Re: Constraint failure; simulation has become unstable

From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Mon Jan 10 2011 - 08:49:07 CST

2011/1/8 Wendy Gonz√°lez <locuswendy_at_gmail.com>:
> But the idea is that if you are comparing three systems (see my first
> email), and A and C run properly, I can not do changes in parameters only in
> simulation B because it did not work. If I changes the parameters for B
> until it runs in 2.7 version, I have to do the same changes later in A and
> C.
> Solution:
> B run in 2.6 version
> A run already in 2.6 version
> I will run C in 2.6 version too.
> All the systems with the same parameters to do comparisons later.

wendy,

you are making a big mistake.
the fact, that one your simulations runs through with
NAMD 2.6 but fails with NAMD 2.7 doesn't necessarily
mean, that the v2.6 simulation is correct and the 2.7
ones are not. you have to have more compelling
evidence to convince people. sometimes you may
have a borderline failing case that completes or fails
with a slightly modified source or due to different
order of summation. that is usually an indication of
a bad model, not a bad code.

you have to provide corroborating evidence.

in general, your simulation strategy, and particularly
the use of "reload charges" seems very strange to me.
reassigning changes can add a significant (and unphysical)
"kick" to the system and perfectly explain your observations,
if the changes to the charges are misdirected or too large.
have you looked at the total energy overall, also in the
v2.6 case? have you looked at a high resolution trajectory
of the atoms giving the constraint failures?
overall, what is the rationale, behind using this feature
and what do you expect to learn from this?
i think particularly the last question is something that should
be discussed _first_ before worrying about whether the
simulation completes or not.

cheers,
    axel.

