SEAMM Model Chemistry Plug-in

A SEAMM plug-in for defining the model chemistry for subsequent steps

• Free software: BSD-3-Clause

• Documentation: https://molssi-seamm.github.io/model_chemistry_step/index.html

• Code: https://github.com/molssi-seamm/model_chemistry_step

History

2026.10.16 – Bugfix: the committee naming in the user guide

• The user guide describes the new names of committees given by path in a job (xnn-step 2026.10.16): Job_012130/2/1/committee is xnn:MLFF@Job_012130-committee-5c20d7.

2026.10.14 – Bugfix: the Checkpoint help did not cover committees

• The help for an MLFF’s Checkpoint and the user guide now say that it may be a committee’s directory as well as a checkpoint file, that the model chosen above then only selects the program, and how the model chemistry is named. The guide also describes what a committee is and that it needs the Active Learning step.

2026.10.12.1 – Machine-learned forcefields: checkpoint and engine options

• For a machine-learned forcefield (xnn:MLFF@...) the step takes a Checkpoint – a path or variable, e.g. a checkpoint written during training – and the Engine options for the run, e.g. --eeq-reuse. The dialog shows them only for an MLFF.

2026.10.12 – Forcefields as model chemistries

• Forcefields (FF@oplsaa, ReaxFF@CHO_cho_2008), offered by the Forcefield plug-in, can be selected. They have no program, so the dialog shows no Program for them.

• A model chemistry’s files, such as a forcefield from your own data directory, are sent with a job submitted to a JobServer, and found there when it runs.

• ORCA’s Hartree-Fock and MP2 are now wavefunction (QC) methods: ORCA:QC@HF/... and ORCA:QC@MP2/.... The old names, ORCA:HF@... and ORCA:MP2@..., are still accepted with a warning, and shown under the new name in the dialog.

2026.10.3 – Bugfix: MDI engines ran the advertised basis instead of the chosen one

• A model chemistry whose basis is not one of the few a program advertises (e.g. ORCA:DFT@B3LYP/def2-TZVP) kept the advertised example’s basis for steps that drive the program as an MDI engine (the Energy step, the Dimer Builder’s energy contact search, Normal Mode Sampling, LAMMPS QM/MD), so ORCA ran def2-SVP whatever basis was chosen. The engine now gets the chosen basis.

2026.10.1 – A model chemistry no installed program offers is refused when built

• The check that a model chemistry is offered by an installed program (ignoring the basis set, which is a free choice) is now shared with SEAMM’s flowchart tools, so an unavailable model chemistry is refused when a flowchart is built or edited, not only when it runs.

• Documented in the user guide. Needs seamm 2026.10.1.

• Internal: CI now installs the package’s declared dependencies with uv rather than a conda test environment.

2026.7.31 – Bugfix: crash and stale picker on Type change / direct-entry edits

• Fixed a crash (“TypeError: object of type ‘NoneType’ has no len()”) when changing the Type selector (e.g. to DFT) in the Model Chemistry dialog.

• Editing the model-chemistry string in direct-entry mode and then switching back to the guided picker now correctly reflects the edit. A component that isn’t a currently offered choice (e.g. a method not advertised by any installed program) is now shown blank, instead of the picker silently substituting some other, unrelated valid choice – which made an edit look like it had been ignored.

2026.7.30 – $variables in the Type/Method/Basis selectors

• The Type and Method (e.g. DFT functional) selectors now accept typed text, including a ‘$variable’ or ‘=expression’, so a preceding Loop step can vary the model chemistry from one iteration to the next.

• A ‘$variable’ typed into any of Type, Method, Program, or Basis is now actually evaluated at run time. Previously only a variable that was the entire model-chemistry value was dereferenced; one embedded in a composed string (e.g. the Basis field alone) was passed through to downstream steps as the literal, unresolved text.

• Added a checkbox to enter the whole model-chemistry string directly as text, instead of using the Type/Method/Program/Basis picker – by far the simplest way to use a ‘$variable’, since it sidesteps the picker’s discovery/validation entirely. The dialog switches to this mode automatically when the stored value can’t be decomposed by the picker (e.g. it is itself a ‘$variable’).

2026.6.28 – Basis-set selection

• Added a basis-set field to the dialog for levels of theory that use a basis (HF, DFT, MP2, coupled cluster), using the shared Basis Set Exchange picker (choose elements on a periodic table, then a covering basis). It is hidden for methods that use no basis (semiempirical, forcefield, MLFF).

• The basis is the user’s free choice – any Basis Set Exchange basis (stored as ‘bse:NAME’), not just the few a program advertises. A level of theory is accepted as long as a program offers its type and method.

• The model-chemistry grammar accepts a ‘bse:’ prefix on the basis.