AMBER Archive (2009)

Subject: Re: [AMBER] simulation blown up

From: Robert Duke (rduke_at_email.unc.edu)
Date: Tue Oct 20 2009 - 22:55:39 CDT


Okay, I have no idea when you really blew up. I am presuming during the
third procedure below. The useful info to include would be the mdout
associated with the run that blew up (the actual mdin, and the last energy
outputs observed). So first of all, a bit of a digression. Don't change
default values unless you have a really good reason. So I see in the mdin
listings a number of settings that while not deadly, may not make a lot of
sense. I just recently complained about someone else using a long cutoff.
Same comment here - don't do this unless you understand why, if you are
using pme (and you are) - it mostly just makes your simulation more
expensive; pme works with a short cutoff because it combines a "direct
space" and a "reciprocal space" computation that essentially does a full,
not cut off, computation of the electrostatics. The only reason I know of
to increase the cutoff, other than some esoteric ones, is that you need to
get better vdw energies, and typically by 9 angstrom, those are getting
pretty reasonable (you CAN see a difference as you increase the vdw cutoff
but increases in accuracy of the vdw energies past 9 angstrom are pretty
small). So the preceding comments don't have that much to do with why you
blew up, but it is important to realize that the defaults exist for a
reason, and you need to understand why you are changing them when you change
them. Now to address the reason why you probably blew up - improper
equilibration design. I don't have a big problem with 2000 step
minimizations - it should work, and it is relatively cheap. A linmin
failure in such a procedure is really not a big deal - the minimizer just
can't make more progress past this point, so it quits. There is no reason
to use shake in the minimizer (ntc = 2, ntf = 2), and I wouldn't. Shake is
for dynamics. So now, somewhere in this process below, you DO run into
problems in shake - that is what the error message is really about - shake
can't come up with an adequate solution. The reason for this, I would
suspect, is that the second restrained step is by no means an adequate
approach to ramping the temperature up from a starting mimimization to
actual production dynamics at 300K. There are a bunch of issues involved in
this, depending on the system, but if you are going to use a ntt 1
thermostat, I think it more appropriate to allow the temperature to
gradually increase over on the order of 50-60 psec (25,000 - 30,000 steps in
your system, this is often what it takes) typically done with nmropt=1 &wt
restraints on the temperature (a temperature "ramp", increments of 50K
usually adequate). The other big omission I see here - there is no use of
NPT md at a low temp - say 10K, to get rid of vacuum bubbles in the system.
Typically, you will spend somewhere between 10 and 50 psec doing this too,
depending on the size and initial density of the system - you want to get
close to a density of 1.00 for the system. I have no idea what your real
system size is here, etc. etc. Not using a big enough solvent box is a bad
idea - at least 12 angstrom is common wisdom; I go bigger than that (so if
you solvate with a boundary of 12 angstrom in leap, you end up with a
significantly smaller boundary after density equilibration, so I start with
somewhere around 15-18 angstrom, depending on the system). So these are
rough, off-the-top-of-my-head metrics, and I am not suggesting an actual
equilibration procedure for you here. What I DO recommend is that you look
at a number of the tutorials on the amber website, ambermd.org, and
understand what is being done there, and then apply them to your system.
Reading some of the references available there is also a great way to
understand what other folks have done to get their systems equilibrated. I
typically spend upwards of a nsec of simulation time after a bit of
minimization to get a system equilibrated; you can maybe spend less using
the ntt 3 thermostat, but you typically need to be careful about how you
increase the system density to 1.00, how you increase the temperature from
0K to 300K, and how you relax the restraints on the system (as well as what
you restrain). Just how long you need to take is dependent on both the
sensitivity of the system you are studying as well as system size.
Regards - Bob Duke

----- Original Message -----
From: <moitrayee_at_mbu.iisc.ernet.in>
To: <amber_at_ambermd.org>
Sent: Tuesday, October 20, 2009 1:41 PM
Subject: [AMBER] simulation blown up

> Dear Amber users,
>
> I am doing a simulation of a protein ligand complex.
>
> I get the the following error and the simulation stops during the
> production run:
>
> ----------------------------------------------------------
> Running from 0170 to 0180 at Tue Oct 20 19:21:39 IST 2009
> ----------------------------------------------------------
>
>
> Coordinate resetting (SHAKE) cannot be accomplished,
> deviation is too large
> NITER, NIT, LL, I and J are : 0 2 4580 9158 9159
>
> Note: This is usually a symptom of some deeper
> problem with the energetics of the system.
> rank 2 in job 15 compute-0-27.local_33789 caused collective abort of
> all ranks
> exit status of rank 2: killed by signal 9
>
> I have checked for this error message in the mailing list. The options to
> solve
> are increasing the size of the box (I increase from 10 to 14 angstroms)
> which in
> my case causes a linmin failure in the minimization step itself. I have
> checked
> the structure and could not find anything unusual.
> I do 2000 steps of minimization followed by 30ps of equillibration before
> the
> production run.
> I am attaching my minimization, equillibration and md scripts as follows:
>
>
> Minimization:
>
> # 2000 steps of minimization, no distance-dependent dielectric constant
> &cntrl
> maxcyc=2000, ncyc=100, imin=1, cut=10.0, igb=0, ntb=1, ntpr=1,
> lastist=8000000,
> &end
>
> Equillibration:
>
> Restraint improvement attempt
> &cntrl
> imin=0,
> ntr=1,
> ntpr=50, ntwr=500, iwrap=1, ntwx=500, ntwe=50,
> ntf=2, ntb=1, igb=0, scnb=2.0, scee=1.2,
> cut=10.0,
> nstlim=5000, dt=0.002,
> tempi=0., temp0=300., ntt=1, tautp=0.5, dtemp=5.0,
> ntp=0, taup=0.5,
> ntc=2, tol=0.0001,
> vlimit=15,
> &end
> &ewald
> vdwmeth=1,
> &end
> Group for restrained atom
> 10.0
> RES 1 560
> END
> END
>
>
> MD:
>
> Equlibriation
> &cntrl
> imin=0,
> ntx=7, irest=1,
> ntpr=50, ntwr=500, iwrap=1, ntwx=500, ntwe=50,
> ntf=2, ntb=2, igb=0, scnb=2.0, scee=1.2,
> cut=10.0,
> nscm=50,
> nstlim=5000, dt=0.002,
> tempi=300., temp0=300., ntt=1, tautp=0.5, dtemp=0.,
> ntp=1, taup=0.5,
> ntc=2, tol=0.00001,
> &end
> &ewald
> vdwmeth=1,
> &end
>
> I am also attaching the pdb file of my interest.
>
> I would be grateful if you please suggest me a way out of this problem.
> Thanks a lot in advance.
>
> Sincere Regards,
> Moitrayee Bhattacharyya
> Molecular Biophysics Unit
> Indian Institute of Science
> Bangalore - 560012
> India
>
>

--------------------------------------------------------------------------------

> _______________________________________________
> AMBER mailing list
> AMBER_at_ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>

_______________________________________________
AMBER mailing list
AMBER_at_ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber