Skip to content

COMPARISON.md

Latest commit

 

History

History
131 lines (113 loc) · 7.77 KB

COMPARISON.md

File metadata and controls

131 lines (113 loc) · 7.77 KB
Raw

Feature & Performance Comparison: GSAS-II · Jana2020 · FullProf

A factual capability comparison of the three mature refinement packages, and an honest assessment of where this browser-native workbench currently stands against them. Cells are Yes / Partial / No with a short qualifier; unclear means it could not be confirmed from primary documentation.

Sources: Toby & Von Dreele 2013 J. Appl. Cryst. 46, 544 (GSAS-II); GSAS-II docs (gsas-ii.readthedocs.io) & k-SUBGROUPSMAG paper (PMC11457102); Jana2020 (jana.fzu.cz), Petříček et al. 2023 Z. Kristallogr. & "Analysis of magnetic structures in JANA2020" (PMC11457105); Rodríguez-Carvajal 1993 Physica B 192, 55 (FullProf), ILL FullProf pages & "Magnetic structure determination and refinement using FullProf" 2025 (PMC12147938).

1. Capability matrix

Data types

Feature GSAS-II Jana2020 FullProf
X-ray powder Yes Yes Yes
CW neutron Yes Yes Yes (core)
TOF neutron Yes (full profiles) Yes (imports) Yes (profiles)
Single crystal Yes Yes (strong) Yes (neutron focus)
Electron diffraction No/unclear Yes (dynamical, unique) No
Total scattering / PDF Yes (built-in) No No

Profile & corrections

Feature GSAS-II Jana2020 FullProf
Gaussian / Lorentzian / pseudo-Voigt Yes Yes Yes
Thompson-Cox-Hastings pV Yes Partial (equivalent) Yes
TOF profiles (α/β + σ) Yes Yes Yes
Background (Chebyshev / Fourier / manual) Yes (all) Yes (poly + manual) Yes (poly + Fourier + interp)
Absorption / extinction Yes / Yes Yes / Yes Yes / Yes
Preferred orientation (March-Dollase) Yes Yes Yes (multi-axial)
Preferred orientation (spherical harmonics) Yes Yes Yes

Symmetry

Feature GSAS-II Jana2020 FullProf
All 230 space groups Yes Yes Yes
Magnetic (Shubnikov/BNS) Yes (1421 BNS) Yes (1651, BNS+OG) Yes (BNS/OG/UNI)
Superspace (3+d) incommensurate Partial (3+1 nuclear) Yes (reference impl.) Yes (via k-vectors)
Wyckoff / absences / symmetry constraints Yes Yes Yes

Refinement engine

Feature GSAS-II Jana2020 FullProf
Least-squares Full-Hessian LM (+SVD) Gauss-Newton LS Marquardt LS
Rigid bodies / restraints Yes / Yes Yes / Yes Yes / Yes
Multi-phase Yes Yes Yes
Multi-histogram / joint Yes Yes Yes
Sequential / parametric Yes (signature) Partial (cyclic) Partial
Simulated annealing (solution) Yes via Superflip Yes
Error / covariance Yes Yes Yes

Intensity extraction & magnetism

Feature GSAS-II Jana2020 FullProf
Le Bail Yes Yes Yes (profile matching)
Pawley Yes unclear unclear
Representation analysis via Bilbao/ISODISTORT Yes (built-in) Yes (BasIreps)
Commensurate k Yes Yes Yes
Incommensurate k (helical/conical/SDW) No Yes Yes
Magnetic form factors / mCIF Yes / Yes Yes / Yes Yes / Yes

Practical

Feature GSAS-II Jana2020 FullProf
Scripting API Yes (Python: GSASIIscriptable) Partial (silent mode) Partial (CrysFML/pyCrysFML)
Platforms Win/Mac/Linux Windows (Wine elsewhere) Win/Mac/Linux
Web / browser No No No
GUI Full (wxPython) Full (OpenGL) WinPLOTR/EdPCR/Studio

2. Strengths & niche

  • GSAS-II — modern Python-native all-rounder; best at automated multi-dataset Rietveld (sequential/parametric via a scripting API), combined X-ray+neutron, and total scattering/PDF. Commensurate magnetism via Bilbao; no incommensurate magnetic.
  • Jana2020 — the reference for aperiodic/(3+d) superspace, incommensurate and composite structures, incommensurate magnetic superspace, and dynamical electron-diffraction refinement.
  • FullProf — the community standard for magnetic structures from neutron powder data: BasIreps representation analysis + k-vector formalism for commensurate/incommensurate/helical/conical order, plus Shubnikov/mCIF.