>
> Regards,
>
> Wendy
>
> 2011/1/8 Athanassios Stavrakoudis <astavrak_at_cc.uoi.gr>
>>
>> Wendy,
>>
>> Though I am not fully qualified to speak about the version differences,
>> I can say that NAMD is a very strong program in MD simulations.
>> Moreover I think that most of the "some thousands" publications made with
>> NAMD
>> came from version betas...
>> Some problems might rise, perhaps due to system setup.
>> But you have to deal with these.
>> The game is not like "push the button Max".
>>
>> All the best,
>> Thanassis
>>
>> > Dear NAMD Users:
>> >
>> > Thanks a lot for your comments. I found already the problem:
>> >
>> > It is NAMD2.7, I run my simulation with NAMD 2.6 and it worked.
>> >
>> > I am very disappointed with NAMD because this situation have been
>> > occurred
>> > several times with other simulations where I can not find a real
>> > solution to
>> > the problem. Then, I run the same simulation  in NAMD2.6 without changes
>> > in
>> > any variable  and it works properly. It was happening to me with
>> > NAMD2.7b2
>> > and  NAMD2.7b3 versions, and I justified the fact due to they were still
>> > Beta versions, but it happened again, and now with 2.7 official version!
>> >
>> > :-(
>> >
>> > Regards,
>> >
>> > Wendy
>> >
>> >
>> >
>> > 2011/1/7 Joshua Adelman <jla65_at_pitt.edu>
>> >
>> > > Hi Wendy,
>> > >
>> > > When you're playing around with your charges in all of your systems
>> > > (and
>> > > with the reloadCharges command), are you making sure that your total
>> > > system
>> > > is neutral? I'm not that familiar with the reloadCharges command, but
>> > > it
>> > > seems like something you should only be messing around with for
>> > > special
>> > > purposes and not for general simulations.
>> > >
>> > >  Also, if you're patching residues in your system, are you making sure
>> > > that
>> > > you've regenerated the angles and dihedrals correctly? If you are
>> > > using the
>> > > autopsfgen gui to build the system, then you need to select the option
>> > > in
>> > > the pull down menu. You can probably tell from inspection if your
>> > > patched
>> > > hydrogens are making strange angles/folding back on the residue. I
>> > > forgot
>> > to
>> > > do this once and I saw similar problems to what you are experiencing,
>> > > even
>> > > for small time steps. You might want to increase the output frequency
>> > > of
>> > the
>> > > dcd and see if you can spot what is going on immediately before things
>> > > crash.
>> > >
>> > > Best wishes,
>> > > Josh
>> > >
>> > >
>> > >
>> > > On Jan 7, 2011, at 4:18 PM, Wendy GonzőĪlez wrote:
>> > >
>> > > Dear Athanassios:
>> > >
>> > > The fact is that I use the same script (see below) in the 3 systems
>> > > (they
>> > > change only in the charge of one histidine residue), and only one of
>> > > them
>> > > have problems, the other run normally.
>> > > This is my script:
>> > >
>> > > #############################################################
>> > > ## ADJUSTABLE PARAMETERS                                   ##
>> > > #############################################################
>> > >
>> > > structure           ionized.psf
>> > > coordinates         DM_5ns_only_70.coor
>> > > velocities          DM_5ns_only_70.vel
>> > > extendedSystem      DM_5ns_only_70.xsc
>> > > parameters          par_all27_lipid.prm
>> > > parameters          par_all22_prot_cmap.inp
>> > > paraTypeCharmm      on
>> > >
>> > >
>> > > #############################################################
>> > > ## SIMULATION PARAMETERS                                   ##
>> > > #############################################################
>> > >
>> > >
>> > > # Force-Field Parameters
>> > > exclude             scaled1-4
>> > > 1-4scaling          1.0
>> > > cutoff              12.0
>> > > switching           on
>> > > switchdist          10.0
>> > > pairlistdist        13.5
>> > >
>> > > # Integrator Parameters
>> > > timestep            0.5  ;# 1fs/step (2)
>> > > rigidBonds          all  ;# needed for 1fs steps (2)
>> > > nonbondedFreq       2
>> > > fullElectFrequency  4
>> > > stepspercycle       20
>> > >
>> > > # Constant Temperature Control
>> > > langevin            on    ;# do langevin dynamics
>> > > langevinDamping     1.0  ;# damping coefficient (gamma) of 10/ps (5)
>> > > langevinTemp        300
>> > > langevinHydrogen    no    ;# don't couple langevin bath to hydrogens
>> > >
>> > > # use lighter damping now that system is equilibrated
>> > > langevinPiston          on
>> > > langevinPistonTarget    1.01325
>> > > langevinPistonPeriod    200
>> > > langevinPistonDecay     500
>> > > langevinPistonTemp      310
>> > >
>> > > useGroupPressure        yes     # smaller fluctuations
>> > > useFlexibleCell         yes     # allow dimensions to fluctuate
>> > > independently
>> > > useConstantRatio        yes     # fix shape in x-y plane
>> > >
>> > > margin 2
>> > >
>> > > wrapAll             on
>> > >
>> > > # PME (for full-system periodic electrostatics)
>> > > PME                 yes
>> > > PMEGridSizeX        103
>> > > PMEGridSizeY        103
>> > > PMEGridSizeZ        88
>> > >
>> > >
>> > > # Output
>> > > binaryoutput        no
>> > > outputEnergies      10000
>> > > outputTiming        10000
>> > > xstFreq             10000
>> > > dcdFreq             10000
>> > > restartfreq         10000
>> > >
>> > > ############################################
>> > >
>> > > if {1} {
>> > > constraints         on
>> > > consRef             restrain_ca_not_filter.pdb
>> > > consKFile           restrain_ca_not_filter.pdb
>> > > consKCol            B
>> > > }
>> > >
>> > > ##############################################################
>> > > outputname          DM_5ns_E70x6times
>> > > run                 10000
>> > > reloadCharges       charges_only_task3_E70x6times.txt
>> > > run                 5000000
>> > >
>> > >
>> > >
>> > >
>> > >
>> > >
>> > > 2011/1/7 Athanassios Stavrakoudis <astavrak_at_cc.uoi.gr>
>> > >
>> > >>
>> > >> Dear Wendy,
>> > >>
>> > >> Multiple reasons may lead to this situation.
>> > >> Most possibly, your system is not well minimized/equilibrated.
>> > >> You need to give a more detailed description of your simulation
>> > >> setup, to get more detailed answer.
>> > >> Also search the NAMD-l archives, almost any user has come with this
>> > >> problem.
>> > >>
>> > >> You can try to decrease the timestep to 1 fs,
>> > >> or to decrease the nonbondedFreq, fullElectFrequency values.
>> > >> Or to experiment with a mixture of all of these.
>> > >>
>> > >> Good luck,
>> > >> Thanassis
>> > >> > Dear NAMD Users,
>> > >> >
>> > >> > I am working with 3 systems, the only difference between them are:
>> > >> >
>> > >> > a) 2 histidines protonated
>> > >> > b) 1 histidine protonated, the second is neutral
>> > >> > c) Both histidines neutral
>> > >> >
>> > >> > I run several nanoseconds of simulations on this systems and they
>> > >> > did
>> > >> not
>> > >> > have problems. I applied the reloadCharges command (increasing the
>> > >> charge of
>> > >> > the same glutamate in the 3 systems) and the results are:
>> > >> >
>> > >> > a) Run normally
>> > >> > b) ERROR: Constraint failure in RATTLE algorithm for atom XXXX
>> > >> > ERROR:
>> > >> > Constraint failure; simulation has become unstable.
>> > >> > c) Run normally
>> > >> >
>> > >> > In case b) I said "atom XXXX", because I run 3 different
>> > >> > simulations and
>> > >> the
>> > >> > error is with a different atom each time.
>> > >> >
>> > >> > Any suggestion??? What could be happening???
>> > >> >
>> > >>
>> > >>
>> > >>
>> > >> Athanassios Stavrakoudis
>> > >> University of Ioannina                web:
>> > >> http://stavrakoudis.econ.uoi.gr/
>> > >> Department of Economics            e-mail: mailto:astavrak_at_cc.uoi.gr
>> > >> GR-45110, Ioannina                    tel: +30 265 100 5935
>> > >> GREECE                                fax: +30 265 100 5092
>> > >>
>> > >
>> > >
>> > >
>> > >
>> >
>>
>>
>>
>> Athanassios Stavrakoudis
>> University of Ioannina                web:
>> http://stavrakoudis.econ.uoi.gr/
>> Department of Economics            e-mail: mailto:astavrak_at_cc.uoi.gr
>> GR-45110, Ioannina                    tel: +30 265 100 5935
>> GREECE                                fax: +30 265 100 5092
>
>

-- 
Dr. Axel Kohlmeyer
akohlmey_at_gmail.com  http://goo.gl/1wk0
Institute for Computational Molecular Science
Temple University, Philadelphia PA, USA.

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 05:23:30 CST