18.8 CCSD(T)-F12

The CCSD-F12 and UCCSD-F12 programs first do DF-MP2-F12/3C(FIX) (closed-shell) or DF-RMP2-F12/3C(FIX) (open-shell) calculations, and then perform the CCSD-F12 (UCCSD-F12) without density fitting. By default, the CCSD-F12a and CCSD-F12b energies are both computed. A specific method can be requested by appending A or B to the -F12 suffix. Furthermore, instead of the 3C(FIX) ansatz, different ansätze (e.g. 3C) can be used. In this case the amplitudes of the explicitly correlated terms are determined in the MP2-F12 calculation and kept fixed in the CCSD-F12.

It should be noted that these methods involve approximations and do not yield the exact CCSD-F12 energies. Extensive benchmarks have shown that the CCSD-F12a method slightly overestimates the correlation energies, while CCSD-F12b underestimates them. For AVDZ or AVTZ basis sets, CCSD-F12a usually gives very good results, but for larger basis sets it may overestimate the basis set limit and converge from below to the limit. Thus, convergence may not be monotonic, and extrapolation of the correlation energies should not be attempted. CCSD-F12b usually converges monotonically from below to the limit and gives best results for AVQZ and larger basis sets. Thus, we currently recommend CCSD-F12a for AVDZ and AVTZ basis sets, and CCSD-F12b for larger basis sets (rarely needed).

The perturbative triples correction can be invoked by using CCSD(T)-F12 or UCCSD(T)-F12. There is no direct F12 correction to the triples, and therefore the basis set error of the triples is not affected by the F12 (small changes of the triples energy arise from the fact that the doubles amplitudes are affected by the F12 terms). In many cases, a simple and pragmatic improvement of the triples energy can be obtained by scaling the triples energy contribution as

\Delta E_{(T*)} = \Delta E_{(T)}*E_{corr}^{MP2-F12}/E_{corr}^{MP2}
\end{displaymath} (4)

This can be done automatically by setting option SCALE_TRIP=1, i.e.


molpro@molpro.net 2018-10-20