CONCURRENT SUPERMASSIVE BLACK HOLE AND GALAXY GROWTH: LINKING ENVIRONMENT AND NUCLEAR ACTIVITY IN z = 2.23 Hα EMITTERS

We obtained a ≈100 ks Chandra exposure consisting of a single 169 × 169 ACIS-I pointing (Chandra ObsID 13976; taken over the 2012 January 13 and 14; PI: B. D. Lehmer) centered on the 2QZ Clus region surveyed by Matsuda et al. (2011; see Figure 1).¹⁰ The observation was centered on the aim point coordinates α_J2000 = 10:03:43.3 and δ_J2000 = +00:13:26.47 and was oriented at a roll angle that was 569 from north. The total duration of the observation was 99.6 ks. For our data reductions, we made use of CIAO v. 4.4 with CALDB v4.5.0. We began by reprocessing our events lists, bringing level 1 to level 2 using the script chandra_repro. The chandra_repro script runs a variety of CIAO tools that identify and remove events from bad pixels and columns, and filter the events list to include only good time intervals without significant flares and non-cosmic ray events corresponding to the standard ASCA grade set (ASCA grades 0, 2, 3, 4, 6).

Zoom In Zoom Out Reset image size

Using the reprocessed level 2 events list, we constructed a first FB image and a point-spread function (PSF) map (using the tool mkpsfmap), which corresponded to a monochromatic energy at 1.497 keV and an encircled counts fraction (ECF) set to 0.393. We constructed an initial source catalog by searching our FB image with wavdetect (run with our PSF map), which was set at a conservative false-positive probability threshold of 1 × 10⁻⁷ and run over seven scales from 1–8 (using a $\sqrt{2}$ sequence: 1, $\sqrt{2}$, 2, 2$\sqrt{2}$, 4, 4$\sqrt{2}$, and 8). To sensitively measure whether any significant flares remained in our observations, we constructed point-source-excluded 0.5–7 keV background light curves for the observation in a variety of time bins (spanning 10–800 s bins). We found no evidence of any significant (≳5σ) flaring events throughout the observation, and therefore considered our reprocessed level 2 events file to be sufficiently cleaned.

Next, using the initial X-ray source catalog and an optical I < 22 mag source catalog from the Sloan Digital Sky Survey Data Release 6 (SDSS-DR6; Adelman-McCarthy et al. 2008), we registered our aspect solution and events list to the SDSS-DR6 frame using CIAO tools reproject_aspect and reproject_events, respectively. The resulting astrometric reprojections gave very small astrometric adjustments, including linear translations of δx = −0.12 pixels and δy = +0.30 pixels, a rotation of −00097, and a pixel scale stretch factor of 1.00018.

Using the reprojected aspect solution and events file (see Section 2.1), we constructed standard-band images, 90% ECF PSF maps (corresponding to 1.497 keV, 4.51 keV, and 2.3 keV for the SB, HB, and FB, respectively), and exposure maps in the three standard bands. The exposure maps were constructed following the basic procedure outlined in Section 3.2 of Hornschemeier et al. (2001); these maps were normalized to the effective exposures of sources located at the aim points. This procedure takes into account the effects of vignetting, gaps between the CCDs, bad column and pixel filtering, and the spatially dependent degradation of the ACIS optical blocking filter. A photon index of Γ = 1.4 was assumed in creating the exposure maps. In Figure 1, we show the adaptively-smoothed FB image and the locations of HAEs and QSOs that are associated with the 2QZ Clus structure. The image was smoothed using csmooth and was "flattened" by dividing the smoothed image by a smoothed exposure map.

We constructed a Chandra source catalog by (1) searching the three standard-band images using wavdetect (including the appropriate 90% ECF PSF and exposure maps) at false-positive probability threshold of 2 × 10⁻⁵ (the equivalent detection threshold adopted by Elvis et al. 2009 in the C-COSMOS field); and (2) adjoining the three bandpass catalogs using cross-band matching radii of 25 and 40 for sources with off-axis angles of <6' and >6', respectively. Through this procedure, we identified a total of 133 unique X-ray detected sources in the 2QZ Clus field. These 133 sources constitute our main Chandra catalog (see Table 1).

