History#

2026.11.11 – Settings fingerprints for xTB’s model-chemistry calculations
  • Calculations through an xTB Model Chemistry (e.g. xTB:SQM@GFN2-xTB) now carry a fingerprint of their settings – the program and its version, the method and the accuracy – like the other codes’ calculations. Steps that reuse results, such as Thermochemistry with its fragment store, can then tell which stored xTB results match the calculation they need.

  • Needs seamm-mdi 2026.11.11 or later for the fingerprints; with an older seamm-mdi, xTB calculations work as before, without them.

2026.11.5 – Tight convergence on request for finite differences
  • Steps that take finite differences of xTB’s gradients over its MDI engine, such as the new Vibrational Analysis step, can now ask for tight convergence: every calculation then runs at accuracy 1e-4. The engine starts each SCF from the previous structure’s, and at the default accuracy that left noise depending on the previous structure (water’s anharmonic force constants: 92 cm^-1 of noise, 0.12 with the tighter accuracy).

  • The MDI engine takes an --accuracy option.

2026.10.20 – Bugfix: an SCF that did not converge ended the MDI engine
  • In the tblite MDI engine (used e.g. by the Dimer Builder), one structure whose SCF did not converge ended the engine, and with it the whole step. A failed SCF is now retried from a fresh start, then with more iterations and stronger damping, then from another starting guess – none of which changes the converged energy. If none converges, the engine returns NaN for that structure and carries on, so the step can leave the point out. For Li+ with ethylene carbonate the failures were at large separations, 7-9 Å, and the retries converged them.

2026.10.9 – Bugfix: GFN2-xTB through tblite crashed
  • Any GFN2-xTB calculation through tblite (the MDI engine, as used by the Dimer Builder) crashed with a segmentation fault: tblite 0.6 and 0.7 need dftd4 4.2, and that combination fails in dftd4’s polarizabilities. The environment now pins tblite-python 0.5.0 (dftd4 3.7), which works; with it the xtb program is 6.4.1. GFN1-xTB was not affected.

  • Internal: requires Python 3.12; CI installs with uv from PyPI (no conda test environment); the unused Docker installation option was removed.

2026.5.2: Plug-in created using the SEAMM plug-in cookiecutter.

2026.5.12: Initial working version
  • Support for Energy, Optimization, and Frequencies

  • GFN0-xTB, GFN1-xTB, GFN2-xTB (the default), and GFN-FF supported

  • ALPB, GBSA, or CPCM-X implicit solvation