Re: [AMBER] Segmentation Fault with pmemd

From: Jason Swails <jason.swails.gmail.com>
Date: Sat, 29 Aug 2015 07:22:42 -0400

On Sat, Aug 29, 2015 at 4:02 AM, Ziheng Wang <zihengwang.sps.edu> wrote:

> Hi all,
>
> I am simulating a small protein system with a Mg2+ GTP bound (Mg2+ params
> from frcmod.) I am using frcmod.ionslm_1264_tip3p for metal parameters and
> http://www.pharmacy.manchester.ac.uk/bryce/amber/ for GTP parameters. I
> was able to generate prmtop and inpcrd for the system. However, I get this
> error message when running pmemd.cuda:
> Program received signal SIGSEGV: Segmentation fault - invalid memory
> reference.
>
> Backtrace for this error:
> #0 0x7FEFF8B0C7D7
> #1 0x7FEFF8B0CDDE
> #2 0x7FEFF8247FEF
> #3 0x46DB1A in __mol_list_mod_MOD_setup_molecule_lists
> #4 0x4AE5F0 in __pme_alltasks_setup_mod_MOD_pme_alltasks_setup
> #5 0x492034 in MAIN__ at pmemd.F90:?
> /tmp/slurmd/job07160/slurm_script: line 27: 19155 Segmentation fault
> (core dumped)
>
> The output file stops after line Energy averages sample interval:
> | ene_avg_sampling = 1000
> with no error message.
>
> The md input file is:
>
> &cntrl
> imin = 0, nstlim = 100000, dt = 0.001,
> irest = 0, ntx = 1, ig = 45678,
> tempi = 100.0, temp0 = 298.0,
> ntc = 2, ntf = 2, tol = 0.00001,
> tautp = 0.1, taup = 0.1,
> ntwx = 1000, ntwe = 0, ntwr = 1000, ntpr = 1000,
> cut = 8.0, iwrap = 1,
> ntt =1, ntb = 1, ntp = 0,
> nscm = 0,
> ntr=1, restraintmask="!:WAT & !.H=", restraint_wt=100.0
> nmropt=1,
> ​
>

​You are using the 12-6-4 parameters for the metal ions, but you are not
using the 12-6-4 potential. Make sure you set lj1264=1 in the &cntrl
section. Also, these restraints are VERY strong. Consider values ~1 to 2
orders of magnitude smaller when running dynamics.



> ​
>
> &end
> &wt
> TYPE="TEMP0", istep1=0, istep2=100000,
> value1=100., value2=298.,
> &end
> &wt
> TYPE="END",
> &end
> /
>
> The system is able to minimize fine with pmemd. The same error occurs
> during MD when I run pmemd without cuda.
>

​It would help to know what your submission script looked like. Make sure
you are running pmemd.MPI for parallel instead of pmemd (pmemd is for
serial, only).

HTH,
Jason

-- 
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Aug 29 2015 - 04:30:05 PDT
Custom Search