THE ELEVENTH AND TWELFTH DATA RELEASES OF THE SLOAN DIGITAL SKY SURVEY: FINAL DATA FROM SDSS-III

MARVELS data collection began in 2008 October and ended in 2012 July. The majority of MARVELS stars were observed 20–40 times (Figure 1) with typical exposure times of 50–60 minutes. These exposure times were designed to reach a signal-to-noise ratio (S/N) sufficient to allow per-epoch RV precisions of tens of m s⁻¹ on stars of $7.6\lt V\lt 12$ mag. The total number of observations was designed to enable the determination of orbital parameters of companions with periods between one day and two years without the need for follow-up RV measurements using additional telescopes. However, the problems in RV calibration, the shortened MARVELS observing period, and the fact that the second MARVELS spectrograph was never built meant that this ideal was not met for all targets. The observing was split into two two-year campaigns: Years 1+2: 2008 October—2010 December; and Years 3+4: 2011 January—2012 July. For any particular star, the time baseline between the first and last observation was thus typically 1.5–2 years.

During its four years of operation, MARVELS obtained 1565 observations of 95 fields collecting multi-epoch data for 5700 stars, with observations of 60 stars per target field.

While we provide all raw data and intermediate data products in this release, the CCF and DFDI results are limited to the 3533 stars with more than 10 RV measurements. The UF1D analysis results include 5513 stars from the 92 fields that pass the basic quality requirements of the pipeline. Restricting to stars with $\geqslant 16$ observed epochs, which might be considered a reasonable threshold for searching for companions in the MARVELS data, yields 3293 stars in DR11 and 3233 stars in DR12 (a small number due to the somewhat tighter quality constraints).

Each spectrographic plate has two sets of 60 fiber holes, corresponding to two different fields to be observed in sequence. Both sets of fibers were plugged at the same time. In between observations of the two fields, the "gang" connector that joins the fibers from the cartridges to the long fibers that run to the MARVELS instruments was switched between the two sets of fibers.

A MARVELS exposure is the result of light from each of 60 fibers passing through a two-beam interferometer with one slanted mirror and then being dispersed in wavelength before being recorded on a $4{\rm{k}}\times 4{\rm{k}}$ CCD. Thus, each MARVELS image contains 120 individual spectra as the beam-splitter produces two interference patterns for each star, one from each beam. The RVs for each star can then be calculated from a comparison of the fringing spectrum observations at different epochs.

In this data release, we provide the 2D raw images, the 2D slices of extracted spectra, the 1D collapsed spectra, and the calculated stellar velocities and associated observational metadata for each spectrum of each star and field.

Target selection for MARVELS is described in full in Paegert et al. (2015). We here summarize the key aspects of the MARVELS target selection in each two-year phase of the survey.

MARVELS aimed to have a target sample in the range of $8\lt V\lt 12$ with a balance of 90% dwarf and subgiant stars with ${T}_{\mathrm{eff}}\lt 6250$ K, and ∼10% giant stars with $4300\lt {T}_{\mathrm{eff}}\lt 5100$ K (spectral types K2–G5). In the first two years of MARVELS, target selection was based on short "pre-selection" observations obtained with the SDSS spectrographs during the first year of SDSS-III to determine stellar surface temperatures and surface gravities. Because these observations used much shorter exposure times than standard SDSS observations, they were not automatically processed with the standard SDSS pipeline. Instead, the SDSS pipeline was used with some custom modifications to provide stellar spectra suitable for processing with the SEGUE Spectroscopic Processing Pipeline (SSPP; Lee et al. 2008). The raw data for these spectra were released as part of DR9. In DR12 we release these custom spectroscopic images, extracted spectra, and derived SSPP parameters as flat files, but due to their specialized and non-standard nature these have not been loaded into the CAS.

Unfortunately, the derived $\mathrm{log}g$ values—needed to discriminate giants from dwarfs—from these moderate-resolution spectra ($R\sim 2000$) were not reliable and the first two years of MARVELS targets resulted in a 35% giant fraction instead of the goal of 10%.

We thus employed a new method for giant-dwarf selection in Years 3 + 4. For this second phase of the MARVELS survey, temperature estimates were derived based on $V-K$ and $J-K$ colors following the infrared flux method of Casagrande et al. (2010), and giants were rejected based on a requirement of a minimum reduced proper motion (Collier Cameron et al. 2007) based on the measured 2MASS J-band proper motion together with the J-band magnitude and $J-H$ color.

From 2011 January onward all MARVELS observations were carried out simultaneously with APOGEE, using plug plates drilled with holes for both sets of targets. The spectroscopic cartridges were adapted to allow connection of both the APOGEE and MARVELS fibers to the long fibers that run to the stabilized rooms that house the respective instruments. This joint observation mode yielded significant overall observational efficiencies, but imposed the restriction that both surveys observe the same fields with the same cadence. This shifted the MARVELS target fields much farther south than originally planned as APOGEE pursued observations toward the center of the Milky Way.

The sky distribution of all observed MARVELS fields is shown in Figure 2.

The MARVELS instrument is designed to be sensitive to wavelength shifts (and thus RV changes) in stellar spectra. It splits each input stellar spectrum into two beams, and then projects a slanted interference pattern of the recombined beams through a spectrograph (see Figure 3).

Zoom In Zoom Out Reset image size

The dispersed slanted interference pattern effectively magnifies the resolution of a moderate-resolution spectrograph ($R\sim 11,000$) by translating wavelength shifts in the dispersion ("x") direction to much larger shifts in the "y" position. This slope is ∼5 pixel pixel⁻¹ for MARVELS. The design goal of the MARVELS analysis is to measure the shift of the interferometric sinusoid in the y direction to determine the wavelength offset due to a RV change.

The key challenges in the processing of MARVELS data are the calibration of the wavelength solution on the detector, identification and extraction of each spectrum, and the measurement of the slant of the interferometric comb and of the resulting interference pattern of the absorption-line features.

Our approach to analyzing the MARVELS data will be described in detail in N. Thomas et al. (2015, in preparation), which specifically describes the UF1D pipeline. The CCF+DFDI and UF1D pipelines follow many of the same steps, but differ in choices of calibration reference sources and complexity of model for instrumental variations. We here outline the important differences in the CCF+DFDI and UF1D processing.

3.4.1. Extraction of Spectra from the 2D Images

3.4.2. Compression to 1D Spectra

3.4.3. Characterizing the Instrumental Wavelength Drift

3.4.4. Measuring the Stellar RV Shifts

As Figure 4 shows, the current data processing results in stellar RV variations of 50 m s⁻¹ or larger even at high S/N, a value several times greater than that expected from photon statistics. This is mostly due to systematic uncalibrated wavelength shifts on timescales longer than a month; repeat observations of stars within the same lunation show much smaller RV variations. However, the figures show that some stars show rms radial velocity variations which approach the photon noise limit, suggesting that with proper calibration, the overall scatter should drop significantly. One possibility currently under investigation is that these stars represent specific fibers that are more stable, while the beams from other stars experienced greater hardware variation across repeated pluggings and fiber connections. Work continues on improving the analysis of the MARVELS data and our understanding of the long-term systematic effects.

Zoom In Zoom Out Reset image size

Despite these challenges, the MARVELS DR11 reductions have been used to study low-mass and substellar companions (Fleming et al. 2012; Wisniewski et al. 2012; Ma et al. 2013), brown dwarfs in the "desert" (Lee et al. 2011), and exotic orbital systems (Mack et al. 2013). Figure 5 shows MARVELS RV measurements of two stars with known exoplanets, showing that MARVELS data are in good agreement with existing orbital models for these systems.

Zoom In Zoom Out Reset image size

However, in general, the MARVELS data and analysis to date have not achieved the survey requirements for RV necessary to discover and characterize a fiducial 0.5-${M}_{\mathrm{Jupiter}}$ planet in a 100 day orbit. Figure 4 shows the achieved RV rms for the current pipelines as a function of stellar magnitude. The upper band of objects with rms from 1–10 km s⁻¹ is predominantly true astrophysical variation from binary star systems. The distribution of objects with rms values in the range of 100 m s⁻¹ is bounded near the photon limit, but the bulk lies several times above these limits.

The BOSS main survey of galaxies and quasars over two large contiguous regions of sky in the northern and southern Galactic Caps was completed in Spring 2014. The majority of the galaxies were uniformly targeted for large-scale structure studies in a sample focused on relatively low redshifts ("LOWZ," with $z\lt 0.4$) and a sample with $0.4\lt z\lt 0.7$ designed to give a sample approximately volume-limited in stellar mass ("CMASS"; B. Reid et al. 2015, in preparation). The total footprint is about 10,400 deg² (Figure 6); the value of 9376 deg² in Table 1 excludes masked regions due to bright stars and data that do not meet our survey requirements.

