[gmx-users] umbrella versus constraint in the pulling code

[Thomas Schlesier]

Finally I had more time to look into this. My impression is that the 
constraint-pulling is somewhat buggy and it only works truly with 
'pull_geometry = distance', the other two options (direction or 
position) are more or less broken. I tested GMX 4.0.7 and 4.6.5 and got 
similar results.
Test system: Two argon atoms in vacuum, cubic box with 10nm length (so 
pbc shouldn't be an issue) at 10 K (v-rescale thermostat) and no 
pressure coupling. Distances between the particles were checked with 
'g_dist'.

'pull_geometry = distance': Everything is fine and the distance 
increases linearly (as expected)

'pull_geometry = direction': For fast pulling (v=0.1 nm/ps) I got the 
expected results, but for faster pulling there were strange fluctuations 
(GMX 4.0.7) or the distance increased too fast (GMX 4.6.5).
But I have no idea why this velocity dependence occurs.

'pull_geometry = position': Is totally broken and one always sees big 
fluctuations, which break the simulation if the distance between the 
particles becomes larger than half the box vectors.
One thing which is strange for the last setup is, that my initial 
temperature is always 0 K, whereas i get an sensible value for the other 
two setups (and using all the time 'gen_seed = 12345').

I will also file a bug report at 'http://redmine.gromacs.org' with the 
input files I used.

Greetings
Thomas


Am 03.12.2013 01:43, schrieb 
gromacs.org_gmx-users-request at maillist.sys.kth.se:
> Hi Thomas -
>
> I would also expect that it works the same way as before and as
> described in the manual. However, thus far I am not very successful.
> The main application of the "constraint" option is probably to
> automatically get evenly spaced windows during pulling, but I am not
> able to achieve this. It is quite mysterious.
>
> Vitaly
>
> Dr. Vitaly V. Chaban
>
>
> On Mon, Dec 2, 2013 at 10:18 AM, Thomas Schlesier<schlesi at uni-mainz.de>  wrote:
>> >Hi,
>> >I never used the pull-code with GMX 4.5.x or 4.6.x,but I would assume that
>> >it'sthe same as in 4.0.x:
>> >'Umbrella': Due to the harmonic potential the force and the distance between
>> >the two particles will fluctuate.
>> >'Constraint': Only the force will fluctuate. The (relative) distance between
>> >the two particles should either stay constant (if v=0) or increase linearly
>> >(v.ne.0). But still the position of both groups can fluctuate, but only in
>> >such a way that the constraint isn't violated.
>> >Greetings
>> >Thomas
>> >
>> >
>> >Am 30.11.2013 15:41, schrieb
>> >gromacs.org_gmx-users-request at maillist.sys.kth.se:
>>> >>
>>> >>Could anyone please explain the difference between how "umbrella" and
>>> >>"constaint" options in the pulling code work. Based on the manual, I
>>> >>expect that the only difference
>>> >>is that umbrella used harmonic potential to bind pulled group to the
>>> >>reference group (point), while "constraint" uses shake algorithm to
>>> >>rigidly bind the groups. However, the simulations shows that this is
>>> >>not true. With all other options being the same, umbrella provides
>>> >>expected pulling, while constraint exhibits rather fluctuations of the
>>> >>pulled group nearby starting position. What do I miss?
>>> >>
>>> >>Dr. Vitaly V. Chaban
>>> >>
>>> >>output from the "constraint" simulation with moving reference point
>>> >>(i.e. pull_geometry = position)
>>> >>z-coordinate
>>> >>   6.38675
>>> >>   5.97254
>>> >>   5.88948
>>> >>   5.86992
>>> >>   5.72523
>>> >>   5.59002
>>> >>   5.84313
>>> >>   5.83232
>>> >>   5.88722
>>> >>   5.83665
>>> >>   6.00198
>>> >>   5.96873
>>> >>   5.57108
>>> >>   5.35213
>>> >>   5.47916
>>> >>   5.46968
>>> >>   5.40241
>>> >>   5.41725
>>> >>   5.4732
>>> >>   5.45464
>>> >>   5.61582
>>> >>   5.66366
>>> >>   5.8005
>>> >>   5.9086
>>> >>   6.10943
>>> >>   6.11296
>>> >>   6.21684
>>> >>   5.95532
>>> >>   5.89226
>>> >>   6.00983
>> >