[gmx-developers] center of mass bug with pulling reference group

Berk Hess hess at cbr.su.se
Tue Apr 27 15:53:34 CEST 2010


This method is computationally problematic for pull groups with many atoms.

O good solution would be not to use a pbc atom, but a pbc coordinate,
which is simply
the COM at the previous step. This would only require a search at start up.
I think I did not implement this yet, because it introduces a dependence
of the pull code
on results from the previous step. I'll think some more about this option.

The paper on the cosine based has been submitted. But I think you would
rather use
the simple, normal center of mass whenever possible.


chris.neale at utoronto.ca wrote:
> Dear Berk:
> you and I discussed this a while ago and I proposed the following
> solution, which I still think is valid. The discussion at that point
> was in regard to pulling with micelles, and you suggested an
> alternative cosine-based method, but I think that the principles are
> the same:
> Every N steps, and separately for each pull and reference group, loop
> through all atoms in the group and, assuming that this atom is the
> pull_pbcatom, determine the RMSD from that atom to each other atom in
> the same group. There must then be one or many "central" atoms that
> return the same minimum RMSD value. One of these "central" atoms is
> then set as the new dynamic pull_pbcatom. In order to allow N>>1, one
> chooses the one of these "central" atoms whose position is closest to
> the center of mass of the entire group after it is put into it's
> appropriate box based on the selected new pull_pbcatom for that group.
> The only downside that I can see with this method is that it might
> offer a false sense of security in that it can fail, although only in
> cases where a static intelligently selected pull_pbcatom would also
> fail. Although it seems to me that this method would be more robust:
> e.g. with the current static pull_pbcmol, when pull_pbcmol is selected
> as the distal carbon of a lipid acyl chain, then if the box is small
> enough that reference atom can fluctuate near the headgroups at the
> same time as a different lipid fluctuates up along the bilayer normal,
> causing a selection of pbc that is other than desired.
> Chris.
> Quoting Berk Hess <hess at cbr.su.se>:
>> Jochen Hub wrote:
>>>>> Thanks a lot Berk,
>>>>> Jochen
>>>> In x-y it should not matter.
>>>> But what might matter is that distances of atoms in the bilayer to the
>>>> pbc atom are larger
>>>> than half the box length, if the pbc atom is in a head group
>>>> instead of
>>>> in the middle of the bilayer.
>>> Aaahh, not again the bloody pbc_atom... :(( Arrrg, I had this problem
>>> already 2 years ago, but I forgot, hmpf.  We fixed the script and now
>>> things are all right. So all my apologies for waisting your time. :)
>>> And thanks a lot for your help!
>>> Best wishes,
>>> Jochen
>> This is a nasty issue that we should really try to automate.
>> The problem is that I am not sure if there is a unique, good solution
>> that would work in 99% of the cases.
>> Berk
>> -- 
>> 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.

More information about the gromacs.org_gmx-developers mailing list