Initial guess for CC3 calculations

Dear users and developers!

I need to compute several lowest excited states for a biheterocyclic compounds like bithiophene or bifuran. The problem is that some of those roots are heavily doubly occupied, so the EOM-CCSD roots, which are used as initial guess are poor approximations of the more accurate solutions of CC3. In other words, the contributions from double replacements are severely underestimated at the CCSD level of theory, and thus often missed from the CC3 solutions due to poor starting point.

Is there a way of manual specification of the starting amplitudes for excited configurations that would be used as initial guess for the CC3 calculations?

Yours sincerely

Unfortunately not. There was some ability to accept user-defined guess vectors for EOM in PSI3. However, that was before my time so I don’t how well it worked, I just know that the code to read them is still around in the source for the EOM code, but all of the calls to it have been removed, because it required functionality from PSI3 that never made it into psi4.

To attempt to capture these missed states here are a few things you can try (have not tried this myself so mileage may vary)

  1. The EOM-CCSD Roots which are used to seed the EOM-CC3 iterations, are themselves seeded by CIS roots which are seeded by a guess generated by single excitations between (HOMO-1, HOMO, LUMO, LUMO+1). You can expand the CIS guess by increasing the EXCITATION_RANGE options. The default value is 2.
  2. In the same spirit as above you can expand the search at the CC3 level by looking for more roots than you need to find. Unfortunately there is no way to simply change the number of CCSD roots that will be fed as guesses to the CC3 iterations, so to start the search with a larger guess space you will need to look for more roots which will take significantly longer at the CC3 level since roots are found one at a time.

Customization of the EOM guess would be a useful feature though, and is something which is on our radar. However we don’t currently have anyone working on it.

Given your use case you might want to consider using QCHEM (if you have it/can get it) as they have a wide range of options that the user can use to tune EOM-CC calculations, or CFOUR where EOM-CC has been implemented up to full quadruples.

Sorry if this is not very helpful.

Thank You for your reply. I was afraid this was the case, but hoping against hope… :slight_smile:
I will have to get back to CFOUR. It is a capable software, but the input format is cumbersome.
Especially the renumberong of orbitals when switching from one module to another, which makes
specification of initiall configurations appear as some kind of strange memory test…
But still, it is perhaps the only software for my purposes, so it is good that it is available :slight_smile:
My very best regards!