OD method vanished from Psi4

I’m wondering why Psi4 doesn’t have the implementation of the OD method that was published published in J Chem Phys 135, 104103 (2011). In the paper, the implementations are said to be done in Psi4, and Psi4 does have the corresponding OO-MP2 algorithm from the same work.

We certainly have olccd. @bozkaya, can you clarify if you have occd (non-l) that got overlooked or if that paper was referring to a linearized ccd?

handy table here if you haven’t met it

In that publication, we used my old implementation which was interfaced to an earliest version of Psi4. It is not working with the current version of psi4. It needs a significant revision to make it compatible with the current version.

However, I am going to implement a fresh DF/CD-OCCD code in the near future, as a requirement of one of my active grants. We may include DF-OCCD in the next release of Psi4!

@loriab @susilehtola

Thanks, @bozkaya. Do you recall which era of Psi4 (one of the betas, maybe) had this in it, in case Susi’s determined to run the method?

You are welcome @loriab. It was during 2010-2011. I am not sure whether I still have that old code or not. That is why I wrote a short Grant project on it, to develop an efficient DF-OCCD version with gradients. Meanwhile, If Susi need an OD code, it is available in Q-Chem.

best regards,

I wanted to use the 2011 variant since the older one in Q-Chem uses the z-vector approach and not the proper, density matrix approach.

psi4 4.0b1 was February 2012 and it looks like we didn’t bundle for sourceforge until b4 (https://sourceforge.net/projects/psicode/files/psi/), not that I recall removal of code being a common activity during the beta years.

I don’t know what, if anything, you want to try, @susilehtola, but if you’d like access to the alpha/betas code, I can get you that. It’ll be GNUMake.

Well I guess that’s alright, I’ll just have to use my own code to try run these calculations, although it’s less efficient to do full CCSD with it.

(To correct my earlier statement, the Z-vector approach is the same as the “\Lambda” approach, so the only difference to Q-Chem is that it uses DIIS, whereas the newer algorithm has quasi-second order convergence. I was able to run my jobs with Q-Chem.)