geo_opt_cp2k

flo fsch... at pci.uzh.ch
Mon Nov 12 16:29:07 UTC 2012


Hi,

Could you please post an input file. I was recently working on the BFGS 
optimizer trying to improve convergence. Unfortunately there seem to be 
some stability issues for some systems. 
If you experience problems there are several things to try.
Try the latest version in trunk. I recently commited some adjustments which 
hopefully do the job. If this is the case, please let me know else try one 
of the following bits:
TRUST_RADIUS 0.1 or smaller (this will prevent from too big steps, which 
can be problematic in the very first steps)
USE_RAT_FUN_OPT TRUE (should help stabilizing the optimization)
USE_MODEL_HESSIAN FALSE (will switch back to the unit matrix guess)

All three together will restore the previous BFGS. 
Anyway, please post your structure so that I can add it to my test set.

Cheers
Flo

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20121112/d761a1c4/attachment.htm>


More information about the CP2K-user mailing list