Strange scf convergence between CCSDT and CCSD(T) scf for Iron

So I'm running into an issue with Cerium. I downloaded the def2-ecp for the def family from basis set exchange. def2-TZVP gives a sensible result. def2-QZVP gives a result 100 hartrees more positive. I checked the core and valence electrons to make sure the same number of electrons are there for TZVP and QZVP.

I also checked to make sure the def2-qzvp basis in the basis folder and on basis set exchange were equivalent.

Attached is attached is output for def2-tzvp and def2-qzvp

Hello!

It seems that the SCF iteration of your QZ calculation did not converge to the global optimum. Note that the calculation starts from the superposition of atomic densities, which is probably not too good for Ce4+. More printout via verbosity=3 option could help identify such cases.

It is generally a good idea not to start your large basis set calculation from the SAD guess.
Since you have the correct TZ solution, you can restart the QZ from that via
scfiguess=restart
That converges to the right solution.

If the RI-JK basis is available, scfiguess=small is also recommended with e.g. basis_sm=def2-svp, or you can do a separate def2-svp run and use scfiguess=restart.

Best wishes,
Peter

For the Ce2+ I'm running into a similar issue where the ROHF result with MRCC for the ROHF differs from MOLPRO. In this case by .15 hartrees. The difference is even using the def2-svp still disagrees from MOLPRO. The scfiguess=restart worked with the Ce4+ but in this case it doesn't seem to be doing the trick.

I have also tried scfiguess=mo and that doesn't work. I switched the initial guess from SAD to ao as well and that didn't change the result. Even the smaller def2-sv(p) is still far away from MOLPRO's ROHF result.

Run a UHF quadratic SCF calculation (scftype=uhf, qscf=aughessg), then restart the ROHF from the UHF orbitals (scftype=rohf, scfiguess=mo).

I'm attaching my Europium atom output

I have tried the previous suggestions. Using scfiguess=ao and the qscf solver I get a sensible UHF answer with the def2-svp set (and using that sensible results with bigger def sets, and atomization energies for the molecular species with UHF). I attempted to use the uhf result as an initial guess and get an answer that is almost .10 hartrees too positive. I tried an initial guess with and without scfiguess=ao for the uhf portion. I also tried to use qscf as the solver with a uhf initial guess and that lead to a segfault.

I downloaded the newest Feb 2020 binary and running into a familiar problem.

It's for AlV. I have attempted the following:

1. uhf/cc-pvdz initial guess for ROHF. I have tried this both with and without qcscf
2. different starting guesses (ao, core)

And I get a very different energy from MOLPRO for the hartree fock portion

From molpro for the ROHF-SCF I get -1184.74971986. The UHF guess gets close to this but when I use the mo's as an initial guess for ROHF-SCF I get a wildly different result

attached is an output file with verbosity=3. Does MRCC have an equivalent of rotating orbitals?