Abstract
We describe the algorithm used to select the luminous red galaxy (LRG) sample for the extended Baryon Oscillation Spectroscopic Survey (eBOSS) of the Sloan Digital Sky Survey IV (SDSS-IV) using photometric data from both the SDSS and the Wide-field Infrared Survey Explorer. LRG targets are required to meet a set of color selection criteria and have $z$-band and $i$-band MODEL magnitudes $z < 19.95$ and $19.9 < i < 21.8$, respectively. Our algorithm selects roughly 50 LRG targets per square degree, the great majority of which lie in the redshift range $0.6 < z < 1.0$ (median redshift 0.71). We demonstrate that our methods are highly effective at eliminating stellar contamination and lower-redshift galaxies. We perform a number of tests using spectroscopic data from SDSS-III/BOSS ancillary programs to determine the redshift reliability of our target selection and its ability to meet the science requirements of eBOSS. The SDSS spectra are of high enough signal-to-noise ratio that at least ~89% of the target sample yields secure redshift measurements. We also present tests of the uniformity and homogeneity of the sample, demonstrating that it should be clean enough for studies of the large-scale structure of the universe at higher redshifts than SDSS-III/BOSS LRGs reached.
Introduction
To understand dark energy and the expansion of the universe, astronomers map the Large-Scale Structure (LSS) using "tracers"—massive, bright objects like Luminous Red Galaxies (LRGs). Previous surveys, such as BOSS (SDSS-III), effectively mapped LRGs up to a redshift of $z \approx 0.6$. However, pushing beyond this boundary is notoriously difficult using optical light alone. As galaxies move to higher redshifts, their characteristic "4000 Å break" shifts into the near-infrared, causing distant LRGs to overlap almost perfectly with cool M-type stars in purely optical color-color space. To build a successful spectroscopic survey like eBOSS, researchers needed a way to filter out millions of foreground stars and low-redshift galaxies.
Methodology
The eBOSS LRG selection algorithm moves beyond traditional optical photometry by integrating data from the Wide-field Infrared Survey Explorer (WISE), specifically the $W1$ band ($3.4\ \mu$m). This approach exploits the 1.6 $\mu$m "stellar bump" in galaxy spectra to clearly separate high-redshift galaxies from stars.
Selection Pipeline and Forced Photometry
The selection process involves a series of filters including signal-to-noise ratio (SNR) selection, flux limits, and specific color cuts for redshift selection and star exclusion. Because many LRGs are too faint to be detected independently by the standard WISE pipeline, the team utilized "forced photometry" (unWISE). By taking the known positions of objects from higher-resolution SDSS images and measuring the infrared flux at those exact coordinates in the WISE data, the usable LRG sample size was increased by roughly 10%.
Additionally, the algorithm uses SDSS imaging re-calibrated against the Pan-STARRS survey, improving global photometric consistency to better than 0.5%. To ensure the sample is uniform enough for cosmological study, target density was modeled against potential systematics like Galactic dust and stellar density to create a mask of the most reliable sky regions.
Key Results
The hybrid selection method successfully extends the reach of LRG surveys beyond the BOSS limit. The algorithm selects approximately 50 LRG targets per square degree, with the majority of targets falling between $0.6 < z < 1.0$ and a median redshift of 0.71.
Efficiency and Purity
Spectroscopic follow-up showed that approximately 89% of targets yield secure redshifts. The inclusion of infrared data acts as a powerful "veto" against stellar contaminants; while an LRG and an M-star might look identical in optical filters, they are distinguishable in the $3.4\ \mu$m band. This resulted in a remarkably low stellar contamination rate of approximately 9%.
Discussion and Limitations
A primary limitation identified is the homogeneity of the South Galactic Cap (SGC), where only ~82% of the area met the requirements compared to 97% in the North, largely due to variations in imaging depth. Furthermore, the $z$-band (the reddest optical filter) proved to be a bottleneck; small shifts in $z$-band zero-points have a disproportionately large impact on selection cuts. Future improvements in automated redshift-finding pipelines could recover many of the failed spectra (the remaining 11%) without requiring new observations, further increasing survey efficiency.
Summary updated