The cutoff value of "0.d0" in pmemd for a pme run is equivalent to requesting that the default values be used - probably 8.0 angstrom, if memory serves, assuming igb 0 for a pme simulation. PME does strange things at small-ish cutoff, but does not allow 0 (in the sense that it says "oh, you wanted 8"), and I presume later catches and disallows cut < 0. In my experience, reducing the direct space cut below roughly 7.0 angstrom causes unacceptable error limits unless you increase grid density for reciprocal space (there is basically a balancing act between reciprocal and direct space in terms of what is necessary to keep error acceptable). And if you think long cutoffs are expensive, play around a bit with greatly increasing nfft1,2,3 in a large system :-}. So all I am really answering here is why pmemd ran - it basically ignored you... I would have to read the paper done with amber 4.1 to have a clue what they were really up to, but pmemd is designed to always use either pme or generalized Born, so resur
recting an old copy of sander may be your best bet..
Regards - Bob Duke
________________________________________
From: Ross Walker [ross.rosswalker.co.uk]
Sent: Sunday, February 23, 2014 9:38 AM
To: AMBER Mailing List
Subject: Re: [AMBER] PME with cutoff = 0
Hi Francois,
Both sander and pmemd almost certainly make assumptions about the cutoff.
Since the days of AMBER 4.1 there have been huge changes in the way the
pairlist is built etc. For example the code use to just rebuild the
pairlist at set intervals. These days it is smart and only rebuilds it
when an atom has moved more than the distance skinnb such that the
pairlist cutoff is really cut+skinnb. Thus reducing cut to zero probably
messes up logic in this part of the code. There are likely other issues as
well, like divisions by zero inside the code that determines the Ewald
coefficient etc. The fact one code crashes and the other doesn't is likely
just dependent on where they hit the first problem. From what you show it
looks like pmemd just hangs while sander crashes with an error - both
could be considered a crash.
Anyway, to cut a long story short neither code officially supports a value
of cut=0 for PME runs and it is something that isn't tested. It may be
possible to run the simulation in GB in sander by setting igb=6 - this
will run a gas phase simulation. I doubt there is anyway to run a periodic
simulation with a cutoff of 0. If you really want to then your options are
likely see if you can find a PRE-PME copy of sander (such as AMBER 4.1) or
(sander-classic from AMBER 6 perhaps?) or crack open the debugger and see
if you can trace through the use of the cutoff value and find out where
the problems are occurring. This may just be a case of going down the
rabbit hole though.
Sorry I can't offer much more help than that. - Note you could also try
cut=0.001 and see if that works. There is almost certainly a 1.0d0/cut^2
somewhere in the code and this might work around that and still give you
what you want from the simulation.
All the best
Ross
On 2/23/14, 9:19 AM, "FyD" <fyd.q4md-forcefieldtools.org> wrote:
>Dear All,
>
>I ran amber10/sander.MPI with PME + cutoff = 0 -> sander crashes
> amber12/sander.MPI with PME + cutoff = 0 -> sander crashes
> (compiled with intel13 + mkl)
> amber12/pmemd.MPI with PME + cutoff = 0 -> pmemd does not crash...
>
>Printed output: printing stops in section 4:
>
>--------------------------------------------------------------------------
>------
> 4. RESULTS
>--------------------------------------------------------------------------
>------
>
>| # of SOLUTE degrees of freedom (RNDFP): 24936.
>| # of SOLVENT degrees of freedom (RNDFS): 0.
>| NDFMIN = 24933. NUM_NOSHAKE = 0 CORRECTED RNDFP =
>24933.
>| TOTAL # of degrees of freedom (RNDF) = 24933.
>
> -> nothing else is printed...
>
>In Fox & Kollman J.Phys.Chem.B 1998, 102, 8070, MD simulation with
>cutoff = 0 is reported & Sander from Amber 4.1 was used.
>
>Why does sander crash, while pmemd does not crash in this case?
> (with the same input & cut = 12, sander does not crash)
>
>thank you, regards,
>Francois
>
>
>
># Cst pressure simulation with cutoff = 0 Angs; 300 K short
>$PARAL -np $NPROC $EXE -O -i md-cstpres0.in \
> -o md-cstpres0b.out \
> -p $mol.parm7 \
> -c md-cstvol.rst7 \
> -r md-cstpres0b.rst7 \
> -x md-cstpres0b.mdcrd
>
>my test input below:
>
> cst pressure MD, nstlim*dt psec, PME with cutoff
> &cntrl
> nmropt = 0, ioutfm = 1, iwrap = 0,
> ntx = 1, irest = 0, ntrx = 1, ntxo = 1,
> ntpr = 100, ntwr = 1000, ntwx = 500,
> ntave = 0, ntwv = 0, ntwe = 0,
> ntf = 2, ntb = 2,
> ntc = 2, tol = 0.00001,
> dielc = 1, ipol = 0,
> cut = 0, ig = 71277, comp = 52.5,
> ibelly = 0, ntr = 0, igb = 0,
> imin = 0, nstlim = 5000,
> nscm = 1000, t = 0.0, dt = 0.002,
> temp0 = 300, tempi = 300,
> ntt = 1, tautp = 0.2, vlimit = 20,
> ntp = 1, pres0 = 1.0, taup = 0.2,
> &end
>
>
>
>_______________________________________________
>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
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Feb 23 2014 - 11:30:02 PST
