Instability with large max_opt_steps
Opened this issue · 1 comments
I am using pyGSM with GFN2-xTB as level of theory in order to model a Claisen rearrangement. I noticed a weird behavior regarding the max_opt_steps and I'm wondering if this is expected or it is a bug.
In particular, when setting max_opt_steps to values "too big" (e.g. 50 or 100), the string optimization makes wild steps until the xTB calculation crashes due to chemically unsound systems. For example, the string optimization might lead to atoms being very close to each other or completely dissociated from the molecule (see example attached). Using a value of 10 or lower does not result in this issue.
I tried changing parameters like the number of nodes or reducing DMAX to 0.01, but the same behavior occurs. This seems to be system-dependent, as I used large max_opt_steps values for other systems without issue (it actually made the overall calculations faster and gave better results).
I am using pyGSM from commit d575ed7, since the later versions crash instantly. I'm also using the latest conda releases of xtb and xtb-python.
Input and output files: GSM_max_opt_steps.zip
Hi Raphael, sorry I just saw this now.