MID-INFRARED SELECTION OF ACTIVE GALACTIC NUCLEI WITH THE WIDE-FIELD INFRARED SURVEY EXPLORER. I. CHARACTERIZING WISE-SELECTED ACTIVE GALACTIC NUCLEI IN COSMOS

The Spitzer-COSMOS survey (S-COSMOS; Sanders et al. 2007) carried out a deep (620 hr), uniform survey of the full 2 deg² COSMOS field in all seven Spitzer bands (3.6, 4.5, 5.8, 8.0, 24, 70, and 160 μm). The IRAC portion of the survey covered the field to a depth of 1200 s in the four bluest bands of Spitzer, with 5σ measured sensitivities ranging from 0.9 μJy at 3.6 μm to 14.6 μJy at 8.0 μm. The longer wavelength observations, obtained with the Multiband Imaging Photometer for Spitzer (MIPS; Rieke et al. 2004), reach 5σ sensitivities of approximately 0.07, 8.5, and 65 mJy for the 24, 70, and 160 μm arrays, respectively (Frayer et al. 2009). These depths are all considerably deeper than WISE.

We identified 6261 unique WISE sources with signal-to-noise ratio S/N ⩾ 10 in WISE band W2 in a region that extends slightly beyond the FoV of the S-COSMOS survey. We used a preliminary version of the second pass data, which co-adds all observations from the WISE mission. In order to avoid spurious and poorly photometered sources, we limited the sample to relatively isolated sources by requiring blend flag NB ⩽ 2. We also avoided contaminated or confused sources by eliminating sources whose W1 or W2 photometry was affected by diffraction spikes (ccflag = D), persistence (ccflag = P), scattered light halos from nearby bright sources (ccflag = H), or optical ghosts (ccflag = O) (for a detailed description of WISE catalog variables, see Wright et al. 2010). The preliminary version of the second pass data we used double counts sources in the overlap regions between processing stripes. Using a 05 match radius, we identified duplicated sources in this preliminary catalog and only retained the source with higher S/N in W1. The conservative (10σ) W2 depth we apply corresponds to an approximate flux density limit of 15.05 mag (∼160 μJy) at 4.6 μm. Most sources are also detected at ⩾10σ at 3.4 μm, corresponding to 16.45 mag (∼70 μJy).

We then identified the nearest S-COSMOS source to each WISE source. We required detections in both IRAC channel 1 (3.6 μm) and channel 2 (4.5 μm) in order to avoid the edges of S-COSMOS, which did not receive full four-band IRAC coverage. We also eliminated saturated stars with the requirement [3.6] ⩾11. Correcting for the small mean astrometric offset between WISE and S-COSMOS, 〈ΔR.A.〉 = 0108 and 〈Δdecl.〉 = 0008, and requiring a conservative 10 matching radius, we find unique, unsaturated, multi-band S-COSMOS identifications for 3618 WISE sources. Most of the WISE sources lacking S-COSMOS counterparts are from outside the S-COSMOS FoV. Within the area of good, unique matches, 4% of WISE sources do not have S-COSMOS counterparts and <1% of WISE sources have multiple S-COSMOS counterparts within the 10 matching radius. Visual inspection shows that confusion is the source of both of these issues, with the lower resolution WISE images merging multiple objects. In the remainder of the paper, we restrict the analysis to the ∼95% of WISE sources within the S-COSMOS area with unique, unsaturated multi-band IRAC identifications.

Figure 3 shows the measured differences between the WISE and S-COSMOS photometry. For S-COSMOS, we use aperture-corrected 29 photometry from the public 2007 June catalog, converted from physical units to Vega magnitudes using conversion factors prescribed by the S-COSMOS documentation available through the Infrared Science Archive (IRSA). As expected given the slightly different central wavelengths and widths of the IRAC and WISE filters, we find slight median offsets between their respective photometric measurements: (W1 − [3.6])_med = −0.01 and (W2 − [4.5])_med = −0.07. The Explanatory Supplement to the WISE All-Sky Data Release¹² finds a color term from analysis of compact sources in the Spitzer SWIRE XMM-LSS field, (W1 − [3.6]) ∼ 0.4([3.6]–[4.5]) (see Figure 2 of Section VI.3.a of the Explanatory Supplement).

Zoom In Zoom Out Reset image size

Figure 4 shows the IRAC color–color diagram for S-COSMOS sources with robust WISE counterparts. The figure shows the expected concentration of Galactic stars with Vega colors of zero. As discussed in Stern et al. (2007) and Eisenhardt et al. (2010), stars warmer than spectral class T3 all have essentially Rayleigh–Jeans continua in the IRAC passbands, leading to similar IRAC colors. Methane absorption causes redder [3.6]−[4.5] colors for cooler brown dwarfs, leading to a vertical extension above the Galactic star locus. Few such sources are found by WISE over an area as small as COSMOS. The sources extending to the right of the stellar locus is dominated by low-redshift star-forming galaxies, where polycyclic aromatic hydrocarbon (PAH) emission causes red [5.8]−[8.0] colors. Finally, as suggested in Eisenhardt et al. (2004) and discussed in detail in Stern et al. (2005), the vertical extension perpendicular to the galaxy sequence is dominated by AGNs. Indeed, Gorjian et al. (2008) show that the majority (65%) of X-ray sources in the XBoötes survey are identified by the Stern et al. (2005) mid-infrared criteria (see also Donley et al. 2007; Eckart et al. 2010; Assef et al. 2011). In IRAC data plotted to deeper depths, a highly populated second vertical sequence is also visible to the left of the AGN sequence. This sequence, due to massive galaxies at z ≳ 1.2 (e.g., Stern et al. 2005; Eisenhardt et al. 2008; Papovich 2008), typically outnumbers the AGN sequence since even very shallow (<90 s) IRAC pointings easily reach well below the characteristic brightness of early-type galaxies out to z ∼ 2, m*_3.6 ≈ 17.5, and m*_4.5 ≈ 16.7 (e.g., Mancone et al. 2010). However, these galaxies are absent with our conservative, W2 ≲ 15 magnitude cut in the much shallower WISE data.

Zoom In Zoom Out Reset image size

In the following analysis, we will adopt the Stern et al. (2005) mid-infrared AGN sample as the "truth sample" in order to explore potential WISE AGN selection criteria. The Stern et al. (2005) method for selecting AGNs was one of the first methods devised to identify AGNs using Spitzer data and has been extensively used by other workers in the field. However, like all AGN selection criteria, it is not without some shortcomings, highlighted below.