Zoom In Zoom Out Reset image size

The main BOSS survey was completed in 2014 February. The additional dark time available through the 2014 summer shutdown was devoted to a portfolio of additional science programs designed to maximize the science return while taking advantage of the unique abilities of the SDSS system. Two of the largest such programs were a variability study of 849 quasars, designed to measure time delays between continuum and emission line variations ("Reverberation Mapping"; Shen et al. 2015a), and an early start on the planned cosmological studies with SDSS-IV (the Sloan Extended QUasar, ELG and LRG Survey, hereafter "SEQUELS," where "ELG" stands for "Emission Line Galaxy" and "LRG" stands for "Luminous Red Galaxy"), together with an exploratory set of plates to investigate the requirements for studies of high-redshift ELGs and other aspects of SDSS-IV. These and other BOSS ancillary programs executed since the DR10 release are described in Appendix.

The DR11 and DR12 releases of BOSS data constitute increments of 35% and 47% in the number of spectra over DR10, respectively, processed using very similar pipelines. These increases were significant enough to warrant a new set of BOSS cosmological analyses for each of these releases. These key papers were one of the motivations for tagging a DR11 data set for later public release along with DR12. The cosmology analyses based on DR11 data include studies of isotropic galaxy clustering (Guo et al. 2015), anisotropic galaxy clustering (Samushia et al. 2014; Song et al. 2014; Sánchez et al. 2014; Gil-Marín et al. 2014a, 2014b; Beutler et al. 2014b; Reid et al. 2014), galaxy clustering in the LOWZ sample (Tojeiro et al. 2014), the baryon oscillations (BAO) in the clustering of the Lyα forest of distant quasars (Bautista et al. 2014; Delubac et al. 2015), the first detection of BAO in the cross-correlation between the Lyα forest and the quasars (Font-Ribera et al. 2014), an updated upper bound to the sum of neutrino masses (Beutler et al. 2014a), a summary BAO galaxy clustering analysis paper (Anderson et al. 2014b), and a joint cosmology analysis paper incorporating all of the BOSS cosmology constraints as well as those from Type Ia supernovae (SNe Ia) and anisotropies in the cosmic microwave background (Aubourg et al. 2014). The BOSS team plans a similar set of papers based on the full DR12 analyses.

The pipeline software for reduction of BOSS spectroscopic data was largely unchanged between DR10 and DR11. The classification and redshift-measurement aspects of this software are described in Bolton et al. (2012).

There were, however, some significant improvements to the spectrophotometric flux-calibration routine for DR12. These improvements were made to mitigate low-level imprinting of (primarily) Balmer-series features from standard-star spectra onto science target spectra. This imprinting was first documented in Busca et al. (2013) in observed-frame stacks of quasar continuum spectra. Although this effect is generally undetectable in any single-spectrum analysis, it has a small but non-negligible effect on the analysis of the Lyα forest across many thousands of quasar spectra. The change implemented for DR12 consists of a simple masking and linear interpolation of the flux-calibration vectors over the observed-frame wavelength ranges shown in Table 2. A more flexible flux-calibration vector model is retained at other wavelengths to accommodate real small-scale features in the spectrograph throughput. This more flexible model was necessary for the original SDSS spectrographs due to time variation in the dichroic filters, although it is likely unnecessary for the improved optical coatings on those surfaces in BOSS (see Smee et al. 2013).

Table 2.  Wavelength Ranges Masked During BOSS Spectrophotometric Calibration

Line	Wavelength Range
	(Å)
H	3888.07 ± 25
$[\mathrm{Ne}\;{\rm{III}}]$	3969.07 ± 30
Hδ	4100.70 ± 35
Hγ	4339.36 ± 35
Hβ	4860.09 ± 35

Note. Observed-frame vacuum wavelength ranges that were masked and linearly interpolated during determination of spectrophotometric calibration vectors.

Download table as:  ASCIITypeset image

In addition, we updated the pixel-response flats used to pre-process the spectrograph frames, improved the bias-subtraction code to catch and correct electronic artifacts that appear in a small number of frames, and updated the CCD bad-pixel and bad-column masks to reduce the incidence of corrupted but previously unflagged spectra. These changes reduce the number of corrupted spectra, and more accurately flag those that remain.

Table 3 gives the full history of significant changes to the BOSS spectrograph detectors and the calibration software to process its data since the BOSS survey began. See also Table 2 of Ahn et al. (2012) for additional changes to the hardware.

As in previous BOSS data releases, a unique tag of the idlspec2d spectroscopic pipeline software is associated with each unique sample of publicly released data.¹⁴² Three tagged reductions of three separate samples are being released at the time of DR12. One (v5_6_5) is the "DR11" version that defines a homogeneous sample of BOSS data taken through Summer 2013; this is the version used in the cosmological analyses described in Section 4.2 above. A second label (v5_7_0) defines the main DR12 BOSS cosmological survey at its point of completion. A third tag (v5_7_2) is associated with the several extra observing programs undertaken with the BOSS spectrographs in Spring 2014 following the completion of the main BOSS survey program (Section 4.1, Appendix). These data-release software versions are summarized in Table 4.

Many of the pipeline changes for the ancillary programs involved bookkeeping and special cases for plates drilled with either fewer or more flux-calibration stars. In addition the SEQUELS plates targeted ELGs at high redshift, so the upper redshift limit of the galaxy template fitting (Bolton et al. 2012) was extended from z = 1 to z = 2. Thus DR12 includes several thousand SDSS galaxy spectra with tabulated redshifts above z = 1.

The APOGEE DR11 data include twice as many stars and spectra as DR10 (53,000 more stars and 200,000 more spectra), analyzed with the same pipeline. The APOGEE DR11 data have been used in several papers, including a determination of distances to and chemical abundances of red-clump stars (Bovy et al. 2014; Nidever et al. 2014), mapping of the Galactic interstellar medium using diffuse interstellar bands measured along the line of sight to APOGEE stars (Zasowski et al. 2015), and an identification of new Be stars and their H-band line profiles (Chojnowski et al. 2015).

APOGEE DR12 represents a further year of data and thus includes another 46,000 stars and 240,000 spectra over DR11. It also uses the updated analysis pipeline described above. The sky coverage of the final APOGEE DR12, covering the bulge, disk, and halo of our Galaxy is shown in Figure 7. The additional observations of stars that already appeared in DR10 improve the S/N of these stars and also provide opportunities for studies of RV and other variations in the observed stellar spectra. Figure 8 demonstrates that we achieved our goal of S/N $\gt 100$ per half-resolution element for the APOGEE sample. Figure 9 shows the distribution of time baselines and the number of observations of each star.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

A succinct overview of the APOGEE survey was presented in Eisenstein et al. (2011) and a full summary will be given by S. Majewski et al. (2015, in preparation). The APOGEE spectroscopic data processing is described in Nidever et al. (2015). The pipeline for deriving atmospheric parameters and abundances from the spectra will be described by A. E. García Pérez (2015, in preparation). The spectra, stellar parameters, and abundances for DR11 and DR12 are described in Holtzman et al. (2015).

Figure 10 shows the observed distribution of the key stellar parameters and abundances for APOGEE DR12. Obtaining robust and calibrated values of ${T}_{\mathrm{eff}}$, $\mathrm{log}g$, and [M/H] along with individual abundances for 15 elements has required development of new stellar libraries (Zamora et al. 2015) and H-band spectral line lists (Shetrone et al. 2015). After describing these fits, we discuss a value-added catalog of red clump stars, then describe specific target classes of APOGEE stars that are new since DR10.

Zoom In Zoom Out Reset image size

In DR12, we provide the best-fitting values of the global stellar parameters, as well as individual elemental abundances for C, N, O, Na, Mg, Al, Si, S, K, Ca, Ti, V, Mn, Fe, and Ni.

The spectra are fit to models based on spectral libraries from astronomical observations combined with laboratory and theoretical transition probabilities and damping constants for individual species. The final measurements and associated uncertainties are calibrated to observations of stellar clusters, whose abundance patterns are assumed to be uniform.

The abundances are most reliable for stars with effective surface temperatures in the range 3800 K $\leqslant \;{T}_{\mathrm{eff}}\;\leqslant$ 5250 K. For cooler atmospheres (${T}_{\mathrm{eff}}\lt 3800$ K), the strengths of molecular transitions are increasingly sensitive to temperature, surface gravity, molecular equilibrium, and other physical details, giving rise to a greater uncertainty in the inferred abundances. Stars with warmer atmospheres (${T}_{\mathrm{eff}}\gt 5250$ K) or at low metallicity ([Fe/H] $\lesssim -1$) have weaker lines, making it more difficult to measure abundances.

