A SPECTROSCOPIC SURVEY OF WISE-SELECTED OBSCURED QUASARS WITH THE SOUTHERN AFRICAN LARGE TELESCOPE

A complete understanding of the evolution of galaxies requires a detailed exploration of the role played by active galactic nuclei (AGNs), highly luminous objects found in the centers of galaxies that are powered by accretion onto a supermassive black hole. The most powerful AGNs, quasars, have been proposed to influence the interstellar gas content and temperature throughout a galaxy, which may lead to reduced star formation (SF; for recent reviews, see Fabian 2012; Alexander & Hickox 2012). The line-of-sight orientation between the central supermassive black hole and any obscuring dust is thought to produce the array of different observed AGNs (Antonucci 1993; Urry & Padovani 1995). This "unification theory" separates AGNs into two broad classes: obscured AGNs, where nuclear and galaxy-scale dust blocks emission from near the central black hole, and unobscured AGNs, where there exists a line of sight down to the central engine. It is far more straightforward to detect unobscured AGNs, while the full population of obscured AGNs is not fully characterized. Historically, unobscured AGNs have been selected by observing broad emission lines emitted from excited gas near the accreting supermassive black hole, and are often called Type I AGNs to differentiate them from Type II AGNs whose spectra only contain narrow emission lines. The difficulty in finding obscured AGNs is important given the evidence for an evolutionary link between obscured and unobscured AGNs. One of the primary proposed methods for fueling a quasar is through major mergers of gas-rich galaxies, where gas is funneled into the central regions of the galaxy (Kauffmann & Haehnelt 2000; Hopkins et al. 2006, 2008). It is thought that quasar fueling can also lead to an obscuration of the UV and optical emission from the accretion process by the in-falling gas and dust, which would eventually be lifted by quasar radiative feedback, setting up a large-scale connection between the supermassive black hole and galaxy-wide SF (Chen et al. 2014). Understanding the full population of obscured AGNs is vital for exploring both the cosmological growth of supermassive black holes as well as the connection to their host galaxies and larger-scale galaxy environments.

Deep X-ray surveys using both XMM-Newton and Chandra have revealed extensive samples of AGNs with a wide luminosity range out to large cosmic distances (Mainieri et al. 2002; Alexander et al. 2003; Mateos et al. 2005; Brandt & Alexander 2010; Xue et al. 2011). AGN population synthesis models, however, require a substantial number of obscured AGNs to explain the shape of the cosmic X-ray background spectrum (Comastri et al. 1995; Gilli et al. 2007; Treister et al. 2009; Ballantyne et al. 2011; Shi et al. 2013), and many of these objects have been missed in existing X-ray surveys (Worsley et al. 2005; Tozzi et al. 2006; Alexander et al. 2008; Burlon et al. 2011). A complimentary approach is to target obscured AGNs selected at mid-infrared (mid-IR) wavelengths, where obscuration effects are minimized and reprocessed light from dust near the central supermassive black hole produces a characteristic red IR power-law spectrum. Lacy et al. (2004) and Stern et al. (2005) described robust color criteria for separating AGNs from star-forming galaxies using the IR photometric bands targeted with the Infrared Array Camera (IRAC) instrument on the Spitzer Space Telescope, which was further explored in Donley et al. (2008, 2012) and Mendez et al. (2013). Hickox et al. (2007) studied a large sample of obscured and unobscured objects in the Boötes field selected to be AGNs by their IRAC colors. They demonstrated a bimodality in optical to mid-IR color that they used to separate obscured and unobscured sources. Lacy et al. (2013) presented an optical and near-IR spectroscopic study of a large sample of 786 objects selected with Spitzer IRAC colors indicative of AGN activity and demonstrated the power in using Spitzer selection techniques to target AGNs and obscured quasars even out to z ∼ 2–3. However, IRAC-based studies like these are limited to only the relatively small fields observed by Spitzer. Recently, the all-sky mid-IR photometric coverage from the Wide-field Infrared Survey Explorer (WISE; Wright et al. 2010) has also been used by many authors to separate active galaxies from star-forming galaxies. WISE covered the entire sky at four wavelengths: 3.4, 4.6, 12, and 22 μm (referred to as W1, W2, W3, and W4 throughout this work), reaching 5σ point-source sensitivities better than 0.08, 0.11, 1, and 6 mJy in each photometric band, respectively. This unprecedented all-sky coverage makes WISE a powerful tool for selecting large numbers of obscured active galaxies for follow-up studies.

There has been a wealth of research into the selection of AGNs by their WISE colors. Assef et al. (2010), Ashby et al. (2009), and Stern et al. (2012) described WISE selection criteria that use only the W1−W2 color to separate AGNs from star-forming galaxies. Stern et al. (2012) used WISE data from the COSMOS field to demonstrate that a WISE color cut of W1 − W2 > 0.8 is successful at recovering 78% of Spitzer-selected AGNs. A less conservative color cut, such as W1 − W2 > 0.5 suggested by Ashby et al. (2009), was shown to increase the completeness of the AGN sample, but the recovered sample suffered from more contamination from star-forming galaxies. Mateos et al. (2012) used a mid-IR-selection method which relies on three of the WISE photometric bands: W1, W2, W3. Using X-ray selected AGNs from the Bright Ultrahard XMM-Newton Survey, the authors demonstrated that the completeness of their MIR-selection method is strongly dependent on AGN luminosity, especially as compared to the luminosity of the host galaxy.

While many of the methods used to define WISE obscured AGN selection rely on exploring the WISE colors of large existing samples of active galaxies, it would be of interest to test these selection methods using a blind survey of obscured objects selected solely by their colors. By performing follow-up spectroscopy, we can test obscured AGN selection criteria, and explore both the type of objects and the redshift distribution of objects with red IR colors. Surveys of this nature are useful for understanding the full population of WISE-selected obscured quasars, as it is necessary to employ the large statistical sample of WISE quasars to study spatial clustering (e.g., Donoso et al. 2013, DiPompeo et al. 2014) as well as the role that AGNs play in galaxy evolution. Additionally, the all-sky coverage of WISE can be used to target rare objects of which there are few in existing spectroscopic fields.

In this paper, we present long-slit optical spectroscopy obtained with the Southern African Large Telescope (SALT) for a sample of 40 objects selected with WISE and Sloan Digital Sky Survey (SDSS) colors indicative of obscured quasar activity. We describe the full sample and our selection criteria in Section 2, examine the optical spectra and redshift distribution for the objects in Section 3, and in Section 4 we discuss the use of various optical emission line diagnostics to examine the physical characteristics of objects that comprise the sample. We present the results of modeling the spectral energy distributions (SEDs) of the sample with a combination of AGN and star-forming galaxy templates in Section 5. We examine individual objects in our sample in Section 6, compare the selection criteria we used to other recent WISE selection criteria in Section 7, and finally, we draw conclusions in Section 8. Throughout, we assume a standard ΛCDM cosmological model with H₀ = 71 km s⁻¹ Mpc⁻¹, Ω_M = 0.27, and Ω_Λ = 0.73 (Komatsu et al. 2011).