X-ray-selected AGNs missed by mid-infrared selection. As pointed out by numerous authors (e.g., Barmby et al. 2006; Cardamone et al. 2008; Brusa et al. 2009), many X-ray sources have mid-infrared colors consistent with normal galaxies, and thus are missed by the mid-infrared AGN color criteria. As first pointed out by Donley et al. (2007) and further expanded upon by Eckart et al. (2010), the fraction of X-ray sources identified as mid-infrared AGNs increases strongly with X-ray luminosity. For example, Donley et al. (2007), using data from the Ms Chandra Deep Field-North (Alexander et al. 2003), find that the mid-infrared selection efficiency increases from ∼14% at L_{0.5–8 keV} < 10⁴² erg s⁻¹ to 100% at L_{0.5–8 keV} > 10⁴⁴ erg s⁻¹. So while low-luminosity AGNs are, unsurprisingly, missed by the mid-infrared color selection criteria (e.g., Figures 1 and 2), such criteria appear remarkably robust at identifying the most luminous AGNs in the universe.

Contamination by star-forming galaxies. Several authors have also pointed out that the mid-infrared color cuts proposed by Lacy et al. (2004) and Stern et al. (2005) extend into regions of color space populated by star-forming galaxies (e.g., Barmby et al. 2006; Donley et al. 2008; Park et al. 2010). In order to minimize such contamination, mid-infrared power-law selection has been suggested (e.g., Alonso-Herrero et al. 2006), though Donley et al. (2012) note that systematic photometric errors from IRAC are often underestimated (e.g., Reach et al. 2005), making power-law selection more vulnerable to the quality of the mid-infrared photometry than simple color–color cuts. Donley et al. (2012) investigate contamination by star-forming galaxies using a combination of galaxy templates and real data from pure starbursts identified from Spitzer IRS spectroscopy. A strong conclusion from this work is that mid-infrared color selection, particularly using the Stern et al. (2005) criteria, has minimal contamination from purely star-forming galaxies below a redshift of z ∼ 1.

Contamination by high-redshift galaxies. The Lacy et al. (2004) and Stern et al. (2005) criteria for identifying AGNs based on their mid-infrared colors were empirically derived from shallow, wide-area Spitzer data. In this limit, the criteria work extremely well. In deeper mid-infrared data, however, significant contamination from faint, high-redshift galaxies becomes problematic and we recommend use of the revised IRAC selection criteria of Donley et al. (2012) for deep IRAC data.

Completeness versus reliability. Finally, in choosing between the IRAC color criteria of Lacy et al. (2004) and Stern et al. (2005), we have opted for the latter. As pointed out by numerous authors (e.g., Eckart et al. 2010; Donley et al. 2012), the less selective Lacy et al. (2004) criteria have higher completeness at the cost of reliability: many more normal galaxies are identified using those criteria, yielding more significant contamination. With the primary goal of identifying a clean sample of powerful AGNs, we therefore adopt the higher reliability IRAC color criteria of Stern et al. (2005).

In summary, we adopt the Stern et al. (2005) mid-infrared-selected AGN candidates as the "truth sample" for analyzing the WISE selection of AGNs. Foremost, identifying a robust truth sample identified at similar wavelength makes logical sense. Radio selection would miss ∼90% of AGNs, while optical and X-ray selection would miss the heavily obscured AGNs which are identified by mid-infrared selection but largely missed by the current generation of optical and X-ray surveys. In principle, a hybrid selection could be adopted, such as identifying all AGN candidates with L_AGN greater than some value. However, a "truth sample" identified in that manner would be vulnerable to spectroscopic incompleteness.

Despite these caveats to mid-infrared selection, we show that the Stern et al. (2005) criteria are quite robust at the shallow mid-infrared depths of WISE. This method identifies the most luminous X-ray-selected AGNs at high completeness. It also identifies a much higher surface density of AGNs than optical and X-ray surveys of comparable depth due to an increased sensitivity to obscured AGNs. Illustrating this point, Hickox et al. (2007) and Eckart et al. (2010) show that mid-infrared AGN candidates individually undetected at high energy are well detected in stacking analyses and reveal a harder-than-average X-ray spectrum, implying significant obscuration. Optical spectroscopy of IRAC-selected AGN candidates also typically reveal type 2 AGN spectra (D. Stern et al. 2012, in preparation). Finally, in shallow mid-infrared data, particularly at the depth of WISE, the Stern et al. (2005) criteria suffer from minimal contamination by Galactic stars, starburst galaxies, or high-redshift galaxies.

Figure 5 shows a hybrid mid-infrared color–color diagram. Rather than plotting IRAC [3.6]−[4.5] color along the vertical axis, we plot W1 − W2. We see similar trends to the IRAC-only diagram (Figure 4), with a stellar locus at zero color, a horizontal sequence of low-redshift galaxies, and a vertical AGN sequence perpendicular to the galactic sequence. While mid-infrared selection of AGNs in even very shallow Spitzer pointings required the longer wavelength IRAC passbands to differentiate AGNs from high-redshift (z ≳ 1.3) massive galaxies, WISE is able to robustly identify AGNs with just W1 and W2 (e.g., see Ashby et al. 2009; Assef et al. 2010; Eckart et al. 2010). Note that these are the two most sensitive WISE passbands, with the highest source counts and the best spatial resolution.

Zoom In Zoom Out Reset image size

We have explored how robustly WISE identifies AGNs using a simple W1 − W2 color cut. Of the 3618 sources in our cross-matched WISE-S-COSMOS catalog, 157 are mid-infrared AGNs according to the Stern et al. (2005) criteria. We consider this the truth sample and Figure 6 shows the completeness and reliability of WISE AGN selection as a function of W1 − W2 color cut. Prior to the launch of WISE, Ashby et al. (2009) suggested that W1 − W2 ⩾ 0.5 would robustly identify AGNs while Assef et al. (2010) suggested a color cut of W1 − W2 ⩾ 0.85. Considering the former, while this criterion is highly (98%) complete at identifying the AGN sample, it suffers from significant contamination from non-active sources (see Figure 6; only 50% of sources appear active according to the IRAC criteria). This is likely, in part, due to the significantly better performance of WISE compared to the prelaunch predictions: Mainzer et al. (2005) reported 5σ point source sensitivity requirements of 120 μJy at 3.4 μm and 160 μJy at 4.6 μm, while we are finding 10σ point source sensitivities of 70 μJy at 3.4 μm and 160 μJy at 4.6 μm. Analysis of Figure 6 suggests that a color cut at W1 − W2 = 0.8 offers an extremely robust AGN sample which is still highly complete. For some uses, a slightly less conservative color cut at W1 − W2 = 0.7 might be preferable, providing a powerful compromise between completeness and reliability for WISE AGN selection.

