Fitbauer

Stable desktop application to load, fold, simulate and fit ⁵⁷Fe Mössbauer spectra.

Current stable version: v5.2.5
Launch: python fitbauer.py
Headless CLI fitting: mossbauer_fit_cli.py (discrete) · fit_bhf_distribution_cli.py (distributions)

Authors: Jorge Sánchez Marcos · Nieves Menéndez González
Department of Physical Chemistry · UAM

---

Contents

• Fitbauer and NORMOS — why it exists, and how it was validated against the original program
• Features
• Screenshots
• Quick start
• Fitting modes — discrete, distribution and relaxation
• Installation — requirements, installer script, manual setup, running, updating, troubleshooting
• Project structure
• Changelog
• License

---

Fitbauer and NORMOS

NORMOS (R. A. Brand, 1990-1994) is the program behind a large part of the published Mössbauer literature. It runs under DOS, it is proprietary and it is no longer maintained. Fitbauer exists so that this body of work —and those files— can keep being used from a current, open, cross-platform program.

That demands two things: producing the same numbers as NORMOS, and speaking its file format. Both have been verified against the original program.

Validated against NORMOS, with numbers

Fitbauer's physics has been checked on two independent benchmarks.

1. Synthetic benchmark. NORMOS generates a spectrum from known parameters, Fitbauer fits it, and the result is compared with the truth.

	spectra	fits	comparisons
Round-trip NORMOS → Fitbauer	411	~1,150	6,497

Median deviation from the true value:

Block	position	BHF	linewidth
First-order sextet/doublet	2·10⁻⁷ mm/s	6·10⁻⁶ T	8·10⁻⁷ mm/s
Full Hamiltonian	6·10⁻⁵ mm/s	1.4·10⁻³ T	5·10⁻⁴ mm/s
Texture and intensities	2·10⁻⁷ mm/s	6·10⁻⁶ T	2·10⁻⁵ mm/s
Multi-site and constraints (up to 10 sites)	7·10⁻⁶ mm/s	1·10⁻⁵ T	7·10⁻⁵ mm/s

In the first-order core the agreement is exact to numerical precision. The residual tail in the Hamiltonian case is the approximation made by NORMOS 1994 itself, not by Fitbauer.

2. Real-job benchmark. 407 distinct jobs fitted in NORMOS over the years —not synthetic: laboratory measurements, with their models and their results— reloaded into Fitbauer and refitted from NORMOS's own solution, on the data folded exactly as NORMOS folds them (whole channel pairs at ⌊PFP⌋ + 0.5) and with the same free parameters; recomputed this way, NORMOS's χ² matches the one in its .RES to about 1 %. A comparison only means something where NORMOS fitted well and both programs see the same spectrum, so only those jobs are kept:

• NORMOS's reduced χ² ≤ 2: 214 jobs where NORMOS did not fit well are discarded;
• Fitbauer reproduces NORMOS's curve at its solution (to 1 % of the peak), both see the same folded data (to 0.3 of the noise) and the .JOB still matches its .RES (19 jobs discarded, mostly with the folding point far from the centre, where NORMOS's folding cannot be reproduced);
• 28 jobs cannot be analysed (distribution jobs, missing spectrum or results).

On the remaining 146 jobs:

• Fitbauer reaches the same or a lower χ² in 146 of 146; median reduced χ²: NORMOS 1.40 · Fitbauer 1.38.

• Median parameter agreement:

  δ	ΔEQ	BHF	Γ	depth
  0.0006 mm/s	0.0015 mm/s	0.0034 T	0.0014 mm/s	0.0001

In units of NORMOS's own errors the differences have a median of 0.15–0.2σ and no bias; the few above 3σ are jobs where NORMOS had stopped short of the minimum (Fitbauer's χ² is lower, and NORMOS reaches it when started from Fitbauer's solution). The figures hardly change with a χ² cut of 1.5 or 3. Among the discarded jobs, two had NORMOS stopped in a local minimum: from the same starting values Fitbauer reaches a reduced χ² of about 1.5 instead of 20–30, and NORMOS confirms it when started there. The jobs are laboratory data and are not distributed; the synthetic bank and the reports are in validacion/informe/.

