CCSD energy difference

Dear MRCC,
In preparation for a longer series, I made a simple run for CCSD energy and MOLDEN files in mrcc and in cfour, expecting the energies to match. The molecule is ScO doublet at a radius near the equilibrium. The HF-SCF energies match, but the CCSD values differ greatly.

The GENBAS file entries came from BSE (aug-cc-pwCVTZ). The system is Ubuntu 22.04 LTS up to date, and Intel oneAPI current level (2023.01, but ifort says 2022.9). MRCC was compiled from the 2022.03.18 tar with .f patches overlaid.  The same results are obtained if mrcc 2022.03.18 is created with no patches.

Is my mrcc input incorrect in some way?
Thanks very much!
Steve Coy
 

Dear Steve Coy,

it appears that the two computations differ (at least) in the frozen core settings. In MRCC it is turned ON by default, while CFOUR correlated all electrons. This indeed can cause such large deviations.

Guessing from the basis set suitable for core-valance (but not particularly for core-core) correlation, you perhaps need a third setting correlating only the 3s3p subvalence electrons of Sc, but not the deep core electrons. You can freeze the 6 deep core orbitals in MRCC via core=6 or you can correlate all electrons via core=0,

Regarding your plans for a longer series, I would recommend to use density fitting for the UCCSD computations for larger systems/basis sets.
Best regards,
Peter

Thanks very much for the info on mrcc-defaulted frozen-core!
A special interest in mrcc is the correctness of the molden file, useful for wavefunction analysis, and the "problematic" nature of the ones from cfour. Can mrcc produce a natural orbital molden file?
Thanks again,
Steve Coy