This APOGEE data release also contains the DR11 and DR12 versions of the APOGEE red-clump (APOGEE-RC) catalog. Red clump stars, helium core-burning stars in metal-rich populations, are good standard candles, and thus can be used as a spatial tracer of the structure of the disk and the bulge. RC stars are selected using the $\mathrm{log}g$, [Fe/H], and near-infrared colors available for each APOGEE star. The construction of the DR11 APOGEE-RC catalog and the derivation of the distances to individual stars were described in detail by Bovy et al. (2014). The DR11 catalog contains 10,341 stars with distances accurate to about 5%, with a contamination estimated to be $\lesssim 7\%$.

The DR12 RC catalog applies the same selection criteria to the full DR12 APOGEE sample, but re-calibrates the surface gravities to a scale appropriate for RC stars; the standard DR12 surface-gravity calibration is not appropriate for RC stars. The calibration starting from the uncalibrated outputs of ASPCAP for surface gravity, $\mathrm{log}{g}_{\mathrm{uncal}.\mathrm{DR}12}$ is

$$\begin{eqnarray*}&&\mathrm{log}{g}_{\mathrm{RC}}=1.03\;\mathrm{log}{g}_{\mathrm{uncal}.\mathrm{DR}12}-0.370,\end{eqnarray*}$$

for $1\lt \mathrm{log}{g}_{\mathrm{uncal}.\mathrm{DR}12}\lt 3.8$ (outside of this range the $\mathrm{log}{g}_{\mathrm{RC}}-\mathrm{log}{g}_{\mathrm{uncal}.\mathrm{DR}12}$ correction is fixed to that at the edges of this range). The DR12 APOGEE-RC catalog contains 19,937 stars with an estimated contamination $\lesssim 3.5\%$ (estimated in the same manner as for the DR11 catalog, see Bovy et al. 2014).

Target selection for APOGEE was described in Zasowski et al. (2013). As with BOSS, the targets for APOGEE are dominated by uniformly selected samples designed to meet the key APOGEE science goals, but also feature additional ancillary programs to take advantage of smaller-scale unique science opportunities presented by the APOGEE instrument. The final distribution of 2MASS magnitudes and colors for all APOGEE targets are presented in Figure 11, both as observed, and corrected for Galactic extinction. Because many of the APOGEE target fields are at quite low Galactic latitudes, the extinction corrections can be quite substantial, even in the infrared.

Zoom In Zoom Out Reset image size

Some of the additional dark time from the early completion of the BOSS main survey was used for the existing APOGEE main program, and allowed the addition and expansion of several ancillary science programs. DR12 adds four additional ancillary target classes to those described in Zasowski et al. (2013) and extends two previous ancillary programs. We briefly describe these additions here:

Radial Velocity Monitoring of Stars in IC 348. The "Infrared Spectroscopy of Young Nebulous Clusters" (IN-SYNC) ancillary program originally observed the Perseus sub-cluster IC 348. Subsequent to those observations a set of stars was targeted for further follow-up to (1) search for substellar companions in bright field stars of all spectral types; (2) search for stellar and substellar companions around low-mass M stars; (3) search for pre-main-sequence spectroscopic binaries in IC 348; (4) study a newly identified Herbig Be object (HD 23478/BD+31 649) and (5) enhance the completeness of the IC 348 sample with 40 additional targets. These 122 stars are labeled with APOGEE_TARGET2 bit set to 18.

Probing Binarity, Elemental Abundances, and False Positives Among the Kepler Planet Hosts. This ancillary project observed 159 Kepler Objects of Interest (KOI; e.g., Burke et al. 2014), 23 M dwarfs, and 25 eclipsing binaries, at high cadence (∼21 observations), over a period of 8 months to study binarity, abundances, and false positives in the planet host sample. This project aims to detect stellar and brown dwarf companions of Kepler host stars, provide detailed abundances for several elements, and understand planet formation in binary systems. KOI targets were selected from the KOI catalog with ${H}_{\mathrm{Vega}}\lt 14;$ "eclipsing binary" targets were selected with $H\lt 13$, periods $\gt 5$ days, and classified as having a "detached morphology" as listed in the catalogs of Prša et al. (2011) and Slawson et al. (2011), plus two systems from Gaulme et al. (2013); and "M dwarf" targets were drawn from the catalog of Dressing & Charbonneau (2013) with ${T}_{\mathrm{eff}}\lt 3500$ K and $H\lt 14$. These 208 stars are labeled with APOGEE_TARGET2 bit set to 19.

Calibration of the Gaia-ESO Spectroscopic Survey Program. A sample of 41 stars was observed to provide improved calibration of stellar parameters in conjunction with the Gaia-ESO Survey¹⁴³ (Pancino & Gaia-ESO Survey consortium 2012). These observations are labeled with the setting of APOGEE_TARGET2 bit 20.

Re-Observation of Commissioning Bulge Stars to Verify Radial Velocity Accuracy. A set of 48 stars in the bulge of the Milky Way that had originally been observed during the early commissioning phase of the APOGEE instrument was re-observed to provide a verification of the APOGEE RV estimates. These observations are labeled with the setting of APOGEE_TARGET2 bit 21.

In addition, two previous ancillary programs were expanded in DR12. The IN-SYNC ancillary program (APOGEE_TARGET2=13) to study young stellar objects (YSO) in the Perseus molecular cloud (see Cottaar et al. 2014 and Foster et al. 2015 for more details) was expanded in DR12 to observe 2634 stars in the Orion A molecular cloud. The APOGEE ancillary program to observe Kepler stars for asteroseismology and stellar parameter calibration (APOGEE_TARGET1=27) proved extraordinarily useful (e.g., Epstein et al. 2014) and was folded into the main APOGEE target selection for DR12.

All new parallel ancillary target classes found in DR12 were given a priority lower than both the primary galaxy and quasar targets and previously approved ancillary programs. The targeted samples for these parallel ancillary programs are therefore not complete. We list these programs roughly in the order of the distance to the targets; in Table 5, we list them in bit order.

Characterizing Low-mass M Dwarfs Using Wide Binaries. M dwarf stars make up $\sim 70\%$ of the stars in the Galaxy by number and have lifetimes longer than the age of the universe. They are thus valuable tracers of the chemical and dynamical evolution of the Milky Way, but their complex spectra dominated by molecular bands make it difficult to determine their ages and metallicities. This program targets earlier-type binary companions to known M dwarf stars; these companions should share the same metallicity and age as the M dwarf but have atmospheres that are easier to interpret. These systems can be used to refine relations between M dwarf properties and spectral signatures (e.g., Stassun et al. 2008; Dhital et al. 2012).

Fibers denoted by the SPOKE2 target flag were assigned to candidate binary companions of spectroscopically confirmed low-mass stars in the Sloan Low-mass Wide Pairs of Kinematically Equivalent Stars (SLoWPoKES; Dhital et al. 2010, 2015) project. A previous ancillary program, the Low-Mass Binary Stars program (Dawson et al. 2013) consisted of systems with angular separations 65–180''. SPOKE2 extends that target sample to late-M spectral types, identifying binaries with separations between 3 and 20 arcsec. No proper motion requirement is imposed (Dhital et al. 2015). Targets have magnitudes in the range $17\lt {i}_{\mathrm{PSF}}\lt 21.3$.

A Census of Nearby Galaxies. We do not yet have a complete catalog of galaxies within 200 Mpc (Kasliwal 2011), hampering studies of nearby transients and the fine detail of the large-scale distribution of galaxies. The Palomar Transient Factory (PTF; Law et al. 2009) is performing a narrow-band survey in two filters, centered at 656 and 663 nm, to complete the catalog of galaxies in the local universe out to 200 Mpc. A sample of galaxies denoted by the PTF_GAL target flag was selected for spectroscopic confirmation. Galaxies without known redshift were observed if they had an SDSS counterpart, a color ${m}_{656}-{m}_{663}\gt 0.7$ mag, and relatively blue broadband color as measured by SDSS (${g}_{\mathrm{model}}-{i}_{\mathrm{model}}\lt 1.3$ mag). Images of all candidate galaxies were first visually inspected to avoid spurious detections.

