Closed mkrompiec closed 4 years ago
Have you checked the time step for the DFT calculation? xTB methods can effort an incredibly large time step of 5 fs, which might be an order of magnitude to large for DFT.
You are right, it works fine provided that the time step is low. It seems the crash on my side was caused by both too high a time step and miscofiguration of TM.
Is your feature request related to a problem? Please describe. xtb's metadynamics implementation is unique, and it would be great to be able to use it with other methods than dftb
Describe the solution you'd like I can see that (at least in some situations) xtb can be used as a driver for Turbomole or Orca. However, attempts to run MTD with Turbomole (xtb --tm) lead to a catastrophe: although Turbomole control files are created correctly and ridft/rdgrad are called, the geometry seems to wander off uncontrollably. Any chance this could be made to work? Alternatively, would it be possible to (re)implement xtb’s RMSD-based MTD in ASE?