Insights into cyclopropanation reactions modeled with GFN2-xTB

[RaphaelRobidas]

This is a long shot, but I figured you might have some insight to share. We have been using GFN2-xTB to study gold(I)-catalyzed reactions and xtb works quite well. We however very often obtain strange transition states (optimized with either ORCA or Pysisyphus) for cyclopropanation steps. While we would expect the concerted formation of the two C-C bonds, GFN2-xTB instead yields two asynchronous transition states (see one of them below). This seems to be because of the PES and not the optimization method, as even the GFN2-xTB reoptimization of a DFT transition states leads to this kind of structure.

GFN2-xTB/ALPB(CH2Cl2) transition state

M06/Def2-SVP/CPCM(CH2Cl2) transition state

Oddly, some GFN2-xTB transition states are somewhat more synchronous:

XYZ files:
xtb_TS_B.txt
xtb_TS.txt
DFT_TS.txt

I imagine this could be a result of the minimal basis set which is not large enough to represent the relatively long distance bond formation. Perhaps you have already observed this behavior and have a better explanation? I would be very interested in understanding why this happens and if there is a way to "fix" it. It would be very helpful for our research, as GFN2-xTB works beautifully otherwise. Thanks!