Re: AMBER: HF/6-31G* ESP Charges do not match with ff02 charges

From: FyD <>
Date: Fri, 19 Jan 2007 09:31:46 +0100


> I do not know what version of NWChem you are using. NWChem at least
> version 4.5-4.7 does
> not properly implement RESP, though it can give deceptively
> good-looking results on some
> molecules. I will guess you want RESP, not plain ESP, as that's how
> AMBER charges were
> assigned. NWChem is not suitable for RESP. I know that NWChem 5.0
> has significantly
> changed its RESP charge calculations, but they too were flawed
> (though in a different
> way) when I used it in beta. If you want to verify that your
> software and methods are
> reasonable, obtain the original RESP papers and confirm that you can
> reproduce results
> for simple cases like butane and n-methylacetamide. NWChem seemed to
> give especially
> unreasonable results on butane when I was doing my own verifications.

Here, is it not possible to ask to NWChem to only (i) optimize a
structure (ii) compute the MEP and NOT derive the charges; letting the
RESP program doing the fitting step to generate RESP or ESP charge
values ?

Did you try to compare the MEP generated by Gaussian, and that
generated by NWChem ? Are they identical ? Thanks, Francois

So far, R.E.D. interfaces GAMESS-US compiled from the sources and
Gaussian (94, 98 & 03). R.E.D.-III now interfaces also PC-GAMESS.

We could also imagine to load optimization output(s) from NWChem or
Q-Chem (as input of R.E.D.-III) for MEP computation using PC-GAMESS,
GAMESS-US or Gaussian. We are going to develop this part...

Quoting Pavan G <>:

> Although the charges add up to 0 in both the cases, there are significant
> differences in numbers and in some cases difference in the sign of charge.
> Could somebody please explain where I've gone wrong and/or the proper way to
> verify the charges produced by AMBER.

AMBER FF charge values are not reproducible. The charge diff. can be
up to 0.07 e in particular for methyl or methylene carbon.

So far, the only way to generate reproducible charge values is to use
the rigid-body algorythm implemented in R.E.D. You can use multi-orientation
RESP fit to get highly reproducible charge values, and
multi-conformation RESP fit to get general charge values more
compatible with MD simulation...

Please see also R.E.DD.B.
where you will find many examples of projects with charge values...

regards, Francois

The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to
Received on Sun Jan 21 2007 - 06:07:34 PST
Custom Search