From: Gumbart, JC (
Date: Wed Jan 29 2020 - 11:54:19 CST

If you look at some of our published examples, the fit usually ends up being quite good. In your case, you still have very low-energy data point on the MM side throwing everything off. I would look deeper into that conformation and try to figure out why it’s so unusual.


On Jan 29, 2020, at 8:57 AM, David Sáez <<>> wrote:

Hi James and Giuseppe,

Thanks for your suggestions. I used the charges provided by CGenFF (much less polarized than those obtained in the calculation) and the profile improved a lot:

The MM potential in the picture has an RMSE of 0.875. Is that a reasonable value, or should I keep searching a profile more similar to the QME?

PS: Regarding the charge optimization I noticed that in different iterations I obtained quite different values for the atomic charges, without signs of convergence. What is the target function in the process? Reproducing the electrostatic potential like in ESP?


On Mon, Jan 27, 2020 at 10:52 PM Gumbart, JC <<>> wrote:
Hi David,

You should look carefully at the dihedral scan that gives the anomalously low values. There is probably some interaction that is too favorable, possibly due to unrealistic charges. If you suspect that, you can modify the charges to see if it makes a difference.


On Jan 25, 2020, at 12:07 AM, David Sáez <<>> wrote:

Hello everyone,

I am working on the parametrization of a small molecule using the superb tool called FFToolkit and Gaussian 16. Up to now, it has worked great. Now, in the dihedral section I am facing problems. When trying to find parameters to reproduce the QM potential, I get this plot:

Is it normal for the preliminary MM potential to be so high in comparison to the QM potential? Should I take a look on the other elements of interaction (charges, angles, bonds) because they could be raising the energy too much?

Thanks for your time.