Choose a Calculator

Every calculation setup except XRD contains a Calculator section. Select the calculator there before editing operation-specific controls. The selection determines which properties are available, which assets must be present, and whether charge, spin, SCF, and solvation settings affect the run.

Physical approximations and validation limits: Choosing a Method. This page follows the visible interface.

Open a calculation setup

Display the structure to be calculated. Press Calc in the left rail and then press the required operation, such as Single Point or Optimization. The same operations are available under the top Calculate menu.

The setup dialog opens with Calculator above the operation-specific sections. Its subtitle identifies the active structure. If the dialog displays Calculation needs one cell, more than one incompatible visible periodic layer is contributing a cell; hide the unrelated layer or select the intended structure before continuing.

The current new-dialog default is g-xTB with its native DFT-revD4 dispersion. It is a balanced starting point, not a recommendation for every calculation. Select and record the method required by the chemistry and observable.

Select Model

Choose MLIP, DFT, or Tight binding in Theory family, then select the method in Model / level. This two-step control keeps the initial choice short without hiding any registered level.

Selection does not assert that a model is accurate for the displayed chemistry.

The lines beneath Model are a static readiness summary of the required module and weight asset — not calculation progress, and not a scientific setting.

Read unavailable property states

Property controls respond immediately to Model. In Single Point Setup, an MLIP leaves Energy, Forces, and periodic Stress available; Charges, DOS, PDOS, Charge density, NCI cubes, and WFN are disabled with a reason such as “MLIP only returns energy, forces, and stress.” Select GFN2 when those electronic outputs are required.

A disabled property is a calculator capability boundary, not an unchecked request that can be restored later. Changing the model may also normalize the requested property set, so re-read the output controls after every model change.

Set charge and unpaired electrons

Total charge and Unpaired electrons remain visible in the calculator section. They are operative electronic-state inputs for GFN2. Enter the signed total charge of the complete calculation structure and the intended number of unpaired electrons before pressing Start.

The same fields are visible with an MLIP selected, but the current MLIP inputs are structure-only and do not consume a molecular charge or spin channel. Their presence in the shared form does not indicate MLIP support for redox or spin-state changes. If the scientific comparison changes charge or spin, use an electronic-structure method that represents the required state.

Configure tight-binding dispersion and SCF

Selecting g-xTB displays None, DFT-D3, and DFT-revD4 in the separate Dispersion selector. DFT-revD4 is the native model and default. The DFT-D3 alternative uses published ωB97M D3(BJ) parameters and records that provenance because no separate g-xTB-specific D3 fit is published. Selecting GFN2-xTB instead offers None and its native D4 model.

For an initial calculation we recommend keeping the defaults unless the system or protocol supplies a reason to change them. Increasing SCF max cycles does not repair an incorrect charge, spin state, or geometry. If SCF fails, inspect those physical inputs before weakening convergence.

Expand Temperature annealing to edit Temperature schedule K, Policy, Start temperature K, Stages, and Stage cycles. Expand Advanced SCF to select Initial guess and SCF mixer. Allow unconverged SCF requests a best-effort electronic result; enable it only for explicitly labelled diagnostic work. An unconverged state is not interchangeable with a converged energy in a comparison.

Configure implicit solvation

With GFN2 selected, check Implicit solvation to display Solvation model, Solvent, Electrostatic model, Born kernel, and the available CDS correction, CM5 correction, and Shift correction controls. When Shift correction is checked, Shift state appears.

Solvation changes the energy model and its reference convention; keep it fixed across every term in one comparison (see Preserve comparability below).

Inspect and prepare assets

Open File → Model Manager when the required module or weight is not stored. Press Download for an absent asset; a partial download exposes a resume action, and a cached asset exposes its management action. Closing Model Manager does not select the model — return to the setup dialog and confirm Model again.

Exact identifiers, capabilities, and assets: Calculator Reference.

After Start is pressed, actual asset-transfer and calculation progress appear in the calculation workspace and Console, not the pre-run byte summary under Model.

Preserve comparability

Before starting a series, keep the following fixed unless the change is the subject of the experiment — same calculator, same settings, same charge/spin state:

• the exact Model entry and model asset version;
• Total charge, Unpaired electrons, and dispersion choice for GFN2;
• SCF, solvation, and temperature/annealing settings;
• the requested property set and the operation’s geometry/cell policy.

Full treatment, including why GFN2 and MLIP energies cannot be subtracted from each other: Comparable Energies and Thermodynamic Cycles.

Confirm the selection

Before pressing Start, re-check the model description, required-asset lines, charge/spin fields, enabled output properties, and operation-specific controls. After completion, open result.json and confirm the recorded level of theory, model/runtime, convergence state, and requested outputs. Cache availability means the files can run; target-domain accuracy still requires a matched validation calculation.

Irregular states