as @gpwood has pointed out, it would be great to compare constraints torsion profiles between DFT training level (or QM ground truth level, even though then we are testing something slightly different) and the trained NNP for a set of models.
This could be included in a general, automatic bond/angle/trosion profile comparison test for a set of molecules.
as @gpwood has pointed out, it would be great to compare constraints torsion profiles between DFT training level (or QM ground truth level, even though then we are testing something slightly different) and the trained NNP for a set of models.
This could be included in a general, automatic bond/angle/trosion profile comparison test for a set of molecules.