[gmx-users] Segmentation fault on HPC

[Justin Lemkul]

On 1/4/15 7:44 AM, Tomek Wlodarski wrote:
> Hi Mark,
>
> Thanks for replay.
> Actually log file ends without any error:
>
> Started mdrun on rank 0 Sun Jan  4 12:30:40 2015
>             Step           Time         Lambda
>                0        0.00000        0.00000
>
>
>     Energies (kJ/mol)
>            Angle    Proper Dih.  Improper Dih.          LJ-14     Coulomb-14
>      1.00844e+05    2.04719e+05    5.69917e+03    6.63396e+04   -4.67549e+05
>          LJ (SR)        LJ (LR)   Coulomb (SR)   Coul. recip. Position Rest.
>      1.46339e+06   -3.74511e+04   -9.86464e+06   -1.81675e+06    4.91794e+01
>        Potential    Kinetic En.   Total Energy  Conserved En.    Temperature
>     -1.03454e+07    8.79250e+05   -9.46610e+06   -9.46610e+06    2.00415e+02
>   Pressure (bar)   Constr. rmsd
>     -5.12367e+03    5.80732e-06
>
> I found something strange - when I changed nstxout value I was able to run
> simulation for more than 90k steps without error (I killed simulation
> because I was running out of space on my disc ;))
> So my simulation is running OK when I have:
> dt                       = 0.002   ; 2 femtosecond time step for integration
> nsteps                   = 50000   ; 100 ps
>
> ; OUTPUT CONTROL OPTIONS
> nstxout                 = 1    ; save coordinates every 2 ps
> nstvout              = 5000    ; save velocities every 2 ps
> nstenergy            = 5000    ; save energies every 2 ps
> nstlog               = 5000
>
> I tired different values of nstxout and simulations always crash...
> changing dt to 0.001 also does not help
> Interestingly when I reduce nstep to 500 my simulation crashes as
> well..even though I have nstxout = 1...
> Is this indication that my simulation is not stable or something else is
> happening?

Your simulation isn't stable.  Output frequency should have no bearing on 
whether or not the simulation runs; you're just flushing the buffer more often 
so you see more output.  Add that to the fact that MD is chaotic, you can't make 
any cause-and-effect relationship here.

Without your full .mdp file and the output of EM (Epot, Fmax), there's not much 
to suggest beyond the "blowing up" link Mark sent you before.

-Justin

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

Justin A. Lemkul, Ph.D.
Ruth L. Kirschstein NRSA Postdoctoral Fellow

Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 629
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201

jalemkul at outerbanks.umaryland.edu | (410) 706-7441
http://mackerell.umaryland.edu/~jalemkul

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