[gmx-users] Inconsistent results in different clusters and cores

[Justin Lemkul]

On 5/12/13 1:17 PM, tarak karmakar wrote:
> Dear All,
>
>       I am simulating a protein in water to check the ligand movement over a
> time span. I have    minimized the system in STEEP and CG and after that
> heated from 0K - 300K within a time span of 300 ps. Then, I performed the
> NPT production runs. In two clusters I have got different results, like in
> one I see the ligand came out of the protein and in other it stayed inside
> the protein cavity, after a same (comparable) time duration. Now, while
> running the same job (same .mdp file) in different numbers of cores (in the
> same machine) I'm getting different behaviors, e.g., in protein ligand
> distance plots etc. I don't know the reason behind this. I'm running all my
> jobs in gromacs-4.5.5 version.
>       Recently I have visited this post [
> http://redmine.gromacs.org/issues/1012 ], but could not able to conclude
> properly. Anyone's help would be greatly appreciated.
>

I don't know if that issue is related, but if it is, then you're using an 
affected version (4.5.5).  Run again using the latest version (4.6.1) and see if 
the problem persists.  Realize too that MD simulations are chaotic, and there is 
no guarantee that one trajectory will be identical to another.  In principal, on 
average, multiple trajectories should produce similar behavior.  If you think 
that core count is a problem, run with mdrun -reprod to see if any optimizations 
are causing issues.  Build a solid test case, one that eliminates the 
possibility that your ligand topology is inaccurate.

-Justin

-- 
========================================

Justin A. Lemkul, Ph.D.
Research Scientist
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

========================================