Quasar Spectrophotometric Calibration. As described in Dawson et al. (2013) and Pâris et al. (2014), the fibers assigned to BOSS CORE and BONUS quasar targets (Ross et al. 2012) were offset in the focal plane to optimize throughput in the blue part of the spectrum, to better observe the Lyα forest. Because the standard stars are not observed with this same offset, the spectrophotometric calibration of these quasar targets is systematically incorrect. The QSO_STD flag denotes an additional sample of spectrophotometric standard stars, from 10 to 25 per plate spread evenly across the focal plane, that were drilled to follow the same offsets in the focal plane as the BOSS quasar targets. These objects are chosen using the same algorithm as for normal spectroscopic standard stars in BOSS, as explained in Dawson et al. (2013). Improved calibration gives improved measurements of quasar spectral energy distributions, important both for constraining quasar emission models, and for interpreting optical depth data in the Lyα forest (Lee et al. 2015).

Spectra of H₂O Maser Galaxies. One current route to the absolute calibration of the luminosities of SNe Ia as standard candles uses the 3%-accurate distance to NGC 4258 afforded by the well-studied H₂O maser in its center (Humphreys et al. 2013). Further improvements, by identifying other maser galaxies with supernovae, could decrease the uncertainty on local measurements of the Hubble Constant (Riess et al. 2011).

There is an apparent correlation between maser activity and host galaxy properties (Zhu et al. 2011); this correlation will be tested with spectroscopy of known maser host galaxies, and spectroscopy of SN Ia host galaxies will be used to identify plausible maser candidates. Targets, identified with the IAMASERS flag, were selected with no previous SDSS spectra and ${i}_{\mathrm{model}}\lt 20$ mag. Objects targeted in the Bright Galaxies ancillary program (Dawson et al. 2013) were also removed from the IAMASERS target list.

Spectroscopy of Massive Galaxy Cluster Members. This program aims to obtain redshifts of candidate member galaxies of X-ray selected clusters. The sources are optical counterparts to X-ray clusters selected as faint sources in the ROSAT All-Sky Survey (Voges et al. 1999, 2000a) identified by applying the redMaPPer (Rykoff et al. 2014) cluster finding algorithm to the position of an X-ray source. The X-ray magnitude limit corresponds roughly to the brightest 30% of clusters that the X-ray satellite eROSITA will find within the BOSS area.

Objects denoted by the target flag CLUSTER_MEMBER are selected from the redMaPPer catalog with ${i}_{\mathrm{cmodel}}\lt 19.9$ mag and ${i}_{\mathrm{fib}2}\lt 21.5$ mag. Roughly 1000 candidate clusters were observed.

Repeated Spectroscopy of Candidate Close Binary Massive Black Holes. Second-epoch spectroscopy was obtained for SDSS I/II quasars that are candidate massive black hole binaries with separations less than 1 pc. The quasars were selected from the DR7 quasar catalog (Schneider et al. 2010) based on having double-peaked broad Balmer lines or significant velocity offsets between broad and narrow line centroids. The SDSS-III spectrum, identified by the DISKEMITTER_REPEAT target class, is separated from the first SDSS-I/II epoch by multiple years and provides a test of binarity by observing changes in the emission line properties. These data should allow new constraints on the close massive black hole binary population in SDSS quasars and will provide a better understanding of the nature of these peculiar broad-line profiles.

Spectroscopy of Hard X-ray Identified Active Galactic Nuclei. This sample, identified with the XMMSDSS target class, was designed to spectroscopically confirm hard (2–10 keV) X-ray selected AGN identified in the serendipitous XMM survey of SDSS(Georgakakis & Nandra 2011). These objects tend to lie at relatively low redshift, $z\lt 0.8$. Objects identified by the XMMSDSS target class were selected with ${f}_{X}$(2–10 keV)$\gt 4\times {10}^{-14}$ erg cm⁻² s⁻¹and SDSS ${r}_{\mathrm{model}}\lt 22$ mag. There was of order one target per square degree. The cross-correlation measurement of those AGN with the SDSS and BOSS galaxy samples will constrain the dark matter halo masses of X-ray AGN as a function of redshift and luminosity.

WISE BOSS: BOSS spectra of Mid-IR bright AGN: Photometry from the WISE All-Sky Data Release catalog was used in combination with SDSS photometry to select a 12 μm-flux-limited sample of quasars that goes beyond the main BOSS quasar sample (Ross et al. 2012). This allows studies of the completeness of the main quasar sample and an exploration of dust obscuration of quasars. The WISE_BOSS_QSO target class was selected as having ${i}_{\mathrm{PSF}}\lt 20.2$, ${\rm{W}}1-{\rm{W}}2\lt 0.30$, ${\rm{W}}1\lt 2.0+0.667{g}_{\mathrm{PSF}}$, ${r}_{\mathrm{PSF}}-{\rm{W}}2\gt 2.0$, ${\rm{W}}2\lt 18.5$, ${\rm{W}}3\gt 12.5$, and, for extended objects, ${\rm{W}}3\gt 10.3$. The resulting sample peaks at redshift $z\approx 1.4$.

Quasar Target Selection with WISE: This was a second sample of WISE-selected quasars, focused on the redshift range $z\gt 2.15$. Candidate quasars, identified with the QSO_WISE_FULL_SKY target class, were identified from SDSS photometry using an artificial neural network as described in Yèche et al. (2010). Point sources are assigned a photometric redshift estimate and a likelihood (NN) ranging from zero (stellar) to one (quasar). Objects with ${NN}\gt 0.3$ were considered targets if they were matched within $1\buildrel{\prime\prime}\over{.} 5$ of a WISE source, had color ${i}_{\mathrm{PSF}}-{\rm{W}}1\gt 2.0+0.8({g}_{\mathrm{PSF}}-{i}_{\mathrm{PSF}})$ and ${i}_{\mathrm{PSF}}-{\rm{W}}2\gt 3.0$, and were brighter than ${g}_{\mathrm{PSF}}=21.5$ mag. These color cuts were designed to identify high-redshift quasars, and indeed almost 3/4 of the candidates have redshifts above 2. Objects satisfying this cut were assigned the QSO_WISE_FULL_SKY flag whether or not they were also targeted by the main BOSS quasar selection.

Quasar Pairs. Candidate quasar pairs separated by angles corresponding to less than a few hundred kiloparsecs were identified for spectroscopic confirmation. When combined with spectroscopy from other programs (e.g., Hennawi et al. 2006, 2010; Myers et al. 2008), this sample will provide a large statistical sample of quasar pairs necessary for small-scale clustering measurements. The target list consists of pairs of quasar targets selected using either the Kernel Density Estimation (KDE) method (Richards et al. 2009) or the XDQSOz method (Bovy et al. 2011). There are both low- and mid-redshift selection samples, both identified by the QSO_XD_KDE_PAIR target flag. The low-redshift selection includes targets with ${g}_{\mathrm{PSF}}\lt 20.85$ mag and a matching target from the same selection within an angular separation, θ, of $1\prime\prime \lt \theta \lt 30\prime\prime$. Objects are selected based on being in the XDQSOz low-redshift selection range ($0\lt z\lt 2.2$) with probability being a quasar, $\mathrm{PQSO}\gt 0.8$; or in the KDE catalog with flags indicating that the object is at low redshift and/or has an ultraviolet excess (lowzts=1 or uvxts=1, as described in Table 2 of Richards et al. 2009).

The mid-redshift selection includes XDQSOz targets with $\mathrm{PQSO}\gt 0.2$ that have a pair (from the same mid-z selection) within $1\prime\prime \lt \theta \lt 20\prime\prime$. These targets are further culled to only retain pairs for which the product of the two XDQSOz probabilities for the pair integrated over $2.0\lt z\lt 5.5$ is ${\mathrm{PQSO}}_{1}\times {\mathrm{PQSO}}_{2}\gt 0.16$. For both low- and mid-z selection, the following algorithm is implemented to clean the sample: (1) target all pairs where one or both of the objects in the pair are in a BOSS tiling overlap region; (2) for pairs where both objects are outside overlap regions, target the object with no existing spectrum; (3) for pairs where both objects are outside overlap regions and neither have existing spectra, target the fainter object; (4) discard all pairs where both objects are outside overlap regions and one of the pair is already a BOSS target; (5) discard all pairs where either object is a spectroscopically confirmed star or is obviously an artifact on visual inspection of the image; and (6) discard all targets (not pairs) that have an existing spectroscopic confirmation.

Because BOSS observations were proceeding ahead of schedule in 2012, a series of plates were added to the SDSS-III program to observe ancillary science programs. These plates do not have primary BOSS galaxy and quasar targets and instead consist entirely of ancillary science targets. The completeness of each dedicated sample is therefore typically higher than the completeness of the samples in the parallel ancillary programs. We describe each of these programs here, again sorted roughly by the distance of the targets. Table 6 summarizes the target categories, listed in order of ANCILLARY_TARGET2 bit. Note that a number of the programs include multiple target classes, each indicated by a separate bit.