The sample described in this paper consists of 40 galaxies observed in three queued campaigns between November of 2012 and March of 2014 (SALT Proposals 2012-2-DC-002, 2013-1-DC-003, 2013-2-DC-003; PI: R. Hickox). For the purposes of exploring WISE and SDSS obscured quasar selection criteria, we chose objects in two different ways. In order to ensure maximum visibility with SALT, we restricted our possible targets to −2° < decl. <2° for the first campaign, and −2° < decl. <0° for the second campaign. Importantly, in both campaigns, all targets have g < 22⁴ in order to increase the likelihood of obtaining a spectrum with high signal to noise with SALT. We further limit the sample to g > 20 to probe higher redshifts and avoid bright objects with existing SDSS spectroscopy. The ramifications of this optical selection criteria are described in Section 3. While the majority of the objects in our sample were chosen for optical followup due to the absence of an existing SDSS spectrum, both J0325−0032⁵ and J0924+0027 were selected so that the SALT spectrum could be compared to the SDSS spectrum, as discussed in Section 3.

Our first selection group consists of all objects with SDSS and WISE coverage with WISE color W1 − W2 > 0.7,⁶ a slightly more relaxed criteria for selecting quasars from what is presented in Stern et al. (2012). In practice, while only one object, J1417+0108, has W1 − W2 = 0.73, the rest have W1 − W2 > 0.8. In order to select objects with the brightest mid-IR emission, we further restricted the possible candidates to only those objects with 7 ⩾ W4 ⩾ 6.5. These relatively rare IR-bright objects are comparable to the Spitzer 24 μm bright objects explored in Lacy et al. (2013). We acquired SALT spectroscopy for 21 objects selected under these criteria, which we refer to as "Group 1."

For the second campaign, we started with only those objects with SDSS and WISE coverage with WISE color W1 − W2 > 0.8, and W4 ⩾ 7.0. We further selected objects with r_AB − W2_AB > 3.1, the criteria introduced in Hickox et al. (2007) that targets obscured quasars. We acquired SALT spectroscopy for 19 objects selected under these criteria, which we refer to as "Group 2." If we only explore the overlapping range in decl. used in selecting our targets for both Group 1 and Group 2 (−2° < decl. <0°) across all R.A.s, we find 144 objects qualify as Group 1 objects, while 1270 objects qualify as Group 2 objects, due largely to the relaxation of the W4 photometry criteria for Group 2. Additionally, in this same decl. range, there are 77 objects that would be selected as Group 1 objects with g > 22, while there would be 8166 Group 2 objects with g > 22, which represents a significant number of candidate obscured quasars that are missed due to our optical photometric cuts. For the full decl. range used to find Group 1 objects (−2° < decl. <2°), there are 277 objects that would be selected under the Group 1 criteria, while an additional 145 objects would be selected under these criteria but with g > 22.

We plot the WISE W1−W2 and W2−W3 colors for both objects in both Group 1 and 2 in Figure 1 along with AGN selection criteria taken from the literature. We also plot the SDSS-selected Type II quasars (with WISE W1, W2, and W3 detections with signal-to-noise ratio (S/N) >3.0) from Reyes et al. (2008) for comparison. The objects in our sample cover a wide range in colors. To indicate the WISE colors of star-forming galaxies as a function of redshift from z = 0–2, we also plot tracks showing the Assef et al. (2010) star-forming ("Sbc") and irregular ("Im") templates, as well as SEDs for the star-forming galaxies M82 and Arp220 from Polletta et al. (2007). We further discuss the implications of our selection criteria in Section 7. The objects in our sample do not have Chandra coverage with any instrument, and while there is XMM-Newton coverage for fields containing the objects J1505+0134 and J1129+0102,⁷ the objects were not detected. We also note that 8 of our 45 objects have detections within 2'' in the Very Large Array (VLA) Faint Images of the Radio Sky at Twenty-Centimeters (FIRST) survey (Becker et al. 1995): J1007+0146 (1.27 mJy integrated 21cm flux, 1.24 mJy beam⁻¹ peak flux), J0852+0137 (3.96 mJy, 3.59 mJy beam⁻¹), J0957−0120 (2.78 mJy, 2.20 mJy beam⁻¹), J1151−0046 (2.85 mJy, 3.16 mJy beam⁻¹), J1333−0126 (1.25 mJy, 1.51 mJy beam⁻¹), J1406−0049 (3.20 mJy, 3.16 mJy beam⁻¹), J1417+0108 (1.38 mJy, 1.64 mJy beam⁻¹), and J1554+0011 (6.15 mJy, 5.78 mJy beam⁻¹). These detections could be caused by both SF as well as AGN activity (Ivezić et al. 2002).

Zoom In Zoom Out Reset image size

The objects were selected using photometry from SDSS Data Release 9 (Ahn et al. 2012) as well as the WISE All-Sky Source Catalog. Recently, the WISE team released the AllWISE Source Catalog (Cutri 2013, made public 2013 November), which combined three WISE surveys: the 4-band Cryogenic Survey (the primary WISE mission that covered the full sky 1.2 times between 2010 January and August), the 3-band Cryogenic Survey (a survey using only the first three WISE wavebands, which covered 30% of the sky between 2010 January and August), and the NEO-WISE post-cryogenic survey (a survey using the first two WISE wavebands, which covered 70% of the sky between 2010 September and 2011 February). Most importantly for the purposes of the current study, the AllWISE Data includes updated photometry more sensitive in the W1 and W2 bands, with updated 5σ point-source sensitivities at 0.054 and 0.071 mJy, respectively (Cutri 2013). For the remainder of this paper, we will report the updated, more accurate AllWISE photometry for the objects in our sample. The average fluxes in the W1 and W2 band for the full selection of objects are 600 and 900 μJy, respectively. For the majority of the objects, the updated AllWISE photometry does not affect whether or not these objects would be selected, although there are a few objects which would move into or out of our selection boxes. The Group 1 object J1417+0108 has updated photometry with WISE color W1−W2 <0.7, which does not satisfy our initial color cut. Similarly, the Group 1 objects J1151−0046 and J1129+0102 have W4-band photometry that excludes them from selection under Group 1 criteria. Finally, J0338−0049 and J1609−0004, two Group 2 objects, have updated color r_AB − W2_AB < 3.1. We further discuss these targets in Section 6. For the W1, W2, and W3 photometry used in our selection, all but one of our objects has AllWISE S/N on the photometry above 3.0 (J1609−0004 has S/N_W1 = 2.9). We only use the observed photometry in the W4 band if the object was observed with S/N >3.0. We report the object identifiers, group number, SDSS and WISE AllWISE Source Catalogue photometry, and the IR and optical-to-IR colors used to select the objects in Table 1.