None of the three has a browser build — that is the niche this project targets.

3. Where this workbench stands

This is an early browser-native workbench. It does not compete on breadth, but it implements a correct, tested core of the shared fundamentals — and a few workflows the mature tools split across several programs (magnetic candidate generation + comparison; the symmetry-mode distortion workflow) as single in-app flows.

Area This app Mature packages still add
Data types CW + TOF X-ray/neutron powder; single-crystal F² (.hkl/.fcf/.int); PDF G(r) (.gr/.sq/.fq) electron diffraction
Profile Gaussian / pseudo-Voigt / TCH + FCJ asymmetry; TOF back-to-back exponential; Chebyshev / Fourier / power backgrounds; March-Dollase PO; displacement / transparency / absorption / roughness corrections; Stephens microstrain, uniaxial size spherical-harmonic texture, Ikeda–Carpenter TOF shape
Symmetry all 230 built-in tables + CIF/mCIF operations; Wyckoff/site constraints; absences; magnetic subgroup enumeration with BNS/OG labels; isotropy + t-subgroup (Bärnighausen) lattices superspace (3+d), full 1651 magnetic tables incl. type IV
Engine Levenberg-Marquardt (SVD, esds, correlations), bounds/ties/restraints, staged controller, multi-start, worker pool, opt-in WebGPU full-Hessian options, rigid bodies, restraint libraries
Uncertainty LM esds + correlations; Bayesian posterior sampling (affine-invariant ensemble MCMC on the same problem seam — prototype, PDF-first; split-R̂/ESS/credible intervals; posterior-vs-esd ratio ≈ 1 validated on the Ni golden) — (none of the three ships posterior sampling; their uncertainties are the least-squares covariance)
Multi-phase / multi-dataset Yes (powder 2-phase; PDF multi-phase + multi-dataset co-refinement; sequential Rietveld + PDF) larger joint-histogram breadth
Intensity extraction Le Bail + Pawley
Magnetism mCIF in/out, single-crystal + powder moment refinement, k = 0 and k ≠ 0 commensurate, k-search, subgroup candidates + comparison representation analysis (built-in), incommensurate/helical
Validation GSAS-II golden values + real-data benchmarks; PDFfit2/PDFgui and diffpy.mpdf goldens decades of community validation
Platform / API Static web app, no install; 33 MCP agent tools desktop; Python scripting ecosystems

Honest gaps (the road to maturity): incommensurate / multi-k and helical/conical magnetism; built-in representation analysis for the magnetic workflow; spherical-harmonic texture; powder extinction and anomalous dispersion; twinning and absolute structure (single crystal); rigid bodies and a restraints library; Pawley extraction. These are laid out in ROADMAP.md and LIMITATIONS.md.

4. Performance notes

The mature tools are compiled (Fortran/C cores; GSAS-II wraps C/Fortran under NumPy/SciPy) with full-matrix solvers; their scaling story is automation and multi-dataset throughput rather than single-fit speed, and none publishes formal benchmarks. This app runs Levenberg-Marquardt in Web Workers in pure TypeScript, with reflection-windowed evaluation, dependency caching, a parallel-Jacobian worker pool (bit-identical to the serial driver), analytic columns for a validated subset of parameters (on the PDF path a fused single-pass ∂G/∂p kernel, measured 2.3× faster on the Ni golden), and opt-in WebGPU f32 kernels for the structure-factor sums — adequate for the pattern sizes shown here; WebAssembly is deliberately skipped (see ARCHITECTURE.md).