Opens and writes .JOB files

File ▸ NORMOS (.JOB)

• Import rebuilds the model and loads its spectrum: a .JOB names its files in the header, and Fitbauer looks for them next to it. Both NORMOS-SITE jobs (discrete sites) and NORMOS-DIST jobs (distributions) work; the latter are detected automatically and open the P(BHF)/P(ΔEQ) panel.
• Export writes the current model in NORMOS format. NORMOS has been verified to accept the file Fitbauer produces, reproducing the original theory with a difference of exactly zero.
• The delicate convention conversions —WID/W13 widths versus Γ₁, D13/D23 area ratios versus depth ratios, the global numbering of NDEX constraints— are handled automatically, and the importer warns about everything it could not carry over.

Fitbauer does not run NORMOS and does not ship it: it only speaks its text format, which is not proprietary.

What Fitbauer does that NORMOS does not

	NORMOS	Fitbauer
2D distributions	—	P(BHF,ΔEQ), P(IS,ΔEQ), P(BHF,IS), as a histogram or as bivariate Gaussians (VBF)
Regularizers	Tikhonov and maximum entropy	plus total variation (edge-preserving)
Choosing α	by hand	L-curve with automatic elbow and click-to-pick
P(IS)	singlet kernel	singlet, doublet or sextet kernel
Distribution shapes	histogram, Gaussian, binomial	plus multi-Gaussian VBF (Rancourt–Ping)
Errors	covariance matrix	plus Monte Carlo bootstrap and asymmetric profile-likelihood intervals
Minimum search	single start	multi-start and automatic global escalation (differential evolution)
Series of spectra	one file at a time	sequential batch fitting with warm start
Superparamagnetism	—	Néel–Arrhenius with lognormal size distribution and global multi-temperature fit
Voigt profile	approximate pseudo-Voigt	exact Voigt
Diagnostics	χ²	plus residual tests (lag-1, runs), correlations (also of P with Γ, δ…) and insufficient-grid warning
Outputs	text	Markdown/PDF reports, TSV with subspectra and complete JSON session
Headless use	—	CLI for discrete and distribution fits
Platform	DOS	Windows, macOS and Linux
Languages	English	8 languages, with integrated help
Licence	proprietary	Apache 2.0, open source

Several parts of the calculation are also measurably more accurate: Hamiltonian diagonalisation in double precision (Hermitian LAPACK versus general EISPACK in REAL*4), a source kernel integrated over each channel instead of sampled, and cubic interpolation when folding instead of truncating to a whole channel.

What it still does not do

Stated just as plainly. None of this blocks routine ⁵⁷Fe work, but it is worth knowing:

• ⁵⁷Fe only. NORMOS also handles ¹¹⁹Sn, ¹⁹⁷Au, ¹⁵¹Eu and ¹²¹Sb.
• Analytical Czjzek / Le Caër distributions. The histogram reproduces their shape, but there is no 2-3 parameter closed form to fit directly.
• External field in Ising relaxation (BEXT): population polarization is there; the line shift it causes is not.
• Emission spectra (source in the sample).
• Two overlapping distribution blocks, each with its own grid. Fitbauer handles one, plus sharp components.
• Octet (ΔmI = ±2): modelled as a sextet plus two singlets, not as a component of its own.
• Preprocessing: channel rebinning, adding several spectra or rescaling counts.
• When importing a distribution .JOB, the LAMDA smoothing parameter is not carried over: NORMOS's is absolute and Fitbauer's is dimensionless, so it has to be set with the L-curve.

The complete inventory, capability by capability and with the exact NORMOS source reference, is in validacion/informe/COVERAGE_NORMOS_EN.md; what remains, with what to touch and how to validate it, is in PENDING_NORMOS_EN.md.

---

Features

• Load local .ws5 and .adt files; download measurements and calibrations from the laboratory web database.
• Spectrum folding with a fractional folding point and cubic interpolation.
• Discrete fitting — singlets, doublets and sextets; Lorentzian/Voigt profiles; Poisson or Gaussian likelihood; robust loss functions; χ²/AIC/BIC.
• Multi-start fitting with configurable restarts and Monte Carlo bootstrap errors.
• Profile-likelihood confidence intervals with adaptive scan.
• Distribution fitting — P(BHF), P(ΔEQ), P(IS) and three 2D modes (P(BHF,ΔEQ), P(IS,ΔEQ), P(BHF,IS)); Hesse-Rübartsch regularization; L-curve α estimation; simultaneous sharp components.
• Advanced quadrupole: first-order, fixed Kündig, powder Kündig; sextet intensity texture.
• Physical constraint presets (3:2:1 powder, tied widths, linked δ/Γ across components).
• Relaxation models: phenomenological, Blume–Tjon two-state, Néel–Arrhenius with lognormal size distribution.
• Parameter limits fully configurable through the GUI (View → Parameter limits…).
• Interactive Matplotlib figure with semi-manual minimum editor.
• Batch fitting across a series of files with warm-start.
• Fit export as TSV with per-component subspectra and an informative header.
• Markdown/PDF reports: full report and condensed short report.
• Complete JSON session save/load; persistent settings across restarts.
• Update checking and one-click download from GitHub Releases.
• Interface and integrated help in English, Spanish, French, German, Portuguese, Russian, Japanese and Chinese.

