search

grimme-lab / xtb

Semiempirical Extended Tight-Binding Program Package
https://xtb-docs.readthedocs.io/
GNU Lesser General Public License v3.0
587 stars 146 forks source link

GFNFF Geometry Optimization Fails and Shows Erratic Behavior for Sulfonamide #519

Closed robpollice closed 3 years ago

robpollice commented 3 years ago

I tried optimizing the geometry of the attached structure using GFN-FF, which is optimized at the GFN2-xTB level and looks reasonable to me. The optimization trajectory shows very erratic behavior (rearranges one N-H bond and inserts it into a C-N bond). I would like to emphasize that optimizing the structure with GFN2-xTB is not a prerequisite for this behavior as it also happens without it at various alternative, but in my view reasonable, starting structures. However, I wanted to make sure that the starting structure has no obvious problem I am missing. gfnff.tar.gz

awvwgk commented 3 years ago

Thanks for reporting.

@sespic Any idea what could be the issue here?

sespic commented 3 years ago

Also big thanks for reporting from my side. It's indeed an issue with the GFN-FF implementation in xtb version 6.4.1. There are two things you can do. A) use version 6.4.0, this works fine. B) set the tdist_thr in line 633 in src/gfnff/gfnff_param.f90 from 12 to 16. This also works. I will fix this issue soon. Besides that, your structure looks fine.

awvwgk commented 3 years ago

I wonder whether it would be worth to revert f63f532eb023642304911129b8a6c99c78d82aaf to restore the original behaviour. While the original 1.0+12 pair distance for sure does look like an oversight, it seems to be more robust in practice than using a huge distance for the pairs and effectively removing the interaction.

I fear that increasing the cutoff to 16 Bohr will only shift the problem to the next larger system at a significant increase of computational cost in the topology setup.