TLUSTY Flux Data Processing

TLUSTY flux files are staged as source tar archives under data/atmospheres/tlusty/. The converter reads *.flux.gz members, parses Fortran-style numeric columns, and writes local processed artifacts without deleting the source tar files.

Use this page when you need to rebuild or validate the local TLUSTY cache. Use Atmosphere capabilities for the higher-level capability matrix and runtime/backend status.

Local layout¶

data/atmospheres/tlusty/
  ostars-2002/raw/
  bstars-2007/vturb-2/raw/
  bstars-2007/vturb-10-cn/raw/
  processed/
    catalog.parquet
    catalog_fragments/
    validation/
    tlusty_flux.zarr/

The released coordinate is frequency in Hz and the released column is Eddington flux H_nu in erg s-1 cm-2 Hz-1. Wavelength ranges in the catalog are diagnostic coverage metadata derived as c / nu; they are not a unit-converted replacement for the raw frequency grid.

The important schema detail¶

The real TLUSTY files are not one rectangular array per named dataset. Frequency grids can differ across spectra, even within one BSTAR or OSTAR product. The processed Zarr layout therefore uses deterministic grid subgroups:

tlusty_flux.zarr/
  bstar_2007_vturb_2/
    grid000/
      frequency_hz
      flux_fnu
      teff
      logg
      vturb_km_s
    grid001/
    ...
  bstar_2007_vturb_10_cn/
    grid000/
    ...
  ostar_2002/
    grid000/
    ...

Each gridNNN subgroup is rectangular because all rows in that subgroup share one frequency_hz array. The catalog row records zarr_group, zarr_subgroup, and zarr_row, so runtime code can find the exact native grid without resampling in the converter.

This is a pedagogical example of the atmosphere policy: preserve the released scientific coordinate first; add interpolation or spectral-density transforms only in a tested runtime backend.

Some released tables contain duplicate printed frequencies. The processed artifact preserves them exactly. At runtime, the TLUSTY adapter mean-coalesces samples at identical published coordinates, records that operation in provenance, converts H_nu to F_nu with 4 pi, converts to wavelength-density surface flux, and resamples each topology vertex onto the requested axis.

Conversion¶

Dry-run a staged directory:

env -u VIRTUAL_ENV uv run --no-sync --extra data python \
  scripts/convert_tlusty_flux.py \
  --raw-dir data/atmospheres/tlusty/bstars-2007/vturb-2/raw \
  --dataset tlusty_bstar_2007_vturb_2 \
  --dry-run

Convert that directory:

env -u VIRTUAL_ENV uv run --no-sync --extra data python \
  scripts/convert_tlusty_flux.py \
  --raw-dir data/atmospheres/tlusty/bstars-2007/vturb-2/raw \
  --dataset tlusty_bstar_2007_vturb_2 \
  --group-name bstar_2007_vturb_2 \
  --overwrite

The three controlled local batches are:

env -u VIRTUAL_ENV uv run --no-sync --extra data python \
  scripts/convert_tlusty_flux.py \
  --raw-dir data/atmospheres/tlusty/bstars-2007/vturb-2/raw \
  --dataset tlusty_bstar_2007_vturb_2 \
  --group-name bstar_2007_vturb_2 \
  --overwrite

env -u VIRTUAL_ENV uv run --no-sync --extra data python \
  scripts/convert_tlusty_flux.py \
  --raw-dir data/atmospheres/tlusty/bstars-2007/vturb-10-cn/raw \
  --dataset tlusty_bstar_2007_vturb_10_cn \
  --group-name bstar_2007_vturb_10_cn \
  --overwrite

env -u VIRTUAL_ENV uv run --no-sync --extra data python \
  scripts/convert_tlusty_flux.py \
  --raw-dir data/atmospheres/tlusty/ostars-2002/raw \
  --dataset tlusty_ostar_2002 \
  --group-name ostar_2002 \
  --overwrite

The current measured processed counts are:

tlusty_bstar_2007_vturb_2      981
tlusty_bstar_2007_vturb_10_cn  551
tlusty_ostar_2002              690

Validation JSON records readback status, source tar hashes, member provenance, and float32 storage roundoff. Values far below the float32 dynamic range may underflow to zero; this is recorded by the roundoff ledger.

Validation¶

After conversion, run:

env -u VIRTUAL_ENV uv run --no-sync --extra data pytest -q \
  tests/validation/test_atmospheres_local_artifacts.py

The TLUSTY portion asserts:

all three datasets appear in the shared processed catalog.
row counts are 981, 551, and 690.
raw source tars still exist.
validation JSON records source tar hashes, readback_ok=true, and raw_deleted=false.
each Zarr subgroup has a positive frequency_hz array and finite sampled flux_fnu.
coverage reports list the datasets as processed with wavelength ranges derived from frequency.

Runtime status¶

TLUSTY is processed and visible to AtmosphereLibrary.from_local("data") through 27 exact composition-scoped products. OSTAR2002 has an accepted linear parameter-interpolation policy. Both BSTAR modes remain POLICY_NOT_VALIDATED because linear and positive-log holdout metrics trade off; the registry refuses those requests until stronger evidence resolves the policy. The processed artifact keeps raw frequency_hz and H_nu intact.

Use Query atmosphere spectra to build an exact request.