[gmx-users] Possible bug in g_hbond with implicit solvent
Mark.Abraham at anu.edu.au
Fri Aug 19 17:57:47 CEST 2011
On 20/08/2011 1:45 AM, Justin A. Lemkul wrote:
> Mark Abraham wrote:
>> On 20/08/2011 1:26 AM, Justin A. Lemkul wrote:
>>> Hi All,
>>> I wanted to solicit some feedback on what I think may be a bug
>>> related to periodicity handling and/or implicit solvent
>>> simulations. I ran some simulations of a few model peptides (some
>>> helices, hairpins, etc) and I got what I believed to be very
>>> inflated results for backbone hydrogen bonds. For instance, a
>>> helical peptide with only 17 residues had, on average, 23.5 hydrogen
>>> bonds within the backbone alone, which, judging from the
>>> configurations, couldn't really be possible.
>> Does the structure file input to g_hbond have a spuriously small box?
> It does, and I suspect this is where the problem ultimately lies. For
> the simulation, with "pbc = no" the box vectors get ignored and
> everything is fine. For analysis, however, it seems that the PBC
> routines are still trying to fit everything into that tiny box and
> thus basically all possible H-bonds are detected as being present
> because the distances are so small.
Sure. In your case, mdrun learns to ignore the box from the .mdp file
setting, but I presume the box values still get copied around in the
same fashion as for NVT PBC... A pre-emptive work-around might be to
generate the simulation .tpr with a spurious large box, which then gets
transferred everywhere else.
Better file formats would be the only reasonable fix, and that would not
be all that reasonable...
I don't know how to avoid people rediscovering this phenomenon in the
> Re-generating the .tpr file with a new, larger, box does not fix the
> problem, likely because the .xtc still has the tiny box. Manipulating
> the trajectory with trjconv -box -center seems to work, though. So I
> suppose this is an acceptable workaround, though it still may be a
> problem worth looking into.
> Thanks for the inspiration; hadn't thought about trying to change the
> box after the fact :)
More information about the gromacs.org_gmx-users