- Perhaps you forgot to send to lammps-users.
- You results do not suggest any “divergence”. You can call it “variance”.
- From 5 independent simulations, you can find an average and an error estimate, reporting the thermal conductivity as k_ave ± k_err.
- You have confused between “correlation time” and “production time”. You need to plot the heat current auto-correlation function and the Green-Kubo integral as a function of the correlation time. You current figure just shows some thermal conductivity values (perhaps the cumulative average of the Green-Kubo integral) as a function of the (cumulative) production time, which does not show the correct physics of Green-Kubo relation. The time scale in your problem should be of the order of 1 ps, not 5 ns.