[lammps-users] Thermal condcutivity

I implemented the calculation of heat flux once. For thermal conductivity it is generally more efficient to set the heat flux and measure the temperature gradient since it takes longer to get an accurate heat flux. Also, with the velocity Verlet integrator, I think I had to accumulate some force times distance terms in the force loop to later multiply by velocity terms. In short, it is both less efficient and more complicated to do it that way.

Brian

For thermal conductivity it is generally more efficient to set the heat flux and measure the temperature gradient since it takes >longer to get an accurate heat flux.

Bingo. This is why people should try fix thermal/conductivity. This is
the reason the Muller-Plathe algorithm was developed, and is what
fix t/c implements.

Also, with the
velocity Verlet integrator, I think I had to accumulate some force times
distance terms in the force loop to later multiply by velocity terms.

And with fix t/c you don't need to do this.

Re: below, using fix nvt to set temperatures in regions. I don't think
it is a sound idea. Fix nvt assumes the collection of atoms it is storing
a history about is a fixed set. You can do it on dynamic regions, but
I don't think that's true Nose'/Hoover.

Steve