THE HOST GALAXIES OF MICRO-JANSKY RADIO SOURCES

SERVS uses deep Spitzer data taken during the postcryogenic Spitzer Space Telescope mission over the course of 1400 hr using the Infrared Array Camera (IRAC) at wavelengths of 3.6 and 4.5 μm, and detects objects out to $z\approx 5$. The survey covers a total of 18 deg² to a detection limit of $\approx 2\;\mu \mathrm{Jy}$ (AB = 23.1) in five fields, including 4 deg² in the Lockman Hole. SERVS catalogs were produced using Sextractor (Bertin & Arnouts 1996) as detailed in Mauduit et al. (2012).

González-Solares et al. (2011) observed the Lockman Hole with the Isaac Newton Telescope Wide Field Camera in the $u,g,r,i,$ and z bands, with AB magnitude limits of 23.9, 24.5, 24.0, 23.3, and 22.0, respectively. Supplementary observations were taken with the Mosaic 1 camera on the Mayall 4 m Telescope of the Kitt Peak National Observatory in the $g,r,$ and i bands. Details of the source extraction are given in González-Solares et al. (2011).

The UKIRT Infrared Deep Sky Survey (UKIDSS) observed the Lockman Hole as part of its Deep Extragalactic Survey using the Wide Field Camera (WFCAM) of the United Kingdom Infrared Telescope (UKIRT) at wavelengths of 1.2 and 2.2 μm, corresponding to the J and K bands, respectively. The survey reaches AB magnitude depths of ${J}_{{AB}}=23.1$ and ${K}_{{AB}}=22.5$. Lawrence et al. (2007) describe the survey and the production of the catalogs.

The SWIRE survey observed the Lockman Hole during its cryogenic mission using the IRAC and the Multiband Imaging Photometer for Spitzer (MIPS). IRAC observed the Lockman Hole using all four channels, 3.6, 4.5, 5.8, and 8.0 μm; and MIPS observed the Lockman Hole with all three of its channels, 24, 70, and 160 μm. The SWIRE catalogs reach depths of $\approx 10\;\mu \mathrm{Jy}$ (${AB}\approx 21.4$) at 3.6 and 4.5 μm. SWIRE detected galaxies out to $z\approx 3$ (Lonsdale et al. 2003). Details of the data release used may be found in the SWIRE data release 2 document.¹⁶

Part of the Lockman Hole was observed at frequencies of 610 MHz and 1.4 GHz using the Giant Metre-wave Radio Telescope (GMRT) and the Very Large Array (VLA), respectively, by Ibar et al. (2009). Their survey covered $\approx 0.5$ deg² in the southern region of the Lockman Hole (centered on R.A., decl. 1630, 5735, generally called the "Lockman Hole East") and contains 1452 radio sources detected at 1.4 GHz, comprised of 1467 discrete components. The GMRT data reach an rms noise of $\approx 15\;$ μJy beam⁻¹, and the VLA data an rms of $\approx 6\;$ μJy beam⁻¹ in the center of the surveys, where the sensitivity is highest, tapering off to $\approx 10\;$ μJy beam⁻¹ at 1.4 GHz toward the edge of the survey region (see Figure 2 for the flux density distribution). The depth of the GMRT data is such that nearly all of the 1.4 GHz sources have counterparts at 610 MHz.

Zoom In Zoom Out Reset image size

Photometric redshifts of the objects observed in the SERVS study were obtained by J. Pforr et al. (2015, in preparation) using the Hyper-z photometric redshift code (Bonzella et al. 2000). A correlation of the photometric redshifts with the spectroscopic redshifts of AGNs in the field (see below) is described in the appendix, and shows that ∼85% of the redshifts are accurate to within an rms of ${\rm{\Delta }}z/(1+z)\approx 0.06$, the remainder being outliers. Highly luminous AGNs at high redshifts, where the hot dust from AGNs in the IRAC 3.6 and 4.5 μm bands can affect the redshift estimate, however, may have poorer photo-zs.

