Dear bkwx97,
Unfortunately, the noted restart options are not yet implemented for LNO-based post-CCSD(T).
What you can do at the moment is to divide a long LNO-post-CCSD(T) computation into a smaller chucks, up to single domain/LMO contributions, as follows:
For example, for LNO-CCSDT(Q), you should make first a
calc=lno-ccsd(t)
ccprog=mrcc
computation with the same basis set.
The MP2 correction for dropped LNOs, HF and LNO-CCSD(T) energy are taken from this LNO-CCSD(T) result file, this could be completed in one piece. (MP2 corrections are obtained using ccprog=mrcc option to remain consistent with the LNO-CCSDT(Q) calculations, otherwise always use the default ccprog=ccsd up to LNO-CCSD(T).)
You need to save the MOCOEF*, FOCK, VARS, SCFDENSITIES files for the next step.
Then, you can compute each the post-CCSD(T) correlation of up to each LMO in separate calculations, using the domainlist keyword (not in the manual yet). Here you use the converged SCF density and LMOs saved from the previous calculation and run:
scfiguess=off
orblocguess=read
domainlist=n
m_1, m_2, ...
where n is the number of LMO domains that are computed in this specific job, and m_1, m_2, ... are the serial numbers of the domains. The rest of the domains are skipped in this computation and can be done parallel in a similar job.
Then, you recover the total LNO-CCSDT(Q) correlation energy by collecting for the single LMOs the domain CCSDT(Q)-CCSD(T) contributions from the separate result files and add them to the first job’s LNO-CCSD(T) result.
We hope this helps,
Peter and Balazs
