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

If you use scftype=uhf, the rohftype keyword is not in effect. It only applies to ROHF calculations.

I'm going to take the compiled patches and move them to the binary to see if this fixes the issue but I'm having an issue where MRCC source + the patches is compiled with Intel 2019 is locking onto a higher energy solution with the pvqz basis set. Attached are my files. The energy is .10 hartrees more positive than pvtz. I'm attaching the CCSD(T)/PVQZ job because it only takes a few minutes on even modest hardware.

I compiled with Intel 2019 with the uccsd.f and ovirt.f patches.

further edit: I tested the patched ovirt.f and uccsd.f with the binary and I can reproduce the issue for pvtz vs. pvqz

In this case the best solution is to restart the cc-pVQZ SCF calculation from the cc-pVTZ density. To that end run an SCF with the cc-pVTZ basis, then run the cc-pVQZ calculation setting scfiguess=restart in the input file.

Feeding it a guess from PVTZ worked fine.

I have an another issue. For Pt (and Pt di and triatomics) the SCF from MOLPRO and MRCC agree to the 8th decimal place. The CCSD(T) energies disagree by .2 Hartrees or more. If I use molpro as the front end this issue disappears for CCSD. But I would like to use the (Q) methods and MRCC standalone due to this particular computer cluster's memory specifications. Attached is the input file for Pt for MRCC.

MOLPRO SCF: -118.511936492627
MRCC SCF: -118.511936493155

I cannot see any problem with the CCSD run.
Could you post the molpro output?

See attached

edit:

Also attached is an MRCC ROHF CCSD(T) computation for Pt titled Ptnew.txt