Skip to content

SED: PHOENIX-ACES bolometric corrections as an alternative model (8.9.12) - #611

Open
jdeast wants to merge 2 commits into
masterfrom
feat/phoenix-bc
Open

jdeast wants to merge 2 commits into
masterfrom
feat/phoenix-bc

Conversation

@jdeast

@jdeast jdeast commented Nov 18, 2026

Copy link
Copy Markdown
Owner

What

Adds PHOENIX-ACES (Husser et al. 2013) bolometric-correction tables as an alternative SED model, selected with model: PHOENIX in the SED file. NextGen stays the default. Review item 8.9.12: the SED could only use NextGen because the BC tables predated the PHOENIX limb-darkening pipeline. That pipeline already streams the PHOENIX archive through the same filter profiles, weighting and extinction law, so the BC tables come from the same stream.

Why PHOENIX is worth having:

  • it reaches Teff 2300-12000 K (NextGen: 2600-10000 K, which severs faint microlensing sources and late M dwarfs);
  • with ld_source: phoenix a star's flux and its limb darkening come from one atmosphere;
  • it opens a route to alpha-enhanced models.

Generator (models/PHOENIX/generate_PHOENIX_LD_Tables.py --products bc)

One stream of the archive now writes the limb-darkening tables, the BC tables and the BC magnitude-system record. The BC tables use the NextGen format (PHOENIX/BCs/{FACILITY}.bc.parquet). The per-node unnormalized band fluxes and bolometric flux go into the intensities intermediate, so the BCs can be re-derived later without the archive.

  • Band flux. The flux is 2 pi int I_band(mu) mu dmu, computed from the same band intensities the limb darkening is fitted to. The filters, DetectorType weighting, extinction law and Vega zeropoints are NextGen's. As a result, PHOENIX.magsys.csv matches NextGen.magsys.csv row for row.

  • Bolometric term: the model's own bolometric flux, not sigma Teff^4. I measured this on 2026-11-18. A PHOENIX flux (and a BT-NextGen one) is given at the model's outer radius, so integrated over 500 A - 5.5 um it falls well short of sigma Teff^4 for low gravities:

    (Teff, logg) integrated flux / sigma Teff^4
    (5800 K, 4.5) 0.985
    (4500 K, 2.0) 0.93
    (4000 K, 0.5) 0.78

    A Planck curve puts only 1-2% of the flux past 5.5 um, so the gap is not missing infrared. Only the ratio of band flux to bolometric flux at one radius is physical. The BC therefore divides by the model's own integral, taken from the archive's flux spectrum of the same node (500 A - 5.5 um plus a Rayleigh-Jeans tail of 0.1-3%). That spectrum equals the intensities' 2 pi int I mu dmu to 1e-4 at the six nodes checked.

  • Coverage. The intensities stop at 2.6 um. A filter gets a column only if at least 99% of its weighted bandpass lies inside that range. That covers every NextGen column from GALEX through 2MASS Ks. WISE W1-W4, Spitzer IRAC 3.6 and Johnson M get no PHOENIX column.

  • Holes. Interior holes are filled by interpolating the BCs linearly along Teff, recorded in meta["padding"].

Subset check (tiny grid, [Fe/H] = 0, run locally against the archive)

BC differences in magnitudes. MIST BCs (ATLAS, normalized to sigma Teff^4) are taken from EXOFASTv2's tables at the same nodes.

node band PHOENIX - NextGen PHOENIX with sigma Teff^4 - NextGen MIST / PHOENIX / NextGen
5800 K, 4.5 V, G, J, TESS 0.027, 0.020, 0.012, 0.007 0.014, 0.008, -0.001, -0.006
4500 K, 2.0 V, G, J, TESS 0.105, 0.090, 0.077, 0.085 0.039, 0.024, 0.011, 0.019 V -0.443 / -0.504 / -0.609
4000 K, 0.5 V, G, J, TESS 0.300, 0.293, 0.291, 0.289 0.036, 0.029, 0.027, 0.026 V -0.894 / -0.990 / -1.290, J 1.601 / 1.606 / 1.315
3000 K, 4.5 V, BP, J, TESS 0.113, 0.143, 0.015, 0.032 0.099, 0.130, 0.002, 0.019

At the two giant nodes, PHOENIX with its own bolometric flux agrees with MIST to 0.005-0.10 mag. The published NextGen tables divide by sigma Teff^4, and the BT-NextGen spectra integrate to 0.77 sigma Teff^4 at (4000, 0.5) and 0.94 at (4800, 2.5). So at the nodes checked, the NextGen BCs of giants look too negative: ~0.3 mag at logg 0.5, ~0.07-0.1 mag at logg 2-2.5. This PR leaves NextGen alone; it is the first thing the follow-up NextGen-vs-PHOENIX comparison has to look at.

