A Simple Likelihood Method for Quasar Target Selection

TL;DR

The paper introduces a Bayesian, flux-space Likelihood method for quasar target selection that explicitly accounts for photometric errors by using training catalogs to compute posterior probabilities for objects being quasars in a given redshift range. Training data are generated from a Monte Carlo quasar catalog (QSO Catalog) built on a Jiang2006 luminosity function and an Everything Else Catalog from Stripe 82, enabling robust likelihoods over the five SDSS fluxes for QSO versus Everything Else. Applied to BOSS Stripe 82 data, the method achieves about 40% efficiency and 65% completeness at a target density of 40 deg^-2, recovering roughly 15.9 z>2.2 QSOs per deg^2 and outperforming simple color-box selections. The framework is flexible for incorporating additional attributes such as variability and more filters, contrasts favorably with XDQSO, and was integrated into BOSS CORE/BONUS target strategies with plans to publish the probabilities as data products in SDSS data releases.

Abstract

We present a new method for quasar target selection using photometric fluxes and a Bayesian probabilistic approach. For our purposes we target quasars using Sloan Digital Sky Survey (SDSS) photometry to a magnitude limit of g=22. The efficiency and completeness of this technique is measured using the Baryon Oscillation Spectroscopic Survey (BOSS) data, taken in 2010. This technique was used for the uniformly selected (CORE) sample of targets in BOSS year one spectroscopy to be realized in the 9th SDSS data release. When targeting at a density of 40 objects per sq-deg (the BOSS quasar targeting density) the efficiency of this technique in recovering z>2.2 quasars is 40%. The completeness compared to all quasars identified in BOSS data is 65%. This paper also describes possible extensions and improvements for this technique

Figures (10)

• Figure 1: Contour plot of the $u-g$ and $g-r$ colors of the Everything Else (red) and QSO (blue) Catalogs. The region of overlap, where target selection becomes difficult, is at $u-g \approx 1$ and $g-r \approx 0$. The error bars are the SDSS single-epoch $g-r$ and $u-g$ magnitude errors at g=22 (black) and g=20 (grey).
• Figure 2: Right Ascension (RA) versus Declination (Dec) of BOSS QSO data used for the likelihood method testing and luminosity function testing. Testing was done in the Stripe 82 calibration band with regions of high ($>90\%$) spectroscopic completeness. The blue points are spectroscopically confirmed quasars and the yellow regions are the sky tiles that were observed. Note that the vertical and horizontal scales are not the same.
• Figure 3: Color-color diagrams of BOSS QSOs recovered by the likelihood method (magenta), false-negative QSOs that were not targeted (missed) by likelihood method (cyan), and false-positive stars that were wrongly targeted by likelihood method (red). These plots show recovered/missed ($z>2.2$) QSOs. It is clear when comparing these plots with Fig. (\ref{['fig:InputCatalogs']}) that the problematic region for likelihood targeting is where the two catalogs overlap near $u-g=1$, $g-r=0.25$. For context the $QSO$ Catalog and $EE$ Catalog contours plot from Fig. (\ref{['fig:InputCatalogs']}) are included in the above plots. The error bars are the SDSS single-epoch $g-r$ and $u-g$ magnitude errors at $g=22$ (black) and $g=20$ (grey). The targeting decisions were computed in flux space rather than the color space shown in the figures.
• Figure 4: The probability ($\mathcal{P}$) distributions of the likelihood method recovered QSOs (magenta, 4617 total), false-negative QSOs that were missed by the likelihood method (cyan, 1566 total), and false-positive stars that were incorrectly targeted by likelihood method (red, 5743 total). The vertical gray dashed line shows the likelihood $\mathcal{P}$ threshold used for targeting ($\mathcal{P} > 0.245$). The spike around $\mathcal{P} = 0$ in the cyan curve are quasars that fall in the midst of the stellar locus and therefore are found by the method to have a very low probability of being QSOs. Most of these quasars are targeted because they are previously spectroscopically confirmed QSOs or by their flux variability. The likelihood distribution of the probabilities for the untargeted stars (true-negative) are not included in the plot, but constitute an additional 742,662 objects.
• Figure 5: The redshift distributions of the likelihood method recovered QSOs (magenta) and false-negative QSOs that were not targeted (missed) by the likelihood method (cyan), compared with SDSS DR5 QSOs (blue).