Zoom In Zoom Out Reset image size

Using W1 − W2 ⩾ 0.8 to select AGN candidates, we identify 130 potential candidates, of which 123 are AGNs according to their IRAC colors (95% reliability, 78% completeness). The less conservative color cut at W1 − W2 ⩾ 0.7 identifies 160 candidates, of which 136 are AGNs according to their IRAC colors (85% reliability, 87% completeness), e.g., this less restrictive color cut identifies ∼10% more AGNs at the cost of tripling the number of contaminants. As seen in Figure 4, several of the "misidentified" AGN candidates have colors very close to the Stern et al. (2005) criteria. In fact, as discussed in Section 3.7, four of the seven "contaminants" for the W1 − W2 ⩾ 0.8 AGN selection have spectroscopic redshifts, two of which are broad-lined quasars. This implies that the above reliability numbers are likely conservative, though complete spectroscopy will be required to determine what fraction of the IRAC-selected AGNs are, in fact, not active. As seen in Figure 5, misidentified AGN candidates often have much redder [5.8]–[8.0] colors, suggestive of low-redshift galaxies with PAH emission. Potential Galactic contaminants are brown dwarfs (cooler than spectral class T1) and asymptotic giant branch stars, both of which have much lower surface densities than AGNs at the depths probed and are not expected to be a significant contaminant, particularly in extragalactic pointings.

In the following section, we use the extensive publicly available data in the COSMOS field to explore the demographics, multiwavelength properties, and redshift distribution of WISE-selected AGNs using the simple W1 − W2 ⩾ 0.8 color criterion.

The effective area of the S-COSMOS survey is 2.3 deg² per passband after removal of poor quality regions around saturated stars and the field boundary (Sanders et al. 2007). Since approximately 8% of the field is only covered by two of the four IRAC passbands, this implies a four-band effective area of approximately 2.1 deg². Our simple W1 − W2 color criterion identified 130 AGN candidates in this area, implying a surface density of 61.9 ± 5.4WISE-selected AGN candidates per deg², up to 5% of which are expected to be T1.

For comparison, the Sloan Digital Sky Survey (SDSS) quasar selection algorithm (Richards et al. 2002) targeted ultraviolet excess quasars to i* = 19.1 (AB mag; 13.0 targets per deg²) and higher redshift (z ≳ 3) quasars to i* = 20.2 (AB mag; 7.7 targets per deg²), yielding a combined list of 18.7 candidates per deg². These depths are comparable to WISE depths for type 1 quasar selection (e.g., Assef et al. 2010). The SDSS algorithm is expected to provide over 90% completeness from simulated type 1 quasar spectra, although at the cost of lower reliability. The overall efficiency (quasars/quasar candidates) was only 66% from initial test data over 100 deg², with the contaminants evenly split between galaxies and Galactic stars. Importantly, optical/UV quasar selection methods are hampered at certain redshifts, especially high ones, where the stellar locus overlaps the quasar locus in color–color space. More sophisticated approaches, such as that presented in Bovy et al. (2011), work better than the old two-color or three-color cuts, though they still have issues. Other methods, such as variability (e.g., Palanque-Delabrouille et al. 2011), also do better, but require more elaborate input data sets. WISE selection is less affected by these problems and has a much flatter selection function as a function of redshift than traditional color-selected UV-excess methods. In particular, our simple W1 − W2 selection is expected to have 78% completeness and 95% reliability assuming that the Stern et al. (2005) mid-infrared selection of AGN candidates from the deeper S-COSMOS data is 100% reliable.

We have studied the longer wavelength properties of WISE-selected AGNs with an eye toward investigating whether the inclusion of W3 or W4 would allow for a more robust WISE selection of AGNs, such as the wedge in W1 − W2 versus W2 − W3 color–color space presented in Jarrett et al. (2011). Our simple W1 − W2 color criterion identified 130 AGN candidates in COSMOS, of which 24 (18%) are detected in W3 (⩾10σ) and only 3 (2%) are detected in W4 (⩾10σ). If we instead use a less conservative 5σ detection threshold, we find that 78 (60%) are detected in W3 and 17 (13%) are detected in W4. These percentages are essentially unchanged when we consider the 123 robust AGN candidates identified by both the WISE and IRAC selection criteria: 24 (20%) are detected in W3 (⩾10σ) and 3 (2%) are detected in W4 (⩾10σ). Using the 5σ detection threshold, these numbers increase to 74 (60%) being detected in W3 and 17 (14%) being detected in W4.

This implies that including the longer wavelength WISE data increases the reliability of the AGN selection, but at the cost of a significant decrease in completeness. For example, requiring a 10σ detection in W3 in addition to the W1 − W2 color criterion provides a surface density of only 11 WISE-selected AGN candidates per deg², but the reliability is expected to be ∼100%. Using the less conservative requirement of a 5σ detection in W3 provides a surface density of 37 WISE-selected AGN candidates per deg² with a reliability of 95%, e.g., the same reliability as our original W1 − W2 ⩾ 0.8 color cut. Any of these mid-infrared selection criteria compare quite favorably with the SDSS quasar selection in terms of surface density, completeness, and reliability. However, in what follows we rely on selecting AGNs using only the single W1 − W2 ⩾ 0.8 color criterion, as this selection, relying on the most sensitive WISE passbands, identifies a much larger candidate AGN population, 2–5 times larger than the samples that also require W3 detections.

Finally, we note that the depth of the WISE survey varies strongly with ecliptic latitude. The COSMOS field was selected to be at low ecliptic latitude in order to make it accessible from both hemispheres, and thus is close to the minimum depth of the WISE survey with a coverage of 12 frames (11 s each). Higher latitude fields have deeper data, reaching a coverage of 250 frames at the ecliptic poles (Jarrett et al. 2011). This increases the surface density of AGN candidates, but with a decrease in robustness since deeper pointings will detect massive galaxies at z ≳ 1 which have similar W1 − W2 colors to AGNs. The inclusion of the deeper W3 data in such fields could be used to separate normal galaxies from AGNs. WISE selection of AGNs in deeper, higher latitude fields is addressed more thoroughly in Assef et al. (2012).