SED side

  • The SED file's model: key is the one spelling, and it is case-sensitive. phoenix raises naming PHOENIX. A name with no published tables raises unless an explicit model_root: holds that tree.
  • bc_grid.PUBLISHED_BC_MODELS dispatches the published-table fetch per model. models/PHOENIX/bc_tables.py follows the NextGen fetch rules: per facility, md5-pinned, a local edit never overwritten, and exozippy-fetch-phoenix-bc-tables for offline use.
  • A filter outside PHOENIX's coverage follows the existing missing-column policy, except that nothing is synthesized for PHOENIX:
    • a .sed row in that filter, or a band whose flux something reads, raises at the grid build. The error quotes the coverage (coverage: in PHOENIX/BCs/PHOENIX.grid.yaml) and names the way out (drop the row, or use model: NextGen);
    • an unread band is skipped with an INFO line, as for NextGen.
  • The paper draft says the BCs come from PHOENIX-ACES (\citep{Husser:2013}).
  • The SED figure keeps the NextGen plotting spectra (there are no PHOENIX plotting spectra), with a WARNING. It now also reads the NextGen grid description, so the axes match the spectra being interpolated.

The Zenodo record does not exist yet

The full table generation (~7000 nodes, 32 workers, about an hour) runs on the cluster (scripts/build_phoenix_bc.job), not here. Until then the record phoenix_bc_tables is registered as PENDING (new zenodo_assets.PENDING / RecordPendingError) and as ON_DEMAND, so neither CI nor the test prewarm fetches it.

While it is pending, model: PHOENIX under the default root fails at construction and names both ways out: NextGen, or locally generated tables through model_root:. Publishing takes three steps:

  1. run scripts/zenodo_publish.py pin --name phoenix_bc_tables --update-registry;
  2. delete the PENDING line;
  3. add the OWNERS line.

A test fails while a name is in both RECORDS and PENDING.

Not in this PR

  • Comparing PHOENIX with NextGen on the shipped examples. That is a follow-up once the record is published. NextGen stays the default until then.
  • WISE W1/W2 and IRAC 3.6/4.5 from the 5.5 um flux spectra. Possible, but those spectra carry no limb darkening, so it would mean a second flux source.

Tests

tests/test_phoenix_bc.py uses synthetic intensities, a synthetic flux spectrum, a fake published record whose "download" is a local copy, and a synthetic PHOENIX tree under model_root:. Nothing touches the network or the archive. It covers:

  • the BC formula against NextGen's;
  • the band-flux and bolometric integrals;
  • hole filling;
  • the intermediate round trip, including an LD-era file;
  • the pending record;
  • fetch and local-edit rules;
  • per-model dispatch;
  • model: PHOENIX reading the PHOENIX grid;
  • the WISE refusal;
  • case-sensitive names;
  • the prose sentence;
  • that PHOENIX.magsys.csv is NextGen's minus the six uncovered columns.

The existing PHOENIX-LD, Zenodo, BC-synthesis, band-filter and model-root suites pass unchanged.

🤖 Generated with Claude Code

https://claude.ai/code/session_01AWjnwpYHm56hSpKBzbPhrG

….12)

`model: PHOENIX` in the SED file selects bolometric-correction tables built
from the PHOENIX-ACES atmospheres (Husser et al. 2013); NextGen stays the
default until the two are compared on the shipped examples (a follow-up,
once the tables are published).

Generator: generate_PHOENIX_LD_Tables.py gains `--products bc`.  One stream
of the archive writes the limb-darkening tables, the BC tables in the
NextGen format (PHOENIX/BCs/{FACILITY}.bc.parquet) and their magnitude-
system record, and keeps the unnormalized band fluxes and the bolometric
flux per node in the intensities intermediate.  The band flux is
2 pi int I_band(mu) mu dmu of the same band intensities the limb darkening
is fitted to.  The bolometric term is the model's OWN bolometric flux (the
archive's flux spectrum of the node, 500 A - 5.5 um plus a Rayleigh-Jeans
tail), not sigma Teff^4: a PHOENIX flux is given at the model's outer
radius, 0.78 sigma Teff^4 at (4000 K, logg 0.5), and with that term the
PHOENIX BCs agree with MIST's where sigma Teff^4 would put them 0.1-0.3
mag off for giants.  Coverage: GALEX through 2MASS Ks; WISE, Spitzer IRAC
and Johnson M get no column.

SED: the model name is resolved once and is case-sensitive; bc_grid
dispatches the published-table fetch per model (PUBLISHED_BC_MODELS); a
filter outside a model's coverage is refused quoting the grid yaml's
coverage sentence and naming `model: NextGen`; the draft names the
PHOENIX atmospheres; the figure keeps the NextGen spectra with their own
grid description.

Zenodo: record `phoenix_bc_tables` is registered as PENDING (new
zenodo_assets.PENDING, RecordPendingError) and ON_DEMAND; models/PHOENIX/
bc_tables.py fetches it with the NextGen tables' rules once pinned, and
says how to proceed until then.  scripts/build_phoenix_bc.job is the
cluster run.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AWjnwpYHm56hSpKBzbPhrG

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant