Excited state pes


#1

Browsing the documentation it appears that for selected levels of theory follow_root should allow to do pes calculations of excited states. However, it does not appear that any of these theories provide analytical gradient calculations in psi4.
Is there something I am missing?
What is more generally the support of psi4 for excited state optimization?
Is there any plan for implementing conical intersection searches?


#2

Psi4 does have EOM-CCSD with analytic gradients.

Longer term, I’m in the process of improving our orbital guess technology so that once you have an SCF converged (potentially at an excited state), you can read that in as a guess for a displaced geometry, even if the displacement causes a loss of symmetry. We’re also adding our first TD-DFT capabilities. While gradients for them would be great, I don’t know if those are in our immediate plans.

I’ve heard rumors of plans to develop a minimum-energy crossing point plugin for Psi, but those remain rumors.


#3

I have program that can do a penalty-function-based CI search, fairly easy to make it work with psi4 gradients.


#4

I was a bit skeptical about relying on eomcc for this, however I stumbled about some work in the literature, e.g.
dx.doi.org/10.1021/ct300759z
J.Chem.TheoryComput. 2013, 9, 284−292
which shades much optimism.
Of course having gradients available for other theories would be fine.


#5

Yes, there are EOM CCSD gradients for both ROHF and UHF approaches. The “follow_root” keyword is intended for CC3 computations where only a single excited state is solved, and one needs the CC3 state to correspond with the previously determined EOM CCSD state. By default, the gradients or properties for EOM CCSD are computed for the last state (irrep and highest energy) obtained. This can be controlled with PROP_ROOT and PROP_SYM. Sometimes it is necessary to carry out the Davidson algorithm for 3 excited states, for example, when it is the 2nd one that one wants to converge and further investigate. The EOM_GUESS (defaulting to the lowest-energy eigenvectors of the Hbar singles block) should allow you to specify other initial-guess states. We seem not to have a test case for this, and if anyone has used it recently, it may be someone in the Crawford group. I read a paper on seeking conical intersections and was very interested, but never got around to implementing it.