[lammps-users] compute heat/flux

Is this true?

I assumed (maybe incorrectly?) that Green-Kubo would give the ensemble average of overall thermal conductivity of the entire system including any subdomains or inhomogeneities, whereas results obtained with NEMD methods might depend on where the inhomogeneity is located within the simulation cell and whether or not there is symmetry in your system. The system I am considering is a nanoparticle in a fluid. With a NEMD method I would need two particles, one left and one right of the hot/cold grid region and since is a fluid system there are no guaranties the particles will stay where you think you want them. Any thoughts on this?


Hi Dave,

Please read Chapter 10.3 in the book “Microflows and nanoflows: fundamentals and simulation” By George Karniadakis, Ali Beşkök and Narayana Rao Aluru. From the book, for Xiaoliang’s case of calculating thermal conductivity in heterostrutures, I think it has not been accepted yet; for your case, if you calculate diffusion in nanoflow, this method is accepted. In my opinion, the G-K formula doesn’t consider the phonon boundary scattering which could be introduced by impurity, interface, void and so on. About NEMD method in your case, I have two thoughts:

  1. You can try to not mirror the original system (one particle in a fluid) and perform EMD with period boundary conditions first. Next, mirror the last equilibrium state you get including atom positions and velocities, then perform NEMD. It is possible that the particle will not move a lot. (I don’t know your exact physical model.)

  2. Actually, someone doesn’t mirror the system when using NEMD. You can just choose two fluid regions at the distance of the half length for heating and cooling regions. In this situation, you need to keep in mind that the heat(or mass) flux passing the particle and the one not passing the particle is not the same. You need to calculate them separately.