Star Formation in the Orion and Taurus Molecular Clouds. This program obtained spectra of candidate YSO in the Orion and Taurus molecular clouds. The data provide a census of YSO into the brown dwarf regime, a measurement of the initial mass function at low masses, and a characterization of circumstellar disks as a function of stellar mass, extending previous studies to fainter magnitudes, to be sensitive to very low luminosity, low-mass objects. Objects were selected mostly from WISE photometry, as well as the 2MASS (Skrutskie et al. 2006) and Spitzer photometry (matching to SDSS imaging where available; Finkbeiner et al. 2004) in the Orion and Taurus regions. Objects were included to the detection limit of the WISE catalog, but those with ${\rm{W}}1\gt 7$ were removed to reduce contamination from luminous, very red, asymptotic giant branch stars. There are several target classes within this program, as detailed in Table 6.

The five plates in this program were designed in a heterogeneous manner due to the different availability of SDSS imaging in each field and the variation in the relative number of IR-excess sources. The latter is primarily related to the age of each star formation complex, as the circumstellar disk fraction decreases with stellar age. When limited SDSS photometry is available in a field, gri magnitudes are derived from the PPMXL/USNO-B1 catalog following the inverse of the transformations tabulated in Monet et al. (2003).

The 25 Ori spectroscopic plate targets WISE-detected stars within 1.5 degrees of the B3 star 25 Ori. It focuses on members of the young 25 Ori group and surrounding pre-main sequence stars in Orion and defines the target classes 25ORI_WISE and 25ORI_WISE_W3.

Objects were selected from the WISE catalog with detections in W1 and W3, with a magnitude limit of ${\rm{W}}3\lt 11.65$, and are assigned a target class of 25ORI_WISE_W3. Sources were required to be fainter than 15 in $g,r$, and i, and brighter than g = 22 and i = 21.

The remaining three Orion plates covering the Kappa Ori, NGC 2023, and NGC 2068 star formation regions were created in an identical manner and define the target classes KOEKAP_STAR, KOEKAPBSTAR, KOE2023_STAR, KOE2023BSTAR, KOE2068_STAR, and KOE2068BSTAR. For all three plates, objects in the *_STAR class are infrared excess sources selected by ${\rm{W}}1-{\rm{W}}2\gt 0$ and a S/N in W1 greater than 10. The *BSTAR objects are other WISE detections within the field.

The Taurus spectroscopic plate targets objects with Spitzer mid-infrared 8 and/or 24 micron excess within 1.5 degrees of the center of the Taurus Heiles 2 molecular cloud. Our sample for Taurus focuses on very low-mass substellar objects with disks and edge-on disks which may have been mistaken for galaxies. The selection used $\mathrm{IRAC}1-\mathrm{IRAC}4\gt 1.5$ and/or $\mathrm{IRAC}1-\mathrm{MIPS}24\gt 1.5$ mag with ${\rm{S}}/{\rm{N}}\gt 10$ for IRAC1 and ${\rm{S}}/{\rm{N}}\gt 7$ for IRAC4 or MIPS24. Here IRAC1, IRAC4, and MIPS24 refer to Vega magnitudes measured through filters centered at 3.5, 8.0, and 24 microns on Spitzer. All science objects on the Taurus plate have a target class of TAU_STAR.

Stars Across the SDSS. This project aims to cross-calibrate the large spectral surveys which are giving us a detailed map of the different stellar populations in the Milky Way. Dedicated stellar spectroscopic surveys such as SEGUE (Yanny et al. 2009), the RAdial Velocity Experiment (RAVE; Steinmetz et al. 2006), APOGEE, the Gaia/European Southern Observatory Survey (GES; Gilmore et al. 2012) and the massive Gaia survey itself (de Bruijne 2012) provide kinematic information and chemical diagnostics for large samples of stars. In addition, there are over 250,000 BOSS spectra of stars (Table 1), mostly targeted as quasar candidates. Derived stellar parameters, such as effective temperatures, surface graft's, and metallicities must be robust and consistent between surveys to use them jointly to build a coherent picture of our Galaxy. Because each survey targets a particular magnitude range, one must be careful to minimize systematic errors in stellar parameters as a function of distance.

This program obtained BOSS spectra of stars observed by the SEGUE1 and SEGUE2 surveys on eight plates (target classes SEGUE1, SEGUE2), GES targets in eight plates (target class GES), and one plate dedicated to stars from the COnvection, ROtation, and planetary Transit mission (CoRoT; Baglin 2002) also observed by GES and APOGEE (target classes COROTGES, COROTGETAPOG). As many CoRoT and GES stars were given fibers as possible, restricted only with the bright magnitude limit of $i\gt 14$ to avoid saturation in the spectrographs. There were not enough targets to fill all the fibers on the BOSS plates, particularly when the GES fields did not overlap with SEGUE-1 or SEGUE-2 plates, so the eight GES plates also targeted stars selected from the SDSS photometry (SDSSFILLER) with the following selection cuts to ensure good S/N and to avoid very cool stars for which it is more difficult to obtain accurate stellar parameters with the SSPP: $0\lt g-r\lt 1.25,g\lt 19,i\gt 15,r\gt 15$. The CoRoT plate had targets chosen from APOGEE (target class APOGEE) and 2MASS (2MASSFILL) as well. Stars were targeted to sample the full parameter space of effective temperature, metallicity, and $\mathrm{log}g$, as much as possible. The GES project (Milky Way survey) targeted stars with $0\lt J-K\lt 0.7,12.5\lt J\lt 17.5$, with near-infrared photometry from the Visible and Infrared Survey Telescope for Astronomy (VISTA; Emerson et al. 2004). In total, the eight GES plates gave spectra with high enough S/N for acceptable SSPP parameters for 296 stars with $-0.25\lt g-r\lt 1.5$ and $14\lt g\lt 19$.

A Galaxy Sample Free of Fiber Collisions. The finite size of the BOSS fiber ferrules means that no two fibers can be placed closer than 62'' apart on a given plate. These "fiber collisions" affect measurements of the small-scale clustering of galaxies from the CMASS and LOWZ samples. CMASS and LOWZ galaxies that were not observed in the main BOSS survey due to fiber collisions with other primary targets were added to ancillary target plates 6373–6398 (North Galactic Cap), 6780–6782 (on Stripe 82), 6369, and 6717. Fibers were also assigned to CMASS and LOWZ targets that suffered redshift failures (ZWARNING_NOQSO $\gt \;0$; Bolton et al. 2012) in previous observations in the data reduction pipeline. These objects are identified with the CMASS or LOWZ target flags in the database; unlike all other objects discussed in this Appendix, they are not assigned a target class in ANCILLARY_TARGET2. This program significantly increases the completeness of these galaxy samples in the region covered by these plates and provides a useful data set for testing the fiber-collision correction methods that are currently used in BOSS clustering analyses (e.g., Guo et al. 2012). A total of 1282 targets were included in this program. These data have been used in an analysis of velocity bias in close pairs of galaxies by Guo et al. (2015).

Quantifying BOSS Galaxy Incompleteness with a WISE-Selected Sample. The CMASS sample is designed to select red galaxies of high stellar mass (${M}_{\mathrm{stellar}}\gt {10}^{11}\;{M}_{\odot }$). This program (target class WISE_COMPLETE) aimed to explore a broader range of galaxy colors in the CMASS redshift range ($0.45\lt z\lt 0.7$), using optical-IR cuts by combining SDSS and WISE. The sample criteria are $17.5\lt i\lt 19.9$, $(r-{\rm{W}}1)\gt 4.165$, and ${i}_{\mathrm{fib}2}\lt 21.7$ (the latter uncorrected for Galactic extinction). Various quality flag cuts were imposed to limit spurious sources. Stars were eliminated using the SDSS morphological classifications for blue objects and a color–color cut in ($r-i$, $r-{\rm{W}}1$) space for red objects. A random subsample of 90% of these objects were selected as targets to meet the required target sky density.

Exploring $z\gt 0.6$ LRGs from SDSS and WISE. WISE and SDSS photometry was used to identify a sample of $z\gt 0.6$ luminous red galaxies, taking advantage of the fact that the 1.6 μm bump in old stellar populations (due to a local minimum in the opacity of the H⁻ ion) is redshifted into the WISE W1 band. This spectroscopic sample will be used to calibrate photometric redshifts in this range and to test target selection techniques for eBOSS.

