AMBER Archive (2007)

Subject: Re: AMBER: Fwd: QM region + cutoff larger that box

From: M. L. Dodson (mldodson_at_houston.rr.com)
Date: Mon Aug 06 2007 - 10:29:43 CDT

Francesco Pietra wrote:
> Hi Bud:
>
> Thanks. I was mislead by the statement on Amber9 manual "does not
currently
> support generalized Born, PME or Ewald". I have just emailed the
signed form
> requesting the required files for scf-dftb. We will see.
>
> Not yet traced the ref to your publications using scc-dftb, though it
will be
> easy.
>
> Cheers
>
> francesco pietra
>

I do not yet have publications using this method. I meant the
publications referenced in the amber9 documentation.

The manual is correct: amber9 does not do PME or Ewald simulations
using DFTB/SCC-DFTB. Amber10 will have that capability. I forgot
about that. Sorry about the confusion. I have been working with
Ross and using a development version of sander from amber10.

You can still do simulations in a spherical cap of solvent. Be
aware that there will not be conservation of energy in such a
system so you may have difficulty with temperature control on
longer runs. If you do this, remember ivcap and fcap for sander.
In particular, read page 125 in the docs, and see solvateCap in
LEaP (page 68) to create the cap.

Bud Dodson

>
> --- "M. L. Dodson" <mldodson_at_houston.rr.com> wrote:
>
>> Francesco Pietra wrote:
>>
>>> Hi Gustavo and Bud:
>>> Yes, as you suspected, the run that failed had qmmask ='1-2'
>>> while there was just one residue (initially I attached the out
>>> file, where the input file could be seen). When qmmask is set
>>> correctly, even this run went to completion up to tried 100ps.
>>>
>>> The far more serious problem is how to run the QM part, i.e.
>>> (1) which theory for the given problem,
>>> (2) what to look at (ESCF, EPtot, ..) to compare the potential
>>> energies of the conformers.
>>> Semiempirical PM3 AM1 MNDO are known to be unable to reproduce
>>> conformational energy differences between conformers. PM3, for
>>> example, gives for methylcyclohexane eq/ax 1.0-1.1, against the
>>> experimental (non-hydroxylic solvents r.t., 1.8). I am
>>> unfamiliar with PDDG or CARBI variants, though probably they
>>> follow the trend of their precursors: where one performs better,
>>> the other one fails. The trend being unpredictable.
>>>
>>> On the other hand, DFT-MM as in Amber9 can't deal with the
>>> medium effect, which is just my interest. Also, DFT requires
>>> very modern functionals and high basis sets to work with my
>>> cases.
>>>
>>> I am now exploring the performance of semiempirical QM-MM with
>>> test cases that embody features of the molecules of my
>>> interest. The conformational ratios for these test cases are
>>> known in condensed phase. I can't see how to approach the case
>>> better for the time being.
>>>
>>> Cheers
>>>
>>> francesco pietra
>>>
>>>
>> I have been using DFTB/SCC-DFTB with or without the dispersion
>> corrections (read the papers). This has the reputation of being
>> very good for QM treatment of chemical reactions in enzyme active
>> sites. This is not very different in concept from your interests
>> in solvent effects. Are you sure this is an inadequate method?
>> This approach is fairly different from DFT calculations in, e.g.,
>> Gaussian. I'm sure others can offer their impressions as well.
>>
>> In my understanding, the dispersion corrections to DFTB/SCC-DFTB
>> are used to apply corrections for Van der Waals interacting
>> systems such as nucleic acids base pair stacking computations.
>>
>> The energy of the system needs to be calculated by averaging over
>> an ensemble of structures as is always true for MD. The ensemble
>> is sampled by the dynamics, so the runs have to be "long enough"
>> to adequately sample the ensembles. How long is "long enough"
>> could serve as the definition of a "hard problem".
>>
>> Good luck,
>> Bud Dodson
>>
>>>
