I am trying to run geometry optimization and MD for Li3Sb strucutre but GFN2-xTB doesn't converge. I tweaked max iterations, fmax, scf accuracy and mixer damping etc. values as much as possible but no success at all. Same parameters work for GFN1-xTB.
I am using TBLite with ASE. Tried BFGS and FIRE optimizers until now.
Using Langevin for MD calculation.
Is it possible to somehow make it work?
You can see the .xyz file attached.
Here are the parameters which work for GFN1-xTB:
--method GFN1-xTB --task opt --structure data/li3sb_2x_vacx2.xyz \ --pbc --cell 13.118 13.118 13.118 --fmax 0.15 --steps 5000 \ --maxstep 0.05 --mixer_damping 1 --scf_accuracy 1 --max_iterations 5000 \ --optimizer FIRE --verbosity 1 \ # --charge 0 --multiplicity 1 \
Greetings Orhan Kıbrıslı
I also have had problems with XTB, in my case, for calculations with Ti atoms. The structure would converge for unrealistic bond lenghts that would render the optimization impossible (something already known in the community for early transition metals). I don't know if there are similar problems for Li and Sb, but it might be possible.
I had more success using semi-empirical methods based on the Neglect of diatomic differential overlap (NDDO) for systems were DFT or Post-HF were not options.
Hope it helps.
- Badges
- Science topic