The necessity of child parameters of t69

Link to the discussion on missing torsion parameter for dialkyl amide: https://openforcefield.atlassian.net/wiki/x/AwDMEw

While inspecting RMSD of MM optimized geometries from QM optimized geometries, found that Parsley fails to locate carbonyl oxygen in C(=O)NR2 (R ≠ H) at a proper position.

Open

Here’s one example of the case. Carbonyl oxygen in the center of the molecule locates in the same side with C1 in QM optimized geometry but after MM optimization with Parsley(transparent green in the figure), due to the low MM rotation barrier of C-N bond, the carbonyl oxygen is moved to the other side during the MM optimization. Simply applying higher rotational barrier could lower the deviation from QM optimized geometries.

Open

(1) The initial k value for t69 is 0.625 in SMIRNOFF99Frosst.

(2)

Open

1. Around 3562 torsions in the first gen training set were assigned to t69, which indicates that there’s a possibility that the parameter is overly general.

2. There are outliers which are supposed to be around 0 degrees but rotate to 180 degree during MM optimization (or 180->0)

(3) fitting child parameter candidates