Performance of the Quasar Spectral Templates for the Dark Energy Spectroscopic Instrument

TL;DR

This paper introduces two DESI QSO template sets—low-z (0.05<z<1.6) and high-z (1.4<z<7.0)—trained on ~2×10^5 SDSS QSO spectra to better capture spectral diversity and redshift evolution. Panels of clustering, redshift re-alignment via Mg II, and EMPCA-derived eigenspectra yield templates that improve QSO completeness and reduce catastrophic failures while preserving the galaxy sample. Compared with the previous DESI templates, the new templates achieve a global redshift-precision improvement (MAD down to ~57 km s^{-1}) and significantly reduce redshift biases across most tracers, with notable gains for BALQSOs when BAL features are masked. These advances meet DESI’s redshift-precision and bias requirements for tracer and Lyα QSOs and are expected to enhance cosmological analyses relying on QSO clustering and Lyα forest measurements; future work includes incorporating Lyα forest optical depth and dedicated BAL templates to further mitigate residual biases.

Abstract

Millions of quasar spectra will be collected by the Dark Energy Spectroscopic Instrument (DESI), leading to a four-fold increase in the number of known quasars. High accuracy quasar classification is essential to tighten constraints on cosmological parameters measured at the highest redshifts DESI observes ($z>2.0$). We present the spectral templates for identification and redshift estimation of quasars in the DESI Year 1 data release. The quasar templates are comprised of two quasar eigenspectra sets, trained on spectra from the Sloan Digital Sky Survey. The sets are specialized to reconstruct quasar spectral variation observed over separate yet overlapping redshift ranges and, together, are capable of identifying DESI quasars from $0.05 < z <7.0$. The new quasar templates show significant improvement over the previous DESI quasar templates regarding catastrophic failure rates, redshift precision and accuracy, quasar completeness, and the contamination fraction in the final quasar sample.

Figures (8)

• Figure 1: Flowchart of template development process.
• Figure 2: Red circles (blue squares) indicate the median redshift correction applied to cluster composite spectra of each low (high) redshift bin, including the $z<0.35$ bin. Error bars indicate the 16th and 84th percentiles of each distribution.
• Figure 3: The high redshift templates are shown on the left in black and cover $1.4<z<7.0$. The low redshift templates are shown on the right in black and cover $0.05<z<1.6$. The previous DESI QSO templates are overlaid in red in both figures. The previous templates are trimmed to the wavelength coverage of the new templates and scaled for clarity in comparison. Prominent QSO emission lines are annotated on the mean spectra. The gray shading indicates the wavelength region that is poorly constrained in both eigenspectra sets owing to low QSO numbers and degraded spectrophotometric calibration at the shortest wavelengths of SDSS spectra.
• Figure 4: The contamination fraction (dashed) and incompleteness (solid) of the QSO Main sample, and of the QSO and galaxy samples from the DESI Truth Table with redshift from Redrock with the new QSO templates. The same metrics for the previous QSO templates are shown in red.
• Figure 5: A subset of spectra from the DESI Truth Table that are classified as QSO by Redrock with the new QSO templates, but not classified as QSO with the previous QSO templates (pre-QSO afterburners). The observed flux is shown in gray, the error on the flux is shown in orange, and the flux after applying a Gaussian smoothing filter is shown in black. The DESI target ID, redshift estimate from the new templates (best z), and basic information on the type of spectrum are annotated at the top of each subplot. Prominent emission lines are indicated in the observer frame with vertical lines and labeled.