[gmx-users] Lincs warnings

[Mark Abraham]

On 20/01/12, Ben Porebski <ben at inoxx.net> wrote:

> Hi all, 
> 
> 
> I've been working on some structures between 2.5A and 3.4A in resolution and all of my structures suffer from lincs warnings.
> From reading the page about lincs warnings, I've performed a more lengthy equilibration:
> 
> 
> I'm using gromacs 4.0.7 single precision and the G53a6 force field with spc water model.
> 
> 
> - Genbox, add ions.
> - EM until converges
> - Positional restraints at 2,000 kJ/mol, 200, 20 and 2 for 100ps each at 100K
> - Followed by simulated annealing with the following parameters:
> 
> 
> 
> ; Simulated annealing @ positional restraint of 2 kJ/mol
> annealing = single      single
> annealing_npoints   = 3           3
> annealing_time = 0 50 250    0 50 250
> annealing_temp = 100 100 300 100 100 300
> 
> 
> 
> - This then undergoes 750ps of md at 300K.
> 
> 
> Once done. I continue the run in my production MD.
> Sometimes my structures will crash within 5 ns, sometimes it will run out for 60 ns before I start seeing any lincs warnings.
> 

 
An important trouble-shooting stage is to observe in a visualization program what happens in your simulation immediately before a crash, and look also at the variation in the energy terms. That might mean saving output more often. This could clue you in to a region of structure that is problematic, either atomic clashes that persist through EM+SA, or a malformed topology. See http://www.gromacs.org/Documentation/Terminology/Blowing_Up. Note also that restraining atoms to positions that are of low resolution can be problematic - you apparently don't want large rearrangements during equilibration, but it seems you need to permit more. Note what grompp -h has to say about position restraints.
 

> 
>  I've attempted to further refine the structures and they behave a little better, but still crash, just less often.
> To continue the simulation, I will not use the .trr output from the previous simulation. If this doesn't work, I'll run the simulation using shake for 2 ns, but this is significantly slower.
> 
> 
> I'm a little concerned that these crashes may be introducing some sort of error into the simulation and I'm not sure of the right way to get around this problem.
> I've played around with changing the lincs order and lincs iteration parameters, which appear to reduce the number of lincs warnings and slows down the simulation slightly.
> 

 
The constraint algorithms are normally not the problem, they're just the machinery most prone to breaking when something else goes wrong. As such, tweaking them is not productive.
 

> 
> Another thing I've played with was the constraints. Changing this from all-bonds to hbonds.
> This stopped my crashes, but a co-worker thought that it may not be accurately modelling the system, or would introduce anomalies into the simulation.
> My impression was that changing from all-bonds to hbonds would improve the accuracy of the system... But I could be wrong.
> 

 
The accuracy of the numerical integration is strongly tied to the size of the time step and the kinds of constraints you are using. See manual section 6.5. One strategy in your case is a lengthy equilibration period with a short time step (say 0.5fs) and no constraints or restraints at all. Monitor the trends in the energy terms over this period to see if changes are occurring.
 
Mark 
 

> 
> 
> If anyone has experience with dealing with lincs warnings or simulating low resolution structures, help would be amazing.
> I've been stuck on this for quite some time now.
> 
> 
> 
> 
> Cheers, Ben
> 
> 

 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20120120/e3b93b01/attachment.html>