CC3/aug-cc-pVTZ oscillator strength have only 3 significant digits

[danielhollas]

Hi all,

first of all, thanks for creating and maintaining QuestDB, it's been an invaluable resource for us. 🙏

In our recent and upcoming benchmarking efforts, we're using CC3/aug-cc-pVTZ as reference values, both for the excitation energies and oscillator strengths. Unfortunately, the oscillator strengths in QuestDB are only reported up to 3 significant digits, which is not enough for many of the darker transitions (e.g. cyanoformaldehyde or formyl fluoride have reported values "0.001").

Would it be possible to provide more accurate values? We'd be also very happy to dig the numbers up and submit a PR ourselves if we were given the original output files.

Thanks!

[pfloos]

Hi @danielhollas,

Thanks for using QUEST! It's great to see that this is useful to the community.

To be fair, three digits, for the level of convergence of these CC calculations, is probably close to numerical precision.
So, I wouldn't trust much more than 3 digits.
If you'd like more accurate values, we would have to rerun the calculations with tighter convergence thresholds, I believe.

Note also that oscillator strengths are very sensitive to the level of theory, especially for the dark states you're looking at.

Cheers, Titou.

[danielhollas]

Hi @pfloos,

thank you so much for such a quick reply. I do agree that one needs to be careful with the precision, and indeed in the paper I linked above we were looking specifically at dark transitions to on the order of f=10^-3 (and below),
and we were careful with setting our convergence criteria.

I guess I should refine our wish here; what we would really like is to have the fourth digit for the transitions that have f~=10^-3. In other words, we'd like to know if f=0.001 is more like 0.00055 or 0.0014. I would think you probably have enough precision for that? At least from our experience with transitions even darker than this, the CC3 results were surprisingly robust with convergence criteria).

Thanks again for consideration.

[pfloos]

Hi @danielhollas,

I think you should ask directly @DMJacquemin if he still has the output files.
Digging all the values to get one additional digit would require a significant amount of time.

Cheers, Titou.

[DMJacquemin]

Hello.

Can start looking at them, but doing all of them is a crazy work...so will (really) not be fast. If you have a few specific you are interested in, I can take a look first for sure.

Best

Denis

[danielhollas]

Thank you so much @DMJacquemin!

We're currently interested in only a handful of molecules / states:

• $^1A''$ formyl fluoride
• $^1A''$ cyanoformaldehyde
• $^1A''$ HPS
• $^1A''$ HCCl
• $^1B_1$ Cyclopropene

We very much appreciate the amount of work, as mentioned above we're also more than happy to dig the data out ourselves if you'd like to share the output files (my email is daniel.hollas@bristol.ac.uk in case)

Best regards,

Daniel

[DMJacquemin]

Hello,

Here there are, rounded on the fourth digit.

Cyanoformaldehyde 0.0005 (the second one is still 0.0000)
Cyclopropene: 0.0015
Formylfluoride: 0.0015
HCCl: 0.0026
HPS: 0.0005

Best

Denis