SP♤DE — a joint spectro-photometric decomposition engine

Fitting imaging, stellar kinematics and stellar populations under one likelihood to separate classical bulges from pseudobulges.

SP♤DE — the Spectro-Photometric Analytic Decomposition Engine — is a bulge+disk decomposition code I developed at ASIAA. Where conventional decompositions fit a single image, SP♤DE fits imaging, stellar kinematics and stellar populations simultaneously under one joint likelihood, so that a galaxy’s velocity dispersion and stellar ages actively constrain its structural decomposition rather than being read off afterwards.

Why it matters

Not all bulges are the same. Classical bulges are dispersion-supported and old — built by ancient mergers. Pseudobulges rotate, are young, and grow secularly as bars and spiral arms funnel disk gas inward. The two form by completely different routes, and telling them apart matters for the question I care about: when a galaxy stops forming stars, is the bulge responsible — and which kind of bulge?

Separating them reliably is hard, because the distinguishing evidence lives in three different datasets. Photometry alone cannot see that a bulge is dynamically hot; kinematics alone cannot see that it is old. SP♤DE’s answer is to stop treating them as separate measurements.

How it works

The model is a single additive log-posterior over three arms — imaging, kinematics, populations — with per-arm weighting so that the ~10⁵ imaging pixels cannot drown out the ~10³ IFU spaxels, plus priors and an anti-swap penalty that keeps the bulge from trading places with the disk.

  • Imaging — DESI r-band, for structure, sizes, shapes and the bulge/disk light fraction.
  • Kinematics and populations — IFU spectroscopy from MaNGA and SAMI, giving V and σ maps alongside stellar ages and metallicities.

The MAP estimate comes from a coordinate loop: stellar populations are solved by regularised NNLS at fixed structure and profiled out, then the structural parameters are optimised with bounded L-BFGS on JAX autodiff gradients, warm-started from GALFITM. ppxf with E-MILES templates pins the absolute age and metallicity calibration, and variational inference provides uncertainties, with full NUTS posteriors as the next tier.

What’s different

Existing tools solve part of this problem. BUDDI decomposes an IFU cube photometrically, wavelength slice by wavelength slice, and extracts component spectra as a post-processing step — the spectra never inform the structural fit. GALFIT fits a single image by least squares. SP♤DE instead evaluates a Bayesian posterior across all three datasets at once, and fuses external high-resolution imaging with the IFU data.

Status

SP♤DE has been run across the KILOGAS sample — roughly 450 galaxies spanning the MaNGA and SAMI footprints — producing bulge/disk structural parameters, component-resolved kinematics and populations, and a classical/pseudo classification for each galaxy. Validation against independent host-galaxy morphology confirms that the classifications track genuine formation channels.

Several papers using these catalogues are in preparation.

Project at ASIAA · 2026–present.