In order to better explore how the change in the WISE photometry affected the selection of targets, we explored all of the objects that would be selected as Group 1 and Group 2 objects in the range 40° < R.A. <185°, −2° < decl. <0° with both WISE All-Sky and AllWISE photometry. For Group 1, 49 total objects would be selected using both the All-Sky and AllWISE photometry, and an additional five objects moved into the selection criteria and 14 moved out of the selection criteria based on the updated AllWISE photometry, primarily because of updated AllWISE photometry with W4 > 7 or W4 < 6.5 for individual objects. For Group 2, 323 total objects would be selected using both the WISE All-Sky and AllWISE photometry, and an additional 56 objects moved into the selection criteria and 28 moved out of the selection criteria based on the updated AllWISE photometry, due to the larger number of criteria used for selecting an object in Group 2. These results demonstrate the increased sensitivity of the updated AllWISE data, and we reiterate that except where noted above, the objects in our sample would be selected to be a part of Group 1 or Group 2 using either the AllWISE or WISE All-Sky photometry.

The full sample of objects were observed using the Robert Stobie Spectrograph (RSS; Kobulnicky et al. 2003; Smith et al. 2006) on SALT in long-slit mode, with a 15 slit and the PG0900 grating, which provides a spectral resolution of 5 Å at a central wavelength of 5000 Å (in this observational set-up, the spectral range was 3600–6000 Å). The seeing was measured using stars observed in the acquisition images taken just before the observations, and was typically ∼2'' for the runs. The slit positions were chosen to coincide with a nearby bright source for ease of object acquisition. While the majority of the objects comprising the sample were observed for 2250 s, for a small number of objects with brighter or dimmer optical photometry, the exposure time was changed accordingly, leading to a full range of observing times between 1950 and 3000 s. Data reduction was performed using standard IRAF scripts.⁸ The data were gain-corrected, bias-corrected, and mosaicked with the SALT reduction pipeline. We subsequently flat-fielded the data, applied a wavelength solution using arc lamp spectra, and then used the three observations made for each target for cosmic ray removal. In our final step, the two-dimensional spectra were background subtracted and combined using a median combine. The spectra were extracted using a 27 aperture centered on the target.

We also observed standard stars in order to perform relative flux calibration on the data and correct for the sensitivity of the SALT CCD. Unfortunately, due to the fixed nature of the primary mirror at the SALT telescope and the strong variation in effective aperture with time and source position, it is not possible to perform absolute flux calibration using the SALT data alone. We will report emission line flux ratios for a subsample of the objects in Section 4. We report the details of the spectroscopic observations, including observation date and total exposure time, in Table 2.

With the assembled, final, reduced spectra, our first task was estimating the spectroscopic redshift of each object, and initial identification of emission lines was done by eye for each object. The strongest optical lines we used to constrain the redshift were Mg ii λ2798, [Ne v] λ3427, [O ii] λλ3726, 3729, [Ne iii] λ3869, Hβ, and [O iii] λλ4959, 5007. Additionally, the two highest-redshift objects in our sample were identified using strong UV emission lines. For the majority of the objects (29/40), we were able to identify two or more emission lines in the spectrum which we used to constrain the redshift, while for J1255−0010 there was only one strong emission line which we identified as [O ii] λλ3726, 3729. Due to a strong observed 4000 Å break to the red of the emission line in the spectrum for this object, we consider the redshift estimate for J1255−0010 to be confident. For five other objects that have a single emission line, we used the fact that there were no other strong emission lines in the spectrum to infer that the single line was most likely Mg ii λ2798, which was also observed to be broad in one of the objects (see Section 6.2 for a discussion of this object). For the remaining objects (5/40), there were no features in the spectrum allowing for accurate redshift identification. One of these objects, J1417+0108, does have a FIRST radio detection, however. In light of this, while we calculated the statistical properties of the sample using all of the estimated redshifts, we will also present statistics calculated without the five objects with a single emission feature.

Redshifts were estimated from the centroids measured using the IRAF task splot to fit Gaussian curves to the emission lines in each spectrum. In order to characterize the systematic error on our redshift estimates introduced due to wavelength calibration we compared SALT RSS spectra observed in a different campaign (as described in Hainline et al. 2013) to the measured SDSS spectra for these objects, and found that the average difference was on the order of ∼0.3 Å. We can also compare the emission line measurements for the two objects in our sample for which SDSS DR9 spectra were taken: J0325−0032 (z = 0.353) and J0924+0027 (z = 0.660). For J0325−0032, the best-fit centroids for the strongest lines [O ii] λλ3726,3729 and [Ne iii] λ3869 agree to within ∼3 Å, while for J0924+0027 the best-fit centroids for the more broad, low S/N emission line [Ne v] λ3427 agrees to within ∼8 Å.

In Group 1, we identified the redshift of 19/21 (90%) of the objects, and in Group 2, we identified the redshift of 16/19 (84%) of the objects (12/19, 63%, were identified with more than one emission line). For those spectra where we were able to estimate the redshift we present the rest-frame spectra in Figures 2 (Group 1) and 3 (Group 2). In these figures, we note those spectra with only one identified emission line with an asterisk. In Figure 4, we highlight our two highest-redshift sources, J0737−0052 (z = 2.565) and J1326−0037 (z = 2.233), both of which are Group 2 objects. Finally, we plot the 5 spectra in the observed frame without reliable redshift information in Figure 5. We list the spectroscopic redshifts for the sample in Table 2.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

One of the fundamental properties of our WISE-selected sources is their redshift distribution, which we plot for our sample in the top panel of Figure 6. We plot the redshift distribution for all of the objects (excluding the two highest-redshift objects) in black, the AGNs from Group 1 in red, and the AGNs from Group 2 in blue. The full redshift distribution has an average of 〈z_all〉 = 0.52, Group 1 has an average of 〈z₁〉 = 0.35, and Group 2 has an average of 〈z₂〉 = 0.73 (the average redshift becomes 〈z₂〉 = 0.49 without including the two z > 2 objects). We performed a Kolmogorov–Smirnov (K-S) two-sample test on the redshift distributions from the two groups and calculated the K-S statistic P = 0.01, indicating that these two samples were drawn from different distributions. We also performed the test without the five objects from Group 2 with only one emission feature, and calculated P = 0.16, however. While this is highly dependent on the objects in our sample where we only identify one emission feature, our results indicate that by targeting objects with fainter W4 magnitudes, the distribution shifts to higher redshifts. This result is not surprising given the significantly shallower depth of WISE observations in the W4 band as compared to the other three photometric bands. By choosing only those objects that are brightest in W4, we preferentially select objects at lower redshift.

