Core-valence calculations with LCCSD(T) and large basis sets

Hello Mihaly,

I would like to find the core-valence correlation energy for isobutene using the cc-pwcvNz basis sets (N = T, Q, 5). A typical input file is attached. The basis sets are taken from the Basis Set Exchange and the auxiliary -RI ones are used both as -RI and -RI-JK.

For the frozen-core calculations, the total energies are close to the canonical ones obtained with CFOUR or MRCC:
TZ: -156.928862 (local) vs. -156.928246 Hartree (canonical)
QZ: -156.970041 vs. -156.968886 Hartree

For full CCSD(T) and TZ, the results are close as well:
-157.122965 vs. -157.122411 Hartree

However, for the QZ basis set, the result does not look right:
 -157.174253 vs. -157.185151 Hartree

The use of lcorthr=vtight or vvtight results in similar values. This problem exists for LCCSD(T) with the 5Z basis set, too. FNO-CCSD(T)/QZ is free of this problem and gives Etot = -157.186725 Hartree using  lnoepsv=1E-5.

Could you please comment on this issue? It would be great to use LCCSD(T) for CBS extrapolation of the core-valence correlation energy instead of the canonical computations.

Regards,

-Eugene