Targets for this program were divided into a higher priority sample denoted HIZ_LRG and a lower priority sample denoted LRG_ROUND3. All objects were required to have

$$\begin{eqnarray*}({i}_{\mathrm{model}}\lt 20.0\ | | \ {z}_{\mathrm{model}}\lt 20.0)\ \&\&({z}_{\mathrm{fib}2}\lt 21.7\ | | \ {i}_{\mathrm{fib}2}\lt 22.0).\end{eqnarray*}$$

Objects in the HIZ_LRG sample were selected to have

$$\begin{eqnarray*}(r-i)\gt 0.98\;\&\&\;(r-{\rm{W}}1)\gt 2(r-i)-0.5.\end{eqnarray*}$$

The LRG_ROUND3 sample used the same $r-{\rm{W}}1$ cut, but the $(r-i)$ color cut was bluer, $(r-i)\gt 0.85$, in order to explore a broader range of galaxy colors.

Tests of eBOSS Target Selection in CFHTLS W3 Field. As a test of target selection algorithms to be used in eBOSS, six plates were dedicated to a selection of LRG and quasars at high density over a region of sky overlapping the Canada–France–Hawaii Telescope Legacy Survey (CFHTLS¹⁴⁹ ) W3 imaging footprint.

Targets selected as potential galaxies in the redshift range $0.6\lt z\lt 0.9$ were denoted FAINT_HIZ_LRG. These objects were selected in a similar manner to the targets that were assigned the HIZ_LRG flag described above, but at fainter magnitudes with a new tuning of color cuts. Targets were required to have

$$\begin{eqnarray*}&&20\lt z\lt 20.5,({z}_{\mathrm{fib}2}\lt 22.2{| | i}_{\mathrm{fib}2}\lt 22.5),\\ &&\quad (r-i)\gt 0.98,(r-{\rm{W}}1)\gt 2(r-i).\end{eqnarray*}$$

Quasar targets, assigned the QSO_EBOSS_W3_ADM target class, were selected from photometry from CHFTLS, SDSS, and WISE, and variability data from PTF. Five selection techniques were applied, and all assigned the same target bit.¹⁵⁰ These selection criteria were as follows.

• •  
  Bit 0: W3 color box selection. These objects were selected from the CFHTLS W3 co-added catalog available at the TeraPix CFHT website.¹⁵¹ The objects were restricted in CFHT magnitudes to $g\lt 22.8$. Stars were excised with the following color cuts (using CFHT photometry):

  $$\begin{eqnarray*}&&(g-r)-0.5(u-g)\lt -0.2| | \\ &&(g-r)+0.7(u-g)\lt 0.6.\end{eqnarray*}$$

  The targets were required to be classified as point sources by SDSS and to have SDSS r magnitudes in the range $17\lt r\lt 22$.
• •  
  Bit 1: SDSS XDQSOz selection. These objects were selected using the XDQSOz selection of Bovy et al. (2012) based on SDSS photometry. Point sources with $17\lt r\lt 22$ were required to have an XDQSOz probability of being a quasar greater than 0.2.
• •  
  Bit 2: SDSS-WISE selection. This program used WISE forced photometry at SDSS source positions (Lang 2014; Lang et al. 2014). A stacked flux was created in SDSS gri (m_opt; with a relative (g, r, i) weighting of ($1,0.8,0.6$)), and a stacked flux was created in WISE W1 and W2 (${m}_{{wise}};$ with (W1, W2) relative weights of $(1,0.5)$). Objects were selected with $17\lt {m}_{\mathrm{opt}}\lt 22$, $(g-i)\lt 1.5$, and ${m}_{\mathrm{opt}}-{m}_{{wise}}\gt (g-i)+3.0$. Extended sources were allowed; the sample was restricted to sources with a difference between SDSS PSF and model magnitudes less than 0.1.
• •  
  Bit 3: CFHTLS Variability selection. Using three years of repeated observation in the one square degree field D3 of CFHTLS, objects were selected based on the variability measured in their light curves. Objects were selected on ${\chi }^{2}$ and structure function parameters A and Γ (Palanque-Delabrouille et al. 2011) averaged over the three bands gri. Using colors $c1$ and $c3$ defined as in Fan (1999): $c1\equiv 0.95(u-g)+0.31(g-r)+0.11(r-i)$ and $c3\equiv -0.39(u-g)+0.79(g-r)+0.47(r-i)$, two selections were applied. The first used only CFHT information, requiring $A\gt 0.08$, ${\chi }^{2}\gt 10.0$, ${\rm{\Gamma }}\gt 0.3$, $c3\lt 0.6-0.33c1$, and $g\lt 23.0$. The second used both CFHT and SDSS, and required that $A\gt 0.08$, ${\chi }^{2}\gt 10.0$, ${\rm{\Gamma }}\gt 0.2$, $g\lt 22.0$, and that the object be classified as point-like by SDSS.
• •  
  Bit 4: PTF variability selection. Using light curves computed from PTF R-band imaging linked to SDSS r with a color correction, quasar candidates were again selected by variability. All structure function or color-term parameters are defined as above (cf., Bit 3). The objects were required to have $A\gt 0.05$, ${\chi }^{2}\gt 10.0$ and ${\rm{\Gamma }}\gt 0.1$. In addition, the objects were limited to $g\lt 22.5$ and had to pass either of the following two criteria based on SDSS photometry: a color and magnitude cut $r\gt 18$ and $c3\lt 1.0-0.33c1$, or a color and morphology cut requiring the object to be classified as point-like by SDSS and to have a probability of being a quasar greater than 0.1 according to the XDQSO algorithm.

eBOSS ELG Target Selection with Deep Photometry. This program used deep photometric data to select ELG candidates, to assess algorithms for eBOSS. Photometry extending to fainter limits than SDSS was used to assess algorithms for selection of Emission Line Galaxies (ELG) for spectroscopic observations. In particular, blue star-forming galaxies in the redshift range $0.6\lt z\lt 1.2$ were selected from the CFHTLS Wide W3 field photometric redshift catalog T0007¹⁵² (Ilbert et al. 2006; Coupon et al. 2009). Targets with the FAINT_ELG target class were selected at a density of nearly 400 objects per square degree, and three plates were observed centered on the same position. The sample was defined to help evaluate the completeness of the targeting sample and redshift success rates near the faint end of the ELG target population.

Selected objects satisfied the constraints:

$$\begin{eqnarray*}&&20\lt g\lt 22.8,-0.5\lt (g-i)\lt 2\;\mathrm{and}\\ &&\quad -0.5\lt (u-r)\lt 0.7(g-i)+0.1.\end{eqnarray*}$$

All photometry was based on CFHTLS MAG_AUTO magnitudes on the AB system. Objects with known redshift were excluded. These data are described in Comparat et al. (2015), which measured the evolution of the bright end of the [O ii] emission line luminosity function.

The TDSS/SPIDERS/eBOSS Pilot Survey. This program carried out pilot observations in two fields for two components of the SDSS-IV eBOSS survey: TDSS and SPIDERS (Section 7). The first field encompasses the existing XMM-Newton Large Scale Survey (XMM-LSS), deep multi-band CFHTLS field imaging, and a Pan-STARRS1 (PS1; Kaiser et al. 2002, 2010) medium deep survey field (MD01) with hundreds of epochs. The second field is also a PS1 medium deep field (MD03) located in the Lynx/IfA Deep Field. Both fields have 3–4 times as many PS1 epochs as does SDSS Stripe 82 (Annis et al. 2014), and PS1 continued monitoring these fields at the time the BOSS spectroscopy of these plates was carried out. There were five target selection algorithms on these plates, as follows:

Objects with the TDSS_PILOT target class were selected from PS1 photometry calibrated as described in Schlafly et al. (2012). Targets were selected by variability within each of the gri filters, with the requirement of a median PS1 magnitude $17\lt {\mathrm{mag}}_{x}\lt 20.5$ and at least 30 observed epochs within that filter. Objects were required to be point-like in SDSS, with the difference between PSF and model magnitude less than 0.05 in each filter, and with no detectable proper motions.

Lightcurves for objects that pass a variability threshold in at least one filter following Kim et al. (2011) were visually inspected in all three filters.

We assign each object a priority based on the number of passed criteria summed over filters, the source brightness, and whether or not a BOSS spectrum already exists.

Objects identified TDSS_PILOT_PM were selected the same way, but this identifier marks objects with significant ($\gt 3\sigma$) total proper motion as measured by SDSS.

Objects identified TDSS_PILOT_SNHOST showed transient behavior in extended objects in the PS1 medium deep photometry, as described in Chornock et al. (2013).

