


Hi, Eric:

Tersoff does not usually give correct thermal conductivity. It’s phonon dispersion is way off at least for GaAs. There was one paper trying to reproduce the correct phonon dispersion but that involves parameter changing and other properties are off as a result. I think current LAMMPS version is compatible with Tersoff as long as the assumption of equal dividing of the contributions (3-body,4-body) is valid.

You can use Muller-Plathe method since you can group the same kind of atom to a certain group for velocity exchange.

Einstein relation has the same problem of equal partitioning the energy contributions. You can out put your energy per atom. For both Green-Kubo and Einstein, may be you need good ensemble average to get convergence.


The GK method in LAMMPS, as part of compute heat/flux, will work
fine with Tersoff potentials - i.e. it does what it says it does for
3-body or many-body
iteractions, not just pairwise. Whether it gives the right answer is
a different question.
I.e. all the points made by Tengfei are valid concerns.