GFNFF geometry optimization

[gnarw]

Describe the bug Using geometry optimization on a 14-mer double stranded DNA using gfnff I don't get the same structure for repeated calculations. I ran the same calculation 10 times and for 4 structure out of 10 the root mean square deviation between structures was about 1.5 - 1.7 angstroms. The other 6 structures were practically identical. xtb version 6.3.3 (5b13467)

To Reproduce xtb --gfnff inputStructure.pdb --gbsa water --opt extreme

Expected behaviour I expect that all 10 geometry optimized structures would be identical I'm I wrong to expect this ?

[awvwgk]

@sespic do you have an idea what might be causing this issue?

[sespic]

This behavior is expected as geometry optimizations are not completely deterministic, especially for large systems with many degrees of freedom. I would also strongly discourage using extreme thresholds. For proteins in general loose optimization thresholds are recommended in combination with GBSA/ALPB.

[gnarw]

Thanks for the answer. Good to have this behaviour confirmed.