Hello everyone,
I'm doing some QSAR analysis with a data set which contains some coordination complexes. In order to incorporate some chemical knowledge into the descriptor set I'm going to use to train my model I want to calculate some quantum chemical descriptors. Since DFT calculations are very costly for a big data set I'm trying to use MOPAC to calculate the QM descriptors.
Now I'm just wondering how reliable are the energies of complexes calculated with MOPAC?
Can someone point me to a source to figure it out by myself? Or does anyone has experience using MOPAC to calculate the energies of complexes?
Best regards,
Markus
The semi-empirical methods supported by MOPAC works reasonably for most organic systems, however the reliability is not quite high (including the latest PM7) if your system contains transition metal, in this case I suggest using GFN-xTB theory instead, which can be approximately regarded as semi-empirical version of DFT and is supported by xtb code. The speed of GFN-xTB is even faster than semi-empirical methods.
The semi-empirical methods supported by MOPAC works reasonably for most organic systems, however the reliability is not quite high (including the latest PM7) if your system contains transition metal, in this case I suggest using GFN-xTB theory instead, which can be approximately regarded as semi-empirical version of DFT and is supported by xtb code. The speed of GFN-xTB is even faster than semi-empirical methods.
High-level QM methods like G3 or G4 produce rms errors on the order of 1 kcal/mol or 4 kJ/mol for small, organic molecules. (Larry Curtiss has spent a great deal of time driving QM calculations towards "chemical accuracy.") DFT methods are a factor of 2-3 worse but much faster. Simple SCF calculations are a hundred times worse. Semi-empirical methods are worse still, although being empirical, they can be refit to specific chemistries.
The G3 test suite results do not include metal clusters, for which high-accuracy methods are hard to find. Semi-empirical methods are fast but not reliable. You could try to fit one to specific chemistry but running the benchmark calculations (high level QM) and fitting the parameters would probably take longer than just running the DFT calculations (in elapsed time).
Canonical semiempiirics are very bad for metalorganic complexes. We tested that first in paper Minenkov-Sharapa-Cavallo few years ago. After we tried pure organics with different semiempirical and tightbinding approaches(Sharapa-Genaev-Cavallo-Minenkov), GFN-XTB2 was really one of the best but still I am afraid for metalorganics it might produce doubtfull results. At the end question is requested quality.
- Badges
- Science topic