Hello all,
I am attempting some geometry optimizations using eom-EE-CCSD of the A state of an open-shell doublet and many of never converge to better gradients or steps than the beginning values, even if I provide an initial hessian using a FREQ calculation.
Is there a way for me to specify an option within the OPT calculation to re-calculate the semi-analytical (b/c only analytic gradients are available) hessian every 5 steps (for example)?
Thanks,
Benj
Hi, Benj. I’m not sure about the Hessian, although we could probably update the code to allow this option in special cases. I suspect that you’re seeing other issues in your optimization, though. We ran into state-swapping issues with EOM recently (see my root-following post). If you plot the electronic state energies at each step, do you see other nearby states?
As of version 5.3, Q-Chem cannot recompute exact hessian in the middle of geometry optimization. It is only possible to begin geometry optimization with exact hessian.
That said, the described behavior suggests that EOM is switching the order of states, which can be checked by comparing the character of the target state in the beginning and end of geometry optimization. The typical workaround in this situation is to restart geometry optimization at the point after the states switch order with an updated value of CC_STATE_TO_OPT=N.
Link to the EOM root following thread: Root-following algorithm in EOM? - Questions - Q-Chem Talk (q-chem.com)
Thank you both so much for the diagnosis. You are correct, the A state is not too far from the X state every time the optimizations run into issues (the EOMEE transition 1/A energies are +0.3 eV to -0.2 eV). I suppose this means I am out of luck with the system approaching a conical intersection between the two states.
I would certainly love to see the option to evaluate hessians throughout a geometry optimization in the future. Many of the open-shell doublet systems I study have a hard time converging without extra help.