ROHF-CCSD(T) using MP2 Natural orbitals

[copied from github.com/psi4/psi4public/issues/165]

Dear Psi4 Developers,

I would like to carry out a single point rohf-ccsd(t) calculation on an anion using mp2 natural orbital. Is this calculation possible in psi4?

Sincerely,
Kaye Archer
Jordan Lab
University of Pittsburgh
Pittsburgh, PA 15260

Certainly psi4 can generate the mp2 natural orbitals and pass them on to Kallay’s mrcc code. But a line in http://www.psicode.org/psi4manual/master/mrcc.html indicates that rohf references and perturbative methods don’t mix. Is that about right, @jturney? Would uhf-ccsd(t) do for your project?

I don’t know about using natural orbitals in Crawford’s CC code.

I don’t think FNOs are hooked into Crawford’s codes yet. With Kallay’s I haven’t implemented UHF FNOs yet. Actually I just checked in code to throw an exception when this is requested. I don’t it would be difficult to generate UHF FNOs for Kallay’s code. I will try to get this done in the near future.

It would be easy to hook NOs into my ROHF-CCSD(T) energy code, though there is also the question of what you mean by MP2 here. Perhaps ROHF-MBPT(2)? Have people tried much with those NOs?

Well, the OPDM isn’t block diagonal for ROHF-MBPT(2) if you include the singles, right? I guess you could just ignore them - I’m sure the NOs would still have enough information to reliably truncate the virtual space.

I am interested in ROHF-MP2 natural orbitals.

Can you clarify what you mean, Eugene? The singles contribution to the ROHF-MBPT(2) OPDM should be trivial to add even though they probably don’t matter for constructing an effective virtual NO space.

Yes, adding singles is trivial, but since the OPDM isn’t block diagonal when you include them, the NOs will mix occupied and virtual orbitals. I think the right way to build the truncated space in this case would be to ignore them so the new virtual NOs won’t have any contributions from the occupied MOs.

Yes, I agree that without the singles there is no OV block of the OPDM, but for building the VNOs wouldn’t you just compute the vir-vir block only, including the singles?

$$\gamma_{ab} \leftarrow t_i^a t_i^b$$

Perhaps we’re talking about the same thing, though.

Sorry, I guess I just don’t like calling them NOs if there is still coupling to the oo block through the ov block. Yes, of course, you can just diagonalize the vv block.