xtb · CREST
User Guide

The Grimme group's semiempirical program xtb and conformer search tool CREST, followed through on a single molecule: optimisation and frequencies, molecular dynamics, metadynamics, conformer ensembles and DFT re-ranking.

Version xtb 6.7 · CREST 3 · Example alanine dipeptide · Run on NUS HPC

About this guide

xtb is a command-line program for the GFN family of tight-binding methods. It is two to three orders of magnitude faster than DFT and gets organic geometries reasonably right, so it is used to narrow down candidates before expensive calculations and to run long dynamics. CREST builds conformer ensembles on top of xtb by repeating metadynamics and optimisation.

This guide is about running xtb and CREST directly on the command line. The same GFN-xTB calculations can also be called from inside an ORCA input file.

xTB inside ORCA → chapter 16 of the ORCA guide

Writing ! XTB2 on the ORCA keyword line makes ORCA call the otool_xtb next to it, which is the same binary as the xtb in this guide. The ORCA/6.1.0-gompi-2023b-avx2 module on Vanda already ships it.

  • ORCA is the easier route when you pre-optimise with xTB and pass the geometry and Hessian to a DFT optimisation or OptTS, when you scan a path first with ! XTB NEB-TS, or when you want ORCA features such as MD or QM/MM with cheap xTB energies underneath.
  • Standalone xtb is the easier route for CREST conformer searches, RMSD metadynamics, long MD with SHAKE and heavy hydrogens, and batches of hundreds of structures driven from a shell script.

Both guides use the same example (alanine dipeptide in water), so the results can be compared.

Topics

Citations

  • GFN2-xTB: Bannwarth, Ehlert, Grimme, J. Chem. Theory Comput. 2019, 15, 1652.
  • The xtb program: Bannwarth et al., WIREs Comput. Mol. Sci. 2021, 11, e1493.
  • CREST: Pracht, Bohle, Grimme, Phys. Chem. Chem. Phys. 2020, 22, 7169.
  • CREST 3: Pracht et al., J. Chem. Phys. 2024, 160, 114110.
Further reading

The official documentation is at xtb-docs.readthedocs.io and crest-lab.github.io/crest-docs.