Objects identified SPIDERS_PILOT were selected as X-ray sources with clear optical counterparts in SDSS DR8 imaging. The X-ray selection was performed on a source catalog constructed from public XMM-Newton data in the XMM-LSS area following the procedure described in Georgakakis & Nandra (2011). The sample was flux-limited in soft X-rays (0.5–2 keV) to the expected limit of the eROSITA deep field survey ($\sim 6\times {10}^{-15}$ erg cm⁻² s⁻¹), and were required to have $17\lt {r}_{\mathrm{PSF}}\lt 22.5$ and not to have been spectroscopically observed by BOSS as of DR9. Objects with higher soft X-ray flux were given higher priority in fiber assignment.

Objects targeted by both the SPIDERS and TDSS algorithms were given higher priority and were assigned the TDSS_SPIDERS_PILOT target class.

Follow-up spectroscopy of wide-area XMM fields. Like the SPIDERS program above, this program targeted X-ray-selected AGN from the XMM-XXL field, now using the full range of sensitivity from 0.5 to 10 keV. SDSS optical counterparts to X-ray sources were identified via the maximum-likelihood method (Georgakakis & Nandra 2011). The main spectroscopic target sample was selected to have ${f}_{X}$(0.5–10 keV) $\;\gt \;{10}^{-14}$ erg cm⁻² s⁻¹and $15\lt r\lt 22.5$, where r is the PSF magnitude in the case of optical unresolved sources or the model magnitude for resolved sources. Targets in this sample are denoted XMM_PRIME. Secondary targets are sources with ${f}_{X}$(0.5–10 keV) $\;\lt \;{10}^{-14}$ erg cm⁻² s⁻¹and $15\lt r\lt 22.5$, or radio sources selected in either 325 or 610 MHz from the catalog of Tasse et al. (2008). These targets are denoted XMM_SECOND.

Multi-Object Reverberation Mapping. The broad emission lines in AGN spectra can have flux variations correlated with variation in the continuum, but with a time delay interpreted as the mean light-travel time across the broad-line region. Measuring this time delay ("reverberation mapping") allows one to study the structure and kinematics of the broad-line regions of AGN. 849 spectroscopically confirmed quasars were observed over 30 epochs to study the variability of this sample.

The observations were scheduled with a cadence of four to five days, as weather allowed, with a goal of five epochs per month between 2014 January and the end of 2014 June. The typical exposure times were 2 hr for this program, and thus the final data from this program comprise a 60 hr effective exposure time for the targets in this field. The survey is described in Shen et al. (2015a).

Previous spectroscopy of the PS1 Medium Deep Field MD07 $(\alpha ,\delta )=(213\buildrel{\circ}\over{.} 704,+53\buildrel{\circ}\over{.} 083)$ provided redshifts of roughly 1200 quasars in the redshift range $0\lt z\lt 5$ over the area of a single plate. The sample was limited to quasars with $i\lt 21.7$. Lower-redshift quasars (whose time delay should be easier to measure) were given higher priority, and are indicated with the RM_TILE1 target class; essentially all of these were assigned a fiber. Higher-redshift targets (RM_TILE2) were tiled with the remaining fibers.

Three plates containing identical science targets were drilled at varying hour angle to ensure that the field was visible for six months. Each plate was given the normal number of sky fibers (80) but was allocated a substantially larger number of standard star fibers (70 rather than 20) to allow more rigorous tests of spectrophotometric calibration.

Early science results from these data include measurements of the velocity dispersions of the host galaxies of low-redshift quasars from the high S/N co-added spectra (Shen et al. 2015b), rapid trough variability in broad absorption line quasars (Grier et al. 2015), and the structure functions and time delays of a number of quasars.

Variability-selected Quasars at $1\lt z\lt 4$ to g = 22.5. The QSO_VAR_LF bit labels a target class designed for studies of the quasar luminosity function to $g\lt 22.5$. The sample is located in Stripe 82 at $36^\circ \lt \alpha \lt 42^\circ$ where multi-epoch SDSS photometry is available, thus enabling a variability selection with the neural network presented in Palanque-Delabrouille et al. (2011). Targets with point-like morphology that passed a loose variability criterion were selected (neural network threshold of 0.5, where 1/0 indicates a quasar-like/stellar-like light curve). Extended sources which satisfied the color selection $c3\lt 0.6-0.33c1$, where $c1$ and $c3$ are linear combinations of SDSS ugriz bands as defined in Fan (1999), were targeted if they passed a tighter variability criterion (threshold of 0.9).

Note that targets previously spectroscopically identified as quasars were not included in the sample and therefore do not have the QSO_VAR_LF bit set, even if they pass the selection criteria for this program.

Faint End of the Quasar Luminosity Function. Targets that have the QSO_DEEP bit set used the same variability selection as for QSO_VAR_LF, but were selected in the range $22\lt g\lt 23.5$ from SDSS Stripe 82 data.

Slightly extended objects with $-0.15\lt ({r}_{\mathrm{PSF}}-{r}_{\mathrm{model}})\lt 0.15$ were selected to a neural network threshold of 0.9.

Additional targets were included in the sample when they had a large probability of being a quasar according to the KDE (Richards et al. 2009). Unresolved objects with $\mathrm{KDE}(1.0\lt {\rm{z}}\lt 2.2)\geqslant 0.999$ or slightly resolved objects with $-0.05\lt ({r}_{\mathrm{PSF}}-{r}_{\mathrm{model}})\lt +0.05$ and $\mathrm{KDE}({\rm{z}}\gt 2.2)\geqslant 0.985$ were included. Targets previously spectroscopically identified as quasars were not included in the sample and therefore do not have the QSO_DEEP bit set even if they pass the selection criteria.

Finally, a sample of candidate Lyman-Break Galaxies was selected in color-space and assigned the LBG bit. These targets are slightly extended objects that lie in one of two color box regions: $0\lt (g-r)\lt 0.15$ && $(u-g)\gt (g-r)+0.2$, or $0\lt (g-r)\lt 1.0$ && $(u-g)\gt (g-r)+1.25$.

SDSS-III Observations of LOFAR Sources. This ancillary program was intended to target radio sources identified in deep observations of the ELAIS-N1 region by the Low Frequency Array (LOFAR; van Haarlem et al. 2013). LOFAR observations were planned with the high-band antenna (HBA: 110–250 MHz) for roughly 10 hours over 9 deg² to eventually reach an rms depth of 100 μJy at 150 MHz. Spectroscopic confirmation of these sources will provide insight into the nature of the LOFAR radio population and aid in the science exploitation of new radio surveys. The LOFAR ELAIS-N1 region is well-studied by optical surveys and contains deep Jansky Very Large Array (JVLA) and Giant Metre-Wave Radio Telescope (GMRT) imaging data near the center of the field.

The LOFAR sample goes considerably deeper near the center of the spectroscopic field, concentrating the targets there and making it impossible to assign sky fibers uniformly over the focal plane. Instead, there were a large number of fiber bundles that did not contain a sky fiber and the usual sky interpolation routine in the automated BOSS reductions could not be applied to the four plates designed for this program. For these plates, the data reduction pipeline was modified to apply a constant sky model across each spectrograph (i.e., fibers 1–500 and 501–1000, respectively). This results in larger sky residuals than the typical calibrated BOSS spectra. With this in mind, users of these data should treat the automated redshift classification and narrow emission lines with caution.

All LOFAR radio sources were matched to SDSS optical counterparts found within 2'' of the radio source position. The SDSS position was used for the fiber placement. The target classes selected for this program are as follows:

ELAIS_N1_LOFAR targets were selected from a preliminary image of the ELAIS-N1 HBA data (115 to 190 MHz) that reached an rms noise level of 333 μJy. Approximately 800 sources were detected to a threshold of 1650 μJy and an additional 400 sources were detected to a threshold of 1000 μJy. These sources are distributed over a field of radius approximately 3° for a total surface area of roughly 30 deg⁻². In addition, 387 fainter LOFAR sources that could be clearly identified by eye in the ELAIS-N1 field were targeted.

ELAIS_N1_FIRST sources lacked a detection by LOFAR but appeared in the catalog of the Faint Images of the Radio Sky at Twenty cm (FIRST) survey (Becker et al. 1995), and had an SDSS optical counterpart with ${r}_{\mathrm{model}}\lt 23.0$. Fibers were placed at the SDSS position.

ELAIS_N1_GMRT_GARN sources were identified from deeper GMRT data at 610 MHz (rms depth of 40–70 μJy) from the Garn et al. (2008) source catalog. These sources are expected to be dominated by AGN.

ELAIS_N1_GMRT_TAYLOR targets were also selected from GMRT data (Taylor et al. 2014), which are even deeper (rms depth of 10 μJy) than that used in the ELAIS_N1_GMRT_GARN sample. The deep GMRT radio catalog includes 2800 sources over 1.2 deg². The positional accuracy from the radio data appears to be better than 05.