---

Screenshots

The interface language is English by default.

Main window

Discrete fit (doublets)

Hyperfine-field distribution P(BHF)

Regularization L-curve

Short Markdown/PDF report

---

Quick start

python3 -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate
pip install -r requirements.txt
python fitbauer.py

Try the included sample data:

1. File → Open… → data_sample/magnetita_Fe3O4.adt
2. File → Load session… → data_sample/Fe3O4_session.json

Typical workflow:

Open spectrum → check folding/Vmax → choose model → fit
  → inspect residuals/areas → export session/report

---

Fitting modes

Discrete fit

Up to ten simultaneous components (singlet / doublet / sextet, plus the relaxation types below). Each component has its own type, parameters and fixed/free status. The Fit button optimises all free parameters; the status panel reports integrated areas, covariance or bootstrap (Monte Carlo) errors, and fit statistics.

For sextets the main parameters are:

Parameter	Meaning
δ (IS)	Isomer shift (mm/s)
ΔEQ	First-order quadrupole splitting (mm/s)
BHF	Hyperfine field (T)
Γ 1,6	FWHM of the outer lines (mm/s)
Γ 2,5 rel / Γ 3,4 rel	Relative widths of lines 2,5 and 3,4
Depth	Global absorption amplitude
int1 / int2	Relative intensities (≈ D13, D23); int3 fixed to 1

Distributions: P(BHF), P(ΔEQ), P(IS) and 2D

Models the spectrum as a sum of many sextets (or doublets) on a regular grid. The Hesse-Rübartsch-style optimisation minimises:

weighted spectral residual² + α · roughness(P)²

Use L-curve α to find a good compromise between residual and smoothness. Besides the histogram there are parametric shapes (Gaussian, VBF multi-Gaussian, NORMOS-style binomial, fixed and, for P(ΔEQ) of glasses and amorphous materials, pure or extended Czjzek); in the Gaussian shapes the width σ can be bounded with σ min / σ max. The Add active sharp components option mixes the distribution with discrete phases (e.g. a broad distribution + metallic Fe at BHF ≈ 33 T). Two-dimensional modes fit P(BHF, ΔEQ), P(IS, ΔEQ) and P(BHF, IS), as a regularized histogram or as bivariate Gaussians. The distribution plot shows a Monte Carlo 1σ band for the histogram and, when δ or ΔEQ are correlated with the grid variable (κ slopes), the δ(x)/ΔEQ(x) line with its uncertainty; each overlay can be switched off.

Relaxation models

Type	Description
Relajacion	Phenomenological blocked/superparamagnetic interpolation
BlumeTjon	Dynamic two-state ±BHF exchange
NeelSize	Néel–Arrhenius + lognormal size distribution (also as a global multi-temperature fit)

---

Installation

Fitbauer is a Python application — there is no compiled .exe. You run it with Python, either through the bundled installer script (recommended) or by setting up the environment by hand. The one-liner is in Quick start; this section covers every case. The standalone reference is INSTALL_EN.md.

1. Requirements

Python	3.11 or newer (CI runs on 3.12). On Windows, tick “Add Python to PATH” in the installer.
pip	Ships with Python; the installer upgrades it inside the virtual environment.
OS	Windows 10/11, macOS 12+, or Linux (X11 or Wayland).
Internet	Needed once to download the Python dependencies, and for Help ▸ Check for updates.
Disk	~400 MB for the virtual environment (mostly PySide6/Qt).

Runtime dependencies, installed automatically: numpy >= 2.0, scipy, matplotlib, requests, PySide6 >= 6.5.

