Re: [ESPResSo-devel] Verlet lists

[Axel Arnold]

Hi!

On 01/10/2013 10:30 AM, Josh Berryman wrote:
> Hello Devs, happy new year.
>
> Anwar and I are looking at the Espresso code and trying to set up a
> means to carry out various pairwise particle comparisons efficiently,
> for purposes of finding reaction coordinates.  So far we have been
> happily adding functions to statistics_cluster.c with the intention of
> eventually submitting anything which turns out to be useful back up.
>
> The obvious tool for these pairwise comparisons is a verlet list, and
> we have been happy enough using one for serial calculations, but the
> code hooks to build one don't seem to be present at the moment for
> every type of parallel calculation (DOMDEC, LAYERED, NSQUARE).
>
> We have found, though, that DOMDEC seems to have something already,
> and that for LAYERED and NSQUARE the functions layered_calc_ia() and
> nsq_calc_ia() do nearly the same job, so our thinking is that it
> should be simple enough to add
>
> layered_build_verlet()
>
> and
>
> nsq_build_verlet()
>
> to the code based on the existing layered_calc_ia() and nsq_calc_ia().
>
> Does anyone have any strong opinions on the merits of doing this, or
> suggestions of a different tack?

There are two points that make things MUCH more complicated than this:

1. the data structures for the Verlet lists are at present in the domain 
decomposition data structures, not the generic cell system data 
structures. Therefore, you would actually need to make serious changes 
to the Verlet list code in order to use one generic Verlet list per 
processor. Otherwise, you need to write specific Verlet lists for each 
cell system, which then would be very unpleasant to use.

2. There is only one Verlet list built to satisfy a particular maximal 
cutoff. The point of the layered and nsquare cellsystems is however, 
that there is no such specific cutoff in some or all directions (that is 
done for some esoteric electrostatics methods, which need N^2 
calculations). Therefore, you would need to give both a new 
infrastructure to specify a cutoff, and that would be solely for 
analysis purpose, not for force/energy/pressure calculation. 
Alternatively, we could make a specific Verlet list infrastructure.

Cheers,
Axel

--
JP Dr. Axel Arnold           Tel: +49 711 685 67609
ICP, Universität Stuttgart   Email: address@hidden
Allmandring 3
70569 Stuttgart, Germany