The COSMOS field has been observed by both the Chandra X-Ray Observatory (Elvis et al. 2009) and XMM-Newton (Hasinger et al. 2007). In the following two subsections, we examine the X-ray properties of WISE-selected AGNs using each of these surveys.

3.3.1. Chandra Observations

The Chandra-COSMOS Survey (C-COSMOS; Elvis et al. 2009) is a large, 1.8 Ms Chandra program that imaged the central 0.5 deg² of the COSMOS field with an effective exposure time of 160 ks per position and the outer 0.4 deg² with an effective exposure time of 80 ks per position. The corresponding point source depths in the deeper portion of the survey are 1.9 × 10⁻¹⁶  erg cm⁻² s⁻¹ in the soft (0.5–2 keV) band, 7.3 × 10⁻¹⁶  erg cm⁻² s⁻¹ in the hard (2–10 keV) band, and 5.7 × 10⁻¹⁶  erg cm⁻² s⁻¹ in the full (0.5–10 keV) band, where these depths assume an average X-ray power-law index Γ = 1.4. C-COSMOS detected 1761 reliable X-ray point sources (catalog ver.2.1; spurious probability <2 × 10⁻⁵). We use a 25 matching radius to cross-identify the WISE and Chandra sources.

A total of 167 of the Chandra X-ray sources have WISE counterparts (Figure 7). Most have relatively blue W1 − W2 colors and are likely associated with low-redshift galaxies harboring low-luminosity AGNs; such sources are common in deep X-ray observations (e.g., Brandt & Hasinger 2005). A concentration of X-ray sources near zero mid-infrared colors are predominantly low-redshift (z ≲ 0.6) early-type galaxies: such galaxies have little or no star formation and therefore lack the dust and PAH emission which causes a horizontal extension in this mid-infrared color–color space. Finally, deep Chandra surveys are also sensitive to X-ray emission from low-mass Galactic stars, which likewise reside in this same region of color–color space (e.g., Stern et al. 2002b).

Zoom In Zoom Out Reset image size

More interesting is the vertical extension seen in Figure 7: 41 of the WISE-selected AGN candidates have C-COSMOS counterparts. Most of the WISE+IRAC AGN candidates lacking Chandra counterparts in Figure 7 are from outside the C-COSMOS FoV. Six (e.g., 13%) of the WISE-selected AGN candidates whose IRAC colors are consistent with the Stern et al. (2005) AGN selection criteria were observed by, but not detected by Chandra. We list these sources in Table 1. All are at least 9'' from the nearest Chandra source. These mid-infrared sources are strong candidates for Compton-thick AGNs (e.g., N_H ⩾ 10²⁴ cm⁻²): sources with so much internal absorption that their X-ray emission below 10 keV is heavily absorbed, undetected in the ∼100 ks Chandra observations. The absorbing material, however, is heated up and is easily detected in ∼100 s integrations with WISE. We also note, as expected, that the small number of WISE AGN contaminants with W1 − W2 ⩾ 0.8 but whose IRAC colors are not consistent with an AGN are undetected by Chandra.

Table 1. WISE +IRAC AGN Candidates Undetected by Chandra

WISE ID	i	W1	W2	W1 − W2	[5.8]−[8.0]	z	Notes
J095855.40+022037.4	19.43	15.41	14.59	0.82	1.39	[0.38]	Bright galaxy
J095937.35+021905.9	22.30	16.00	14.70	1.30	1.17	0.927
J100006.19+015535.3	20.90	15.53	14.19	1.34	1.05	0.661
J100043.70+014202.5	21.77	15.89	14.89	1.00	1.40	0.741
J100046.91+020726.5	21.86	14.87	13.14	1.73	1.18	1.158	Faint
J100135.61+022104.8	25.11	16.98	14.97	2.01	1.49	...	Faint

Note. Bracketed redshift indicates photometric redshift.

Download table as:  ASCIITypeset image

Figure 8 shows the relationship between hard X-ray (2–10 keV) fluxes of S-COSMOS sources and their hardness ratios HR ≡ (H − S)/(H + S), where H and S are the numbers of hard and soft X-ray photons detected, respectively. We compare the full sample (smallest black dots) to the subset with WISE counterparts (small black circles) to the smaller set of sources with IRAC and/or WISE colors indicative of AGN activity (larger symbols). Note that many of the brightest X-ray sources in the field are identified as AGN candidates by WISE, regardless of their hardness ratio.

Zoom In Zoom Out Reset image size

3.3.2. XMM-Newton Observations

The XMM-Newton wide-field survey of the COSMOS field (XMM-COSMOS; Hasinger et al. 2007; Brusa et al. 2010) observed the entire 2 deg² COSMOS field to medium depth (∼60 ks). The survey detected nearly 2000 X-ray sources down to limiting fluxes of ∼5 × 10⁻¹⁶ erg cm⁻² s⁻¹ in the 0.5–2 keV (soft) band and ∼3 × 10⁻¹⁵ erg cm⁻² s⁻¹ in the 2–10 keV (hard) band. Thus, XMM-COSMOS covers a wider area than C-COSMOS, albeit to shallower depth. We use the 2008 November XMM-Newton point-like source catalog, available thru IRSA, which contains 1887 sources. Using a 35 matching radius—slightly larger than used for C-COSMOS to account for the poorer spatial resolution of XMM-Newton—we identify WISE counterparts for 244 XMM-COSMOS sources, of which 92 have W1 − W2 ⩾ 0.8. The mid-infrared color distribution of XMM sources is similar to what was seen for C-COSMOS, though more of the mid-infrared AGN candidates have X-ray detections courtesy of the wider spatial coverage of this survey (Figure 9). Thirty-three WISE sources whose IRAC colors suggest an active nucleus (out of 123) are undetected by XMM-Newton, though seven are from outside the XMM-Newton coverage; the other 26 are listed in Table 2. All six of the sources from Table 1 remain undetected by XMM-Newton.

