[gmx-users] System almost blowing up

Dayhoff, Guy gdayhoff at health.usf.edu
Mon Apr 3 20:33:53 CEST 2017


>> Hi,
>> 
>>   I?m receiving a couple (~15) EM did not converge warnings as well as a few (~3) 1,4 interaction
>> warnings during my run. It looks like it starts down the path to blowing up but recovers. Is this ?recovery?
>> system dependent? Should I take these messages (even without the subsequent crash) to indicate an
>> underlying issue with my system/topology or equilibration? I?m not ignoring or circumventing any warning
>> during pdb2gmx, or any other commands.
>> 
>> For more context: I?m running the Drude-2013 forcefield, with a dt of .5fs emtol of 1.0 and niter of 150
>> using the V-rescale thermostat and Berendsen pressure coupling while my systems cell shape relaxes
>> as a continuation from position restrained NVT then NPT ensembles. The starting structures were minimized
>> in vacuum, then solvated and minimized once again prior to the posres equilibration.
>> 
> 
> Such a short dt and strict niter should not be necessary in practice.  The 
> failure of SCF to converge/LINCS warnings is what we typically refer to as 
> polarization catastrophe, so your system is on the brink of instability.  This 
> is one of the inherent problems of SCF in polarizable systems; it often fails to 
> converge and leaves the system potentially unstable.  The reflective hardwall is 
> much more reliable (and faster).  If the instability is happening in your 
> equilibration, that may be OK but it is something you should try to troubleshoot 
> to make sure there's nothing else going wrong.
> 
> -Justin
> 
> -- 
> 

Got it. I have been spending some time investigating the issue i’m having and
after reading shellfc.cpp I was pretty confident the issue was with SCF not converging,
and so my niter and dt were an attempt at looking into that. 

The workflow i’ve been trying to use is…
EM -> Solvate -> EM -> posres NVT V-rescale -> posres berendsen NPT isotropic 1bar w/ V-rescale tcoupl ->
berendsen NPT semi-isotropic various pressures w/ V-rescale tcoupl

then once my cell shape has reached an equilibrium I finally switch to NH thermostat and PR barostat for
the production run. 

If I switch to employing the hardwall (and the lagrangian dynamics for drudes) then I understand it is not
compatible yet with NPT ensembles. So, if I change my work flow as follows would it be appropriate still?

EM->Solvate->EM->posre NPT isotropic 1 bar w/ V-rescale tcoupl->NPT semi-isotropic varying pressures
w V-rescale tcoupl until cell shape is equilibrated using PR barostat, then switch to NVT without pcoupl
to use the hardwall during my production run?

For some reason, i’ve got this idea that I need to be tcoupling and pcoupling during my production runs
so i’ve been apprehensive to do this. 

My Best,
  Guy Dayhoff


More information about the gromacs.org_gmx-users mailing list