Of the photometric redshifts for the radio galaxies, 20 (1%) failed to find a solution. All of these failed solutions had only limits on their detection in either, or both, of the J-band or K-band. To ensure a good quality photometric redshift, we thus excluded the 254 objects (20% of the 1245 objects matched between the radio and SERVS) lacking a J- or K-band detection from any analysis that needed redshift information, ensuring that we had a minimum of four bands to estimate a redshift (J, K, [3.6], and [4.5]). In Figure 2, we show the 1.4 GHz flux density distribution of the survey for objects with and without photometric redshifts, indicating that there is not a very strong bias toward, for example, only the brighter radio sources having photometric redshift information. The photometric redshifts extend out to $z\approx 5$, as seen in Figure 3 (though some of the redshifts at $z\gt 4$ use only the minimum four bands, and thus may be unreliable). There is a dip in the redshift distribution at $z\approx 1.4$, which has no clear origin. Two possible explanations are that it may be a degeneracy in the photometric redshifts in the region of the "redshift desert" where the optical bands, in particular, do not contain strong spectral features to constrain the redshifts, or that it may reflect real large-scale structure in the field. Spectroscopic redshifts were available for 62 of the AGNs from Lacy et al. (2013) or the SDSS quasar survey (Ahn et al. 2014), including the normal type-1 AGN, whose optical emission is dominated by quasar light. We have used these in preference to the photometric redshifts where applicable.

Zoom In Zoom Out Reset image size

We used the SKA Simulated Skies (S³) extragalactic model (Wilman et al. 2008) that uses measured luminosity functions combined with an underlying numerical dark matter simulation to obtain a predicted redshift distribution for μJy radio sources. This predicted redshift distribution is compared to our measured one in Figure 3. Both distributions peak at $z\approx 0.7$, though our measured distribution has fewer sources in the lowest redshift bin. This deficit is also seen in the smaller (but much more spectroscopically complete) sample of Simpson et al. (2012), and thus is unlikely to be due to large-scale structure fluctuations, though we cannot rule out incompleteness at very low redshifts due to galaxies being resolved out in the radio survey. We see fewer sources in the $z\gt 2.5$ bins than the predictions, though this may be explained by the 20% of faint sources we excluded due to missing J- and/or K-band detections. The median redshift we observe, 0.99, is very similar to the median redshift predicted by the simulations (1.04).

When looking for the relative contributions of AGNs and star-forming galaxies, bright AGNs are simpler to isolate than star-forming galaxies. For the 479 objects that were detected in all four of the 3.6, 4.5, 5.8, and 8.0 μm bands of the IRAC in the SWIRE data, we were able to isolate rapidly accreting, cold-mode AGNs using the method proposed by Lacy et al. 2004, 2007), which selects AGNs based on their color relative to star-forming or quiescent galaxies. Cold-mode AGNs have a characteristic red power law in the mid-infrared that stands out well in a ${\mathrm{log}}_{10}({S}_{5.8}/{S}_{3.6})$ versus ${\mathrm{log}}_{10}({S}_{8.0}/{S}_{4.5})$ color–color plot, (Figure 4, left). We used ${S}_{3.6}$ and ${S}_{4.5}$ from SERVS and ${S}_{5.8}$ and ${S}_{8.0}$ from SWIRE. Three cuts were made: ${\mathrm{log}}_{10}({S}_{5.8}/{S}_{3.6})\gt -0.1$, ${\mathrm{log}}_{10}({S}_{8.0}/{S}_{4.5})\gt -0.2$, and ${\mathrm{log}}_{10}({S}_{8.0}/{S}_{4.5})\leqslant 0.8{\mathrm{log}}_{10}({S}_{5.8}/{S}_{3.6})$, which produce the wedge-shaped area in the upper right corner of Figure 4. This area represents the objects likely to be cold-mode AGNs. These 226 objects make up 15% of the radio sources in the Ibar et al. survey (Figure 5). Spectroscopic follow-up of AGN samples selected in this manner show that this particular "wedge" selection is reliable in the sense that 78% (527 out of 672) of objects for which spectroscopic redshifts and classifications could be obtained using optical or near-infrared spectra are confirmed as AGNs (Lacy et al. 2013). (Many of the remainder may also be highly obscured AGNs, but cannot be confirmed as such from optical/near-infrared data alone.) Among the cold mode AGNs there are a total of six type-1 quasars from the Sloan Digital Sky Survey Data Release 10 (Ahn et al. 2014; five of which are in the Lacy et al. 2013 sample).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Lacy et al. (2006), Afonso et al. (2011), Gurkan et al. (2014), and Singh et al. (2014) also find that some weak radio AGNs appear outside the AGN or star-forming galaxy selection region, with IRAC colors similar to quiescent galaxies. These galaxies are likely the low accretion rate, "hot mode" AGNs, whose radio jets are highly efficient. These radio sources can be found in the area beneath the cut formed by the line ${\mathrm{log}}_{10}({S}_{8.0}/{S}_{4.5})=-0.2$ in the left-hand panel of Figure 4.

