Wrong (T) correction

 Dear MRCC developers,

Here are two calculations: the same closed-shell system, canonical CCSD(T), no RI, RKS and UKS BP86 reference orbitals. In both cases, SCF converges to the same solution (same KS and semicanonical energies, no spin contamination for UKS). The CCSD energy is also the same, but the (T) correction differs by 3.6 kcal/mol. Compared to another program, the UKS energy is the correct one. The issue is only for non-self-consistent orbitals; i.e., RHF and UHF references yield the same CCSD(T) energy.

Best regards, Alex.

 

 

Dear Alex,

The explanation is that RKS and UKS activates the closed and open shell routes, respectively. (cf. Executing ccsd and uccsd)

In unrestricted CCSD(T), there is a term in (T) that is proportional to the Fock off-diagonals, which is non-zero for the unrestricted (T), but zero (and thus not implemented) for RHF-based closed shell (T). The difference comes from this, unrestricted UKS-based (T) code includes this, while the RKS-based restricted (T) does not (although it would be better if it was included there too).

With the current version, you should use consistently the uccsd version, e.g., via scftype=UKS. The HF and CCSD terms are unaffected.

Note: for closed-shell molecules, I am not aware of many advantages of using RKS over RHF orbitals. So I am not sure why RKS is useful to you over RHF, the CCSD(T) results are usually very close.

I hope this helps,
Peter