eBOSS Quasar Target Selection

The eBOSS quasar target selection algorithms are described in Myers et al. (2015), together with early results of applying these algorithms to SEQUELS. Further details about variability-selected eBOSS quasar targets are provided in Palanque-Delabrouille et al. (2016). The eBOSS quasar target selection IDL code is available in full in the SDSS-IV SVN Software Repository.

Overview

As noted in Dawson et al. (2016), cosmological models project that eBOSS can achieve a 2% constraint on the cosmological distance scale at redshift ~1.5 by detecting the baryon acoustic oscillation (BAO) feature in the clustering of 58 deg-2 quasars at redshifts of 0.9 < z < 2.2 spread over an area of 7500 deg2. In addition, quasars at z > 2.1 can be used to improve critical high-redshift BAO constraints via clustering in the Lyman-α Forest.

In view of this, the main or CORE eBOSS quasar sample targets quasars at z > 0.9 over the entire eBOSS footprint. CORE quasars are targeted uniformly, such that they can be used as direct clustering tracers. Above z > 2.1, the CORE sample of quasars can also be used for studies of clustering in the Lyman-α Forest. A sample of quasars selected using variability of sources in Palomar Transient Factory (PTF) imaging (see also Palanque-Delabrouille et al. 2016) is targeted to supplement Lyman-α quasars in the CORE sample. PTF quasars do not need to be selected uniformly to study the BAO, as Lyman-α quasars can be used merely as distant beacons to illuminate unassociated large-scale structure at high redshift.

Previously known targets

The eBOSS CORE quasar sample does not re-target known sources that have believable spectra, confident classifications, and good measurements of redshift. This is because if a potential target is known to either not be a quasar, or known to be a quasar with as good a redshift measurement as can be plausibly obtained with the SDSS-III spectrographs (Smee et al. 2013), then eBOSS already has as much information as it needs or can acquire for that source.

eBOSS quasar targeting defines an object to be “known” if it was observed as part of SDSS-I, II or III. A target is considered to have been observed as part of SDSS-I or II if it appears in the SDSS Data Release 8 line-by-line parallel spectroscopy and imaging catalogs. An object is considered to have been observed as part of SDSS-III if it appears in the circa May 30, 2014 BOSS SpAll file. A “known” object in these files is considered to have a believable spectrum if neither of the LITTLE_COVERAGE or UNPLUGGED bits are set in the ZWARNING mask (see also Table 3 of Bolton et al. 2012). A known object with a believable spectrum is considered to have a confident classification and a good redshift if it is included in the twelfth data release of the SDSS quasar catalog (Paris et al. 2014) and is not labeled QSO? or QSO_Z? in that catalog.

CORE quasar targets

Requirements for the density of other eBOSS targets constrained the eBOSS quasar samples to be selected from no more than 115 deg-2 fibers. Choosing not to target previously known SDSS targets released ~25 deg-2 fibers. This means that the eBOSS CORE sample is designed to recover at least 58 deg-2 0.9 < z < 2.2 quasars from an average of ~90 deg-2 fibers. These targets are intended to be selected in a reproducible and relatively statistically homogeneous fashion, such that they can be used as direct tracers of large-scale structure.

The eBOSS CORE sample is initially selected from objects in SDSS imaging that meet the following criteria:

  • SURVEY_PRIMARY must be set in the RESOLVE_STATUS bitmask
  • Bits 3-9 (BAD_ROTATOR to NOISY_CCD) must not be set in the IMAGE_STATUS bit mask
  • Targets must be unresolved in imaging (objc_type==6)
  • Targets must be brighter than g < 22 OR r < 22 (using PSF magnitudes corrected for Galactic extinction)
  • Targets must be fainter than i > 17 in FIBER2MAG magnitudes

Next, targets are run through the XDQSOz algorithm of Bovy et al. (2012). This version of Extreme Deconvolution takes ugriz fluxes in the SDSS filter system and returns a probabilistic estimate of whether an object is a quasar in any range of redshift. The eBOSS CORE sample uses a probabilistic cut of PQSO(z > 0.9) > 0.2.

Finally, a cut in mid-IR-optical color is used to remove contaminating stars. The mid-IR imaging is taken from the 3.4 and 4.6 micron bands of the WISE survey (Wright et al. 2010), which are typically referred to as W1 and W2. Weighted optical and mid-IR stacks are created in flux space:

  • fopt = (fg + 0.8fr + 0.6fi)/2.4
  • fWISE = (fW1 + 0.5fW2)/1.5

and then targets are accepted as part of the eBOSS CORE sample if their magnitudes adhere to mopt – mWISE ≥ (g – i) + 3.

Ultimately, then, the eBOSS CORE sample comprises the union of the following two criteria:

  • PQSO(z > 0.9) > 0.2
  • mopt – mWISE ≥ (g – i) + 3

which were chosen to achieve an average target density of ~90 deg-2.

