Molcas Forum

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 86

SA-RASSCF molecular dymanics

Dear Community,

I am running some molecular dynamics with RASSCF.
With a CIROOT = 1 1 1, everything works as expected.
But, I want to do an excited state dynamics.

- CIROOT = 1 N; N does not work: states intermix and it does not converge.
- CIROOT = N N 1 works as expected, but with MDRLXR=N Dynamix starts using numerical hessian (calls MLCR), which is very slow.

This led me to a hack: after the main calculation with CIROOT = N N 1, I run another CIONLY calculation with CIROOT = 1 N; N. It does not optimize the orbitals, and makes a single-state (non-SA) wavefunction.
That does work, that does not introduce much of energy difference (when tested for ground state), but I do not like the results.

Is there another way?

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: SA-RASSCF molecular dymanics

If you run with CIONLY, the orbitals are not optimized, and even if analytical gradients/hessians run without crashing, they'll probably be wrong since the assume the orbitals are optimized (i.e., that the energy is at a minimum with respect to orbital changes).

A state-specific calculation for an excited state will probably not converge, especially in a dynamics where the geometry changes: You may be lucky, or be able to find a solution at some particular geometry, but it will most likely fail at some point.

A state-averaged calculation is much easier to converged, but since the orbitals are not optimized for any specfic root, an additional term is needed for the gradients, which is what MCLR does (it's not for hessians). If it's slow, there's not much you can do. Use RICD if you're not already using it, reduce the active space and/or basis set...

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 86

Re: SA-RASSCF molecular dymanics

Thanks.

Is there maybe a way to converge a RASSCF to a local minimum? I.e. say I prepare a SA WF that is kinda close to a single-root excited state WF (with fine-tuned weights, lets not get into that). I then run another RASSCF calculation with orbital optimization but somehow instruct it to fall to the nearest minimum, local, not global, hopefully resembling the SA WF of the excited state.

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 86

Re: SA-RASSCF molecular dymanics

Also. If I understand correctly, the problem with individual excited state calculation is such that, as it is being optimized, it falls below the states that used to be lower, and thus, technically, becomes not the state of interest.

Can this be solved by a simple shift in energy? The shift will make the energy appear higher than it actually is, and (fingers crossed) the state of interest will keep its position in the list of states (i.e. root 3 will still be root 3 even if it's true energy fell below that of root 2).

Last edited by andrewshyichuk (2024-09-24 15:30:17)