CERIDWEN

Comprehensive SED Estimation Routine Involving Data-driven WEight calculatioNs. A JAX-native, GPU-capable spectral energy distribution (SED) fitting package for galaxies, with variational-inference preconditioned Hamiltonian Monte Carlo, nested sampling, and native redshift support.

The forward model is a single differentiable XLA graph. It builds a galaxy spectrum from stellar populations, applies dust attenuation and emission, adds nebular continuum and line emission, attenuates the intergalactic medium, and projects the result to the observed frame at the galaxy's redshift. Inference runs on CPU or GPU.

Features

• Star formation history (non-parametric continuity or parametric)
• Metallicity history (constant or time-varying)
• Dust attenuation (Kriek & Conroy diffuse, power-law birth-cloud, multi-component age-dependent)
• Dust emission (Draine & Li grids)
• Nebular continuum and emission lines (CLOUDY grids)
• Broadband photometry, spectra, and emission-line fluxes, fit on their own or jointly
• Redshift-aware forward model with cosmological flux normalisation
• IGM attenuation (Madau 1995), extensible through an IGMModel base class
• NUTS, nested sampling, and VI-preconditioned NUTS

Where to next

• Installation: Python 3.11, dependencies, and the FSPS / $SPS_HOME setup.
• Quick start: build a model and fit it end to end.
• Conventions & gotchas: the unit and indexing conventions that catch people out. Read this before fitting real data.
• API reference: the public classes and functions.

Citing

If you use CERIDWEN, please cite it. See CITATION.cff in the repository; the methods paper reference will be added on publication.