Dear lammps user’s:
i would ask if it is possible to calculate directly velocity autocorrelation function using the command:
fix ave/correlate and how to do this if it is possible
because i find in archive that the velocity must dumped firstly and then analysed to calculate VATF
best regards
Been a while since I looked at this, but I believe
the answer is no. Fix ave/correlate is for global
quantities, not per-atom ones. The reason we've
opted not to try to do this on-the-fly is it requires lots of old
data be stored. E.g. many old values of velocity
per atom. Which is much easier to access in a post-
processing mode. Maybe someone else wants to comment.
Steve
Been a while since I looked at this, but I believe
the answer is no. Fix ave/correlate is for global
quantities, not per-atom ones. The reason we've
opted not to try to do this on-the-fly is it requires lots of old
data be stored. E.g. many old values of velocity
per atom. Which is much easier to access in a post-
processing mode. Maybe someone else wants to comment.
i think it would make sense to do auto-correlations
of per-atom quantities on the fly, particularly if one
is not after relaxation times, but wants to integrate
the resulting ACFs accurately. doing this in postprocessing
would require storing an insane amount of data *and*
would need to be parallelized in a similar way as LAMMPS
already is. with clusters now having 2 GB RAM per
core or more, the memory consumption should not
be that much of a problem.
that being said, if it is about computing spectral densities
alone, there is a way to do this in a less memory consuming
way, by doing a convolution and transform to fourier space
while accumulating the data as it is done in this VMD plugin.
http://www.ks.uiuc.edu/Research/vmd/plugins/signalproc/
the only downside of this method is that one has to decide
up front on resolution, oversampling, and maximal frequency
and won't get intermediate results (except when there is
oversampling). this method also has the benefit that no
arbitrary cutting off of the ACF is needed or a maximum
entropy fit or something similar to get the spectrum.
cheers,
axel.
How many old velocity values do you need (per-atom)
to do a VACF?
Steve