We performed aperture photometry for our 133 sources in each of the three standard bands. For each X-ray source, we extracted source plus background counts S_src and exposure-map values T_src using circular apertures with radii corresponding to the 90% ECF, which were determined using the corresponding PSF-map value at the source location. We visually inspected these regions and found that only two source-pairs had very minor overlaps in the HB PSF (and no overlaps in the smaller SB and FB PSFs). The overlapping regions of these source pairs contained ⩽1 HB counts; therefore, we did not perform any corrections to the photometry for these sources.

For each source, local background counts and exposure values were then extracted from each of the three bandpasses. This was achieved by creating images and exposure maps with all sources masked out of circular masking regions of a radius 1.5 times the size of the 95% encircled energy fraction (estimated using each local PSF). Using these masked data products, we then extracted background counts S_bkg and exposure values T_bkg from a larger background extraction square aperture that was centered on the source. The size of the background extraction square varied with each source and was chosen to contain ≳30–100 background counts in all bands. For each bandpass, net source counts N were computed following N = (S_src − S_bkgT_src/T_bkg)/γ_ECF, where γ_ECF is the ECF appropriate for the source extraction region and bandpass. We computed 1σ level Poisson errors on the net counts following the methods described in Gehrels (1986).

To calculate vignetting-corrected count-rates, we made use of count-rate maps, which were constructed by dividing the images by the exposure maps. The advantage of count-rate maps is that they allow for an accurate accounting of the source intensity when gradients in the exposure are present in a source extraction region (e.g., near chip gaps, image edges, and bad pixels). We made use of the same source and background regions described above to extract on-source count-rates ϕ_src and background count-rates ϕ_bkg. Net count-rates were then computed following ϕ = (ϕ_src − ϕ_bkgA_src/A_bkg)/γ_ECF, where A_src and A_bkg are the source and background extraction areas that contain exposure.

For each source, we converted our vignetting-corrected count-rates to fluxes using conversion factors of [1.18, 0.659, and 1.96] × 10⁻¹¹ erg cm⁻² s⁻¹ (cnts s⁻¹)⁻¹ for the FB, SB, and HB, respectively. These factors assume a power-law SED with Γ = 1.4 that is corrected for Galactic extinction. In Table 1, we provide the basic X-ray properties of the 133 main-catalog sources in the 2QZ Clus field. The survey reaches ultimate 5-count sensitivity limits of ≈6.0 × 10⁻¹⁶ erg cm⁻² s⁻¹, ≈3.4 × 10⁻¹⁶ erg cm⁻² s⁻¹, ≈1.0 × 10⁻¹⁵ erg cm⁻² s⁻¹ for the FB, SB, and HB, respectively; at z = 2.23, these limits allow for the detection of a source with rest-frame 2–10 keV luminosity L_X ≳ 10⁴³ erg s⁻¹. These limits are nicely compatible with those achieved by the C-COSMOS survey, which reaches a factor of ≈1.4–1.7 times deeper in ultimate sensitivity (Elvis et al. 2009).

