openmaterials.ai / codes / xtb
The free xtb program runs the GFN family, mostly semiempirical tight-binding methods with element-specific parameters up to radon (Z = 86), fitted on gas-phase reference data.
Second-order density fluctuations enter as damped interactions of cumulative atomic multipoles, with charge-dependent D4 dispersion solved self-consistently. No hydrogen or halogen bond corrections (Bannwarth 2019).
GFN1-xTB and GFN2-xTB with self-consistent charges, non-iterative GFN0-xTB, and the GFN-FF force field. GFN0 is 2 to 20 times faster than GFN2 (Bannwarth 2020).
Global and element-specific parameters only, available for all elements up to Z = 86, with an f-in-core treatment for lanthanides (Bannwarth 2019).
Geometry optimization, harmonic frequencies and thermodynamic functions run from the geometry file alone. ANCopt uses a Lindh model Hessian (Bannwarth 2020).
A generalized Born model with surface area terms is coupled to all GFN methods, avoiding the integration grid of a polarizable continuum (Bannwarth 2020).
crest runs parallel xtb instances for conformer searches and enso re-ranks ensembles. ASE and mGSM call xtb. AMS, CP2K, DFTB+ and ORCA implement GFN (Bannwarth 2020).
A geometry file. Set charge with --chrg and unpaired electrons with --uhf. No input file.
GFN2-xTB by default, GFN1, GFN0 or GFN-FF for larger systems, GBSA for solvent.
xtb <geom> --ohess optimizes with ANCopt and computes the Hessian in one run.
The THERMO block gives enthalpy, entropy and heat capacity at the chosen temperature.
crest drives biased MD runs with xtb and re-ranks the ensemble.