Modeling the Spectral Diversity of Quasars in the Sixteenth Data Release from the Sloan Digital Sky Survey

TL;DR

This paper introduces a novel spectral modeling approach for quasars using pixel-level clustering and eigenspectra, improving redshift estimation and spectral reconstruction for large quasar datasets from SDSS.

Contribution

The study develops a new quasar spectral modeling method based on clustering and eigenspectra, enhancing redshift accuracy and spectral diversity representation.

Findings

Improved spectral reconstruction accuracy over SDSS pipeline models.

Redshift estimates are more consistent with Mg II emission line redshifts.

The method enhances automated classification and absorption feature analysis.

Abstract

We present a new approach to capturing the broad diversity of emission line and continuum properties in quasar spectra. We identify populations of spectrally similar quasars through pixel-level clustering on 12,968 high signal-to-noise spectra from the Sloan Digital Sky Survey (SDSS) in the redshift range of $1.57<z<2.4$. Our clustering analysis finds 396 quasar spectra that are not assigned to any population, 15 misclassified spectra, and six quasars with incorrect redshifts. We compress the quasar populations into a library of 684 high signal-to-noise composite spectra, anchored in redshift space by the Mg II emission line. Principal component analysis on the library results in an eigenspectrum basis spanning 1067 to 4007 $\r{A}$. We model independent samples of SDSS quasar spectra with the eigenbasis, allowing for a free redshift parameter. Our models achieve a median reduced chi-squared on non-BAL quasar spectra that is reduced by 8.5% relative to models using the eigenspectra from the SDSS spectroscopic pipeline. A significant contribution to the relative improvement is from the ability to reconstruct the range of emission line variation. The redshift estimates from our model are consistent with the Mg II emission line redshift with an average offset that displays 51.4% less redshift-dependent variation relative to the SDSS eigenspectra. Our method for developing quasar spectra models can improve automated classification and predict the intrinsic spectrum in regions affected by intervening absorbers such as Ly$\alpha$, C IV, and Mg II, thus benefiting studies of large-scale structure.