NORMOS interoperability

⚙️ Generated page — do not edit here. Source: docs/normos_job_en.md. Edit it in the repository and the wiki rebuilds itself.

NORMOS interoperability: .JOB files

Module: core/normos_job.py (pure functions) · GUI: File ▸ NORMOS (.JOB)

---

What it is and why

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 It runs under DOS, it is proprietary and it is no longer maintained, yet many laboratories keep years of work stored as .JOB files.

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

---

Importing

File ▸ NORMOS (.JOB) ▸ Import NORMOS job…

It rebuilds the model in the panels and loads the spectrum. A .JOB names its files in the first four lines, with no path, because NORMOS ran under DOS with everything in one directory:

Fe080725.ws5      ← spectrum
distcri1.JOB      ← the job itself
Fe0807di.res      ← results NORMOS will write
Fe0807di.plt      ← plot
 &DATA
 NLTEXT=4, VMAX=-11.966, TRIANG=.true.,
 &END

resuelve_fichero_de_datos() looks for the spectrum next to the .JOB, ignoring upper/lower case —the names come from DOS and rarely match the case on disk— and if the declared one is missing it tries the job's own name and, as a last resort, the only spectrum in the folder. It never returns a NORMOS output (.RES/.PLT).

Keep every file of the job in the same folder. That is what NORMOS expects and what makes the import work in one go.

Two families are recognised automatically:

Family	What its subspectra are	Where it ends up
NORMOS-SITE	discrete sites	singlet / doublet / sextet components
NORMOS-DIST	the points of a grid	P(BHF)/P(ΔEQ) panel

For DIST jobs Fitbauer translates the grid (origin and step), the shape (histogram, Gaussian, binomial or fixed), the δ(x) correlation and the edge anchors; the "crystalline" subspectra (NXLS) become sharp components.

Exporting

File ▸ NORMOS (.JOB) ▸ Export NORMOS job… 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.

---

Convention conversions

This is the delicate part, and getting it wrong raises no error:

NORMOS	Meaning in Fitbauer
WID, W13, W23	WID is the width of lines 3,4 and W13/W23 are relative to it; gamma1 is that of lines 1,6. The conversion is gamma1 = WID·W13
D13, D23	Area ratios. int1/int2 are depth ratios. They agree only when the widths are equal
DEP (or ARE)	The subspectrum area in mm/s, not a depth
NDEX/FACTOR/CONST	Constraints in NORMOS's global numbering, 13 + 15·(n−1)

The BHF scale

NORMOS derives the sextet line positions from the nuclear moments; Fitbauer uses the published α-Fe pattern. They do not differ by a simple scale factor. To reproduce one of its BHF values exactly, fit with the NORMOS convention active:

from core.constants import sextet_pattern

with sextet_pattern("normos"):
    ...   # the fit uses NORMOS line positions

The difference is about 0.1 T.

---

The folding point is not imposed

The PFP carried in &DATA is the seed of the folding-point search, not its result: NORMOS refines it over two cycles, and in real jobs it ends up more than one channel away from what the file asked for. Fitbauer runs its own search —the correct counterpart— and reports the PFP as information only.

There is a second subtlety. The refined point NORMOS prints in its .RES is not where it folds either: its final routine (normospr.for:601-604) truncates it and adds whole channels,

IPFA = PFA + 1.0E-4          ! real→integer assignment: truncates
IPFP = PFP + 1.0E-4
DO 602 L=1,NP
  TEMP(L) = Y(IPFA-L+1) + Y(IPFA+L)

The pairs sum to 2·IPFA+1, so the symmetry axis falls at ⌊PFP⌋ + 0.5. It lives in core.normos_job.punto_de_doblado_normos(), and taking it into account is what makes its fits reproducible.

---

What is not carried over

The importer warns about each of these, because what was not translated matters as much as what was:

• Czjzek / Le Caër distributions (DISTRI=4) and the Billard–Chamberod neighbour model (METHOD=3).
• Several overlapping distribution blocks: Fitbauer handles one.
• The LAMDA smoothing parameter. NORMOS's is absolute and Fitbauer's alpha is dimensionless, so there is no one-to-one conversion: set it with the L-curve. The BETA/LAMDA ratio IS preserved and becomes the edge anchor.
• DTQ in field distributions. The distcalf.for loops for METHOD 1-5 compute RH = BHF+PP*DTB and RI = ISO+PP*DTI and do not touch ΔEQ, so carrying it over would introduce a correlation NORMOS never applied.

Watch out for inherited .JOB files. The DIST format does not accept SITE keys such as NLINE, DEP, W13 or W23. If they were copied from another job, NORMOS reads and discards them without a word, so that subspectrum never entered its fit. Fitbauer does warn about it.

---

From the command line

The discrete-fit CLI accepts a .JOB as a template, detected by content rather than by extension:

python mossbauer_fit_cli.py --template MY_JOB.JOB --spectrum measurement.ws5
python mossbauer_fit_cli.py --template model.json --spectrum measurement.ws5 \
       --export-job OUTPUT.JOB

---

Validation

The equivalence with NORMOS is not a statement of intent: it is measured on two independent benchmarks —411 synthetic spectra and 564 real fits made with the original program— with the full reports in validacion/informe/.

In 355 of 503 comparable jobs (71 %) Fitbauer matches or improves on NORMOS's reduced χ².