I’m trying to use custom MO coefficients within a CCSD calculation, but am obtaining results which are both wrong and inconsistent between different QC modules for a trivial example. This may be related to Post-SCF with custom matrices, but I don’t need it to work for open-shell references. Here is a minimal working example, using pseudocanonicalized MP2-natural orbitals. (I.e. the Fock matrix is only diagonal within the occupied-occupied and virtual-virtual subspaces.)
min_example.py (2.7 KB)
It’s a 2-electron problem, so CCSD should be exact, regardless of orbital choice, but that’s not what I’m seeing:
Exact energy: -1.1400734808025696
Old CCSD energy: -1.1400734808003057
FNOCC CCSD energy with new orbitals: -1.1407206561955716
CCENERGY CCSD energy with new orbitals: -1.1400706133026943
Do I need to be doing something other than wfn.Ca().copy(…) to make CCSD care about the off-diagonal part of F? Is one of the qc_modules preferred for this?