Jason,
Thanks for your input. Just a quick note, there is no "vlimit exceeded" warnings from serial pmemd run. Everything looks normal. If I disable SHAKE (ntc=1, ntf=1), pmemd.MPI job will run, but energies blow up in a short period of time.
DX
________________________________
From: Jason Swails <jason.swails.gmail.com>
To: AMBER Mailing List <amber.ambermd.org>
Sent: Wednesday, August 17, 2011 8:56 PM
Subject: Re: [AMBER] Coordinate resetting cannot be accomplished, deviation is too large
A couple things:
First, I don't think pmemd.MPI accepts restraintmask='.CA,C,O,N', it should
only accept GROUP inputs (unless of course one of the CUDA patches updated
that silently without me realizing, I added the mask parsing functionality
to pmemd well after Amber 11 was released, but maybe not before the latest
CUDA upgrade, so if it wasn't dumped into pmemd by an earlier patch then it
will at least be part of Amber 12).
Outside of that, my best advice will be to look at the trajectory
(*specifically* atoms 86 and 119, since those are the atoms SHAKE failed
for). Does anything weird happen? You should see those atoms bouncing
around a lot (you may have to reduce ntpr in order to visualize that).
My guess is that even though you may not be getting that error with the
serial pmemd, it may still be suffering from the same issue (do you see any
"vlimit exceeded" warnings in either case?).
On another note -- why do you have nscm = 0? You should leave that at its
default, especially in GB solvent, I would imagine. Left to its own
devices, any system in implicit solvent will move freely around in space, so
it's always a good idea to eliminate translational motion via the nscm flag.
Of course it could be that you have a complicated system that either falls
through one of the error traps (if your system isn't quite valid) or that it
triggers a rare bug in parallel due to its peculiar behavior, but I think
this is probably unlikely...
HTH,
Jason
On Wed, Aug 17, 2011 at 10:13 PM, Danny Xu <quantum_mania.yahoo.com> wrote:
> Hi,
>
> When using pmemd.MPI, I get
>
>
> Coordinate resetting cannot be accomplished,
> deviation is too large
> iter_cnt, my_bond_idx, i and j are : 1 1 86 119
>
>
> whereas serial pmemd works just fine...
>
> Here's my heating.mdin
>
> &cntrl
> imin = 0, !no mininisation
> irest = 0, !no trajectory restart
> ntx = 1, !only coordinates are read in
> dt = 0.002, !2fs time step
> ntc = 2, !SHAKE; bonds involving H are constrainted
> ntf = 2, !all H-atom bonds forces are not evaluated
> nsnb = 25,
> tol = 0.000001, !Shake tolerance 10 times tighter
> ntb = 0, !no periodicity and PME off!
> ntp = 0, !No pressure control
> cut = 9999.0, !9999A cut off
> igb = 2,
> rgbmax = 9999.0, !default max dist between atom pairs Born radii
> calc.
> saltcon= 0.2, !salt concentration (M)
> ntt = 3, !constant temp using langevin
> tempi = 0.0, !initial temp = 0K
> temp0 = 300.0, !final temp = 300K
> gamma_ln = 1.0, !1.0 per ps for collision freq
> ntpr = 500, !print details to log every 500 steps
> ntwx = 500, !write coordinates to mdcrd every 500 steps
> ntwr = 50000, !write restart file to heat.rst every 50000 steps
> iwrap = 0, !wrap periodic molecules.
> nstlim = 850000, !run for 850,000 steps
> ntr = 1, !restraints
> restraint_wt=2.0, ! Weight
> restraintmask=".CA,C,O,N", !Backbone atoms selection
> nscm = 0, !No removal of COM motion
> nmropt = 1, !Scale temp target.
> ig = -1, !Random seed based on wallclock.
> ioutfm = 1, !NetCDF MDCRD.
> /
> &wt type='TEMP0', istep1=0,istep2=600000,
> value1=0.0, value2=300.0
> /
> &wt type='TEMP0', istep1=600001,istep2=850000,
> value1=300.0, value2=300.0
> /
> &wt type='END'
> /
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
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
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Aug 17 2011 - 21:30:02 PDT