To measure the AGN activity in the 2QZ Clus structure at z = 2.23, we made use of the catalog of 22 HAEs from Matsuda et al. (2011) and four known z = 2.23 QSOs. After accounting for overlap in the HAE/QSO populations, there are a total of 23 unique z = 2.23 source candidates (see Figure 1). The one QSO that does not have an HAE counterpart lies outside the footprint of the narrow-band survey and is likely to be an HAE itself based on its selection from optical emission lines. In the HAE survey, Matsuda et al. (2011) selected candidate z = 2.23 sources that satisfied the following criteria: (1) K −H₂S1 ⩾0.251 (EW_obs ⩾ 50 Å) and (2) the Hα flux excess significance (i.e., the ratio between H₂S1 excess and the uncertainty in the K −H₂S1 color) Σ ⩾ 2.5. The narrow-band imaging reaches a H₂S1 5σ depth of ≈19.9 mag (AB). This selection is inherently subject to minor (at ≲10%–20% level) contamination from Paα emitters at z = 0.13, Paβ emitters at z = 0.65, and [O  iii]5007 emitters at z = 3.24; however, for many cases we can make use of B − z' and z' − K colors (i.e., the BzK criteria) to constrain the redshifts to z ∼ 1.4–2.5 and rule-out such contaminants. The BzK criteria includes (z' − K)_AB −(B − z')_AB ⩾ −0.2 or (z' − K)_AB > 2.5 (Daddi et al. 2004). From Matsuda et al. (2011), 17 out of the 23 z = 2.23 are either spectroscopically confirmed 2QZ Clus structure member QSOs or satisfy the BzK criteria (see Figure 1). Interestingly, the three QSOs with B, z, K photometry do not satisfy the BzK selection criteria, indicating that AGN-dominated SEDs can be missed by these criteria. This is likely due to the presence of strong emission lines, and indeed the three QSOs lie only ≈0.1–0.2 mag away from the BzK selection line (see Figure 3 of Matsuda et al. 2011).

One HAE of the 23 z = 2.23 sources is located outside of the Chandra field of view. We matched the remaining 22 sources (21 HAEs and four z = 2.23 QSOs) to our 2QZ Clus Chandra catalog using a 25 matching radius (see annotations in Figure 1). Given the HAE and X-ray source densities, we estimate that this choice of matching radius will lead to a negligible number of random associations (i.e., a total of ≈0.02 false-matches expected). We find a total of seven matches, giving an initial AGN fraction of ≈32⁺¹⁷₋₁₂% (1σ intervals based on Gehrels 1986). Four of the seven X-ray detected sources are the optically-identified QSOs that were used to originally select the 2QZ Clus for study (see Matsuda et al. 2011 for details); therefore, three additional z = 2.23 AGN are identified here as a result of our Chandra observations. If we exclude the four pre-selected QSOs, the AGN fraction of the underlying HAE population is ≈17⁺¹⁶₋₉%. Six of the seven X-ray detected sources are either spectroscopically confirmed 2QZ Clus member QSOs or BzK galaxies and are unlikely to be interlopers. The one HAE (XID = 49) that does not satisfy either criteria is inferred to be very luminous log L_X/erg s⁻¹ ≈ 44.7 and may have an AGN-dominated SED that does not satisfy the BzK criteria (similar to the other known QSOs); however, we cannot rule out the possibility that this source is a low-redshift interloper. In Figure 2, we show the HB-to-SB count-rate ratio versus 2–10 keV luminosity for X-ray-detected HAEs and z = 2.23 QSOs in the 2QZ Clus field, and in Table 2, we list the properties of these sources. These seven X-ray sources have HB-to-SB count-rate ratios and X-ray luminosities that are consistent with powerful unobscured QSOs.

Zoom In Zoom Out Reset image size

In Figure 3, we show the distributions of Hα fluxes for the 2QZ Clus HAEs and highlight the subset of HAEs with X-ray detections. We find that the AGN occupy the bright-end of the Hα flux distribution (f_Hα ≳ 2 × 10⁻¹⁶ erg cm⁻² s⁻¹), indicating (not surprisingly) that the AGN themselves have a dominant contribution to the intensity of the Hα emission lines in these systems. The mean Hα flux of the 2QZ Clus sample overall appears to be higher than any of the non-AGN HAEs, indicating that the bulk of the Hα power from the 2QZ Clus population is likely to be from AGN. However, once the pre-selected QSOs are removed, the mean Hα flux of the 2QZ Clus sample is within the range of the star-forming galaxies. A K-S test reveals that both the 2QZ Clus and C-COSMOS Hα flux distributions are consistent with each other whether or not pre-selected QSOs are included.

