search

openmopac / mopac

Molecular Orbital PACkage
http://openmopac.net
GNU Lesser General Public License v3.0
115 stars 31 forks source link

NMR calculations and pH #84

Open flatstik opened 2 years ago

flatstik commented 2 years ago

A great feature would be to print out NMR data prediction with J-couplings with/from the generated output with keyword "NMROUT(C, H, F...) etc". EPS would be mandatory to simulate the solvent (with keywords or better defined numerical values) and if EPS is used, optimization to that solvent with the most rigorous precision before optimization should be used).

pH function would be also be great for ionization of molecules. This is nice function in openbabel for adding hydrogens: -p adds hydrogens appropriate for pH (which use transforms in phmodel.txt)

godotalgorithm commented 2 years ago

MOPAC does have a pKa feature already, but it is only for protonation of oxygen atoms in molecules, and it involves a limited, custom-built semiempirical model on top of a PM6 calculation. Without additional fits/corrections, pKa predictions suffer in accuracy from MOPAC's relatively poor description of the charge density around hydrogen atoms (which is why MOPAC needs explicit corrections for hydrogen bonding in PM6-D3H4 and PM7). Jimmy Stewart does not recommend "consistent" pKa estimates that are calculated directly with PM6/PM7 and without any additional correction factors. I have not explored this myself yet to have a more informed opinion.

NMR is a feature in the suite of tools associated with xTB. I am not familiar with that feature, but it appears that ab-initio quantum chemistry data is required as part of their prescribed process - they do not appear to trust NMR calculations based purely on semiempirical model calculations.

In both cases, the technical implementation might not be that hard (especially piggy-backing on some tools from the xTB suite), but it will be non-trivial research to fit and validate these features to be of usable accuracy in a purely semiempirical setting. Also, the result may simply be that MOPAC's models are just too crude to describe these effects to sufficient accuracy for any practical purpose. These are use cases for which MOPAC's models may very well be near their Pauling points, and something substantially more sophisticated might be needed for reliable semiempirical versions of these property calculations. Whatever the outcome, this would be a research project, not straightforward software development.