Local CCSD from localized HF orbitals

Dear developers,
I have a localized HF orbitals. Is it possible to use these orbitals to do Local CCSD calculations? And subsequently can I run dens=2 for this local CCSD calculation?

Dear prasanta13,

you can generate various types of localized HF orbitals within MRCC, see orbloc* keywords.
If for some reason you need something else, you can read localized orbitals via the "read" option of orblocguess keyword. Those have to correspond to the other files coming from an MRCC HF calculation.

dens=2 is only available at the moment for the general order ccprog option of mrcc, but such dens calculation does not have local approximations.
Best regards,
Peter

Thank you very much Dr. Nagy. Is there any chance in the future that one can obtain the 2pdm from localized CCSD?

How many atoms/AOs do you have in your system of interest? If this is not very large, there are developments in progress.

It seems to be rare that one needs CC 2-RDM for all atoms of a large molecule. Are you sure that you need 2-RDM and not a specific property, like force?

If 2-RDM is required, what is the purpose of it? Do you need CCSD accuracy for all atoms of the large molecule?

Currently we are working with ~300-350 primitives (Water cluster with aug-cc-pVDZ(decontracted)) basis set with CCSD. With the help of LCCSD, we aim to reach the at least twice the number of primitives that we employ currently. The corresponding 2RDM will then be huge, ofcourse and that is why we worked to develop a localization scheme that can give sparse and/or fragmented 2PDM (yet to be tested though).
You are right to say that it is very rare to calculate the 2RDM rather than properties. However, we need the total energy and the 2RDM (must) for our analysis of atomic level interaction of correlation energies. Please look at the Eq - 1 of this paper link.springer.com/article/10.1007/s00214-023-03057-x .This is the master equation that needs the 2RDM. This is the purpose of the 2RDM.
Now, the target level of theory is always CCSD. We can think of MP2, but the effect of 1-e correlation terms are not present with MP theory. The strategy is to remain consistent with desired level of accuracy, such as CCSD. We can target larger systems with the advent of better algorithms and architectures. What do you think?

We can run with the development MRCC code CCSD gradient with 600-700 orbitals. That does not require local approximations, DF enables this even in canonical orbitals without any additional approximations. However, the 2-RDM is not directly computed in DF algorithms, only 2-RDM*3-centerDFintergral. One can extract 4-center-2-RDM from this, although that requires considerable work. Than it can in principle be transformed to localized MO basis to compress it.
A better compression would be some factorization of 2-RDM or rewriting your equations to work with the much more compact 2-RDM*3-centerDFintergral. All of the above would require a lot of extra work from us. If this would be very important for you, feel free to write me an e-mail to discuss further.