Zoom In Zoom Out Reset image size

We can interpret our redshift distribution in light of the results on WISE AGN selection described in the literature. Assef et al. (2013) describe the spectroscopic redshift distribution for a sample of WISE-selected AGNs (W1 − W2 ⩾ 0.8) in the Boötes field, and demonstrate that the distribution is double-peaked for those objects in their sample with W2 < 15.73. The main peak is between 1 < z < 2, while there is a smaller peak at z ∼ 0.25–0.5. The authors propose that the WISE sensitivity results in recovering obscured AGNs at lower redshifts, where emission from these objects dominate the W1 and W2 photometric bands. As the redshift increases, the W1−W2 color criteria is less efficient at selecting obscured quasars since the portion of the SED probed by the W1 and W2 photometric bands moves to shorter wavelengths where the obscured AGN emission is not as dominant compared to host galaxy emission. This effect leads to a minimum in the distribution at z ∼ 0.75, although there is an increase beyond that due to a larger comoving volume being observed. The distribution of redshifts for our SALT targets is very similar to the peak at z ∼ 0.25–0.5 from the Assef et al. (2013) distribution. The vast majority of our objects have z < 0.75, and the two bright z ∼ 2 AGNs also agree with the higher-z tail of the Assef et al. sample.

As we probed objects with fainter W4 magnitudes, between Groups 1 and 2, we found that our distribution moved to higher redshifts, but the lack of objects at 1 < z < 2 in our sample is most likely a result of our initial optical selection criteria of 20 < g < 22. For obscured quasars, the g-band probes host galaxy emission, and so this cut limits our galaxy selection to lower redshifts. To test this assumption, we explored a large catalogue of ∼4000 galaxies with photometric redshifts and SDSS identifications from the Boötes field. For all objects with 20 < g < 22, the average (median) photometric redshift is only 0.43(0.37) ± 0.24, demonstrating the impact a SDSS g-band magnitude cut can have at selecting high-redshift galaxies.

These redshifts, along with the WISE photometry in the W2, W3, and W4 bands, can be used to estimate an AGN IR luminosity. Following the analysis presented in Hainline et al. (2013), we used the WISE photometry to estimate a luminosity at rest-frame 8 μm (L_8 μm). For this estimate, we modeled the AGN mid-IR emission with a power law and do not account for the individual filter response functions, although, based on the WISE colors for these objects, any flux corrections would be on the order of a few percent (Wright et al. 2010). We also assume that the flux is measured at the central wavelength for each filter. In Section 5, we describe the use of galaxy and AGN templates to model the broadband SEDs for these objects, and we can also use the best-fit AGN template to estimate the 8 μm AGN luminosity for each object in our sample. We find that luminosities measured from the best-fit templates agree with those estimated from the WISE photometry (and presented in Table 2) to within 0.2 dex. Excluding the two high-redshift sources, the objects in our sample have a range in 8 μm luminosity of log (L_8 μm) = 44.0–45.0, with an average (median) of 〈log (L_8 μm)〉 = 44.9(44.8). In order to compare the luminosities to those from quasars in the literature, we plot the 8 μm luminosity against the redshift for our sample in Figure 6. For comparison, we also plot the estimated 8 μm luminosities from the SDSS Type II quasar sample (with W1, W2, W3, S/N >3.0) from Reyes et al. (2008) (〈z〉 = 0.324, z_median = 0.279), with an average (median) of 〈log (L_8 μm)〉 = 44.7(44.3), as well as the estimated 8 μm luminosities for the SDSS Type I quasar sample (with W1, W2, W3, S/N >3.0) from Shen et al. (2011; for the quasars at z < 0.84, 〈z〉 = 0.519, z_median = 0.529), with an average (median) of 〈log (L_8 μm)〉 = 45.0 (44.8) (we restrict the objects used in this average to those with z < 0.84, the upper limit on the redshifts from the Reyes et al. 2008 sample). Based on these results, if the mid-IR is dominated by AGN emission for the objects in our sample, these AGNs would lie in the quasar regime. Notice, however, that at each redshift, the objects that comprise our sample are at systematically higher IR luminosities compared to the optically selected sample, as would be expected from WISE selection.

The objects that comprise our sample were selected from their IR colors to be AGNs. To test the success of our selection criteria, we sought to examine whether excitation of the narrow emission lines in our objects was driven by AGN activity or SF. We can compare the strong optical emission lines present in our spectra in order to explore the source of the excitation. We fit emission lines in our observed spectra with a single Gaussian. For the narrow emission lines, we forced the width of each fitted line to be the same. To estimate uncertainties on the line fluxes, we used a Monte Carlo approach. For each object we generated 500 artificial spectra by perturbing the flux at each wavelength of the true spectrum by a random amount consistent with the 1σ error spectrum. For each artificial spectrum, we measured the line fluxes using the same procedure as was done on the true spectrum. The standard deviation of the distribution of line fluxes measured from the artificial spectrum was used as the error on the true line flux measurement. For each object, if a line was detected with less than 3σ significance we used a measure of 3σ as the maximum flux for the line, and report an upper limit.

For nearby galaxies the optical emission-line ratios of [O iii] λ5007/Hβ as well as [N ii] λ6584/Hα are commonly compared to separate Type II AGNs from star-forming galaxies (the Baldwin, Phillips & Terlevich, or "BPT diagram"; Baldwin et al. 1981). Collisionally excited emission lines [O iii] λ5007 and [N ii] λ6584 are much stronger in AGNs, which produce a much harder ionizing spectrum, relative to the recombination lines Hα or Hβ. In the BPT diagram, the position of an object falls into one of three commonly defined areas based on the emission-line ratios of [O iii] λ5007/Hβ as well as [N ii] λ6584/Hα: an AGN region, a star-forming locus, and a region where "composite" objects with evidence for both AGN activity and SF are found. The usage of these emission lines is limited, however, to redshifts where the emission lines are still present in the wavelength range covered by the spectrograph. The SALT RSS set-up that was used for this study spans an observed wavelength range to 6100 Å, which excludes [N ii] λ6584 and Hα, and we can only observe [O iii] λ5007 to z < 0.22.

There are three objects in our assembled sample (J1151−0046, J1007+0146, J1320−0102) with z < 0.22, and for these objects, we observe [O iii] λ5007 and Hβ with high S/N. For these sources, log([O iii] λ5007/Hβ)⩾0.9, a ratio which is strongly indicative of AGN activity. For the two objects with SDSS spectra, we calculated the [O iii]/Hβ ratio as well. For J0325−0032, log([O iii] λ5007/Hβ) =1.1 and log([N ii] λ6584/Hα) =−0.11, which puts this object firmly in the AGN regime on the BPT diagram (Kauffmann et al. 2003; Kewley et al. 2006). For J0924+0027, log([O iii] λ5007/Hβ) = 0.7, and as z = 0.660, the SDSS spectrum does not cover [N ii] or Hα. Additionally, we can use the [O iii] luminosity for those objects in our sample with [O iii] detections to confirm the results for the luminosity regime spanned by our sample. While the fixed primary mirror of SALT prevents us from measuring the luminosity for the [O iii] lines in the spectra where the line appears (due to our inability to perform absolute flux calibration), we measured the [O iii] luminosity for the two objects with SDSS spectra. For J0924+0027, we measure log(L_[O iii]/erg s⁻¹) = 42.79, while for J0325−0032, we measure log(L_[O iii]/erg s⁻¹) = 42.63. Both of these values are well above the limit of log(L_[O iii]/erg s⁻¹)>41.88 used by Reyes et al. (2008) to select Type II quasars from SDSS.