2. Get the code

Clone the repository:

git clone https://github.com/sullymike/Fitbauer.git
cd Fitbauer

…or download a release ZIP from the Releases page, unzip it, and open a terminal in the resulting folder. The release ZIP is the source, not a binary — you still run one of the installs below.

3. Install — option A: installer script (recommended)

From the project folder:

Linux / macOS

python3 install.py
./fitbauer

Windows

py install.py
fitbauer.bat

If py is not recognised, use python install.py.

install.py does everything in one step:

• creates a local virtual environment in .venv/;
• installs the dependencies from requirements.txt;
• writes the launchers fitbauer (Linux/macOS) and fitbauer.bat (Windows);
• runs a quick byte-compile smoke test;
• registers Fitbauer in the system application menu (per-user, no admin rights) so it opens from the menu with its icon:
  • Linux — ~/.local/share/applications/fitbauer.desktop (Education category);
  • Windows — a Fitbauer folder in the Start Menu;
  • macOS — menu registration is skipped; launch with ./fitbauer.

Installer flags:

python install.py               # full install + menu registration
python install.py --menu-only   # only (re)register the menu entry
python install.py --no-menu     # install, leave the menus untouched
python install.py --uninstall   # remove the menu entry

4. Install — option B: manual virtual environment

If you would rather manage the environment yourself:

python3 -m venv .venv
source .venv/bin/activate            # Windows: .venv\Scripts\activate
pip install --upgrade pip
pip install -r requirements.txt
python fitbauer.py

For the test suite, also install the dev requirements:

pip install -r requirements-dev.txt
QT_QPA_PLATFORM=offscreen pytest -q   # add "xvfb-run -a" on headless Linux

5. Run it

• Installer route: ./fitbauer (Linux/macOS) or fitbauer.bat (Windows), or the application-menu entry.
• Manual route: activate .venv and run python fitbauer.py.
• Headless fitting, no GUI: python mossbauer_fit_cli.py (discrete) or python fit_bhf_distribution_cli.py (distributions).

First run, with the bundled sample data:

1. File ▸ Open… → data_sample/magnetita_Fe3O4.adt
2. File ▸ Load session… → data_sample/Fe3O4_session.json

6. Update

• From inside the app: Help ▸ Check for updates… downloads the latest release ZIP. A stable/beta channel can be chosen in the update settings.
• From source: git pull, then re-run python install.py — it reuses .venv and refreshes the dependencies. With a release ZIP, unzip the new one over the folder and re-run python install.py.

7. Build a standalone executable (optional)

pip install pyinstaller
pyinstaller Fitbauer.spec        # → dist/Fitbauer/

The folder dist/Fitbauer/ then runs without any Python installation.

8. Troubleshooting

Symptom	Fix
python3: command not found	Install Python 3 from python.org or your distro's package manager.
GUI does not start	Check the venv is active, then pip install -r requirements.txt and python -m py_compile fitbauer.py mossbauer_qt.py.
ImportError for PySide6, or a Qt "platform plugin" error on Linux	Install the system libraries Qt needs — on Debian/Ubuntu: libgl1, libxkbcommon-x11-0, libegl1. On a headless machine set QT_QPA_PLATFORM=offscreen.
PDF report fails	The Markdown report is still written; PDF export needs optional rendering libraries on some systems.
Menu entry not created	Re-run python install.py --menu-only (on macOS this step is intentionally skipped).

---

Project structure

core/          Physics and fitting engine (no GUI dependency)
gui/           Modular Qt/Matplotlib GUI — thin controllers only
locales/       Translations: en / es / fr / de / pt / ru / ja / ch
data_sample/   Sample spectra and sessions
tests/         Physics, fitting, CLI and Qt tests

The physics and fitting engines live exclusively in core/; the GUI is a thin client. See docs/architecture.md for details.

---

Changelog

See CHANGELOG.md.

---

License

Fitbauer is released under the Apache License 2.0 — see LICENSE and NOTICE.

This program uses the Qt toolkit through the PySide6 bindings, under the terms of the GNU Lesser General Public License v3 (LGPLv3). Qt is not modified; the full licence text ships in licenses/LGPL-3.0.txt and licenses/GPL-3.0.txt. In the binary builds, the Qt libraries are bundled as ordinary shared libraries and can be replaced by a compatible build, as LGPLv3 §4 provides.

Every third-party component and its licence is listed in THIRD-PARTY-LICENSES.md.

© Jorge Sánchez Marcos, Nieves Menéndez González — Department of Physical Chemistry, UAM.