About Powder 6

Commands and Results have independent tabs. Each coloured Datasets tab contains that dataset’s values and refine/fixed controls, including background X/Y, resolution and phase contributions. Bounds, sharing and links are edited there too. Phases contains shared definitions and starting models; Method chooses the next calculation; Settings contains numerical conditions. Results has separate Le Bail, WPPF and Rietveld tabs showing one selected run. Choosing a previous run preserves the working setup; Use as start restores it explicitly. Press FIT or F9 from any view. Use ⇄ Panels to swap the two panels.

The dataset checkboxes define both the fit and plot selection. Choose All datasets, or a range with Shift+click, for large batches. Click a dataset name to inspect it without changing the checked datasets. Shift-click selects a range; Ctrl-click toggles inclusion. Plots offers Overlay, Overview and Inspect one dataset. Anchors, Backgnd, Phases, Refl and Subtract have independent display controls.

Background… chooses one background type for all datasets. Chebyshev points are estimated automatically; Fit to points fits and holds the polynomial coefficients. Their Use checkboxes select point targets, independently of the polynomial order. Anchors remains visible for both background types. With spline backgrounds, Ctrl+click places an anchor at both cursor coordinates; right-drag moves it, Escape cancels the move. Edit exact X (2θ), Y (intensity) and the Y refinement checkbox in Datasets → dataset tab → Background. Magenta filled anchors are refinable; grey hollow anchors are fixed. A yellow line identifies the anchor being edited. Ctrl+Shift+drag edits exclusions.

Each method retains its own setup and current result. History stores completed fits by method, with grouped individual batches and complete global fits, and every charge-flipping and Wyckoff solution. For the chosen run it lists everything: settings, phases, statistics, every parameter with its uncertainty, instrument, and for solutions their sites. Set restores a record; Keep protects it from automatic pruning; Delete asks first. Save History… and the report buttons write the records, with all their results, as JSON or text. The default automatic budget is 100 MB, with 20 recent unkept records for each method and, separately, for charge flipping and for the Wyckoff search; current and kept results are protected. Ctrl+Z undoes recent editing. Log contains detailed messages.

Save a compressed .project file (projects saved as .powder6 or .archive open too) to retain observations, model definitions, method states, groups, history and structure candidates. Observations, the WPPF precision block and each structure solution are stored once. Compact history and CF density-map inclusion are optional. A saved project is the state the page compares against: it asks before leaving only when something changed after it. Open the file later using File or drag and drop; this is how History is read back.

Commands → Structure → Space group contains Test for the systematic absences of powder intensities and ADDSYM for an atomic model. ADDSYM’s Find higher symmetry… calls the public, locally bundled Moyo WebAssembly engine (0.21.0, MIT license). ADDSYM is the interface label; the engine is Moyo rather than PLATON’s ADDSYM implementation. Choose either the starting model or a dataset’s Rietveld coordinates and cell, including its saved Rietveld model while using Le Bail or WPPF. The report gives the returned group and atomic mismatches; conversion is previewed with the selected model. ADDSYM works without a WPPF fit. See atomic-model symmetry and the bundled Moyo code and license.

Convert… next to the space group of a phase describes the structure in another space group or setting: the cell, the atomic positions and the values of every dataset are transformed with it, sites split for a subgroup and merge for a supergroup (only when the structure has that symmetry). The transformation is found automatically or entered as new axes and origin.

Instrument resolution belongs to explicitly selected datasets. Sample presets target a phase and datasets. CF and Wyckoff candidates can be copied to an existing or new phase, with affected datasets shown before transfer. The source solution remains in History.

Datasets: Constraints, Restraints and Links. Constraints impose exact atomic equivalences, sums and holds; sum refine checkboxes act on the whole equation. Restraints apply soft distance and angle rules whenever their textbox contains rules. Helper collects proposals in Geometry, Molecule and Suggest tabs: Copy puts them on the clipboard, and only pasting into the dataset textbox applies them. Duplicate pasted rules are rejected. Links impose target = multiplier × source + offset across chosen parameters and datasets; the source refine flag controls refinement, and either value can be edited within limits. Limits and applicable global sharing controls are in Parameters, immediately before Links. Add the first parameter link from Parameters; Links appears when there are existing relationships to manage. The last Copy accordion copies values or flags once; it does not create a link. See the complete guide and examples.

