In a recent blog post, +Jan Jensen wrote that work on a semiempirical PCM interface was underway. At that time the energy calculations did work and solvation free energies (dG) we obtained for ammonium type cations are listed in the table below where we show the difference to Chudinov et al. (ddG) has an RMSD of 1.33 kcal/mol (a similar table with oxonium type cations the RMSD is roughly 2 kcal/mol).
The original plan was to use numerical gradients to do optimizations but because I failed at naming a variable iat and not jat, numerical issues kept me from truly optimizing anything. Furthermore, the calculation of the electric field gradient using a numerical approach was very expensive. Instead of renaming iat to jat, I derived the analytical field gradient and implemented it only to find that the speed of calculating the gradient had improved greatly (a least two orders of magnitude) the optimization failures persisted because the field gradient I obtained numerically and analytically is pretty much the same.
Lo and behold. Renaming iat to jat fixed the rest of my problems and I've optimized chignolin and tryptophan cage using PM3/PCM at a speed of about 10 sec/step and 60 sec/step, respectively. Both proteins required roughly 200 steps to be optimized from their crystal structure.
Here is a plot of the energy, maximum gradient element and the root mean square of the gradient for an optimization of Chignolin using PM3/PCM.
So it looks like we're in good shape I guess. I know I am pleased.
Note to self: what about anions?