Dear Dr Case,
Ups, excuse me, I missed your email...
>> that being said - an error of +/- 0.001 i.e. +/-
>> 0.0010 for a charge accuracy of 1.10-4 (i.e. 0.0001) is _not_ normal
>
> Are we sure what the error we are talking about is? Is there a simple
> example that shows this?
In PyRED we have encountered the following problems (2 examples
below), and PyRED should ;-) now correct these problems:
(CH3)3-CH: i.e. 2-methylpropane
1) Let's guess we choose a charge accuracy of 1.10-4 e (to be printed
in the .mol2 output) and an error of +1.10-4 (because of the rounding
off procedure at the chosen accuracy - from the 1.10-6 one printed in
the RESP output) is generated during the charge fit for each of the 9
hydrogen atoms of the tertiobutyl group and to the CH hydrogen; thus
one ends up with: 9*1.10-4 + 1.10-4 = 0.0010 e: this case is 'normal'.
Correcting (a posteriori to the resp fit) the partial charge values
using -1.10-4 (i.e. using the chosen charge accuracy) of a selection
of the atoms of 2-methylpropane in agreement with (i) chemical
equivalencing and (ii) the total charge value (equals zero) is
possible here (but a little tricky).
(similarly a chosen charge accuracy of 1.10-3 e can potentially lead
to an error of 0.010 e for 2-methylpropane; such 1.10-2 errors were
present in the old GLYCAM force field).
2) Then, still for 2-methylpropane we were able to generate errors of
3.10-4 because of the use of additional an intra-molecular charge
constraint (intra-mcc). Thus here correcting this 3.10-4 error in
agreement with (i) chemical equivalencing (ii) the total charge value
and (iii) the intra-mcc using the chosen charge accuracy is
impossible; thus we had to choose to correct using the double of the
chosen charge accuracy; i.e. -1.10-4 + -2.10-4 (loop twice in the algo).
uff not easy to describe...
> In my experience, the internal charges for am1-bcc always add to zero;
> if they are only printed out to three decimal places, then the usual rounding
> effects may mean that the sum of the *printed charges* is different
> from zero.
Yes exactly.
> But you seem to be referencing some deeper error, and an example
> would be most helpful.
> Apologies if I missed the post that had a good test case.
Personally I am lost with the errors reported by the users for the
different versions of Antechamber, and I do not know the source code
of Antechamber: I remember some Antechamber versions did not apply
chemical charge equivalencing: the 'sustiva_?' molecule and the 3
fluorine atoms), then chemical charge equivalencing was applied, and
then examples with these 'strange' errors were reported. I wonder if
an example of such an error was not discussed with Michael Crowley in
the amber developer mailing list: I will try to find the reference to
that email, uff...
regards, Francois
F.-Y. Dupradeau
---
http://q4md-forcefieldtools.org/FyD/
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Jan 16 2017 - 07:00:02 PST
