Hi David,
Negative bond parameters in Paramfit are an indication of a poor fitting
result. Try increasing the number of input structures, or not fitting bonds
simultaneously with other parameters. You may want to fit bond parameters
using a different method (simple quantum scan of the bond of interest, fit
function directly to that) or not fit those parameters at all. The program
is designed more for angles and primarily dihedrals.
A more detailed explanation of this is that Paramfit's fitting function
doesn't understand anything about chemistry, and while the initial
generation of random parameter sets is produced in a reasonable range,
there is nothing to prevent the algorithm from wandering outside of that
range trying to minimize the cost function (least squares difference of
energy). Often if it is allowed to mess with bond terms it will do so in
order to compensate for other deficiencies in the fit. Also, check your
final function value. If it is quite high, this is also indicative of a
poor fit.
Best,
Robin
On Thu, Feb 11, 2016 at 3:53 PM, David Poole <thepoole.ucdavis.edu> wrote:
> Hello all,
>
> Thanks for you past advice, I've not been replying to posts much but I have
> greatly appreciated your help. It has been a long road for this side
> project of mine and this should hopefully be the last issue in
> parameterization.
>
> I am using the paramfit tool to fit parameters to a number of minimized
> structure energies for the liganded ferrous heme systems. But whenever I
> use the simplex it sometimes produces negative bond force constants (which
> are bad). This almost always occurs if both algorithms are used in
> sequence.
>
> What can I do to prevent this? Or is this a bug?
>
> -David
> _______________________________________________
> 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 Thu Feb 11 2016 - 17:00:03 PST
