[lammps-users] computing time correlations in LAMMPS

Since there is a lot of traffic on the mail list about compute thermal
via Green-Kubo - there is new 18Aug patch with a new fix ave/correlate
and some time-integration functions added to variables, to make this easier.

Thanks to folks from Materials Design and to Reese Jones (Sandia) for
help with this.


Let me add my two cents' worth.

When computing the time auto-correlation function of some particular
time series data, such as the instantaneous heat flux, it is much more
efficient (and statistically sound) to use Fast Fourier Transform
methods, especially for large data sets, rather than doing the double
loops. The time series data can be dumped and then analyzed with any
software (Matlab, Mathematica, home-brew, etc) with FFT. See pp.
188-191 of Allen and Tildesley where this is explained in detail.