ELAIS_N1_JVLA sources were also selected to be much fainter than the other samples. The deep JVLA radio catalog includes 483 sources over 0.13 deg² at an angular resolution of 25 and rms noise of 1 μJy (Taylor et al. 2014). The positional accuracy is similar to the ELAIS_N1_GMRT_TAYLOR sample. Both this sample and the ELAIS_N1_GMRT_TAYLOR sample should include a significant fraction of star-forming galaxies at $z\lt 1$.

SEQUELS serves both as a pilot program for the eBOSS survey of SDSS-IV and as a stand-alone science program within SDSS-III. SEQUELS also encompasses two SDSS-IV sub-programs to obtain spectra of variability-selected objects and X-ray detected objects, which are pilot studies for the TDSS and SPIDERS programs within eBOSS described in Section 7.

The main SEQUELS footprint lies in the North Galactic Cap. Targets were selected over the region covering $120^\circ \lt \alpha \lt 210^\circ$ and $45^\circ \lt \delta \lt 60^\circ$ within the nominal BOSS footprint, but only 300 deg² of this area were observed. The targets in the primary SEQUELS program have the SEQUELS_TARGET bit set in the ANCILLARY_TARGET2 bitmask. Plates that were drilled but not observed before DR12 will be observed as part of eBOSS.

SEQUELS targets fell into four broad categories, which we describe in detail below: (1) luminous red galaxies (LRG), designed to extend the BOSS CMASS redshift coverage, yielding a median redshift of ∼0.72; (2) quasars both as direct tracers of the cosmic density field at redshifts $0.9\lt z\lt 2.2$, and as probes of the Lyα forest; (3) X-ray targets as a SPIDERS precursor, and (4) variability-selected targets as a TDSS precursor. Several other target classes do not fall neatly into any of these categories and are listed at the end.

In addition, SEQUELS incorporated a pilot program to identify high-redshift ELGs. The ELG targets are listed with the ANCILLARY_TARGET2 bitmask (Table 6). The bitmasks for all other SEQUELS programs are listed in Table 7 and are described in detail in what follows. Note that some of these bits (such as bit 0, DO_NOT_OBSERVE) do not indicate programs per se, but rather give information about the target selection process.

A.3.1. LRGs in SEQUELS

A.3.2. Quasars in SEQUELS

The main sample of SEQUELS quasars is assigned the QSO_EBOSS_CORE target class and is designed to meet the eBOSS sky density goal of ∼70 $0.9\lt z\lt 2.2$ quasars deg⁻².

The target selection makes no attempt to filter out higher-redshift quasars, so objects from this sample will also be useful for Lyα forest studies. Quasars in the CORE are selected by a combination of XDQSOz (Bovy et al. 2012) in the optical and a WISE-optical color cut, as detailed in A. Myers et al. (2015, in preparation); see also the description of bit 1 and 2 of the QSO_EBOSS_W3_ADM target class above. This sample (and all the SEQUELS quasar candidates which follow, unless otherwise indicated) are restricted to objects classified as point sources, with faint-end magnitude cuts of $g\lt 22$ or $r\lt 22$.

We also selected quasars via their variability as measured by the PTF; these are given the target class QSO_PTF. This sample is less uniformly selected, given the availability of multi-epoch PTF imaging, but that is acceptable for Lyα forest studies. These objects are limited in magnitude to $r\gt 19$ and $g\lt 22.5$.

Targets that have the QSO_EBOSS_KDE bit set in SEQUELS consisted of all objects from the KDE catalog of Richards et al. (2009) that had uvxts=1 set (indicating that they had a UV excess, and thus were likely to be at $z\lesssim 2.2$) within that catalog. Only KDE objects that matched to a point source in the DR9 or the custom SDSS photometry used to select SEQUELS targets were included.

The QSO_EBOSS_FIRST bit indicates quasars that are targeted in SEQUELS because there is an SDSS source within 1'' of a source in the 2013 June 05 version¹⁵³ of the FIRST point source catalog (Becker et al. 1995).

An object is flagged QSO_BOSS_TARGET if it has been previously observed by BOSS and does not have either LITTLE_COVERAGE or UNPLUGGED set in the ZWARNING bitmask (see Table 3 of Bolton et al. 2012). Similarly, an object from SDSS DR8 is flagged QSO_SDSS_TARGET if it is included in the SDSS DR8 spectroscopic database, and similarly has neither of those flags set in ZWARNING.

We separately flagged those quasars with QSO_KNOWN whose spectra had been visually confirmed, as listed in the SDSS sample used to define known objects in BOSS (see Ross et al. 2012), and a preliminary version of the DR12 BOSS quasar catalog of I. Paris et. al. (2015, in preparation).

As part of SEQUELS, we also re-observed a number of high-redshift ($z\gt 2.15$) quasars that had low S/N spectroscopy in SDSS DR7 or BOSS, to improve the measurement of the Lyα forest.

Objects flagged QSO_REOBS had $0.75\leqslant {\rm{S}}/{\rm{N}}/\mathrm{pixel}\lt 3$ in BOSS. This target class also included objects which have a high probability of being quasars based on their photometry, but had no signal in the BOSS spectra because of dropped fibers or other problems.

In the same spirit, BOSS spectra of some objects are of low enough quality that their classification as quasars, or measurements of their redshifts, are uncertain upon visual inspection. Such objects are designated as QSO? or QSO_Z? in the DR12 quasar catalog (I. Paris et. al. 2015, in preparation). Those objects in the SEQUELS footprint are re-observed, and given the QSO_BAD_BOSS target class. A preliminary, but close-to-final version of the DR12 catalog was used to define this sample for SEQUELS.

We set a flag bit, DO_NOT_OBSERVE, to indicate which previously observed quasars should not be re-observed, even if they were selected by one of the SEQUELS algorithms. It is determined by the following combination of target flags:

$$\begin{eqnarray*}&&({\text{}}{QSO}\_{KNOWN}| | {\text{}}{QSO}\_{BOSS}\_{TARGET}| | {QSO}\_{SDSS}\_{TARGET})\\ \quad {\text{}}\&\&\\ &&\quad {\text{}}!({QSO}\_{BAD}\_{BOSS}| | {QSO}\_{REOBS}).\end{eqnarray*}$$

SEQUELS targeted quasars were selected in both the DR9 imaging used for BOSS and an updated DR12 imaging calibration intended for use in eBOSS targeting. The DR9_CALIB_TARGET bit signifies quasars that were selected for SEQUELS using the DR9 imaging calibrations instead of (or as well as) the updated DR12 imaging.

A.3.3. SPIDERS Targets within the SEQUELS Program

A.3.4. TDSS Targets within the SEQUELS Program

A.3.5. Other Target Classes in SEQUELS

Galaxies from the main BOSS target selection, both LOWZ and CMASS, that were not assigned fibers due to fiber collisions were observed in SEQUELS and given the target class SEQUELS_COLLIDED. Observing these galaxies in SEQUELS creates large contiguous areas that have 100% spectroscopic completeness in the final BOSS data sample. A similar sample was described in Section A.2.

Variable targets selected from the PTF survey are targeted with the SEQUELS_PTF_VAR target class in three classes: hosts of supernovae detected in the PTF supernova program, RR Lyrae stars, and additional sources whose light curve built from PTF data show variations by 0.4 mag or more.

Emission-line galaxy candidates tend to have blue colors and thus are relatively bright in the u band. The South Galactic Cap U-band Sky Survey¹⁵⁴ (SCUSS) was carried out over the SEQUELS area using the 2.3 m Bok Telescope at Kitt Peak to obtain deeper data ($u\approx 23$ for $5\;\sigma$ detections of point sources) than SDSS (X. Zhou et al. 2015, in preparation; H. Zou et al. 2015, in preparation). We used these data together with SDSS $g,r,i$ photometry to select ELGs in the redshift range $0.4\lt z\lt 1.6$ in a region of the sky of 25.7 deg² around $(\alpha ,\delta )\sim (23^\circ ,20^\circ$).

The brightest and bluest galaxy population (SEQUELS_ELG) is selected by

$$\begin{eqnarray}-0.5\lt u-r\lt 0.7(g-i)+0.1\;\&\&\;20\lt u\lt 22.5\end{eqnarray}$$

To fill the remaining fibers we also observed targets satisfying broader color cuts (SEQUELS_ELG_LOWP):

$$\begin{eqnarray*}(20\lt u\lt 22.7\;\&\;-0.9\lt u-r)\\ \quad \&(u-r\lt 0.7(g-i)+0.2| | u-r\lt 0.7).\end{eqnarray*}$$