Dear Lammps users,
I am trying to use lammps and Green-Kubo method to study the
thermal transport in thin films. I started with the bulk silicon. I
obtained an average thermal conductivity of around 60 W/m-K at 1000 K
using the SW potentials and a 6*6*6 simulation cell with periodic
boundary conditions, as reported in the literature. The repeatability
of the calculation is very good. I ran 5 simulations and the deviation
was less than 10%. However, when I checked the thermal conductivity
values along the x,y,z directions, I found k11, k22, and k33 were not
equal. Although their average was always around 60 W/m-K, the three
values along different polarizations could hardly reach the same
value, and were not even close. For example, k11, k22, and k33 can be
40, 75 and 68, respectively. I ran several simulations with the same
sampling length and correlation length. It seems that in different
simulations, the magnitudes of the three thermal conductivity
components vary randomly in the range between 30-90 W/m-K while
maintaining almost the same average thermal conductivity. I thought
that the three components should be the same because silicon has a
cubic unit cell. I also ran the example of Green Kubo method (solid
Ar) shown on lammps website and found the similar phenomena. I am
wondering why this phenomena happens. If I want to study the
anisotropies of phonon transport, what can I do to reduce the
uncertainty?
Any comments or suggestions? Your help is greatly appreciated.
Thanks,
Baoling