Convergence of thermal conductivity: Green-Kubo

Dear LAMMPS users,

I’m trying to calculate thermal conductivity of uranium dioxide (10x10x10 at 500-3500K) using the Green-Kubo formalism. The obtained numerical values of thermal conductivity at these temperatures are acceptable. I visualized the data of averaged autocorrelation function of the heat flux from the file J0Jt.dat. Graph of ACF is good, but the convergence of integral of ACF is poor in comparison with other MD calculations. I attached pictures of my integrals and the graphs of the other results for comparison. I tried to change the settings fix ave /correlate (Nevery Nrepeat Nfreq) but it not produced significant results in improving the convergence.

What else can I do to improve the convergence of ACF integral using lammps?

Any comment or suggestion would be highly appreciated.
Thank you

integral of ACF 500K.jpg

integral of ACF 1500K.jpg



Have you tried reducing timestep while keeping the same correlation time? If the ACF has oscillations that are fast compared to your sampling, then the trapezoid integration won’t work very well. So a smaller timestep might allow better sampling of the ACF…