[gmx-users] mdrun error after 2.6nS

[Mark Abraham]

Blaise Mathias-Costa wrote:
> Hello Mark and Tsjerk, Thanks a lot for your suggestions. I have gone 
> through that web site and it doesn't say anything very specific, for 
> example if the system is not minimized properly it should have 
> complained at the beginning, but the problem arises after 2.6nS!

No. The numerical equations are well-defined whether or not you 
minimize. Whether the system is numerically or physically stable is 
another matter, and things could fairly easily take many picoseconds to 
nanoseconds in a feedback loop to get bad enough to cause LINCS warnings.

The website can't be more specific, because proper system preparation is 
a bit of dark art and system-dependent.

> I have searched in the archive, when somebody had similar problem with 
> PR run David suggested to look for the "center-of-mass motion removal", 
> to give POPC+Protein, but I dont have POPC in my system. In another case 
> advised to reduce the time step, does it mean that I should reduce the 
> time step from 0.002 to 0.001 and do the simulations from the beginning?

Yes. Making the integration time step smaller can help you relax away 
from bad numerical space into good ones... once equilibrated you can 
then try relaxing the time step back to 2fs.

> Another suggestion was the computing power may not be sufficient, but I 
> don't think this will be applicable for me, b'se I am running in 8 nodes 
> each node is with 4 processors, and the unit cell size is ~96000 atoms 
> in total.
>  
> What could be the most appropriate way to solve this problem?  Many 
> thanks in  advance!

Find a reasonable equilibration protocol in the literature for a system 
similar to yours, try it out, and experiment.

Mark