[gmx-users] LINCS warning

[Mark Abraham]

> Hi everyone,
>
> I'm having trouble with performing 1 ns simulations on a 318 residue
> long protein with 'mdrun'. The run.log file, included at the end of
> this message, gives a 'LINCS warning' within a minute or two after
> the simulation begins:
>
> Step 148, time 0.296 (ps)  LINCS WARNING
> relative constraint deviation after LINCS:
> max 0.151505 (between atoms 4353 and 4355) rms 0.002175
> bonds that rotated more than 60 degrees:

In plain English, this means the bond or angle constraints implemented
with the LINCS algorithm are having to be enforced after very large
deviations. This suggests that there is something numerically unstable
going on. Visualising the trajectory is always a good idea at this point,
as it can suggest whether the whole world is exploding (re-visit the
preparation protocol) or just part of it (be more cunning).

> I've have been minimizing the structure of the protein prior to
> trying to perform the simulation (no problems here). Also, when I use
> 'grompp' to produce a .tpr file for a 1 ns MD simulation with the
> minimized structure, I have no problems with this either. The .mdp
> file that is fed into the .tpr file will work with different protein
> structures (single residue mutants) without incident. So, it would
> appear that the .mdp file used to compile the .tpr file is working.

Yes... the .mdp file describes the conduct of the simulation, the .top
file describes the topology of system, and the structure (.gro) file
describes the original conformation. All three are combined to produce the
.tpr file. There are some constraints on the three, but subject to those
constraints you can swap different .mdp/.top/.gro files.

> What is really puzzling to me is that the simulation will sometimes
> run properly and sometimes it won't even though it would appear
> that  I use the exact same .gro file.  Does this sound familiar to anyone?

There is more needed to prepare a system for MD than energy minimization
to relieve bad contacts. Read any MD paper and they will describe some
kind of regime for generating initial velocities and heating the system to
the required temperature, and then leave it to sit for a while. Often
temperatures will be equilibrated in NVT, before relaxing to NPT when that
ensemble is desired for the production calculation. Since you don't
describe doing this, my guess is that you are not doing it (well enough).

Mark