SLAM¶
SLAM (Stellar Labels Machine) estimates stellar parameters for M dwarfs from BOSS coadded spectra. It was developed by Bo Zhang and adapted for SDSS-V by Zach Way.
What it does¶
SLAM determines the following stellar labels for M dwarf stars:
Effective temperature (
teff)Surface gravity (
logg)Iron abundance (
fe_h)Iron abundance from NIU calibration (
fe_h_niu)Alpha-element abundance (
alpha_fe)
How it works¶
Target selection: Spectra are filtered by photometric and program criteria before fitting. Objects must satisfy at least one of:
Magnitude cut: Gaia
G - RP > 0.56and absolute G-band magnitude > 5.553 (with valid parallax).Program match: The source is assigned to the
mwm_ysoormwm_sncprogram.
Objects failing both criteria are flagged and skipped.
Spectral preparation: Observed BOSS spectra are rebinned onto a standard wavelength grid used during SLAM training. For coadded spectra, no radial velocity correction is applied; for visit spectra, the XCSAO radial velocity is used to shift to the rest frame.
Continuum normalization: Spectra are block-normalized using the
laspecnormalization routines over the wavelength range 6002–8957 Angstroms with a polynomial approach (quantile 0.7).Label prediction: Labels are predicted in two passes:
A fast initial prediction (
predict_labels_quick) provides starting guesses.A refined prediction (
predict_labels_multi) uses the initial guesses to perform a full optimization, returning labels, covariance, and optimizer status.
Post-processing flags: After fitting, results are flagged if the effective temperature or metallicity falls outside the expected bounds, or if the optimizer reports a bad status.
Output fields¶
Field |
Description |
|---|---|
|
Effective temperature (K) |
|
Uncertainty on effective temperature |
|
Surface gravity (log cm/s^2) |
|
Uncertainty on surface gravity |
|
Iron abundance [Fe/H] (dex) |
|
Uncertainty on [Fe/H] |
|
Iron abundance from NIU calibration (dex) |
|
Uncertainty on NIU [Fe/H] |
|
Alpha-element abundance [alpha/Fe] (dex) |
|
Uncertainty on [alpha/Fe] |
|
Correlation coefficients between pairs of labels |
|
Initial guesses for each label from the quick prediction |
|
Whether the optimizer converged |
|
Optimizer exit status code |
|
Whether the solution satisfies optimality conditions |
|
Chi-squared of the fit |
|
Reduced chi-squared of the fit |
|
Bitfield summarizing result quality |
Correlation coefficients¶
The model provides pairwise correlation coefficients between the fitted labels
(e.g., rho_teff_logg, rho_teff_fe_h, etc.), which quantify covariances in
the label estimates.
Flags¶
Flag |
Bit |
Meaning |
|---|---|---|
|
2^0 |
Optimizer returned a bad status (not 0 or 2) |
|
2^1 |
Teff is outside [2800, 4500] K |
|
2^2 |
[Fe/H] is outside [-1.0, +0.5] dex |
|
2^3 |
Source photometry is outside expected M dwarf range |
|
2^4 |
Source fails the photometric magnitude cut |
|
2^5 |
Source is not in the |
Composite flags¶
flag_warn: Set ifflag_bad_optimizer_statusorflag_outside_photometry_rangeis set.flag_bad: Set if any offlag_teff_outside_bounds,flag_fe_h_outside_bounds,flag_outside_photometry_range, orflag_bad_optimizer_statusis set.
Key caveats¶
SLAM is trained on M dwarf spectra. Applying it to stars outside the M dwarf parameter range (approximately 2800–4500 K) will produce unreliable results, which is indicated by
flag_teff_outside_bounds.The photometric filtering criteria (Gaia color and absolute magnitude) are designed to select M dwarfs. Stars without valid Gaia parallax will only be analyzed if they belong to the
mwm_ysoormwm_sncprograms.The
fe_h_niulabel is an alternative metallicity calibration; both metallicity estimates are provided for comparison.Spectral data arrays (model flux, continuum) are stored in intermediate pickle files and can be accessed through the result object.