>>>
>>> --- Gustavo Seabra <gustavo.seabra_at_gmail.com> wrote:
>>>
>>>> Hi Francesco,
>>>>
>>>> It's really hard to pinpoint the problem without more details. But
what
>> this
>>>> message is telling you, in general, is that your QM region is
"expanding"
>>>> too far. QM atoms share one big pair list, and are not allowed to
interact
>>>> with their periodic images. So, the QM region + cutoff must be kept
>> smaller
>>>> than the box size.
>>>>
>>>> (See a brief explanation in J. Phys. Chem. A 2007, 111, 5655-5664.
A more
>>>> detailed explanation will be in an upcoming article)
>>>>
>>>> This could be, as was suggested, one solvent molecule mistakenly
included
>> in
>>>> the QM region and that is diffusing away. It also did happen to me
before
>>>> that, as the QM region moves, it assumes an extended position and
suddenly
>>>> it's too large. Or, depending on the problem, you molecule could even
>>>> dissociate and then the parts separate too much.
>>>>
>>>> Just to clarify one point, the QM region is recentered at every MD
step,
>> so
>>>> you don't have to worry about it diffusing out of the box, as long
as it
>> is
>>>> in one piece.
>>>>
>>>> HTH,
>>>>
>>>> Gustavo.
>>>>
>>>>> -----Original Message-----
>>>>> From: owner-amber_at_scripps.edu
>>>>> [mailto:owner-amber_at_scripps.edu] On Behalf Of Francesco Pietra
>>>>> Sent: Saturday, August 04, 2007 4:33 AM
>>>>> To: Amber
>>>>> Subject: AMBER: Fwd: QM region + cutoff larger that box
>>>>>
>>>>> Although I suspected the problem does not arise from the box
>>>>> size, I have rerun md for the eq conformer. I started from
>>>>> prepin used successfully for the ax conformer in chcl3 and
>>>>> for both conformers in h2o. Setting the chloroform box 20A
>>>>> and md nstlim = 20,000.
>>>>>
>>>>> I have reexamined the min.in and md.in, they seem to be correct.
>>>>>
>>>>> It bombed at step 19150 with same indication "QM part +
>>>>> cutoff larger than box". As on previous runs, the mdcrd file
>>>>> shows the solute correctly immersed in molecules, allegedly
>>>>> chloroform. Say "allegedly" because of what said previously (below).
>>>>>
>>>>> Thanks
>>>>>
>>>>> francesco pietra
>>>>>
>>>>>
>>>>> --- Francesco Pietra <chiendarret_at_yahoo.com> wrote:
>>>>>
>>>>>> Date: Fri, 3 Aug 2007 10:02:35 -0700 (PDT)
>>>>>> From: Francesco Pietra <chiendarret_at_yahoo.com>
>>>>>> Subject: QM region + cutoff larger that box
>>>>>> To: Amber <amber_at_scripps.edu>
>>>>>>
>>>>>> QM-MM run successfully (20,000 steps) for axial
>>>>> methylcyclohexane in
>>>>>> chloroform along the lines of Tutorial A2 with Amber9.
>>>>>>
>>>>>> Equatorial methylcyclohexane also run successfully for
>>>>> 10,000 steps.
>>>>>> When I attemped to double nstlim (20,000), run bombed because QM
>>>>>> region + cutoff
>>>>>> (8.0)
>>>>>> larger than box (15.0). I repeated the run, instead of two
>>>>> consecutive
>>>>>> 10,000 steps, just a single run of 20,000 starting from where the
>>>>>> 10,000-step run had been successful. Same error bombing.
>>>>> out file for
>>>>>> the latter run attached. I can't grasp where my error is.
>>>>>>
>>>>>> An additional query: I was unable to delete chloroform from VMD.
>>>>>> Curiously, if I indicate to show carbon only, this is shown
>>>>> correctly
>>>>>> for the methylcyclohexane molecule, while chloroform is shown as an
>>>>>> ammonia-like umbrella with four identical atoms. I.e., chlorine is
>>>>>> shown as if it were carbon. Should I select a color for chlorine? I
>>>>>> was unable to trace where to deal with chlorine atom. I used the
>>>>>> chloroform box in Amber9.
>>>>>>
>>>>>> All that because I am checking convergence for ESCF with a simple
>>>>>> test, flexible molecule.
>>>>>>
>>>>>> Thanks
>>>>>>
>>>>>> francesco pietra
>>>>>> 

-- 
M. L. Dodson
Email:	mldodson-at-houston-dot-rr-dot-com
Phone:	eight_three_two-56_three-386_one
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu