Error with LR-LCC2

Hello, dear MRCC community!

I'm seeking an accurate adsorption spectrum for [Ru(terpyridine)2]2+ with CC2. The computation with the def2-TZVP basis ends with a fatal error in drpa (file attached). The same computation with the def2-SVP completes correctly. What is the problem? I'm not entirely familiar with LNO and redcost_exc concepts, so this computation was supposed to be a "starting point" for further variation of parameters (maybe you can also kindly advise something in this respect).

Latest MRCC of December 31, 2024; pre-compiled binaries from the site; not patched yet.

 

Dear Alex,
Thank you for your interest and for the report. I'll look into the issue and get back to you if I find anything.
All the best,
Dávid

Well, the same input with nstate=1 gives more informative error:
 

nstate=1 means that ground-state calculation is requested. The fact that the program stops here is expected and is not related to the original issue.

Dear Alex,

Thank you for the report. I have fixed the bug, and the patch has been uploaded to the website.

Regarding the recommendations, here are my comments:

1. The CC2 transition density is not implemented, so adsorption spectrum calculations at this level are not possible. I recommend using the ADC(2) or ADC(2)-based double-hybrid TDDFT methods instead.

2. Using the local infrastructure (localcc=2018) for CC2/ADC(2) calculations up to 60 atoms does not provide any advantages, especially with the localthr=tight option. The entire molecule would be included in the local domain, meaning the calculations would not significantly speed up. Accordingly, I would remove this option.

3. If you still insist on using it, the local fitting approximation for the CIS calculation is unnecessary. This can be disabled with the redcost_tddft=off option. For molecules of this size and with the available memory capacity (mem=100GB), the CIS problem can be solved without any obstacles. In my calculation, using the cialg=disk option, it took 45 min, with a significant portion (2/3) of the time spent on integral transformation. If less memory is available on the hardware or a larger molecule is being studied, the cialg=direct option can be used to reduce the memory requirement, though this calculation would take longer due to the integral direct algorithm.

4. At this point, the reduced-cost algorithm is still enabled (redcost_exc=on). This option effectively combines the frozen virtual natural orbital (VNO) and natural auxiliary function (NAF) approximations. For a 60-atom molecule using def2-TZVP basis sets, I would also disable these approximations. In my calculation, a single ADC(2) iteration for 9 excited states took approximately 3 h. This means the complete calculation should take around 2 days. If you still want to speed up the calculations, I recommend applying only the NAF approximation (redcost_exc=3 or . This would prevent the state-selective optimization from struggling to converge for higher-energy states.

5. Please keep in mind that the default cutoff parameters (lnoepsv and naf_cor) for the frozen VNO and NAF approximations were determined for the aug-cc-pVTZ basis sets. If you intend to use a different basis set, it may be advisable to experiment with the thresholds on smaller test cases.

I hope you find this helpful. Feel free to reach out if you have any further questions.

Best regards,
Dávid

Thank you! Your recommendations are really helpful! Everything works correctly now, so it'll take a while to test these options.