[gmx-developers] unintended fatal error and pull code

[Berk Hess]

I put this check in some time ago.
Before that there was no check.
You should not pull groups up to a distance of more than half the 
shortest periodic image distance.
If you don't do that, you don't need a check.
My bug report, and I assume this post, were for cases were the distance 
was smaller than 0.5 * s.p.i.d.,
so a false positive.
Note that also the maximum distance allowed was reported incorrectly.

The chance on false negatives is extremely small and would in most cases 
cause a crash within a few steps.

Also note that shifts to periodic images have nothing to do with this. 
Those are taken into account correctly.

Cheers,

Berk

On 12/09/2011 04:28 PM, chris.neale at utoronto.ca wrote:
> That sounds like a pretty serious error. I checked the redmine report. 
> Are you sure that this would lead to visible artefacts or crashes? It 
> doesn't take too many highly non-boltzmann sampling values to totally 
> change a PMF.
>
> Sorry to hijack the original post, I realize that the original 
> question has not been answered yet.
>
> Chris.
>
> Quoting Berk Hess <hess at kth.se>:
>
>> Hi,
>>
>> There is a bug in this distance check. It should check dr[m]*dr[m],  
>> not dr[m].
>> Two weeks ago I filed an issue and fixed it for the upcoming 4.5.6:
>> http://redmine.gromacs.org/issues/844
>>
>> Berk
>>
>> On 12/09/2011 05:34 AM, Alan Chen wrote:
>>> Hi all:
>>>
>>> I just wanted to report what I thought was an overly restrictive fatal
>>> error case in the 4.5.5 pull code. Previously, I've been very happily
>>> simulating an optical-tweezers type dissociation experiment (i.e.
>>> non-equilibrium pulling) using version 4.0.5 where I have a
>>> bimolecular complex in solvent aligned in a long, narrow box that I
>>> dissociate by applying a constant force to two atoms on opposite ends
>>> of the complex (i.e. two pull groups pulled in opposite directions
>>> along the long box dimension).  This is ideal, as the actual single
>>> molecule experiment I'm comparing to applies a constant force to two
>>> tethered groups of an aligned complex (albeit at a much longer
>>> timescale).
>>>
>>> Upgrading to 4.5.5 I found my simulations quickly aborted with the
>>> error "Fatal error:
>>> Distance of pull group 2 (2.687620 nm) is larger than 0.49 times the
>>> box size (7.520599)"
>>>
>>> searching the code referenced in the error message, I see this line
>>> was tripped in pull.c
>>>
>>>  if (max_dist2>= 0&&  dr2>  0.98*0.98*max_dist2)
>>>     {
>>>         gmx_fatal(FARGS,"Distance of pull group %d (%f nm) is larger
>>> than 0.49 times the box size (%f)",g,sqrt(dr2),max_dist2);
>>>     }
>>>
>>> Now this sanity check makes sense if I was applying a distance-based
>>> bias, as the minimum distance vector will abruptly shift to one of the
>>> periodic images and not do what you would intuitively want a
>>> distance-based bias to do. However, in my case of simulating a force
>>> microscopy style experiment with constant force along the Y axis there
>>> is nothing wrong with having such widely separated pull groups.  For
>>> my own purposes, I can easily just comment out this line and recompile
>>> but I just wanted to suggest that maybe in future versions this error
>>> should only be tripped if there is a distance-based bias being
>>> invoked?
>>>
>>> Cheers,
>>>
>>> Alan Chen
>>>
>>>
>>> chena7 at rpi.edu
>>> Postdoctoral Fellow
>>> Garcia Group
>>> Biotech 2nd Floor RM 3237
>>> Rensselaer Polytechnic Institute
>>> 110 8th Street
>>> Troy, NY 12180
>>
>> -- 
>> 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.
>
>
>