This question was actually asked and answered 2 days ago, but the web
archives appear to have stalled the past month (no updated threads since
May 24). So I'll paste the previous answer below
On Wed, Jun 19, 2013 at 5:35 AM, 涂志萍 <tuzhiping711.gmail.com> wrote:
> Dear all,
> I’m using amber12 pmemd.cuda_SPFP( Version 12.3) to run MD simulation to
> study the protein folding problem. But the running process is always
> terminated soon. There was nothing error in the md2.out. The log.txt
> exports: Nonbond cells need to be recalculated, restart simulation from
> previous checkpoint with a higher value for skinnb.and I can not understand
> what does it mean.
>
This 'error' results from the simulation cell shrinking a lot (your
starting box was way too big). The system is divided into cells for
performance reasons, and if the cells shrink too much, then they become
smaller than the region defined by the cutoff. This breaks energy
conservation and creates a whole host of other problems.
I would suggest leaving skinnb alone (you can see what it does in the Amber
manual), and just restart the simulation from the last good restart file as
suggested. This will recalculate the nonbonded cells, getting rid of the
problem.
> My input file
> irest=1, ntx=5,
> imin=0, ! run MD
> ntb=2, ntp=1, pres0=1.0, taup=2.0, ! periodic box, constant pressure
> ! ntb=1, ! constant volume
> ntpr=10000, ntwr=10000, ntwx=10000,
> ntt=3, gamma_ln=1.0, tempi = 298.0, temp0 = 298.0,
> ig=177559, ! rand number
> ntf=2, ntc=2, ! SHAKE
> nstlim=30000000, dt=0.001, ! steptime
> cut=12.0,
>
This is a very large value of 'cut' and slows your calculation down quite a
bit without providing any measurable improvement in accuracy. I would
suggest a value closer to the Amber default of 8 to 10. Note that
regardless of the cutoff, electrostatic interactions are computed fully due
to the use of the Ewald summation.
HTH,
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Jun 19 2013 - 03:30:03 PDT
