Diffrence in frequencies from ORCA and autode

[tkpiskorz]

Describe the bug The frequencies printed from Molecule.frequencies are different from ORCA output file generated by autodE.

To Reproduce

orca = ade.methods.ORCA()
mol = ade.Molecule(smiles='CCCC')
mol.optimise(method=orca, keywords=['wB97X-D3', 'def2-TZVPP', 'tightOPT', 'freq'])
print(mol.frequencies)

Expected behavior The frequencies from python script using autodE should be the same as in ORCA output generated by the same script.

Environment

• Operating System: Linux
• Python version: 3.7
• autodE version: 1.3.2 and 1.4.0

Additional context The above example is the minimalistic, the difference is small (5-10 cm-1). I observed it for metal complexes, where with autodE I get negative frequencies (-30 cm-1), while in ORCA there are all positive. I noticed that the difference is smaller with PBE0/def2-SVP.

[t-young31]

Hey @tkpiskorz

Can't say I'm too surprised with this. Neither the opt or normal mode analysis is fully deterministic. ORCA also has access to higher precision quantities. What's the ORCA<->Gaussian difference like?

Also, frequencies at DFT are wrong by ~50 cm-1 so 🤷🏼 (ref. e.g. https://arxiv.org/pdf/2302.04448.pdf)

[tkpiskorz]

Thanks for quick answer. I am using frequencies to determine bond force constants for MM. The small diffrences are not that important. However, when it changes sign to negative, I am worried that negative frequencies will result in negative bond force constants. Maybe I should not be that worried (as they will contribute only a little).

[edit] :sorry did not mean to close it

[t-young31]

However, when it changes sign to negative, I am worried that negative frequencies will result in negative bond force constants

Oh yes – that would be very bad!

What's the reason for not computing the force constants numerically and so avoiding horrible coupling(?)

[t-young31]

did you get any further on this @tkpiskorz ? I'm inclined to say this isn't a bug..

[tkpiskorz]

No, I have not managed that, but thanks!