For objects with z > 0.22, we cannot use [O iii] λ5007/Hβ to test the ionization source. Instead, we used a different emission line diagnostic, [Ne iii] λ3869/[O ii] λλ3726+3729. This line ratio is used in the Trouille, Barger & Tremonti diagram ("TBT diagram"; Trouille et al. 2011), a plot that compares [Ne iii]/[O ii] to the rest-frame g − z color (which we will refer to as ^0.0(g − z)) for galaxies at intermediate redshifts, and can be used to separate AGNs and star-forming galaxies with a high degree of confidence. While [Ne iii] λ3869 is weaker than the strong [O iii] λ5007 emission line, it is also indicative of highly ionized gas excited by an AGN, and separates AGNs from metal-rich star-forming galaxies. The rest-frame g − z color helps to separate AGNs from metal-poor star-forming galaxies, as they tend to be dominated by their bulge components, and are significantly redder. Trouille et al. (2011) demonstrate the success of the TBT diagram in recovering both SDSS BPT-selected AGNs (98.7% of the objects in their sample) as well as BPT-selected star-forming galaxies (97%).

Our usage of the TBT diagram is limited to z < 0.59 where [Ne iii] λ3869 appears on the RSS chip in our observational set-up. We measured the line flux ratios between the [Ne iii] λ3869 and [O ii] λλ3726+3729 emission features, where we fit both lines simultaneously. For six objects, either [Ne iii] λ3869 or [O ii] λλ3726+3729 fell on a gap between the detector chips or on a bright sky feature, and as a result, we cannot report a line flux ratio. For J0325−0032, we can also compare the line flux ratio measured from our SALT spectrum against the ratio measured from the SDSS spectrum. For this object, we calculate log([Ne iii]/[O ii]) =−0.36 ± 0.02 from the SALT spectrum, and log([Ne iii]/[O ii]) =−0.39 ± 0.02 from the SDSS spectrum, consistent within the given uncertainties.

To determine ^0.0(g − z) we start with the SDSS ugriz photometry and follow the methodology from Trouille et al. (2011), using kcorrect v4_2 (Blanton & Roweis 2007) to estimate the rest-frame g and z magnitudes. Uncertainties on the rest-frame colors are estimated by creating 500 artificial sets of photometry for each object, where each artificial magnitude was created by randomly altering the observed SDSS photometry by an amount consistent with the uncertainties. We then ran the kcorrect software on each set of artificial magnitudes, and report the standard deviation on the distribution of resulting ^0.0(g − z) values as the uncertainty in the true value. We report both the [Ne iii] λ3869/[O ii] λλ3726+3729 line ratios as well as the ^0.0(g − z) values for each of the objects in our sample in Table 2.

We plot those targets with [Ne iii] λ3869/[O ii] λλ3726+3729 ratios on the TBT diagram in Figure 7. This figure also includes contours for a large sample of SDSS emission-line galaxies at 0.02 < z < 0.35 from Trouille et al. (2011), with emission-line fluxes measured by MPA-JHU (Aihara et al. 2011). Contours for BPT-selected AGNs are shown in green, BPT star-forming galaxies are shown in blue, and BPT composite objects are shown in gray. The objects in our sample are plotted with red and blue circles depending on their selection group, and we plot those objects with upper limits on the flux ratio with an arrow.⁹ Our sources are mostly found clustered to the top of the figure, located in the region of the diagram dominated by BPT AGNs, and are distinctly separate from star-forming galaxies.

Zoom In Zoom Out Reset image size

For those objects for which we can use emission lines (either optical or UV) to estimate the source of the ionizing radiation, there is strong evidence that the WISE selection criteria is targeting AGNs. However, there does exist a sample of objects without optical evidence for AGN activity. There are four objects with upper limits on the ratio of [Ne iii]/[O ii]: J1136+0152, J1255−0010, J1320−0102, and J1609−0004. As discussed above, J1320−0102 is at a low redshift and we measure the flux ratio of log([O iii]/Hβ) =0.9, which is indicative of an AGN. For the other three objects, if they are indeed star-forming galaxies rather than AGNs, this implies a contamination fraction in our sample of 3/23 (∼13%) (We only compare these three objects to those galaxies where we can measure the [O iii]/Hβ, [Ne iii]/[O ii], or UV emission lines). This contamination fraction agrees with the fraction of non-active galaxies observed in the Spitzer-selected AGN sample presented in Lacy et al. (2013; 22%). If we include the five objects in our sample without spectroscopic redshifts, assuming that they are star-forming galaxies, the contamination fraction would have an upper limit of ∼30% (8/28). In the next section, we discuss the use of SED modeling to further explore the AGN activity in our sample.

Existing optical through infrared photometry for the objects comprising our sample allows us to explore their SEDs. SED decomposition is a powerful tool for gaining physical insight into the AGN and host galaxy properties of our sample, and it can also be used to estimate the redshift for those objects where we do not observe emission features in the spectrum. In order to best model the SEDs, we supplement the SDSS and WISE photometry described in Section 2 with near-IR Y, J, H, and K photometry from the UKIRT Infrared Deep Sky Survey (UKIDSS; Lawrence et al. 2007) Large Area Survey. We use the extended source aperture-corrected magnitudes with a 57 aperture measured with >3σ significance. We have UKIDSS data for 23 objects. For an additional three objects where UKIDSS data was not available, we also used near-IR J, H, and K photometry from the Two Micron All-Sky Survey (2MASS; Skrutskie et al. 2006) Point Source Catalogue where an object was detected in an individual filter with >3σ significance. As this SED modeling is only designed to explore the relationship between the host galaxy and quasar emission for each object in our sample, we use the catalogue uncertainties for estimating the best fits but, as we are not accounting for all of the observational errors, we follow Chung et al. (2014) and assign a conservative minimum photometric error of 0.05 mag on each measurement, which is comparable to or smaller than the photometric calibration uncertainties for each survey (Hewett et al. 2006; Skrutskie et al. 2006; Ahn et al. 2012; Cutri et al. 2013).

