[gmx-developers] distance restraints and domain decomposition

[David van der Spoel]

Den 2018-06-07 kl. 16:33, skrev Bert de Groot:
> On 06/07/18 16:22, Berk Hess wrote:
>> If you are sure the x, y and z components are shorter than half the 
>> box diagonal, your results should be fine.
> 
> so far, so good. The involved distances behave as expected based on the 
> restraint input in the simulations thus far so the restraints seem to be 
> doing what they should.
> 
>>
>> If you think such a setup is important to run efficiently in parallel, 
>> we can think of ways to make this work.
> 
> thanks. Right now I cannot judge how frequently we'll need a setup like 
> this in the future. For the current set it appears we'll be able to get 
> decent throughput on available openMP nodes.
> 
I implemented in genrestr the option to generate distance restraints 
between sets of atoms, e.g. between all C-alpha in a protein (which I'd 
like to use in the context of SAXS refinement), so there may be more 
applications.

One way around the periodicity problem would be if the distance 
restraint code stored the distance vector from the previous step and 
then at the next application of the code checks whether the vector has 
flipped. This assumes the input structure should be correct.



> cheers
> 
> Bert
> 
> 
>>
>>  Cheers,
>>
>> Berk
>>
>>
> 
> 


-- 
David van der Spoel, Ph.D., Professor of Biology
Head of Department, Cell & Molecular Biology, Uppsala University.
Box 596, SE-75124 Uppsala, Sweden. Phone: +46184714205.
http://www.icm.uu.se