Zoom In Zoom Out Reset image size

One of our key goals is to compare the AGN activity of z = 2.23 HAEs in the 2QZ Clus structure with that of HAE field samples. As described in Section 1, the 2QZ Clus structure was initially selected as an overdensity of QSOs, and therefore represents a region biased toward both active SMBH mass accretion and galaxy stellar growth. To put into broader context the AGN activity in the 2QZ Clus structure, we make use of an independent sample of z = 2.216–2.248 HAEs from HiZELS in the wide-area ≈1.6 deg² COSMOS survey field (Sobral et al. 2013; see also Geach et al. 2012). The HiZELS HAEs were selected using the same telescopes, cameras, and filters as used for the 2QZ Clus HAE selection; however, unlike 2QZ Clus, the COSMOS field was not selected to have any QSO overdensity at z = 2.23. In total there are 353 z = 2.23 HAE candidates with f_Hα ≳ 5 × 10⁻¹⁷ erg cm⁻² s⁻¹ (complete at the ≈90% level) that satisfy the criteria used to select the 2QZ Clus HAEs (i.e., K −H₂S1 ⩾0.215; H₂S1 ⩽19.9 mag). In this selection, we do not remove non-BzK galaxies for fair comparisons with 2QZ Clus HAEs and to avoid z = 2.23 AGN being removed (see discussion in Section 3.1 regarding z = 2.23 2QZ Clus QSOs that do not satisfy BzK selection). As such, our selection is somewhat more liberal than the final selection of galaxies presented in Sobral et al. (2013), which do include a BzK filtering (slightly modified), together with another color–color selection (UBR), high quality photometric redshifts, and information on double/triple line emitters (see Sobral et al. 2013 for details).

Of the 353 HiZELS HAEs, 210 overlap with the footprint of the Chandra-COSMOS survey (C-COSMOS; Elvis et al. 2009; Puccetti et al. 2009). C-COSMOS is a contiguous ≈0.92 deg² Chandra survey that reaches 70–180 ks depths (vignetting-corrected) across ≈84% of the surveyed area. The combination of the HiZELS HAE and C-COSMOS surveys therefore constitutes a powerful data set by which direct comparisons can be made with the 2QZ Clus data. We made use of the BzK technique to make a first-order assessment of the contamination fraction for our HAE sample. Of these 210 HAEs, we found that 160 had B, z, and K photometry that would allow for a BzK redshift assessment. Of these 160 HAEs, 132 (≈83%) have colors satisfying the BzK color selection. We note that this BzK fraction is larger than that reported by Sobral et al. (2013) for the full HAE sample. The difference here is that we have limited our analysis to the relatively bright HAEs in the HiZELS sample, which has a lower contamination fraction. Furthermore, the BzK fraction for our HAEs is consistent with that found for the 2QZ Clus HAE sample. We have tested the effects of excluding all non-BzK HAEs from our samples and find that this choice has an insignificant impact on the results presented throughout this paper.

Using a matching radius of 25, we obtain a total of 10 Chandra sources matched to the 210 HiZELS HAEs that were within the C-COSMOS footprint, giving an initial AGN fraction of 4.8^+2.0_−1.4%; a factor of 6.7^+4.7_−3.2 times lower than the initial 2QZ Clus AGN fraction of 32⁺¹⁷₋₁₂%. Given the QSO pre-selection in the 2QZ Clus field, this difference in AGN fraction may not be so surprising. When we exclude the four pre-selected QSOs from our computation, we find the 2QZ Clus AGN fraction is ≈17⁺¹⁶₋₉% (i.e., 3 AGN out of 18 HAEs); still a factor of 3.5^+3.8_−2.2 times higher than the HiZELS AGN fraction. If we assume that the C-COSMOS HAE AGN fraction is representative of the HAE population in general, we estimate that the binomial probability of detecting 3 or more AGN out of 18 HAEs is ≈6%.