For the fitting, we follow the methodology we model each galaxy as a linear combination of a single galaxy template and a reddened AGN template (note that this is different from the methodology discussed in Assef et al. 2010, where each galaxy was modeled as a linear combination of all three galaxy templates plus an AGN template). We use the empirically-derived models from Assef et al. (2010), which consist of three galaxy templates (a spiral "Sbc," elliptical "E," and irregular "Im" galaxy), as well as an unobscured AGN template. In order to simulate the effects of AGN obscuration, we follow Assef et al. (2010) and apply an extinction model to the AGN template that consists of a Small Magellanic Cloud (SMC) like extinction curve at λ < 3300 Å (Gordon & Clayton 1998) and a Galactic extinction curve at longer wavelengths (Cardelli et al. 1989), with R_V = 3.1 for both, and in this paper we parameterize the extinction using E(B − V). Because we are attempting to fit an unobscured AGN template to objects with varying amounts of dust obscuration, we follow Mainieri et al. (2011) and Hainline et al. (2012) and employ an additional prior of E(B − V)_AGN > 1.0 for those objects with evidence indicating that they are an obscured AGN. We use this prior on objects with an SDSS TYPE flag of "3," indicating that the galaxy is resolved in optical imaging. For these objects, we assume that the optical portion of the SED is dominated by stellar emission rather than that of the central AGN. Setting a lower bound on the best-fit extinction value was only required for fitting six objects out of the total 35 that we modeled in this way, and the fits were not made significantly worse by using this prior. For each object, we calculated the monochromatic flux densities from the observed photometry,¹⁰ and compared these values to the flux densities from the best-fit templates passed through the SDSS, UKIDSS, 2MASS, and WISE bandpass filters, and used χ² minimization to find the best fits for each object. The output parameters are then two coefficients for the linear combination (a galaxy and an AGN coefficient), and a value for E(B − V)_AGN. The results from our fitting are shown in Table 3 and Figures 8–9.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 3. Best-fit SED Parameters

Object	Group	Templatea	Gal. Coeffa	AGN Coeffa	E(B − V)AGNa	$\chi ^2_{{\rm red}}$a	fAGN, 1 μmb
J032533.31−003216.4	1	Sbc	9.21e −16	2.47e −16	4.72	21.86	<0.01
J033820.71−004935.6	2	Sbc	6.01e −16	6.82e −17	0.35	2.16	0.23
J035726.82−002724.9	2	E	2.32e −16	7.90e −17	0.26	1.46	0.32
J073745.20−005229.8	2	Im	1.05e −16	1.16e −16	0.10	11.88	0.51
J085259.35+013715.1	1	Sbc	1.12e −15	1.38e −16	15.61	16.91	<0.01
J092435.55+002716.4	1	Sbc	1.44e −15	4.03e −16	1.00c	15.25	0.26
J095718.06−012049.1	2	Sbc	1.29e −15	1.14e −16	3.98	8.97	<0.01
J100711.52+014627.1	1	Sbc	1.29e −15	2.03e −16	5.53	2.64	<0.01
J110722.11+013336.8	1	E	6.42e −16	1.90e −16	5.53	6.24	<0.01
J112931.47+010254.7	1	Sbc	1.27e −15	2.98e −16	5.52	9.86	<0.01
J113635.48+015252.9	1	Sbc	1.63e −15	2.17e −16	1.00c	8.09	0.14
J115158.63−004641.2	1	Sbc	9.75e −16	1.38e −16	9.38	15.55	<0.01
J125521.24−001018.2	1	Sbc	1.54e −15	1.16e −16	8.69	11.35	<0.01
J130500.31+005422.1	1	Sbc	1.84e −15	6.10e −16	47.72	16.30	<0.01
J130845.53+015542.0	1	Sbc	4.30e −16	3.16e −16	1.00c	9.80	0.48
J132031.04−010248.3	2	E	2.47e −16	1.50e −16	0.61	11.96	0.36
J132648.81−003757.6	2	Im	1.14e −16	1.82e −16	0.54	6.07	0.47
J133331.15−012653.3	1	Sbc	1.78e −15	7.53e −16	53.38	31.81	<0.01
J135423.71−000314.5	1	Im	5.82e −17	3.24e −16	27.50	4.18	<0.01
J135534.66−002206.1	2	E	7.85e −16	2.24e −16	1.00c	43.56	0.14
J140618.16−004923.1	2	Sbc	1.40e −15	1.54e −16	1.86	13.59	0.05
J143459.27−014432.8	2	Sbc	1.20e −15	2.23e −16	14.01	10.31	<0.01
J144006.46−011624.7	1	Sbc	1.34e −15	2.42e −16	1.00c	13.79	0.18
J144625.94−015721.9	2	Sbc	8.03e −16	1.24e −16	1.00c	14.45	0.16
J150539.97+013433.9	1	Sbc	1.34e −15	2.05e −16	11.63	4.02	<0.01
J152736.35−001007.7	2	E	6.11e −16	8.50e −17	0.22	16.89	0.17
J153846.30+012951.9	1	Sbc	8.47e −16	2.76e −16	7.34	13.15	<0.01
J154503.94−010010.4	1	Sbc	1.80e −15	3.31e −16	1.76	11.29	0.09
J155421.94+001115.0	1	E	9.63e −16	2.20e −16	6.71	26.44	<0.01
J160903.75−000426.2	2	Sbc	7.58e −16	8.31e −17	2.36	0.96	0.03
J160928.56−013344.1	2	E	9.26e −16	1.65e −16	4.12	5.67	<0.01
J163445.76−010808.6	2	E	1.76e −16	1.16e −16	0.26	15.50	0.48
J173114.67−003708.1	2	Im	5.04e −17	2.64e −17	0.56	1.29	0.22
J194745.03−003603.8	2	E	8.56e −17	7.04e −17	0.36	2.31	0.50
J202952.16−010805.3	2	Sbc	8.06e −16	9.83e −17	0.45	6.45	0.22

Notes. ^aParameters from the SED fits using the templates from Assef et al. (2010). Each fit was done using one galaxy template, along with the dust obscured AGN template with obscuration parameterized by E(B − V)_AGN. We also provide the reduced χ² value for each fit. ^bFraction of the total emission at rest-frame 1 μm that is estimated to arise from AGN emission. ^cFor these objects, the fits were restricted to E(B − V)_AGN ⩾ 1.0 due to the objects being resolved in optical SDSS imaging, and most likely unobscured AGNs.

Download table as:  ASCIITypeset image

As can be seen from the best-fit χ² values, the majority of the objects in our sample are well fit by the Assef et al. (2010) templates. J1355−0022 has $\chi ^2_{{\rm red}} > 40$, due primarily to the Assef et al. AGN template being redder than the observed W3−W4 color. We observe a large variety in the optical SED shapes for the objects in our sample, and all three galaxy templates are used in our best-fit models. For each object, the best-fit AGN coefficients are above zero, indicating that galaxy light alone is unlikely to be the source of the infrared emission seen in the SEDs. For the three objects described in the previous section that do not show evidence in their optical spectra for an AGN, we do recover non-zero AGN coefficients, but these SEDs would also perhaps be fit using a more extreme star-forming template than the one derived in Assef et al. (2010). Indeed, the SEDs of both M82 and Arp220 taken from the SWIRE template library (Polletta et al. 2007) have stronger mid-IR emission from dust heated by SF, which fits the WISE photometry for the objects in our sample with the highest values for E(B − V)_AGN. For each object, we also calculate f_{AGN, 1 μm}, which is defined as the fraction of the total emission in the best-fit model at rest-frame 1 μm that is contributed by the AGN template (Hainline et al. 2012). We list these values in Table 3. There are 16 objects with f_{AGN, 1 μm} < 0.01, which we note in the table. To check the validity of our SED fits, we examined the far-infrared luminosities and star-formation rates (SFRs) implied by the best-fit galaxy coefficients. The Assef et al. (2010) templates only extend to a rest-frame wavelength of ∼30 μm, and so we used Chary & Elbaz (2001) far-infrared star-forming galaxy templates to predict the far-infrared luminosity from SF for our sample. We passed the observed best-fit galaxy template (ignoring the AGN component) through the Spitzer MIPS 24 μm passband, and then used this flux to constrain the Chary & Elbaz (2001) templates, which were integrated to produce the far-infrared luminosity from SF for each object. We then used the conversion from Kennicutt (1998) to calculate the SFRs. We measure an average (median) SFR for the sample (without the two high-redshift objects) of 16 (6) M_☉ yr⁻¹, in agreement with expectations for massive star-forming galaxies at the redshift range probed by our sample. We note that this method is highly uncertain, especially without longer-wavelength observations to better constrain the far-infrared templates.