Zoom In Zoom Out Reset image size

Table 2. WISE +IRAC AGN Candidates Undetected by XMM-Newton

WISE ID	i	W1	W2	W1 − W2	[5.8]−[8.0]	z	Notes
J095733.79+020943.1	19.66	16.33	15.08	1.25	1.07	1.441	SDSS QSO
J095736.56+020236.7	21.31	16.12	14.74	1.38	1.35	...
J095752.32+022021.2	18.80	15.75	14.47	1.28	1.32	2.050	SDSS QSO
J095756.65+020719.5	20.27	15.67	14.87	0.80	0.78	[0.31]
J095821.40+025259.0	19.53	15.67	14.50	1.17	1.29	...
J095855.40+022037.4	19.43	15.41	14.59	0.82	1.39	[0.38]	Bright galaxy
J095905.55+025145.0	19.85	15.95	15.08	0.87	1.35	...
J095937.35+021905.9	22.30	16.00	14.70	1.30	1.17	0.927
J095945.60+013032.2	20.34	15.97	15.01	0.97	0.99	1.106	SDSS QSO
J100006.19+015535.3	20.90	15.53	14.19	1.34	1.05	0.661
J100008.42+020247.4	20.24	15.87	14.82	1.05	1.12	0.370	Type-2 AGN
J100008.93+021440.5	18.93	15.93	14.72	1.20	1.26	2.536	QSO
J100013.51+013739.2	19.94	16.23	14.83	1.40	1.12	1.608	SDSS QSO
J100043.70+014202.5	21.77	15.89	14.89	1.00	1.40	0.741
J100046.91+020726.5	21.86	14.87	13.14	1.73	1.18	1.158	Faint
J100115.38+024231.4	20.44	16.21	15.02	1.20	1.13	...
J100135.61+022104.8	25.11	16.98	14.97	2.01	1.49	...	Faint
J100137.11+024650.6	21.58	16.36	14.87	1.49	1.16	0.143
J100142.22+024330.7	22.69	15.74	14.02	1.72	1.28	[1.62]
J100231.86+015242.3	21.01	15.85	14.68	1.17	0.99	...
J100244.77+025651.5	20.26	15.86	14.72	1.14	1.09	...
J100246.35+024609.6	19.08	15.63	14.69	0.94	1.37	...
J100253.31+020222.1	20.84	15.32	14.24	1.08	1.61	0.902
J100303.46+022632.0	20.69	15.68	14.82	0.85	1.14	...
J100305.98+015704.0	19.46	14.63	12.91	1.72	1.32	0.370	Type-2 AGN
J100322.00+014356.5	...	16.21	15.08	1.12	1.05	...

Note. Bracketed redshifts indicate photometric redshifts.

Download table as:  ASCIITypeset image

Therefore, 75% of the WISE-selected AGN candidates have XMM-Newton counterparts, but that still leaves a significant number of WISE-selected AGNs—including those whose IRAC colors indicate an AGN—that are undetected by XMM-Newton. We also note that two of the sources with W1 − W2 ⩾ 0.8 but outside of the Stern et al. (2005) IRAC wedge are detected in the X-rays; both have IRAC colors very close to the wedge defined in that paper.

Similar to the Chandra results, we find that the WISE-selected AGNs are brighter and have softer spectra than typical XMM-Newton sources in XMM-COSMOS. However, Figure 10 also shows quite clearly that the brightest X-ray sources tend to be identified as AGN candidates from their WISE colors, regardless of X-ray hardness ratio. For comparison, we also plot the expected point source sensitivity of the all-sky eROSITA telescope. The brightest soft X-ray sources, F_{0.5–2 keV} ≳ 2 × 10⁻¹⁴ erg cm⁻² s⁻¹, will basically all already have been identified as AGN candidates by WISE. At the sensitivity limit of eROSITA, however, large numbers of X-ray sources are expected that are not identified by WISE; these are likely lower luminosity AGNs at lower redshifts (e.g., Eckart et al. 2010; Donley et al. 2012). WISE also detects a significant population of fainter, harder spectrum X-ray sources, below the sensitivity limit of eROSITA. These results, particularly Figure 10, emphasize the complementarity of X-ray and mid-infrared AGN selection: each selection technique identifies samples of AGNs missed by the other technique.

Zoom In Zoom Out Reset image size

Finally, using the greater statistics of the XMM-Newton sample, we consider if there are any trends between X-ray hardness ratio and mid-infrared W1 − W2 color. Though no strong correlation is evident, we do find the expected general trend of redder mid-infrared sources having harder X-ray spectra. Splitting the X-ray sample at HR = 0, the softer mid-infrared AGN candidates (e.g., HR < 0) have a mean WISE color of 〈W1 − W2〉 = 1.18. In contrast, the harder mid-infrared AGN candidates (e.g., HR > 0) have a mean WISE color of 〈W1 − W2〉 = 1.32. However, there are examples of very hard X-ray sources with relatively blue WISE colors, as well as very soft X-ray sources with relatively red WISE colors.

The Very Large Array (VLA) obtained deep radio images of the COSMOS field at 20 cm. The goals, observing strategy, and data reductions for this large program, called the VLA-COSMOS survey, are described in Schinnerer et al. (2007). The survey entailed nearly 350 hr of exposure time, primarily in the highest resolution, or A, configuration. The Large project imaged the full 2 deg² COSMOS field with a resolution of 15 to a sensitivity of ∼11 μJy (1σ) (Bondi et al. 2008). We use the joint catalog of Schinnerer et al. (2010), which combines an improved analysis of the Large project with data from the Deep project that doubled the integration time in the central 0.84 deg² region of the survey. The Joint catalog includes 2865 sources, of which 131 consist of multiple components.

We match the full Joint catalog to the full WISE source list of 3618 sources using a 15 matching radius. We find 333 matches, of which 33 have multiple components in the VLA data. Figure 11 shows the mid-infrared colors of the VLA sources, which shows many radio-detected sources on both sides of our W1 − W2 = 0.8 color cut. This is unsurprising, as these extremely deep radio data detect emission related to stellar processes (e.g., supernova remnants) as well as AGN activity at a range of Eddington ratios. Considering the WISE-selected AGN candidates, 55 of the 130 candidates (42%) have radio matches; five of these are flagged as consisting of multiple components in the radio data. Of the 123 AGN candidates identified by both WISE and IRAC, 52 (42%) are detected by these very deep radio data. For the WISE-selected AGNs not flagged as likely AGNs by IRAC, three (43%) are detected by the VLA. All of these fractions are much higher than the 8% of sources with W1 − W2 < 0.8 which are detected by the VLA-COSMOS survey.

