Frequencies and geometries calculated in NWChem differ from other programs

[Matthew Graneri]

While testing NWChem 7.2.3 on my Mac, I noticed something concerning. When I compared the frequencies of water calculated in NWChem to other programs (Gaussian 16, Orca 6.1 and Molpro 2022.2), NWChem deviated from those other programs by a non-negligible amount. I also tested this on a Linux build of NWChem but got the same result. I then tested calculations on ammonia and again found that NWChem disagreed with other programs.

While I am aware that the results from calculations can sometimes be different between programs when grids and different implementations are concerned, these calculations were all conducted at MP2/cc-pVDZ level of theory, and should therefore not be influenced by grid effects at all. I also made sure that all programs were using the most abundant isotopic masses rather than the average atomic masses in their frequency calculations, so that is also not the cause.

It appears that the geometry optimisations in NWChem are also not converging to the same geometries as the other programs. While the deviation isn't much, it is still usually an order of magnitude greater than the differences between the other programs.

The results of my tests are summarised in an excel document attached below. I will also attach the output from the nwchem calculation on NH3. If there's anything else you need, just ask.

Regards,

Matthew

[Edoardo Aprà]

Your input file is missing the standard MP2 freeze input option.

mp2

 freeze atomic

end

https://nwchemgit.github.io/MP2.html#freeze-freezing-orbitals

[Matthew Graneri]

Thank you, that fixed it! I didn't realise the default would be a full MP2 calculation.