Hello Jérémie,
I am not sure anyone is going to tell you which is the best choice. For
one, I am not sure what is the reference geometry that you are comparing to
- is it an experimental geometry or QM optimized geometry. Is a 0.02 Å
deviation within the error of the QM theory (or experiment) that you are
comparing to? How reliable is your chosen QM theory level for predicting an
accurate experimental geometry. What accuracy to do you want your model to
achieve? This can be difficult to determine, especially if you have an
experimental reference crystal structure, but you are doing gas-phase
optimization or condense-phase simulations.
As you are seeing, the bonded parameters are somewhat dependent upon the
partial atomic charges set when you have 1-4 or more intramolecular
interactions. You could always adjust the bond parameter that you are
looking at to obtain the length that you think it should be for a given
parameter set. However, I know from experience that other parameters (e.g.
angles and torsions) will likely need to be reoptimized a bit because of
parameter coupling. Maybe this will be a big effect, maybe not.
Ultimately it is best to validate your model by comparing to experimental
results or high QM theory levels. Otherwise, you want to justify your model
based on what others in the community have done. You need to prove/convince
others that the eventual MM/MD results are reasonable. To this end when you
are presenting your work to the community, it is always good to talk about
what approximations were made when creating your model, and what are the
possible errors that these approximations will give rise to.
Hope this helps,
Karl
On Mon, Nov 2, 2015 at 2:20 PM, Jérémie Knoops <jeremie.knoops.gmail.com>
wrote:
> Sorry for multiple emails, but I want to make it clear.
>
> I first tried to use the RESP charges obtained with the iop(6/33=2)
> iop(6/42=6) iop(6/50=1) keywords ("automated way"). The absolute values of
> those charges seems to be to high, and when I used them, the bond lengths
> between the central carbon and the nitrogens are 0.06 Å shorter than the
> reference value or the bond length obtained by a quantum calculation
>
> Then I tried to use the RESP charges obtained with the iop(6/50=1) keyword
> only, which seem to be "well balanced" with the gaff and ff12SB forcefields
> (for example the charges of the arginine guanidinium group). It's still
> impossible to parametrize the molecule (the bond lengths are still more
> than 0.02 Å shorter, and I get other problems with angles).
>
> The absolute values of the am1-bcc charges seem very low in comparison with
> the arginine guanidinium, but I can get a reasonable equilibrium geometry
> with them.
>
> It seems that I have to choose between a good geometry and well balanced
> charges... What's the best choice?!
>
>
> *Jérémie Knoops*
>
> *PhD StudentLaboratory for Chemistry of Novel Materials
> <http://morris.umh.ac.be/default.aspx>*
> *University of Mons <http://portail.umons.ac.be/en2/>*
>
>
>
> 2015-11-02 10:06 GMT+01:00 Jérémie Knoops <jeremie.knoops.gmail.com>:
>
> > Hi,
> >
> > The antechamber "automated way" to determine RESP charges, as reported
> > here: http://www.ub.edu/cbdd/?q=content/gaussian-09-and-antechamber12 ;
> > create gaussian input files with the following keywords: iop(6/33=2)
> > iop(6/42=6) iop(6/50=1).
> >
> > But it seems that using iop(6/50=1) only is a better way to get well
> > balanced charges for the amber & gaff forcefields.For example, only this
> > keyword is mentionned In the gaussian manual (
> > http://www.gaussian.com/g_tech/g_ur/k_population.htm ).
> >
> > Why are those extra keywords iop(6/33=2) & iop(6/42=6) added in
> > the automated way?
> >
> >
> > *Jérémie Knoops*
> >
> > *PhD StudentLaboratory for Chemistry of Novel Materials
> > <http://morris.umh.ac.be/default.aspx>*
> > *University of Mons* <http://portail.umons.ac.be/en2/>
> >
> >
> > 2015-10-27 11:02 GMT+01:00 Ye Mei <ymei.itcs.ecnu.edu.cn>:
> >
> >> RESP charge fitting suffers from ill-conditioning of the linear
> equations,
> >> which means that you may get "degenerate" solutions for the atomic
> >> charges.
> >> Some might be unphysical.
> >> Usually, AM1 charges are fine. Or I should say they are compatible with
> >> existing AMBER force fields.
> >> If you REALLY want to use ESP-based charge, you can use our dRESP charge
> >> fitting method ( doi: 10.1002/jcc.23208) taking AM1 charge as an initial
> >> guess.
> >> Or you can hack the resp code by replacing LU decomposition with SVD
> >> analysis and setting the null space to zero.
> >>
> >> Good luck,
> >>
> >> Ye
> >>
> >> On 10/27/2015 04:38 PM, Jérémie Knoops wrote:
> >> > Dear Amber users,
> >> >
> >> > I'm trying to determine the charges for phenylguanidinium.
> >> > I normally use the RED servers (and very often multiconfigurational
> fit)
> >> > but the q4md website is still down.
> >> > Using RESP and Gaussian with one conformation (minimized structure), I
> >> get
> >> > too high absolute values for them (up to 1.1 for a carbon atom,
> >> > see GdPhenyl_resp.mol2).
> >> > It's impossible to parametrize the molecule with those charges, the
> >> > equilibrium bond length is 0.06 Å shorter than the reference bond
> >> length.
> >> >
> >> > I tried the am1-bcc scheme, also with one conformation and the
> charges
> >> > seem reliable (see GdPhenyl_am1-bcc.mol2). I can parametrize the
> torsion
> >> > with them.
> >> > Do I have to use the am1-bcc charges? Are they really reliable? Are
> >> there
> >> > other options?
> >> >
> >> > Thank you for your advice.
> >> >
> >> > *Jérémie Knoops*
> >> >
> >> > *PhD StudentLaboratory for Chemistry of Novel Materials
> >> > <http://morris.umh.ac.be/default.aspx>*
> >> > *University of Mons* <http://portail.umons.ac.be/en2/>
> >> >
> >> >
> >> > _______________________________________________
> >> > 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
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Nov 02 2015 - 08:00:05 PST