In Figure 2, we show the count-rate ratio versus 2–10 keV luminosity for the HiZELS z = 2.23 HAEs that are X-ray detected in C-COSMOS. The HiZELS HAEs host an almost equal blend of moderately obscured (N_H ≳ 10²² cm²) and unobscured AGN as measured by their band ratios. This appears to differ somewhat from the "softer" AGN found in 2QZ Clus (see Figure 2). If we assume that the ≈50% ratio of obscured to unobscured AGN in the C-COSMOS HAEs is typical of z = 2.23 HAEs, then the probability of finding all three uniquely X-ray selected AGN (i.e., excluding the four pre-selected QSOs) in 2QZ Clus to be unobscured is ≈25%. We therefore do not conclude that the differences between the X-ray spectral slopes in 2QZ Clus HAEs are significantly different from those of C-COSMOS HAEs. In Figure 3, we display the Hα flux distribution of the C-COSMOS HAEs and highlight the subset that are X-ray AGN. In contrast to 2QZ Clus HAEs, we find that the mean C-COSMOS HAE flux is within the distribution of non-AGN HAEs, indicating that the total Hα power output is likely dominated by the star-forming galaxy population.

To assess the role that environment plays in driving SMBH accretion in the HAE population, we made use of the wide-area HiZELS HAEs in the C-COSMOS survey to estimate AGN fraction as a function of source density. For the broader HiZELS HAE sample (i.e., the ≈1.6 deg² sample), we measured local HAE overdensities for each source following ρ/〈ρ〉 ≈ 4/(πr²₄)/〈ρ〉, where r₄ is the separation between a given source and its fourth nearest neighbor and 〈ρ〉 is the mean HAE source density over the entire HiZELS COSMOS field. Since the HiZELS COSMOS field has wider aerial coverage than the subpopulation of HAEs that lie within the C-COSMOS footprint, we expect that ρ/〈ρ〉 should provide a good estimate of the local source environment for all C-COSMOS sources without suffering from uncertainties due to survey edge effects. Focusing on only the HAEs that were within the C-COSMOS footprint, we divided the sample into three bins of source overdensity, log ρ/〈ρ〉 = [−0.35, 0.35, 1.12] ±0.38. For each bin, we computed the fraction of HAEs hosting luminous AGN detected in the X-ray band. Figure 4(a) shows the AGN fraction as a function of overdensity ρ/〈ρ〉 for the C-COSMOS sources, and in Table 3 we tabulate the AGN fractions of each subset. We find suggestive evidence for an increase in AGN fraction with local HAE source density, such that the f_AGN(log ρ/〈ρ〉 ≈ 1.12) = 3.4^+7.0_−2.7 × f_AGN(log ρ/〈ρ〉 ≈ −0.4). A similar trend was noted for Lyα emitters in the z = 3.1 SSA22 protocluster (Lehmer et al. 2009a).

Zoom In Zoom Out Reset image size

Table 3. X-Ray Stacking of HAE Samples

Sample	log (ρ/〈ρ〉)	Ngal	NAGN	fAGN	$\phi _{{2\hbox{--}7\,\rm keV}}/\phi _{{0.5\hbox{--}2\,\rm keV}}$	log LX	$\dot{M}_{\rm BH}$	S250 μm	SFR
						(erg s−1)	(10−3 M☉ yr−1)	(mJy)	(M☉ yr−1)
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
C-COSMOS low density	−0.38 ± 0.38	61	2	0.03+0.04−0.02	0.62 ± 0.47	42.52 ± 0.13	11 ± 4	3.9 ± 1.6	68 ± 29
Medium density	0.38 ± 0.38	122	5	0.04+0.03−0.02	0.53 ± 0.66	42.79 ± 0.23	21 ± 15	3.4 ± 0.6	60 ± 10
High density	1.12 ± 0.38	27	3	0.11+0.11−0.06	0.45 ± 1.03	42.50 ± 0.17	11 ± 5	<2.3	<42
2QZ Clus including QSOs	0.31 ± 0.10	21	6	0.29+0.17−0.11	0.48 ± 0.10	44.01 ± 0.14	362 ± 142	6.0 ± 2.6	111 ± 73
Excluding QSOs	0.31 ± 0.10	18	3	0.17+0.16−0.09	0.59 ± 0.18	43.76 ± 0.19	204 ± 110	3.6 ± 1.3	63 ± 23