Working with it

Dataset setup is in Commands → Datasets. The expandable Datasets for fitting and plotting list selects patterns; coloured tabs choose the dataset to inspect and edit. Hundreds of independent fits are queued with only the active worker jobs prepared in memory. Busy calculations lock editing; Stop or Escape cancels a fit or structure search.

1. File, read powder patterns from the Datasets tab, or drop them on the window (Bruker, PANalytical, Rigaku, Philips, GSAS, CIF, two-column ASCII). New datasets appear at the top of the list. Set the wavelength, polarisation and anomalous scattering in Instrument (Datasets tab), the background in Background, the regions to leave out in Exclusions.
2. Read a CIF for each phase, or make a new phase from its space group and cell: without atoms it is fitted by Le Bail or WPPF.
3. In Datasets, use Phases… (or Refine → Phases of the datasets…) to define which phases each dataset contains.
4. In each dataset tab, set values and tick individual parameters or use the group checkboxes: background/scales, zero/lattice, Sample and Resolution; Rietveld also offers positions, B and occupancies. Sample selects phase broadening; Resolution selects instrument broadening independently. Press Fit (F9). Le Bail or WPPF first, to settle the cell and the peak shape, is the usual start. Global fits the selected patterns together: select A and C to refine what they share with those two only.
5. With a WPPF fit done, open Commands → Structure to extract intensities, test the space group and solve the structure. Inspect output in Results → Structure, export the atoms to the phase, switch Method to Rietveld and fit again.
6. Results shows the selected completed run: R factors, weight fractions, atoms, reflections and extracted intensities. Its run selector opens previous calculations; exports and Report use the selected run. History offers Set, Keep, Delete, reports and storage controls.

The model

y_calc(2θ) = background + Σ_phases Σ_hkl I_hkl · Ω(2θ − 2θ_hkl), Ω the peak shape of unit area, Kα2 included. Rietveld: I = S · m · Lp · P · |F|², with S the scale of the phase in the pattern, m the multiplicity, Lp the Lorentz-polarisation factor, P the March–Dollase factor and F the structure factor from the atoms. Le Bail: I is the share of the observed counts that the reflection gets in proportion to its calculated profile (Le Bail, Duroy and Fourquet 1988), partitioned again from flat at every cycle. WPPF: I is a free variable of the least-squares fit, non-negative, solved exactly at every step (variable projection, Golub and Pereyra 1973), so its standard uncertainty and its correlation with the cell and profile are those of a full Pawley refinement. The background is a Chebyshev series over the fitted range or a monotone cubic spline through anchors; points in excluded regions are calculated but not fitted. Choose Levenberg–Marquardt (LM), or parallel tempering (PT) followed by LM, in Settings → Numerical conditions. Both respect limits and weights 1/σ²; standard uncertainties come from the normal matrix scaled by χ². Weight fractions (Rietveld phases) follow Hill and Howard.

Mouse and keys

Requirements

Open index.html through a web server (it reads sg/all.json and scatters/), for example python -m http.server in its folder, then http://localhost:8000/index.html. Fits, charge flipping and the Wyckoff search run in Web Workers. Parallel-tempering fits and charge flipping can use WebGPU and have CPU fallbacks. The Wyckoff search requires WebGPU for sampling and fitness evaluation, with CPU control and final coordinate refinement; see GPU and CPU execution.

The methods (WPPF, Le Bail, exact overlaps, constraints, the space-group test, charge flipping, the Wyckoff search) are explained in the documentation, with where each one stops being trustworthy. Powder 6 descends from PowderV2 (Dragoe, N. (2001). J. Appl. Cryst. 34, 535).

Full documentation: Powder6_documentation.html (commands, file formats, parameters, structure solution, troubleshooting).

Charge flipping: Oszlányi and Sütő (2004). Space-group test: Markvardsen, David, Johnson and Shankland (2001). Peak shapes: Thompson, Cox and Hastings (1987).