In order to estimate redshifts for those objects in our sample without detected emission lines, we ran the SED analysis as described above on each of the objects in our sample but allowed redshift to be a free parameter. Estimating redshifts from AGN photometry is difficult as the power-law shape of an AGN SED with minimal extinction creates a degeneracy between color and redshift which requires long wavelength photometric coverage to break (Brodwin et al. 2006; Salvato et al. 2009, 2011). We examined the photometric redshifts recovered for those objects with spectroscopic redshifts, as well as those with existing SDSS photometric redshifts (Oyaizu et al. 2008; photo-z's were estimated for each object that was resolved in SDSS imaging, and these values largely agree with the spectroscopic redshifts for these objects). For our own photometric redshift calculation, we used the χ² values to explore the 2σ confidence intervals and find the best physical fit to the data for each object, incorporating the E(B − V)_AGN > 1.0 prior discussed above for objects that were resolved in SDSS imaging. In a few cases, we rejected best fits at large redshifts (z > 2) both due to the low likelihood of a galaxy with 20 < g < 22 to be at higher redshift (see our analysis of the AGNs in this optical photometric range in the Boötes field in Section 3) and the low likelihood of a quasar to be this luminous at such high redshift, based on the observed quasar luminosity function (Hopkins et al. 2007). The final photometric redshifts estimated from this procedure are given in Table 2. In Figure 10, we plot the spectroscopic redshifts we derived from our optical spectra against the photometric redshifts we derived from this process for objects with z < 1.0. Overall, there is excellent agreement between the spectroscopic and photometric redshifts for the majority of the objects in our sample, although our recovered photometric redshifts are often low in comparison to the corresponding spectroscopic redshifts. The two most discrepant objects, J1634−0108 and J1947−0036 have SEDs that are dominated by AGN emission (f_{AGN, 1 μm} ∼ 0.5, and E(B − V)_AGN < 0.4).

Zoom In Zoom Out Reset image size

With these uncertainties in mind, we estimated the photometric redshift for the five objects in our sample without spectroscopic redshifts seen in Figure 5. We plot the best-fit SEDs in Figure 11 and report the calculated photometric redshifts for these sources in Table 2. As can be seen, these objects have a range of 0.1 < z_phot < 0.9, in broad agreement with the spectroscopic redshifts observed for our sample. If these photometric redshifts are roughly correct, then we calculate an average redshift of z = 0.52 for all of the objects in our sample, while Group 1 has an average redshift of z = 0.36, and Group 2 has an average redshift of z = 0.71 (the average redshift is z = 0.52 without including the two z ∼ 2 objects), supporting our conclusions that selecting objects based on fainter W4 photometry leads to a higher-redshift sample. It would be of special interest to extend this photometric redshift estimation to a larger sample of SDSS+WISE-selected AGNs in order to support quasar clustering analyses (e.g., DiPompeo et al. 2014).

Zoom In Zoom Out Reset image size

While we discussed the full population of objects in Sections 3 and 4, some of the individual objects that comprise the sample warrant a more in-depth discussion of their optical spectroscopic and SED properties. In this section, we will describe our highest-redshift sources, as well as those sources with broad emission lines in their spectra.

6.1. High-redshift Sources

6.2. Unobscured Sources

Based on our examination of the optical spectroscopic properties of our sample in Section 4, we conclude that the WISE IR selection criteria we used was largely successful at recovering obscured quasars. We compare our simple selection criteria, which was based on the W1−W2 color cut from Stern et al. (2012), to other criteria discussed in the literature. Recently, Mateos et al. (2012) used AGNs selected using the Bright Ultrahard XMM-Newton Survey to create a WISE selection criteria that attempted to maximize the fraction of X-ray selected AGNs while minimizing contamination from star-forming galaxies. In a follow-up paper (Mateos et al. 2013), the authors examine the effectiveness of their revised WISE selection criteria in selecting obscured quasars using the SDSS Type II quasar sample described in Reyes et al. (2008). Their results indicate that it is AGN luminosity, especially with respect to host galaxy luminosity, that is the driving factor in selecting AGNs by their mid-IR colors. In this section we describe where our objects lie with respect to the Mateos et al. (2012) selection criteria.

Based solely on the selection criteria described in Mateos et al. (2012), 32/40 (80%) of the objects in our SALT spectroscopic sample fall in the Mateos et al. color–color wedge, as can be seen in Figure 1. We calculated the redshift of 28/32 (88%) of these objects, although 4 only have one emission feature in their spectrum. If we exclude objects where the redshift precludes us discerning any AGN activity in the object (due to both the wavelength coverage of the SALT chip, or either [Ne iii] or [O ii] falling on a chip gap or sky feature), then the Mateos et al. criteria were successful at identifying 14/17 (82%) objects with strong evidence of AGN activity based on the objects position on the TBT diagram, the BPT diagram, or the presence of high-ionization emission lines in the UV spectrum for the two highest-redshift objects. The other three objects have upper limits on the [Ne iii]/[O ii] ratio, indicating a non-detection of [Ne iii] in the spectrum.

Eight objects in our SALT sample would not be selected under the Mateos et al. (2012) criteria. Two of these objects have unusual W2−W3 colors, J1527−0010 (W2 − W3 = 1.80), and J1354−0003 (W2 − W3 = 6.24). There is only one emission line identified in the spectrum of J1527−0010 and while J1354−0003 is in the AGN region of the TBT diagram, the position of this object is in the region also occupied by AGN/SF composite objects.