Zoom In Zoom Out Reset image size

Assuming a typical quasar radio spectral index α = −0.5 (S_ν∝ν^α) and adopting the Gregg et al. (1996) cutoff value for the 1.4 GHz specific luminosity, L_1.4 GHz = 10^32.5 h⁻²₅₀ erg s⁻¹ Hz⁻¹ (≈10²⁴ h⁻²₅₀ W Hz⁻¹ sr⁻¹), to discriminate radio-loud and AGN radio-quiet populations, only 2 of the 98 WISE-selected AGNs with spectroscopic redshifts are radio loud.¹³ The results are unchanged if we assume α = −0.8, as might be more typical for quasars without the jet aligned along our line of sight. Both of these sources are SDSS quasars at z > 1 (WISE J095821.65+024628.2 at z = 1.405 and WISE J095908.32+024309.6 at z = 1.318). These two sources come from a total of 45 WISE-selected AGNs that are optically bright and have spectroscopy from the SDSS; all are classified as broad lined, and 35 are at z > 1, implying that they are clearly luminous quasars. Considering just this SDSS subsample of WISE-selected AGNs, our results are statistically consistent with the canonical value of ∼10% of quasars being radio loud (e.g., Stern et al. 2000b). However, the fact that none of the 53 other sources with spectroscopic redshifts are radio loud is surprising, suggesting that the radio-loud fraction might be different for type 2 AGNs. We note, however, that Zakamska et al. (2004) found no change in the radio-loud fraction for their sample of SDSS-selected obscured quasars.

Note that radio-loud AGNs do not fall below the 5σ VLA-COSMOS sensitivity until beyond z ∼ 10, e.g., the survey is sensitive enough to detect all radio-loud AGNs that WISE is likely to detect. All of the WISE-selected AGN candidates without redshifts that were detected by the VLA-COSMOS survey have S_ν < 0.72 mJy, implying that they would have to be at z > 3.4 in order to be radio loud.

We next consider the optical properties of the WISE-selected AGN candidates. Figure 12 shows the r-band magnitude distributions of all WISE sources in the COSMOS field (open histogram) as well as the AGN candidates with W1 − W2 ⩾ 0.8 (solid histogram). All photometry is in the AB system and comes from the COSMOS photometry catalog of Capak et al. (2007), which is available through IRSA. The left panel shows r-band photometry from the second data release (DR2) of the SDSS (Abazajian et al. 2004). Objects as bright as 10th mag have good photometry in the SDSS imaging, and the imaging depth, defined as the 95% completeness limit for point sources, is r ∼ 22.2. Similar data are available over more than 11,000 deg². However, many of the WISE sources are fainter than this limiting depth, so in the right panel of Figure 12 we show the Subaru r⁺ photometry in the COSMOS field obtained with the Suprime-Cam instrument (Komiyama et al. 2003). These data reach a 5σ depth (3'' aperture) of 26.6 and detect all of the WISE-selected AGN candidates. Many of the brighter sources in the Subaru data are unresolved, leading to saturation issues. This causes the truncation seen at r⁺ ≲ 18.

Zoom In Zoom Out Reset image size

Figure 12 shows that very few of the WISE-selected AGN candidates are brighter than r ∼ 18, and they represent less than 5% of the WISE source population at bright optical magnitudes. However, the AGN candidates represent an increasing fraction of the optically fainter WISE sources, accounting for ∼20% of WISE sources at r ∼ 21 and more than 50% of WISE sources with r ≳ 23. In order to explore the optical brightnesses of WISE-selected AGN candidates with higher fidelity, we identified AGN candidates over 4000 deg² of the SDSS (cf., Donoso et al. 2012; Yan et al. 2012). Figure 13 shows the resultant r-band distribution. While the majority of AGN candidates are well detected in the SDSS imaging, their optical brightness distribution peaks at r ∼ 19.5, making them fainter than the typical spectroscopic limits of SDSS, i ∼ 19.1 for quasars and r ∼ 17.8 for galaxies (as discussed in the following section, approximately half of our WISE-selected AGN candidates are spatially resolved).

Zoom In Zoom Out Reset image size

Figure 14 shows an optical color–magnitude diagram of WISE sources. At bright magnitudes, the distribution is dominated by Galactic stars with r − i ∼ 0.1, while a second galaxy sequence becomes evident at r ≳ 18. Mid-infrared AGN candidates are, on average, bluer than typical galaxies, though the color distribution clearly overlaps with the galaxy sample. AGN candidates identified by both WISE and Spitzer/IRAC, which represent the most robust AGN sample with the highest rate of X-ray detections, tend to be bluer than galaxies at i ≲ 21, though at fainter magnitudes where obscured AGNs become more prevalent, the distribution fans out. This is partially due to photometric errors, but also because a large fraction of the AGN candidates have colors similar to galaxies. Not surprisingly, AGN candidates identified by Spitzer but not by WISE—e.g., sources at the blue corner of the Stern et al. (2005) wedge and close to the galaxy locus in mid-infrared color–color space—tend to have galaxy-like optical colors.

Zoom In Zoom Out Reset image size

Conventional wisdom states that the most luminous AGNs in the universe are associated with unresolved point sources at optical wavelengths. While this is true for the vast majority of unobscured, type 1 quasars, this is not the case for obscured, type 2 quasars. For instance, luminous high-redshift radio galaxies often have a clumpy, irregular morphology at rest-frame ultraviolet wavelengths, with the emission generally elongated and aligned with the radio source axis (e.g., McCarthy et al. 1987). At rest-frame optical wavelengths, where stars dominate the galaxy luminosity, the hosts of most luminous radio galaxies are normal elliptical galaxies with r^1/4-law light profiles (e.g., Zirm et al. 2003).

The cornerstone data set for the COSMOS survey is its wide-field Hubble Space Telescope Advanced Camera for Surveys (ACS) imaging (Scoville et al. 2007; Koekemoer et al. 2007). With 583 single-orbit F814W (I₈₁₄ hereafter) observations, these data cover (or define) the 1.8 deg² COSMOS field and constitute the largest contiguous Hubble imaging survey to date. The data are extremely sensitive, with 009 resolution (FWHM) and reaching a 50% completeness limit of I₈₁₄ = 26.0 (AB) for sources 05 in diameter.

