Modeling the Spectral Diversity of Quasars

The Dark Energy Spectroscopic Instrument (DESI) will measure the spectra from tens of millions of extragalactic objects, of which approximately three million will be from quasars. The large spectroscopic sample requires a reliable, comprehensive model capable of automatically classifying these quasar spectra with high purity and completeness. The model must also provide precise redshift estimates to reduce the impact of redshift errors on cosmological measurements. We present a new approach to develop quasar spectra models using pixel-level clustering algorithms. Our algorithm captures quasar spectral diversity by compressing a large sample into discrete populations that each exhibit low internal diversity while rejecting peculiar and misclassified objects. Redshift corrections are applied to the quasar sample through fitting the high signal-to-noise MgII emission line in the error-weighted mean spectrum of each sub-population. The correction aims to place all spectra in the same rest frame and reduce redshift dependent bias in the final model. The redshift-corrected clustering results inform a quasar spectra model derived through an iterative implementation of PCA. Our model consists of two PCA basis sets: a low redshift set specialized in capturing the contributions from host galaxies (z < 1.3) and a high redshift set characterizing the diversity of quasars at the highest luminosities (z > 1.0). We present the classification, modeling, and redshift performance of the new model on the visual inspection catalogs of DESI.