[gmx-users] Re: Freeze Groups in Gromacs 4.5.x (bug?)!!!

[Vitaly Chaban]

Hey, Miguel -

Hmm... I think, frozen particles will give unnatural pressure of your
system, that's why barostats do not do their jobs as expected. Maybe
it could be interesting to rescale the coordinates of the frozen atoms
with the box vectors (with certain corrections in pressure
calculation)... but it should be another keyword to control such
trick.

--
Dr. Vitaly Chaban
University of Rochester


> Indeed, the source of the problems comes from both the constraints and
> the pressure bath. The temperature bath did not seem to interfere.
>
> Now, the problem is clear, to remove constraints is not only compulsory
> but also recommendable for such simulations.
>
> However, the pressure bath (which is the major source of the problems)
> should not play such a role. This is a limitation of the current
> implementation of GROMACS or just a plain BUG.
>
> I think most people would agree that one should not be limited to NVT
> simulations when using freeze groups.
>
> A previous version (3.2.1) that we have been using in our group was
> completely immune to both pressure bath and constraints when simulating
> a frozen protein in water. As far as I remember, this older version did
> a normal step integration with all its requisites and, in the end, just
> took back the previous positions for the frozen group (protein).
>
> My question remains...
>
> Will we be able to use freeze groups together with a NPT simulation in
> Gromacs 4.5.x???
>
> Any comments from the developers are welcome...
>
> Regards,
> Miguel
>
> On 11/11/2010 1:07 AM, Mark Abraham wrote:
>> On 11/11/2010 3:14 AM, Miguel Machuqueiro wrote:
>>> Dear users and developers,
>>>
>>> In GROMACS 4.5.x when using "freeze groups" on a Protein in a
>>> Protein+Solvent system, I noticed that the coordinates of a few atoms
>>> (not many, but a few) are changed after a short period of MD (~0.2
>>> ps). Of course, this is wrong!
>>>
>>> Are there any options in MDP/GROMPP/MDRUN that can be originating
>>> this? Or it is some bug in the code?
>>>
>>> A copy of my input file follows below.
>>>
>>> Note that this version of GROMACS crashes if one uses "constraints" in
>>> LINCS together with "freeze groups". Hence the "constraints = none" line.
>> Indeed, because imposing the constraints can perturb the (frozen)
>> positions slightly, leading to numerical issues... Similarly, if you
>> couple the frozen groups to a heat bath, you're going to find that
>> they're not actually frozen. And pressure-coupling is not known to
>> always play nicely with frozen groups either (search the archive for more).
>>
>>> Any help is appreciated.
>>>
>>> Regards,
>>> Miguel