Thermal conductivity : Green Kubo

hi Deena,
we used NVT since the ensemble averages < J(0) . J(t) > are in the NVT ensemble. I think the practice of using initial conditions from Boltzmann and then running NVE dynamics is justifiable but we chose not to.


Actually, using NVT dynamics is not a rigorous way of evaluating a Green-Kubo relation in MD. A thermostat such as Nose-Hoover or Andersen is only proved to give you correct equilibrium properties. There does not exist a proof for any of these thermostats that guarantees correct calculations of dynamical properties (such as time correlation functions) in general (there are systems that seem to work, but there are systems that definitely don’t work).

The original derivation of the Green-Kubo relations assumed classical Hamiltonian mechanics with fixed volume with the assumption that the equilibrium distribution for the systems is canonical. Therefore, a rigorous way to evaluate a Green-Kubo relation is to average over multiple NVE trajectories from systems that are chosen from a canonical distribution (the systems should first be equilibrated with NVT dynamics, then run NVE).