Lyman-α quasar targets and the supplemental PTF sample

The eBOSS CORE sample already contains Lyman-α quasars, as eBOSS targeting makes no attempt to limit the upper redshift range of CORE targets. The vast majority of the ~25 deg-2 eBOSS fibers that remain available after CORE quasar targets are assigned are reserved for supplemental Lyman-α targets selected using Palomar Transient Factory (PTF) imaging (Law et al. 2009; Rau et al. 2009).

The eBOSS PTF sample is initially selected from objects in SDSS imaging that meet the following criteria:

  • SURVEY_PRIMARY must be set in the RESOLVE_STATUS bitmask
  • Targets must be unresolved in imaging (objc_type==6)
  • Targets must be brighter than g < 22.5 and fainter than r > 19 (using PSF magnitudes corrected for Galactic extinction)
  • Targets must be fainter than i > 17 in FIBER2MAG magnitudes

Light curves based on SDSS r-band imaging and customized stacks of PTF R-band imaging are then constructed and fit with a structure function model of variability of the form

  • σ2(Δm)=[A(Δtij)γ]2 +(σi2 + σj2)

and sources are selected as PTF targets in eBOSS if they meet both of the following criteria:

  • γ > 0
  • γ > -30A + 1.5

Sources are further rejected based on how the χ2 of this model compares to a fit in which flux does not vary with time. This χ2 is allowed to change across the sky as the PTF and SDSS imaging quality changes. Note that Lyman-α quasars selected as part of eBOSS are intended to be used to illuminate the (unassociated) Lyman-α Forest in their foreground, and so there is no requirement that they be selected in a homogeneous fashion.

Finally, SDSS PSF magnitudes are used to calculate color contours, as in Fan (1999), in order to reject the “fundamental plane” of SDSS stars and low-redshift galaxies:

  • c1 = 0.95(u – g) + 0.31(g – r) + 0.11(r – i)
  • c3 = -0.39(u – g) + 0.79(g – r) + 0.47(r – i)

eBOSS PTF quasar targets are then restricted to objects that meet the following criteria:

  • c3 < 1.4 – 0.55c1
  • c3 < 0.3 – 0.1c1

In addition, r < 20.5 objects are rejected as eBOSS PTF quasar targets if they fall in the following region of (c1, c3) space:

  • 0.85 < c1 < 1.35
  • c3 > -0.2

This region tends to be populated by variable stars that can mimic quasar variability in sparsely sampled light curves.

Other Target Classes

The few remaining eBOSS quasar fibers per square degree are used to improve Forest signal in known BOSS Lyman-α quasars, and to target radio-selected quasars to search for bright, Lyman-α quasars that may have been color outliers in u – g and so missed by the BOSS radio-selected quasar sample.

Confirmed BOSS quasars are reobserved by eBOSS if they lie in the eBOSS footprint and have 0.75 < SNR pixel-1 < 3 or if they have SNR pixel-1 = 0. Here, SNR pixel-1 is the mean signal-to-noise per Lyman-α Forest pixel calculated over the rest-frame wavelength range of 1040-1200 Angstroms.

eBOSS also targets all SDSS point sources that are within 1″ of a radio detection in the 13 June 05 version of the FIRST point source catalog (e.g. Becker et al. 1995). The radio-selected eBOSS sample undergoes the same initial cuts as the CORE sample (PRIMARY, good IMAGE_STATUS, r or g < 22 etc.)

Results from SEQUELS and projected BAO constraints

The selections outlined above were guided by the requirements of an average total fiber allocation of 115 deg-2 fibers and of recovering at least 58 deg-2 0.9 < z < 2.2 quasars in a uniform fashion. These were the projected desiderata in order to constrain the BAO feature near redshift z~1.5 to ~2%. A superset of these selections was applied as part of the SEQUELS survey, allowing the ultimate expected characteristics of the eBOSS quasar target sample to be inferred.

Applying the CORE quasar selection to SEQUELS resulted in a density of 72.0 deg-2 quasars spectroscopically confirmed to be at 0.9 < z < 2.2, once the survey area was weighted for spectroscopic completeness. The design of the eBOSS survey footprint means that fibers can be assigned to roughly 95% of eBOSS targets. This implies that the eBOSS CORE selection will identify (0.95 × 72.0 =) 68.4 deg-2 0.9 < z < 2.2 quasars. About 80% of these 0.9 < z < 2.2 quasars will be new discoveries in eBOSS, with about 20% having been identified in previous iterations of the SDSS.

Similarly, the eBOSS CORE algorithm should identify ~95 deg-2 quasars in total at any redshift. About 70% will be new discoveries, with about 30% having been identified in previous iterations of the SDSS. The higher fraction of previously known quasars at redshifts of z > 2.2 reflects the extensive targeting of quasars in this redshift regime by BOSS. SEQUELS implies that, for Lyman-α quasars at z > 2.1, the eBOSS CORE selection should newly discover ~6-7 deg-2 quasars, and the eBOSS PTF selection should newly discover an additional ~3-4 deg-2 quasars.