Notes. Column 1: description of z = 2.23 HAE sample being stacked. Column 2: HAE source overdensity, ρ/〈ρ〉, computed as the ratio of the local HAE source density ρ to mean HAE density across the entire HiZELS COSMOS survey area 〈ρ〉 (Sobral et al. 2013). For the shallower 2QZ Clus HAE sample, we computed ρ/〈ρ〉 using bright HAEs in both the 2QZ Clus and HiZELS COSMOS fields (see Section 4 for details). Column 3: number of galaxies in the relevant stacking sample. This number reflects the exclusion of HAEs that were outside the C-COSMOS footprint, as well as sources that were in the near vicinity (within ≈15'') of unrelated X-ray detected sources. Column 4: number of sources that were detected in the X-ray band. Due to the relatively high survey luminosity limits for z = 2.23, these sources are expected to be AGN. Column 5: AGN fraction of sample. Column 6: ratio of stacked 2–7 keV to 0.5–2 keV count-rates ϕ. Column 7: logarithm of the mean 2–10 keV luminosity calculated using the stacked 0.5–2 keV emission. Column 8: estimate of the mean BH accretion rate based on the X-ray luminosity (see Section 4 for details and assumptions). Column 9: mean 250 μm flux in mJy based on stacking analyses. Column 10: mean SFR based on 250 μm flux (see Section 4.2).

Download table as:  ASCIITypeset image

Given that the 2QZ Clus HAE survey is somewhat shallower and less complete than the HiZELS COSMOS HAE survey, we cannot directly compute the equivalent HAE source density all the way down to the Hα flux limit. Therefore, to estimate an average ρ/〈ρ〉 for all sources in 2QZ Clus, we computed the survey-wide 2QZ Clus HAE source density of a highly complete (≳90%) subsample using only relatively bright HAEs (f_Hα > 6.3 × 10⁻¹⁷ erg cm⁻² s⁻¹), and then compared this with the source density of relatively bright COSMOS HiZELS HAEs. The relative bright-source densities indicate that the 2QZ Clus HAEs have on average 〈ρ〉^bright_2QZClus/〈ρ〉_HiZELS^bright = 2.0 ± 0.5. This value is somewhat higher than, although consistent with, the ≈1.5 ± 0.4 value found by Matsuda et al. (2011) when comparing the 2QZ Clus HAEs with a smaller HAE control sample. We note that the relatively small area in 2QZ Clus limits us from effectively measuring the local densities of each galaxy in the way we did for the HiZELS HAEs. A wider area HAE survey of the 2QZ Clus structure would mitigate this limitation. In Figure 4(a), we highlight the AGN fraction of the 2QZ Clus HAEs, both including and excluding HAEs that were pre-selected to be QSOs. After excluding the pre-selected QSOs, we find that the 2QZ Clus AGN fraction is a factor of ≈4.1^+5.0_−2.8 times higher than the AGN fraction of C-COSMOS HAEs in similarly overdense environments.

