RVS approximations for CC2 or ADC(2)

I don’t think this is currently possible so I guess this counts as a feature request.

Is there currently any way to use the RVS (restricted virtual space) approximation with either CC2 or ADC(2) for calculating excited states? I’ve got some situations where TDDFT is possibly a bit questionable (large amounts of charge-transfer character to electronic transitions) but CC2 and ADC(2) are just out of reach with the systems I’m interested in…although they may be if it were possible to

Essentially, it allows a cutoff to be set (typically, 50 or 60 eV), above which virtual orbitals are not included in the coupled-cluster calculations, although clearly it’s only of any use for the lowest excited states.

I think it’s possible to do similar to some extent through FNOCC, using suitable values of OCC_PERCENTAGE or OCC_TOLERANCE, but I don’t think it’s possible with CC2 or ADC(2). It looks something that would be easy to incorporate but that’s probably down to my naivety of the code / algorithms!

E.g. CC2 lit: J. Chem. Phys. 134, 214114 (2011)

ADC(2) lit: J. Comput. Chem. (2017)

You’re correct. The cc* modules (e.g., ccenergy, cctriples) are from Psi3 and haven’t yet adopted some of the modcons of Psi4. There’s been talk of adapting them so that they can take in an orbital-truncated object, but there’s some other infrastructure changes that will probably happen first. I’ve added it to the infrastructure wish list.

It is really not a good sign that the reliability of RVS-ADC2 depends on the nature of the excited state (though I only read the abstract).

I would not put much hope into this approach. A significant speed-up can be obtained if CC2/ADC2 is formulated with density fitting. E.g the Turbomole RI-CC2 implementation is fastest I know.

I must admit, I don’t really know a lot about the pitfalls of ADC(2) but there’s quite a few studies using RVS-CC2 (also including RI, I think) to study biological chromophores surrounded by protein residues (up to perhaps 200 atoms: these studies have generally used development versions of Turbomole).

Is the RI (or DF) part of RI-CC2 in forming the reference wavefunction or within the coupled-cluster part? Currently, it’s possible to use DF to form the reference wavefunction for CC2, rather than the default that’s selected for PK (presumably to get a “better” reference.

I’ll have a bit more of a fiddle with the current options…

The SCF part can be done either way, RI-CC2 refers to RI in the coupled-cluster part (including the AO-MO trafo just as in RI-MP2)

That makes sense (having had a play with Turbomole that I haven’t touched in years!). A full RI-CC2 implementation with the ability to freeze both core and virtual orbitals in psi4 would be amazing!