We also use the ${q}_{24}={\mathrm{log}}_{10}({S}_{24}/{S}_{1400})$ value to classify objects without four-band IRAC detections (e.g., Ibar et al. 2008). As the SWIRE 24 μm flux density limit, $\approx 150$ μJy, is five times ${S}_{1.4\;\mathrm{GHz}}$ of our faintest radio sources, and given that the typical value of q₂₄ for star-forming galaxies is $\approx 1$ (Appleton et al. 2004; Ibar et al. 2008), we expect to detect the majority of star-forming galaxies and cold-mode AGNs at 24 μm. We indeed find that the fraction of radio sources detected at 24 μm is high (overall, 874/1245 (70%) of matched radio sources are detected at 24 μm). Thus, the radio-bright objects not detected at 24 μm are likely to be hot-mode AGNs. We place the dividing line between radio-loud and radio-quiet at ${q}_{24}\approx 0$ at z = 0, then use the template k-corrections shown in Figure 1 of Ibar et al. (2008) to approximate the change in redshift of this dividing line as ${q}_{24}\lt -0.15(1+z)$ (the dashed line in Figure 1, right-hand panel). For those objects not detected at 24 μm, we assume a limit of 150 μJy to classify them. We separate the hot mode AGNs selected using the wedge-based and q₂₄-based methods in Figures 3 and 4, but thereafter include them all in a single class, that comprises about 4% of the total radio source population in this sample.

For the objects lacking photometric redshifts, we approximated a classification by assuming a mean redshift of unity (dotted line in the right-hand panel of Figure 4). As can be seen, this should result in $\lesssim 10$ misclassifications.

The remainder of the objects in the infrared color–color diagram—the upper left corner in Figure 4—correspond to star-forming galaxies, and make up 15% of the radio sources. In addition, 30% of objects are detected at 24 μm with q₂₄ above the dashed (or dotted, in the case of objects lacking photometric redshifts) line in the right-hand panel of Figure 4, but not detected in all four IRAC bands. These most likely comprise a mix of cold-mode radio AGNs and star-forming galaxies, and we henceforth classify them as "ambiguous AGN/SF."

32% percent of the identified radio sources remain unclassifiable using either the infrared color–color diagram or q₂₄, because they lack four-band IRAC detections, are undetected at 24 μm, and their flux density limit at 24 μm is insufficiently deep compared to their radio fluxes to classify them as radio-loud or radio-quiet.

For the most part, our classification scheme is self-consistent between the wedge technique and the q₂₄ one. Among the objects with spectroscopic redshifts, however, 15/37 of the hot mode AGNs that are classified as such by the wedge technique show up as radio-quiet in the q₂₄ plot. Furthermore, in eight cases, objects initially classified as hot-mode AGNs are detected in Herschel HerMES (Oliver et al. 2012) images at 250 μm and also have photometric redshifts $\gt 1$. These are most likely misclassified star-forming Ultraluminous Infrared Galaxies, which, due to their high redshifts, no longer have the PAH features in the IRAC 8 μm band resulting in them falling out of the star formation region in Figure 4 (left). They have been reclassified as star-forming galaxies. There is also one low redshift (z = 0.04) galaxy that appears to be a HerMES detection, we also reclassify it as a starburst. The remainder tend to plot in the radio-louder side of the infrared-radio correlation. They may have significant residual star formation, or 24 μm emission from their AGN (although hot-mode radio AGN tend to be faint in the mid-infrared, some do have significant mid-infrared emission, e.g., Ogle et al. 2005). Similarly, there are 4/207 galaxies classified as star-forming in the wedge selection that appear radio-loud in the q₂₄ one, these may be radio-loud AGNs in starburst galaxies. Further, more accurate, classification could be attempted with full SED fitting, including Herschel data from the HerMES survey (Oliver et al. 2012), but we will defer this activity to a future paper.

Our classification results can be compared to the study of Bonzini et al. (2013), who used a survey of similar depth in the Extended Chandra Deep Field South to classify a sample of 883 radio sources as 19% radio-loud AGNs, 24% radio-quiet AGNs, and 57% star-forming galaxies. The main difference between our classification techniques and those used by Bonzini et al. is their use of a less conservative ${q}_{24}$ criterion for defining radio-loud AGNs (called "hot mode" AGNs in this paper), and also their X-ray data and deeper Spitzer data that allows them to resolve more ambiguities between cold-mode AGNs and star-forming galaxies.