The 68.4 deg-2 0.9 < z < 2.2 quasars that we expect to comprise the eBOSS CORE sample easily meets the requisite 58 deg-2 such quasars needed to obtain a 2% constraint on the BAO at redshift z~1.5. Zhao et al. (2016) project that the eBOSS CORE sample of clustering quasars will therefore be large enough to constrain the BAO at z~1.5 to ~1.6%.

Quasar targeting bits

Ultimately, the following bits are set in the EBOSS_TARGET1 bitmask by the eBOSS quasar targeting algorithm (see the wider description above and/or Myers et al. 2015 for more details):

Bit 10: QSO_EBOSS_CORE: Targets that meet the eBOSS CORE criteria. These targets are intended to be uniformly selected for clustering studies. CORE quasars At 0.9 < z < 2.2 can be used to constrain the BAO at z~1.5. CORE quasars at z > 2.1 can also be used for Lyman-α Forest studies.

Bit 11: QSO_PTF: Targets that have been selected based on their variability over time in SDSS and Palomar Transient Factory imaging. PTF quasars are not homogeneously selected, and are intended for Lyman-α Forest studies to aid BAO measurements at z > 2.

Bit 12: QSO_REOBS: Quasars that were previously confirmed in BOSS that are of reduced (but not prohibitively low) signal-to-noise. These targets are intended to increase signal in the Lyman-α Forest in order to improve BAO constraints at z > 2.

Bit 14: QSO EBOSS FIRST: Targets that are detected as point sources in SDSS imaging and that are within 1″ of a radio detection in the 13 June 05 version of the FIRST point source catalog (e.g Becker et al. 1995).

Bit 15: QSO_BAD_BOSS: Known quasars that have uncertain classifications or redshifts upon visual inspection. Such objects are designated as QSO? or QSO_Z? in the twelfth data release of the SDSS quasar catalog ( DR12QSO; Paris et al. 2014). The purpose of this bit is to ensure that ambiguous BOSS quasars are always reobserved, regardless of which other targeting bits are set. Prior to the eboss6 tiling, a close-to-final but preliminary version of the DR12QSO catalog was used to define this sample.

Bit 16: QSO_BOSS_TARGET: Targets that have a sufficiently high quality spectrum from BOSS (Data Release 9 or after) that they should not need to be reobserved. “Sufficiently high quality” in this context means that any such targets that are quasars should already have been correctly identified as quasars.

Bit 17: QSO_SDSS_TARGET: Targets that have a sufficiently high quality spectrum from SDSS-I/II (Data Release 8 or before) that they should not need to be reobserved. “Sufficiently high quality,” here, has the same meaning as for QSO_BOSS_TARGET.

Bit 18: QSO_KNOWN: This bit is used to track which previously known objects have a good visually inspected (or otherwise highly confident) redshift and classification from prior SDSS quasar catalogs. This bit is purely informational, in the sense that eBOSS does not use it to change how fibers are assigned to targets. The eBOSS quasar targeting algorithm determines whether to reobserve a known object on the basis of whether the spectrum of that object is of sufficient quality to identify a quasar. Whether a spectrum should be of sufficient quality to identify a known quasar is recorded in the QSO_BOSS_TARGET and QSO_SDSS_TARGET bits. As with QSO_BAD_BOSS, the QSO_KNOWN bit was set using a preliminary version of the DR12QSO catalog prior to the eboss6 tiling.

Bit 0: DO_NOT_OBSERVE: The bits QSO_REOBS, QSO_BAD_BOSS, QSO_BOSS_TARGET, QSO_SDSS_TARGET and QSO_KNOWN are used in concert to determine a sample of objects for which eBOSS does not need to obtain an additional spectrum. In general, a target is intended to not be reobserved if a previous spectrum of the target should have been good enough to identify it as a quasar, and for Lyman-α quasars from BOSS, if that spectrum had sufficiently high signal-to-noise in the Lyman-α Forest. In Boolean notation, DO_NOT_OBSERVE is set if:

(QSO_KNOWN || QSO_BOSS_TARGET || QSO_SDSS_TARGET) && !(QSO_BAD_BOSS || QSO REOBS)

REFERENCES

Becker, R. H., et al., 1995, ApJ, 450, 559

Bolton, A. S., et al. 2012, AJ, 144, 144

Bovy, J., et al., 2012, ApJ, 749, 41

Dawson, K. S., et al. 2016, AJ, 151, 44

Fan, X., 1999, AJ, 117, 2528

Law, N. M., et al., 2009, PASP, 121, 1395

Myers, A. D., et al. 2015, ApJS, 221, 27

Paris, I., et al., 2017, A&A, 597, 79

Palanque-Delabrouille, N., et al. 2016, A&A, 587, 41

Rau, A., et al., 2009, PASP, 121, 1334

Smee, S. A., et al. 2013, AJ, 146, 32

Zhao, G., et al., 2016, MNRAS, 457, 2377