Griffith & Stern (2010) have recently analyzed the optical morphologies of AGNs in the COSMOS field identified from a variety of methods: radio selection, X-ray selection, and mid-infrared selection (see also Gabor et al. 2009). They find that the radio-selected AGNs are likely to be hosted by early-type galaxies, while X-ray- and mid-infrared-selected AGNs are more often associated with point sources and disk galaxies. Considering just the brighter X-ray and mid-infrared subsamples, approximately half of the AGNs are optically unresolved and a third are associated with disk galaxies. These morphological results conform with the results of Hickox et al. (2009) who studied the colors and large-scale clustering of AGNs, and found a general association of radio-selected AGNs with "red sequence" galaxies (an old, well-known result; e.g., Matthews et al. 1964), mid-infrared-selected AGNs are associated with "blue cloud" galaxies, and X-ray-selected AGNs straddle these samples in the "green valley."

We find similar results here. Of the 130 WISE-selected AGN candidates, 94 are located within the portion of COSMOS imaged by ACS. A bit more than half (52/94, or 55%) of the sources are spatially resolved; the other AGN candidates are associated with point sources. As an aside, we note that one of the contaminants, WISE J100050.63+024901.7, was flagged by Faure et al. (2008) as one of the 20 most likely strong lensing systems in the COSMOS field (see also Jackson 2008). The ACS I₈₁₄ image of this system shows four faint arcs surrounding a bright early-type galaxy, with a radial separation of 19. Inspection of the IRAC images for this system shows that the mid-infrared data are still dominated by the optically bright lensing galaxy.

Figure 15 shows Hubble/ACS I₈₁₄ images of four example WISE-selected AGNs. All four examples were also identified as AGN candidates by their IRAC colors. Several are X-ray and/or radio sources as well. Panel (a) shows WISE J095834.75+014502.4, a bright SDSS quasar at z = 1.889. It is unresolved in the ACS image; approximately half of the WISE AGN candidates have similar morphologies. Panel (b) shows WISE J100109.21+022254.2, one of the optically faintest WISE-selected AGN candidates in the COSMOS field, with I₈₁₄ = 22.9. As discussed in the next section, we were unable to obtain a redshift for this source from deep Keck spectroscopy, though subsequent to our observations, Brusa et al. (2010) reported that this source is as a narrow-lined AGN at z = 1.582. Panels (c) and (d) show two z ∼ 0.9 type 2 (e.g., narrow-lined) AGNs from Trump et al. (2007). The former is WISE J100005.99+015453.3, which is a spiral galaxy with a very bright nucleus. The latter is WISE J100013.42+021400.4, which has a more irregular morphology.

Zoom In Zoom Out Reset image size

Figure 16 presents a color–magnitude diagram of WISE sources in the COSMOS field, with optical-to-mid-infrared color plotted against 4.6 μm brightness (W2). For the AGN candidates, we only plot the ∼70% of sources covered by the ACS imaging. There are several things to note from this plot. First, the WISE-only AGN candidates (e.g., WISE-selected AGN candidates not identified as AGN candidates from their Spitzer colors) clearly reside on the right side of Figure 16, with no contaminants brighter than W2 = 14.8. This implies that caution should be applied before extending our simple WISE color criterion to fainter limits. Indeed, in Assef et al. (2012) we investigate the interloper fraction as a function of W2 magnitude and derive a magnitude-dependent WISE AGN selection criterion applicable to higher ecliptic latitude (e.g., deeper) portions of the WISE survey.

Zoom In Zoom Out Reset image size

Second, optically unresolved AGN candidates tend to have bluer r − W2 colors, consistent with the expectation that they suffer less extinction at optical wavelengths. Quantitatively, the unresolved WISE+IRAC AGN candidates in Figure 16 have 〈r − W2〉 = 5.18. Similarly selected sources that are resolved in the Hubble imaging have 〈r − W2〉 = 7.29. The unresolved AGN candidates are also slightly brighter, with 〈W2〉 = 14.54 as compared to 〈W2〉 = 14.69 for the resolved WISE+IRAC AGN candidates. Importantly, however, we note that both resolved and unresolved sources are found across the full W2 range probed.

Finally, we also consider the optical properties of the WISE+IRAC AGN candidates that are undetected by XMM-Newton (Table 2). These sources have an even fainter median mid-IR brightness, 〈W2〉 = 14.72, and also have redder optical-to-mid-IR colors than typical WISE+IRAC AGN candidates. Quantitatively, 〈r − W2〉 = 6.58 for the X-ray-undetected AGN candidates, while 〈r − W2〉 = 5.85 for the X-ray-detected WISE+IRAC AGN sample. This is consistent with the X-ray-undetected sources being associated with more heavily obscured AGNs, diminishing both their optical and X-ray fluxes.

In order to understand the redshift distribution and properties of WISE-selected AGN candidates, we have both matched the candidate list to publicly available spectroscopy in the COSMOS field and obtained new observations. Published spectroscopy come from several papers: bright targets have spectroscopic redshifts from the SDSS (Abazajian et al. 2009); Prescott et al. (2006) report on MMT/Hectospec follow-up of optically selected quasar candidates in COSMOS; Lilly et al. (2007) report on zCOSMOS, a large VLT/VIMOS I-band magnitude-limited survey of the COSMOS field; Trump et al. (2007) and Trump et al. (2009) report on Magellan/IMACS spectroscopy of X-ray- and radio-selected AGN candidates in the COSMOS field; and Brusa et al. (2010) report on spectroscopy of X-ray sources from the XMM-Newton wide-field survey of the COSMOS field, synthesizing both previously published results and new spectroscopy from Keck.

We obtained additional spectroscopy on UT 2010 March 12–15 using the Low Resolution Imaging Spectrometer (LRIS; Oke et al. 1995) and the DEep Imaging Multi-Object Spectrograph (DEIMOS; Faber et al. 2003). We observed three Keck slitmasks in the COSMOS field. On UT 2010 March 12 we observed cos10b for 5200 s using the dual-beam LRIS instrument. We used the 400 ℓ mm⁻¹ grism on the blue arm of the spectrograph (blazed at 3400 Å; resolving power R ≡ λ/Δλ ∼ 600), the 400 ℓ mm⁻¹ grating on the red arm of the spectrograph (blazed at 8500 Å; R ∼ 700), and 6800 Å dichroic. On UT 2010 March 13 we observed cos10a for 1200 s using LRIS. The red CCD was non-functional that night, so we channeled all of the light to the blue arm of the spectrograph and again used the 400 ℓ mm⁻¹ grism blazed at 3400 Å. On UT 2010 March 14 we observed cos10d for 3600 s with DEIMOS in cloudy conditions, using the 600 ℓ mm⁻¹ grating (blazed at 7500 Å; R ∼ 1600) and the 4000 Å order-blocking filter. Masks all used ∼12 wide slitlets. Data reduction followed standard procedures, and we flux calibrated the data using observations of standard stars from Massey & Gronwall (1990). Note that the DEIMOS data were taken in non-photometric conditions, resulting in an uncertainty in the flux scale of those sources.

