I went through the Schelling et al. 2002 paper in Physical Review B (attached) in which they describe their “direct approach” for computing thermal conductivity in silicon. Their approach does appear to be a variation of the reverse NEMD approach implemented in LAMMPS (fix/thermal conductivity).
In the Schelling et al. approach, they “add” kinetic energy to a slab/layer of atoms at a certain location in the domain, and “remove” the same kinetic energy from a different slab/layer of atoms in the domain. This kinetic energy addition/subtraction at each timestep is done via a velocity rescaling of the atoms in the relevant layers/slabs, while ensuring that energy is conserved.
In LAMMPS (fix/thermal conductivity), the kinetic energy of the hottest atom in the “cold” layer is exchanged with the kinetic energy of the coldest atom in the “hot” layer.
It will be interesting to try out the Schelling et al. approach in LAMMPS. Can you please advise on how I can get started with this?
Schellingetal-2002.pdf (147 KB)