[gmx-developers] MC with PME

[hess at sbc.su.se]

> On 2010-06-14 21.09, André Assunção S. T. Ribeiro wrote:
>> Dear all,
>>
>> I have been working to implement a Monte Carlo "integrator" for GROMACS.
>> For the last couple of months I've been trying to improve performance of
>> the algorithm. For that end, I have made adaptations to the ns and force
>> evaluation routines. Energies for each interaction are stored
>> separately, so one can just calculate the needed energy terms in each
>> trial move. When treating nonbonded interactions with a cutoff, MC
>> performance is more than 10-fold greater than MD for a simple system
>> containing 216 SPC waters (by performance I mean number of simulation
>> steps/CPU time, not actual sampling efficiency). When I use PME, the
>> situation is far worst, with an approximate 3-fold increase. When I try
>> to call gmx_pme_do for only a small number of atoms performance
>> decreases when compared to the total calculation. For simple Ewald sum,
>> there is a 10-fold increase when a single molecule is calculated,
>> however, it is still a lot slower than the worst PME result. Would
>> anyone have any ideas on how to increase performance here?
> If you use MC you do not need to evaluate forces, this will remove 30%
> of the PME time roughly. It used to be coded in there at some stage, but
> it might have been removed.
>

Furthermore, I implemented functionality to calculate the PME mesh
energy at a single location only. This makes the PME cost for multiple
trial moves nearly identical to a single PME evaluation.
But I would still think it is far more expensive than MD,
since you move only one or a few particles at a time.

Berk

>>
>> Cheers,
>> Andre.
>>
>
>
> --
> David van der Spoel, Ph.D., Professor of Biology
> Dept. of Cell & Molec. Biol., Uppsala University.
> Box 596, 75124 Uppsala, Sweden. Phone:	+46184714205.
> spoel at xray.bmc.uu.se    http://folding.bmc.uu.se
> --
> gmx-developers mailing list
> gmx-developers at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-developers
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-developers-request at gromacs.org.
>