DLPNO-CCSDT Calculation

More
2 months 1 day ago #913 by pradeep
Is it possible to utilized the ORCA with MRCC to perform a DLPNO-CCSDT calculation? Orca can do only DLPNO-CCSD(T), while MRCC can perform CCSDT or CCSD(Q) calculations. In that case, can one used the pair natural orbitals from ORCA and perform a CCSDT iteration from MRCC?

Please Log in or Create an account to join the conversation.

More
2 months 1 day ago #914 by kallay
Replied by kallay on topic DLPNO-CCSDT Calculation
No, it is not possible. You should run LNO-CCSDT or LNO-CCSDT(Q) calculations with Mrcc without Orca.

Best regards,
Mihaly Kallay

Please Log in or Create an account to join the conversation.

More
2 months 1 day ago #915 by nagypeter
Replied by nagypeter on topic DLPNO-CCSDT Calculation
Dear Pradeep,

in a bit more detail, LNO is a local natural orbital method, in some ways similar to DLPNO. Both approaches have CCSD(T) implementation, so you can also use LNO-CCSD(T), and both LNO and DLPNO converges to the same CCSD(T) limit without local and other approximations. There are, however, significant differences in the details of the formulation, implementation, performance, etc.
To our knowledge DLPNO-like implementations are not available in any code yet for CCSDT, CCSDT(Q) or higher-order, so in this category LNO seems the only choice.

You can access it in MRCC via, .e.g,
calc=lno-ccsdt(q)

Some notes:
- you should expect at least a 100x difference in the runtime of LNO-CCSDT(Q) compared to LNO-CCSD(T) for the same systems and basis set.
- the accuracy of the LNO approximations is less explored for CCSDT(Q), while the accuracy and performance of LNO-CCSD(T) is well-known (see Ref. 41 of the manual). Thus is it highly recommended to check if the pre-defined Normal, Tight, etc. provide sufficient accuracy for your case.
- if or rather when the LNO-CCSDT(Q) become too expensive, you can try to compute the difference of LNO-CCSDT(Q) and LNO-CCSD(T) with a smaller basis set and add and LNO-CCSD(T) based basis set correction

If you share the details of your system of interest (structure, basis set, quantity to compute...) I could probably give more specific advice.

Best wishes,
Peter

Please Log in or Create an account to join the conversation.

More
2 months 10 hours ago #918 by pradeep
Replied by pradeep on topic DLPNO-CCSDT Calculation
Thanks for your valuable response. I am planning to use it for some open-shell system. Recently I used DLPNO-CCSD(T) for OH+HCl reaction and I found some interesting results. The paper is under review, once accepted I will share the paper. In an earlier paper, I compared the CCSD(T) and CCSDTQ results for the OH+HCl system (Impact of Post-CCSD(T) Corrections on Reaction Energetics and Rate Constants of the OH• + HCl Reaction J. Phys. Chem. A 2018, 122, 36, 7151–7159). Is LNO-CCSDT possible for open-shell systems?

Please Log in or Create an account to join the conversation.

More
2 months 7 hours ago #919 by nagypeter
Replied by nagypeter on topic DLPNO-CCSDT Calculation
Thank you for sharing your paper and project.
This systems seem quite small, so I expect that the gain from local approximations is rather limited, and the speedup comes mostly from the use of natural orbitals (NOs). So you might also want to consider frozen NO approaches besides the local + NO ones.

So far only closed shell LNO-CCSD(T), LNO-CCSDT,... are released.
The open-shell codes are written in the development version up to LNO-CCSD(T), are we are planning to implement the open-shell general order LNO methods soon. I will contact you when we have something useful. Let me know if you can make use of the open shell LNO-CCSD(T) now.

Best wishes,
Peter

Please Log in or Create an account to join the conversation.

More
1 month 4 weeks ago #921 by pradeep
Replied by pradeep on topic DLPNO-CCSDT Calculation
Thanks once again for your kind reply. Presently If you can provide me the open-shell LNO-CCSD(T) , I can at least use it and compared it with DLPNO-CCSD(T) method for some systems. Once I got the results, I will mail you all the results for your advice. May I politely asked for your email id?

Please Log in or Create an account to join the conversation.

Time to create page: 0.021 seconds
Powered by Kunena Forum