On Fri, Jul 13, 2012, Albert wrote:
>
> I am using the following command and input file to run NPT
>
> *mpiexec -np 2 $AMBERHOME/bin/pmemd.cuda.MPI -O -p apo.prmtop -c m6.rst
> -ref m6.rst -i npt1.in -o npt1.out -r npt1.rst -x npt1.mdcrd -inf npt1.info*
> *
> *
> *equilibration*
> * &cntrl*
> * imin=0, irest=0, ntx=1,*
> * nstlim=500000, dt=0.002,*
> * ntc=2, ntf=2,*
> * cut=10.0, *
> * ntb=2, ntp=3, taup=2.0, pres0=1,*
> * ntpr=5000, ntwx=5000, ntwr=50000,*
> * ntt=3, gamma_ln=1.0, ig=-1,*
> * iwrap=1,*
> * temp0=310.0,*
> * csurften=3, ninterface=2, gamma_ten=17,*
> * restraintmask='1-396 &!.H',*
> * restraint_wt=20,*
> * /*
>
> however, I found that the npt1.out and npt1.info stacked there after
> several hours of running:
Usual instructions for debugging MD:
1. Set nstlim=50 and ntpr=1. If that works, you can start increasing
both variables. If it fails, you may have more information, or go to the next
step.
2. For the short run cited above, start with a single thread cpu run. Then
try in parallel on a cpu; then in serial mode on the GPU; see if there is
something that works and something that fails.
Specific to your situation:
Equilibrate first a constant volume until you get a temperature somewhat near
your final temperature; then turn on constant pressure. You are asking
for constant pressure simulations with an intial temperature of zero, which
can lead to unstable behavior.
FWIW, I always do initial system preparation and some amount of equilibration
on a cpu...lots of bad things can happen right at the beginning (sometimes
driven by my poor choices of input variables.) Once I'm confident that the
input files are giving acceptable outputs, I'll start increasing nstlim and
consider moving things to a GPU.
...good luck....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Jul 13 2012 - 18:30:03 PDT
