Geometric optimization vs L-BFGS criteria

[Andreas Döll]

Hello,

I have the exact same script, but with different molecules and I am doing geometric optimization. One time it says Geometric optimization completed as usual, but for the other xyz file it says before reaching all convergence criteria at an optimization step (Max grad not converged) 

 ***********************************************
 * Specific L-BFGS convergence criteria
 * WANTED_PROJ_GRADIENT and WANTED_REL_F_ERROR
 * satisfied .... run CONVERGED!
 ***********************************************

and it stops the optimization.

I found https://groups.google.com/g/cp2k/c/ERSHqK6yAHg/m/iQEhlX57AAAJ where "Tomas Kuehne" explains that LBFGS optimizer is working as intended and does not use RMS/MAX values, but what can I change to reach my geo opt criterion before the optimizer terminates? Just making other geo opt criteria tighter should lead to the desired result? Anything else?

Cheers, 

Andreas

[Krack, Matthias]

Hi Andreas

You can try to add  TRUST_RADIUS 0.1 in the &LBFGS section or increase MAX_H_RANK to 10 or more.

If the system is not too large, you can also resort to BFGS which is more robust.

HTH

Matthias

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/f1495f1c-4534-4475-9cb3-07256b928293n%40googlegroups.com.