[gmx-developers] Velocity and force trajectory restricted to a group

Andreas Baer andreas.baer at fau.de
Wed Oct 10 13:30:04 CEST 2018


thank you for the comments. I will think about it.


On 09.10.2018 22:42, David van der Spoel wrote:
> Den 2018-10-09 kl. 18:01, skrev Berk Hess:
>> Hi,
>> We don't have such functionality planned.
>> But velocities are extremely correlated over several time steps. You 
>> should be able to write output every 10 steps without significantly 
>> reducing the statistics.
> Depends what one is trying to sample of course. If hydrogen vibrations 
> are being sampled both short saving times and decent precision are 
> needed. However with a time step of 2 fs there likely are no hydrogen 
> motions...
>> Alternatively, if you are running in without domain decomposition, 
>> you can quickly hack a printf statement in the code.
>> Cheers,
>> Berk
>> On Oct 9, 2018 17:48, Andreas Baer <andreas.baer at fau.de> wrote:
>>     Dear developers,
>>     within my simulations, I encounter the problem, that I have to save
>>     the forces and velocities of the whole system, even though I need
>>     only those for very few atoms (in my case: a small particle e.g. a
>>     fullerene, further details: see below). This generates a lot of
>>     unnecessary data and also slows down the simulation a lot, as I have
>>     to write out the trajectory every step (2 fs).
>>     A possible solution would be to write forces and velocities only for
>>     a group of atoms, as is possible for the compressed trajectory for
>>     positions (.xtc). So far, I did not find any possibility doing that
>>     and also no upcoming features.
>>     Did I miss anything? If no, is this, or a similar functionality
>>     planned for a future release?
>>     If neither of this, is there a chance to do it myself, ether only
>>     for my personnel use or as an official contribution?
>>     Kind regards,
>>     Andreas Baer
>>     Some further details to my simulation:
>>     A nanosized particle (artificial or fullerene) is dissolved in a box
>>     of solvent (10^^5 to 10^^6 molecules). Transport coefficients should
>>     be calculated using the Green-Kubo formalism. Thus, velocities and
>>     forces are required at a very high frequency, so that I write out
>>     every step. As I only need the information for the particle, I get a
>>     huge amount of unused data, that I have to postprocess, to retrieve
>>     the desired information. The biggest problem is, that this slows
>>     down my simulation by a factor of roughly 20, depending on the
>>     hardware used.

More information about the gromacs.org_gmx-developers mailing list