The other six objects not selected by the Mateos et al. (2012) wedge have red W2−W3 colors that put them below and to the right of the wedge. Five of these objects (J1305+0054, J0852+0137, J1554+0011, J1107+0133, and J1333−0126) are identified as AGNs based on their position on the TBT diagram, and we have not identified the spectroscopic redshift of the sixth object, J1417+0108. The presence of AGNs outside of the Mateos et al. (2012) selection region is not surprising; these authors used an analysis of AGN mid-IR templates to demonstrate that heavily obscured AGNs, as well as higher-redshift AGNS (z ∼ 1–1.5) would occupy this position in WISE color–color space. The positions of heavily obscured AGNs in WISE color–color space was further discussed in Mateos et al. (2013), where the relative contribution of an AGN and its host to the mid-IR SED was the primary factor that determines whether or not an AGN was found in the Mateos et al. selection wedge. For the five objects in our sample outside the wedge with spectroscopic redshifts, the best-fit SEDs indicate that these objects have a best-fit value of E(B − V)_AGN > 6.0 (three of these objects have E(B − V)_AGN > 16.9), indicating that the AGN only becomes dominant in the mid-IR; this is expected based on the red W2−W3 colors for these objects. As mentioned in the previous section, these objects would not be chosen under the Group 2 selection criteria, as they have r_AB − W2_AB < 3.1. For selecting the most highly obscured quasars, these results indicate the necessity in using an optical to infrared color cut with the W3 or W4 photometric bands. These high values for E(B − V) indicate that these objects are likely obscured quasars, but it should be noted that while we do apply a simple prescription for obscuration in our SED fitting, we do not know the geometry of the obscuring medium. As we do see strong infrared emission, we know that there must be dust near the central black hole, but galaxy-scale dust can also contribute to the observed obscuration (see Merloni et al. 2014; Chen et al. 2014). We cannot rule out extreme SF as the origin of the rest-frame IR photometry, supported by the measurement of f_{AGN, 1 μm} ⩽ 0.01 for these objects and the position of the Polletta et al. (2007) Arp220 template occupying a similar position on the color–color diagram at low redshift as can be seen in Figure 1. However, this figure as well as Mateos et al. (2013) demonstrate that a significant number of SDSS Type II quasars from the Reyes et al. (2008) sample also fall into this region and would not be selected as IR-AGNs using the Mateos et al. (2012) criteria.

We can also use the SED modeling results to test whether our full sample of objects would be selected under the Spitzer selection criteria outlined in Lacy et al. (2004), Stern et al. (2005), and Donley et al. (2012). We passed the best-fit combined stellar and AGN template through the Spitzer IRAC transmission curves to estimate Spitzer photometry. We find that all objects but one fall into the Lacy et al. (2004) selection box. The lone outlier, J1354−0003, has an SED with a very strong AGN power law as compared to the stellar population, as seen in Figure 8. Donley et al. (2012) revised the Lacy et al. (2004) criteria in order to account for contamination to the IR from star-forming galaxies, and we find that only three of our targets would lie outside of the Donley et al. (2012) selection box. These objects, J1333−0126, J1305+0054 and J0852+0137, are objects that would also not be selected under the Mateos et al. (2012) criteria and have near-IR emission dominated by host galaxy light, with our SED modeling showing that an AGN contribution is only prominent at longer wavelengths. The bulk of the targets that comprise our sample, then, compare very well to the Spitzer-selected objects described in Lacy et al. (2013).

The all-sky coverage of WISE has made it possible to find large numbers of obscured quasars based solely on their observed infrared colors. We have observed a sample of 40 objects with WISE colors indicative of AGN activity in two groups. Group 1 objects were initially selected to have W1 − W2 > 0.7 and 7 ⩾ W4 ⩾ 6.5, while Group 2 objects were initially selected to have W1 − W2 > 0.8, W4 ⩾ 7.0, and r_AB − W2_AB > 3.1. Both groups were selected to have 20 < g < 22. We used optical spectroscopy from SALT RSS to uncover the redshift distribution and ionization properties of obscured quasars. We use the ratio [O iii]/Hβ, and [Ne iii]/[O ii] to determine that the majority of our objects in our sample are AGNs: of those objects with redshifts that allow us to use emission lines to identify the ionization source, only ∼13% of the objects in our sample do not have evidence in their observed spectra of being an AGN. For the AGNs, we measure [O iii] and IR luminosities that are comparable to SDSS-selected quasars. The objects in our sample have an average redshift of z ∼ 0.5, and the majority have z < 1.0, which is most likely a result of our optical photometric cut. Our sample also includes two interesting high-redshift sources, a BAL quasar at z = 2.565, and a quasar with Lyα and C iv λ1549 emission at z = 2.233.

We also use a simple two-component SED analysis to explore the relationship between AGN and SF as a function of the position of the objects on the WISE W1−W2 versus W3−W4 color–color diagram. The majority of the objects in our sample have strong AGN emission in the mid-IR. We also compared our selection criteria to that of Mateos et al. (2012). While their criteria are successful at recovering the AGNs in our sample, there are five objects outside of their selection wedge at red W3−W4 colors with optical line ratios indicative of AGN activity. SED modeling of these objects demonstrates that these are the most highly obscured quasars in our sample, in line with the conclusions made in Mateos et al. (2012). These results indicate the selecting objects based on red optical to W3 or W4 colors is required to select objects with the highest levels of obscuration.

The results from observing and analyzing this small sample of obscured quasars are currently being used to understand the entire population of obscured quasars with both WISE and SDSS photometry. We can use our results from the photometric redshift estimation for these objects to aid future WISE-selected AGN clustering studies. The most heavily obscured quasars in our sample (such as J1354−0003, with a strong AGN component to the SED that only becomes apparent at the longest mid-IR wavelengths) may also be excellent targets for observations in the hard X-ray portion of the spectrum with The Nuclear Spectroscopic Telescope Array (NuSTAR) High-energy X-Ray telescope (Harrison et al. 2013). We are also using SALT to probe portions of the WISE W1−W2 versus W2−W3 color–color diagram that are not spanned by our current sample. Together, our results along with future analysis help demonstrate the power of mid-IR WISE selection for recovering large samples of obscured quasars.

We acknowledge the referee for a constructive report which significantly improved the paper. K.N.H., R.C.H., A.D.M., and M.A.D. were partially supported by NASA through ADAP award NNX12AE38G and by the National Science Foundation through grant numbers 1211096 and 1211112.

Some of the observations reported in this paper were obtained with the Southern African Large Telescope (SALT). This publication also makes use of data products from the Wide-field Infrared Survey Explorer, which is a joint project of the University of California, Los Angeles, and the Jet Propulsion Laboratory/California Institute of Technology, funded by the National Aeronautics and Space Administration. Funding for SDSS-III has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, and the U.S. Department of Energy Office of Science. The SDSS-III Web site is http://www.sdss3.org/.

SDSS-III is managed by the Astrophysical Research Consortium for the Participating Institutions of the SDSS-III Collaboration including the University of Arizona, the Brazilian Participation Group, Brookhaven National Laboratory, Carnegie Mellon University, University of Florida, the French Participation Group, the German Participation Group, Harvard University, the Instituto de Astrofisica de Canarias, the Michigan State/Notre Dame/JINA Participation Group, Johns Hopkins University, Lawrence Berkeley National Laboratory, Max Planck Institute for Astrophysics, Max Planck Institute for Extraterrestrial Physics, New Mexico State University, New York University, Ohio State University, Pennsylvania State University, University of Portsmouth, Princeton University, the Spanish Participation Group, University of Tokyo, University of Utah, Vanderbilt University, University of Virginia, University of Washington, and Yale University.