In Figure 6, we plot the radio luminosity versus the photometric redshifts of the radio sources. The radio luminosity was calculated for objects within the μJy population using the flux densities and spectral indices of the 1.4 GHz objects within the Ibar et al. (2009) survey. Most of our radio sources have ${L}_{1.4}\sim {10}^{22.5}-{10}^{24.5}\;{\mathrm{WHz}}^{-1}$, spanning the traditional boundary in luminosity between radio-loud and radio-quiet objects. With few exceptions (probably misclassifications) all the star-forming galaxies selected via Figure 1 have ${L}_{1.4}\lt {10}^{24.5}\;{\mathrm{WHz}}^{-1}$, corresponding to star formation rates $\lesssim 2000\;{M}_{\odot }{\mathrm{yr}}^{-1}$, assuming a Salpeter IMF.

Zoom In Zoom Out Reset image size

The Ibar et al. catalog includes spectral indices between 610 MHz and 1400 MHz (${\alpha }_{1400}^{610}$) for objects detected at both frequencies. The mean ${\alpha }_{1400}^{610}$ for the objects matched to SERVS is −0.78. The mean spectral indices for the cold-mode AGNs and star-forming galaxies are very similar to the overall mean (−0.77 and −0.79, respectively). That for the hot-mode AGNs is significantly flatter, −0.44, indicating a large fraction of flat-spectrum AGNs in that class.

K-band emission from radio galaxies tends to consist primarily of stellar light (Lacy et al. 1995; Best et al. 1998; Simpson et al. 1999; Seymour et al. 2007), though in very luminous cold-mode AGNs there can be significant contributions from reddened quasar continuum (Rawlings et al. 1995) or emission lines (Eales & Rawlings 1993). Previous results generally found that the $K-z$ relation corresponded to a curve representing a stellar population that formed its stars at very high redshifts ($z\gt 5$) and passively evolved afterward (e.g., Lacy et al. 2000; Jarvis et al. 2001; Willott et al. 2003). In Figure 7, we plot the K-magnitudes against the logarithms of their photometric redshifts. The dashed curve in Figure 7, which has a function of $K\;=17.37+4.53\;{\mathrm{log}}_{10}{(z)-0.31\;({\mathrm{log}}_{10}(z))}^{2}$, was proposed by Willott et al. (2003) as the polynomial of best fit for luminous steep-spectrum radio galaxies within a standard 63.9 kpc metric aperture. Willott et al. show that their $K-z$ relation fit closest to a model in which most stars formed in a single burst at redshifts of z = 10, and evolved passively afterward. However, the scatter toward brighter magnitudes at $z\gt 2$ apparent in both Willott et al. and in Figure 7 cannot eliminate a model in which many radio source hosts formed at a lower redshift, $z\sim 5$. This is discussed further in Section 6.3 below.

Zoom In Zoom Out Reset image size

Figure 7 shows that most of the hosts of μJy radio sources are fainter in the K-band than the mean for the hosts of the much radio-brighter sources from the 3C, 6C, and 7C samples (with typical radio flux densities at 1.4 GHz ∼50–1000 mJy) studied by Willott et al. (2003). This is perhaps not surprising for the star-forming galaxy population, which has a very different physical mechanism for radio emission than the AGN populations that dominate the Willott et al. sample, but it also seems true for the cold-mode AGN population, at least at low redshifts/luminosities (see also Barger et al. 2014). In Figure 8, we plot the offset from the $K-z$ relation for different types of objects (especially AGNs) as a function of radio luminosity and redshift. In the lowest radio luminosity bin, ${10}^{22.5-23.5}\;{\mathrm{WHz}}^{-1}$, the mean and error in the mean of the offset is 0.93 ± 0.06 for cold-mode AGNs (significant at 15σ), but insignificant (0.05 ± 0.11) for hot-mode AGNs. In the next bin, ${10}^{23.5-24.5}\;{\mathrm{WHz}}^{-1}$, the offsets are barely significant (0.22±0.07) for the cold-mode AGNs and again insignificant (0.10±0.16) for the hot-mode AGNs, and, similarly, we see no significant offsets in the highest luminosity bin (${10}^{23.5-24.5}\;{\mathrm{WHz}}^{-1}$).

