Hi Yong,
just one brief comment.
When for example we try to compare quality of AMBER
force field with benchmark QM data for stacking, H-bonding,
RNA tertiary interactions.... we usually derive the charges
at the high MP2/large_basis_set level.
The purpose is to verify the phys. chem. validity of the ESP
point charge + vdW model.
Thus we want to make the AMBER model as close as possible
to the benchmark QM data, to pick up QM - AMBER differences
stemming from the incompleteness of the model.
The use of standard HF charges would introduce an undesired systematic
error into such gas phase comparisons.
With complete basis set MP2 calculations and CCSD(T) corrections
the QM description of systems with rigid nucleobases is close
to ultimate and the performance of AMBER (as a phys. chem. model)
is very good and the agreement even better
than with medium level QM benchmarks from nineties.
It translates to very good AMBER performance in RNA simulations.
For molecules with flexible chains, exactly as you describe,
the geometry is pretty sensitive to the calculational level,
and even the electronic distribution can be, and subsequently
the ESP. With Dan Svozil, my wife, Tom Cheatham also participates,
we now play with the NA backbone and the only thing
we can tell now is that it is a pretty complex issue....
The electronic structure is pretty sensitive to various aspects
of computations and is yet to be seen what will be at the very end.
It looks like a Pandorra's box.
Best wishes, Jiri
>
> Just chip-in my 2 cents ...
>
> Ross made a number of excellent points. But I do not think Jiri quite meant
> what Ross thought he meant :).
>
> As for optimization, although the accepted approach, following the tradition
> of Cornell et al, is to use relatively in-expensive (in today's scale)
> optimization. Don't forget, Cornall et al's paper was published in last
> millennium when HF/6-31G* was considered an expensive and accurate model. In
> today's standard, it is just a model. As for peptide chemistry, I am not
> convinced that we should be bound by this limitation which was more or less
> due to the speed (or lack of it) of supercomputers in last century. If we
> also use the term "supercomputer" as the "moving" standard, we might be more
> inclined to think beyond what was available a decade ago.
>
> However, if the purpose is to calculate the electrostatic potential to fit
> the charges, the "conventional wisdom" thinks optimizations do not have to
> be as thorough. Having said that, it is still a matter of unsettled issue as
> to how and how much that subtle differences in charge distribution would
> affect the property besides the obvious dipoles and other-poles. Not to
> drive this discussion awfully off the track, molecular properties, as for
> peptide-chemistry in solvent, are probably more influenced by local
> chemistry than long-range field. In other words, the influence of charge
> distribution to the 1-4 energy, which is intricately related to the torsion
> energy (as a matter of fact should be considered part of it), is probably
> more important.
>
>
> yong
>
> >
> >
> >
> > With regards to the
> > optimisation part
> > of the calculation I think one could discuss this for ever.
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber.scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Sun Nov 05 2006 - 06:07:13 PST
