MRCC correlation is up to 2 milli-Hartree above CFOUR for Si(^3P)

[ndattani]

First, the aug-cc-pV(D+d)Z on EMSL basis set exchange has fewer contraction coefficients in MOLPRO format than in ACESII format. GENBAS has the MOLPRO version now, but even with the MOLPRO basis set, the correlation in CFOUR and MRCC is several micro-Hartree lower than in MOLPRO.

nH agreement for ROHF:

ROHF/MOLPRO          -288.847 062 536 595
ROHF/MRCC            -288.847 062 536 595 6153
ROHF/CFOUR           -288.847 062 536 595 615

8 uH dis-agreement for FC-UCCSD: (this is due to MOLPRO not transforming core during AO->MO)

ROHF-FC-UCCSD/MOLPRO -288.918 180 594 504
ROHF-FC-UCCSD/MRCC   -288.918 180 594 505   (with molpro integrals)

ROHF-FC-UCCSD/MRCC   -288.918 188 577 032   (EMSL basis in molpro format) 
ROHF-FC-UCCSD/CFOUR  -288.918 188 516 574 79 (EMSL basis in molpro format) 
ROHF-FC-UCCSD/MRCC   -288.918 188 510 312   (with CFOUR integrals, EMSL basis in molpro format)

ROHF-FC-UCCSD/MRCC   -288 918 188 596 860   (EMSL basis in ACES II format)
ROHF-FC-UCCSD/CFOUR  -288.918 188 536 40410 (EMSL basis in ACES II format)

10 uH dis-agreement for FC-FCI: (this is probably due to MOLPRO not transforming core during AO->MO)

FC-FCI/MOLPRO        -288.920 287 049 762
FC-CCSDTQ/MRCC       -288.920 310 541 826
FC-CCSDTQ/MRCC       -288.920 310 521 320 (EMSL basis in molpro format)
FC-CCSDTQ/MRCC       -288.920 310 541 827 (with CFOUR integrals, EMSL basis in molpro format)
FC-CISDTQ/MRCC       -288.920 310 592 803 (with CFOUR integrals, EMSL basis in molpro format)

mH dis-agreement for AE-CCSD:

ROHF-AE-UCCSD/MOLPRO -288.958 425 442 665
ROHF-AE-CCSD/MRCC    -288.958 425 442 665 (with molpro integrals)
ROHF-AE-CCSD/MRCC    -288.960 375 742 453 
ROHF-AE-CCSD/CFOUR   -288.958 425 442 663
ROHF-AE-CCSD/MRCC    (with cfour integrals)

[ndattani]

Check if this is due to not using symmetry equivalenced orbitals.

[ndattani]

Seems symmetry equivalenced orbitals won't help because FCI doesn't even match (though that's with frozen-core, though all-electron results also don't match with CCSD), and because for S -> S+ we have a similar problem and even the energy differences are 3cm-1 different between MOLPRO and CFOUR or MOLPRO and MRCC.

[ndattani]

For S, the EMSL basis set for cc-pVDZ has one fewer digit for the contraction coefficients when in ACESII format than in MOLPRO format (this is also the case in Grant Hill's database, and also if you look at the MOLPRO database, the basis set there also lacks the extra digit). GENBAS now has VDZ-MOLP which has the extra digit, and the HF now agrees better (lowered by 7 nH) with MOLPRO and the correlation energies are about 70 nH lower. But this didn't help the fact that MOLPRO's energy is so much higher.