Target selection was done prior to access to the WISE data in the COSMOS field, though we had already anticipated that red W1 − W2 colors would be an effective method to identify a large population of AGNs. We sought to test that hypothesis using Spitzer/IRAC imaging from the S-COSMOS survey (Sanders et al. 2007), assuming that W1 ∼ [3.6] and W2 ∼ [4.5]. Additional targets were selected using the Stern et al. (2005) IRAC AGN wedge selection criteria. Figure 17 and Table 3 present the results for the six COSMOS targets that subsequently were found to match our W1 − W2 ⩾ 0.8 AGN candidate selection criterion. All six would also be selected by the Stern et al. (2005) IRAC criteria. Table 4 in the Appendix presents the results for the additional COSMOS targets observed on these masks. Our new Keck results are occasionally slightly discrepant with previous results, but typically with Δz ⩽ 0.01. The S/N of these new data are quite high and the data were taken at relatively high spectral dispersion, suggesting that these redshifts should take precedence over previous results.

Zoom In Zoom Out Reset image size

Table 4. Additional Results from Keck Observations

I814	R.A.	Decl.	z	Q	Slitmask(s)	Notes
17.32	10:00:11.83	+02:26:23.1	0.440	A	D[35]	[O ii], [Ne iii], Hζ, H, Hδ, Hγ, Hβ, [O iii]
16.26	10:00:14.89	+02:27:17.9	0.728	A	D[38]	[O ii], Hβ, [O iii]
19.30	10:00:22.79	+02:25:30.6	0.349	A	D[3]	CaHK, Hα, [N ii]
17.99	10:00:24.28	+02:27:36.2	1.243	B	D[39]	[O ii]
17.14	10:00:24.51	+02:26:18.1	1.129	A	D[34]	[O ii]
18.36	10:00:28.56	+02:27:25.8	0.248	A	D[4]	CaH, Hγ, Hβ, [O iii], Hα, [N ii]
17.15	10:00:29.55	+02:26:35.9	0.348	B	D[36]	[O ii], Hβ, [O iii], Hα, [N ii] (could be serendip)
16.93	10:00:32.46	+02:27:59.3	1.405	B	D[41]	[O ii]
16.80	10:00:33.23	+02:27:59.3	0.981	B	D[42]	[O ii]
17.93	10:00:44.50	+02:23:54.0	1.299	B	D[17]	[O ii], CaHK
18.43	10:00:50.15	+02:26:18.5	3.730	A	D[33]	QSO: Lyα,C iv
18.56	10:00:50.58	+02:23:29.3	3.093	A	D[13]	QSO: Lyα,C iv, C iii]
19.08	10:00:56.65	+02:26:35.5	0.344	A	B, D[0]	Mg ii absn,[O ii], CaH, Hγ, Hβ, [O iii], Hα, [N ii]
17.55	10:00:58.07	+02:26:16.8	0.425	A	D[32]	[O ii], Hβ, [O iii], Hα
16.55	10:00:58.70	+02:25:56.2	0.694	A	D[28]	QSO: [Ne v], [O ii], Hβ, [O iii]
17.09	10:00:59.00	+02:24:17.9	1.193	B	B	[O ii]
19.15	10:00:59.81	+02:24:30.7	0.541	A	B	[O ii]
16.94	10:01:08.35	+02:23:42.0	1.930	A	B, D[15]	QSO: Lyα,C iv, C iii],Mg ii
18.22	10:01:08.65	+02:23:14.1	0.503	A	D[12]	[O ii], D4000
16.52	10:01:13.93	+02:25:48.1	0.373	A	B	QSO: broad Mg ii,[Ne v], [O ii], Hα
17.18	10:01:14.68	+02:24:49.5	1.656	A	B	LBG: C ii, C iv,[O ii]
16.36	10:01:17.00	+02:27:31.2	0.518	A	B	Hα
16.25	10:02:17.42	+02:29:59.7	1.100	A	A	QSO: C iv, C iii]; odd broad lines at ∼1650 Å
17.56	10:02:28.18	+02:30:15.4	0.344	B	A	[O ii]
15.23	10:02:29.89	+02:32:25.1	0.431	A	A	AGN: [Ne v], [O ii], [Ne iii], [O iii]
17.76	10:02:31.90	+02:35:07.4	0.880	A	A	AGN: C iii],Mg ii

Notes. Q indicates the quality of the redshift (see text for details). Masks A and B were observed with LRIS. Mask D was observed with DEIMOS; the bracketed numbers indicate the DEIMOS slitlet number.

Download table as:  ASCIITypeset image

Four of the sources show prominent AGN features, such as broadened Mg ii 2800 emission. WISE J100036.06+022830.5 does not show obvious AGN features; the spectrum shows narrow emission lines from [O ii], [Ne iii], Hβ, and [O iii], as well as Balmer absorption lines indicative of a relatively young stellar population. We find log  ([O iii]/Hβ) ∼0.12, which is consistent with both star-forming and AGN activity in the Baldwin et al. (1981) diagram; spectral features redward of our data are required to distinguish the principle line excitation mechanism. We did not obtain a redshift for WISE J100109.23+022254.5 with our data, though Brusa et al. (2010) identify this source as a narrow-lined AGN z = 1.582 on the basis of their deep Keck/DEIMOS spectroscopy; the quality of the redshift is not indicated. Our spectroscopy does not show any features such as redshifted C iv emission or absorption to confirm that redshift. However, we note that strongest feature at this redshift is likely to be [O ii] emission at 9623 Å. This is beyond the wavelength coverage of our Keck spectroscopy.

Figure 18 presents the distribution of spectroscopic redshifts for the WISE AGN candidates. We have spectroscopic redshifts for 101 of the 130 candidates (72%); the median redshift is 〈z〉 = 1.11. Seven of these candidates are from outside the IRAC AGN wedge, four of which have spectroscopic redshifts. Two are broad-lined quasars at z ∼ 1; the other two are galaxies at z = 0.27 and z = 0.75 from zCOSMOS. This suggests that the 95% reliability rate derived in Section 3.1 is actually a lower limit; some of the WISE-selected candidates are indeed AGNs despite not being identified as such by their IRAC colors.

Zoom In Zoom Out Reset image size