Zoom In Zoom Out Reset image size

Figure 9 shows the Petrosian magnitudes¹⁸ of the radio sources overlaid on the $K-z$ distribution from the entirety of the SERVS Fusion catalog, including optical and infrared sources which were not matched to radio sources. Despite the offset from the $K-z$ relation for the most luminous radio galaxies that we observe, the radio matched objects are, at any given redshift, among the most luminous objects, and are generally much more luminous than the objects found only in the optical, infrared, or ultraviolet surveys.

Zoom In Zoom Out Reset image size

We estimated the approximate stellar masses of the Lockman Hole radio sources using the 3.6 μm flux density and a 1.4 Gyr stellar population template from Bruzual & Charlot (2003) to calculate the rest-frame K-band luminosity, L_K. The 3.6 μm wavelength is a good compromise between being close to rest-frame K-band at $z\sim 1$, while at the same time not usually containing significant contamination from hot dust emission from the AGNs (Caputi 2013). The choice of stellar population template makes only a small difference to the k-corrections in the near-infrared. We then used K-band mass-to-light ratio estimates at $z=0\text{-}1$ from Drory et al. (2004) and at $z\sim 2$ from Borys et al. (2005) to parameterize the evolution of the K-band mass-to-light ratio in a typical massive galaxy as $M/{L}_{K}\approx 1.0/(1+1.5z)$. The result is shown in Figure 10. Some of the largest indicated masses for the cold-mode accretion AGNs are probably spurious, produced by contamination of the 3.6 μm flux density by AGN hot dust emission. For the remainder, we see a fairly well-defined cutoff at $\sim {10}^{11.5}\;{M}_{\odot }$.

Zoom In Zoom Out Reset image size

We further estimated the stellar masses of the optical and infrared sources that were not matched to radio sources, plotting the distribution of stellar masses of the radio-loud (${L}_{1.4\;\mathrm{GHz}}\gt {10}^{24}\;{\mathrm{WHz}}^{-1}$) and radio-quiet populations in Figure 9. This figure shows that the radio-loud sources tend to be among the objects with the largest stellar masses at any redshift (see also Bonzini et al. (2013), who find that the radio-loud AGNs in their work are also in host galaxies significantly more massive than cold-mode AGN or star-forming galaxies), though the distribution of masses of our "hot mode" AGNs lacks the tail to low masses (below $\approx {10}^{10.5}\;{M}_{\odot }$ seen in Figure 9 of Bonzini et al.). A high mean mass for the radio-loud population is expected, based on the work of Best et al. (2005) and Simpson et al. (2012).

Effective AGN feedback depends on a high duty cycle of activity (Croton et al. 2006), so a high fraction of radio-loud AGNs is needed if radio-jet powered AGN feedback models are to be plausible. Among objects with stellar masses greater than or equal to ${10}^{11.5}\;{M}_{\odot }$ with redshifts $0.5\lt z\lt 2.5$, we found 19/69, or 28%, of all massive galaxies are radio-loud (Figure 11), consistent with this requirement. At $z\gt 2.5$, the space density of galaxies with masses $\sim {10}^{11.5}{M}_{\odot }$ is very low, $\sim {10}^{-6}\;{\mathrm{Mpc}}^{-3}$ (Duncan et al. 2014), and remains similar to the space density of radio sources. It is thus quite plausible that, when a galaxy (and its black hole) reaches a certain critical mass, it is able to switch on as a radio-loud object with a duty cycle of a few tens of percent. Better statistics are needed at high-z, however, both on the radio source population and on the numbers of massive galaxies to understand the evolution of radio source hosts at the earliest epochs.

Zoom In Zoom Out Reset image size

The star-forming galaxies also have relatively high stellar masses, $\sim {10}^{11}\;{M}_{\odot }$. The most likely explanation for this is that these star-forming galaxies are at the high end of the radio luminosity function for star-forming objects, with typical radio luminosities $\sim {10}^{22.5}\;{\mathrm{WHz}}^{-1}$, forming stars at rates of $\sim 10{M}_{\odot }\;{\mathrm{yr}}^{-1}$ (Chabrier IMF; Karim et al. 2011). Karim et al. (2011) and Zwart et al. (2014) show that, at the redshifts at which these objects are mostly seen in our study ($z\sim 0.3-0.5),$ these objects have a typical specific star formation rate $\mathrm{SSFR}\sim 0.1$ Gyr⁻¹. Although less massive objects tend to have higher SSFR, the trend is relatively weak, therefore, fewer massive galaxies would be expected to have lower absolute star formation rates and thus would be missing from the survey.