From the above analyses, it appears as though the enhanced AGN fraction in the 2QZ Clus cannot be explained by an enhanced source density. However, given that the HAE source selection is limited to the detection of sources with high SFR (SFR ≳ 14 M_☉ yr⁻¹) or AGN activity, it may be possible that there are large numbers of 2QZ Clus galaxies that are lower SFR and/or high Hα extinction leading to an inaccurate characterization of the local source environment. For this to explain the relatively large AGN fraction in terms of environment would require that the underlying galaxy population in 2QZ Clus be very different from that of C-COSMOS. Future observations sensitive to these galaxy populations would be needed to address this.

If we assume that the enhanced AGN fraction in the 2QZ Clus structure is due solely to the biased QSO-based selection and that our estimates of the local source overdensities are representative of the true environments in which the galaxies are found, we can assess how rarely we should find such elevated AGN fractions. From our analysis of the C-COSMOS HAEs, we know that the probability of an HAE hosting an AGN in an environment like that of 2QZ Clus is ≈4%. Assuming this incidence fraction is representative of the broader HAE population, we used the binomial probability distribution to compute the rarity of finding AGN fraction enhancements like those of 2QZ Clus. We find that the probability of detecting 3 or more AGN in 18 HAEs (i.e., after removing pre-selected QSOs) is ≈4.1%. This suggests that there is likely to be something inherently "special" about the 2QZ Clus region; perhaps some unaccounted for physical mechanism is responsible for the enhanced AGN activity that is unique to 2QZ Clus. It is also possible that the elevated AGN fraction is a product of the presence of more massive host galaxies and SMBHs. For example, Xue et al. (2010) found that the AGN fraction increases rapidly with stellar mass. From their AGN fraction versus stellar mass relations for luminous AGN (see Figure 14(B) of Xue et al. 2010), we infer that the 2QZ Clus galaxies would have to be on-average a factor of ≈5–8 times more massive than the C-COSMOS galaxy population to explain the enhanced AGN fraction that is observed. The 2QZ Clus and C-COSMOS HAEs were selected at the same SFR limits and have similar mean SFRs (see below). Since SFR is in general correlated with stellar mass, with a dispersion of a factor of ≈2 (e.g., Elbaz et al. 2007; Salim et al. 2007), it is unlikely that the mean stellar masses of the HAEs in 2QZ Clus would be ≈5–8 times higher on-average than those in C-COSMOS. Better characterization of the 2QZ Clus HAE source properties, large-scale environment, and underlying non-HAE galaxy population will help us understand this. This could be achieved via HST WFC3 infrared observations of the HAE population and/or spectroscopic follow-up of BX/BM and BzK galaxy candidates.

To estimate the mean SMBH growth rates of 2QZ Clus and C-COSMOS HAEs, we performed X-ray stacking following the techniques described in Lehmer et al. (2008), with errors measured as 1σ intervals of bootstrap resampling (see, e.g., Basu-Zych et al. 2013 for details). In 2QZ Clus, we stacked the total HAE samples both including and excluding the pre-selected QSOs. For C-COSMOS, we stacked each of the three subsamples that were divided based on source density (see Section 4.1). In Table 3, we tabulate our stacking results for the five subsamples. We find that the stacked 2–10 keV luminosities range from log L_X/(erg s⁻¹) ≈ 42.5–44, indicating that AGN clearly dominate the mean stacked X-ray emission over normal galaxy emission by more than an order of magnitude for galaxies of similar SFRs and redshifts (e.g., Lehmer et al. 2010; Basu-Zych et al. 2013). In all cases the 2–7 keV/0.5–2 keV count-rate ratios indicate the mean stacked spectrum is consistent with unobscured AGN (see Figure 2 and Column 6 in Table 3). We converted our mean 2–10 keV luminosities into bolometric luminosities L^AGN_bol using a correction factor of 22.4, which corresponds to the median bolometric correction for local AGN with L_X ≈ 10⁴¹–10⁴⁶ erg s⁻¹ (Vasudevan & Fabian 2007). We then estimated the mean SMBH accretion rates following $\dot{M}_{\rm BH} \approx (1-\epsilon)L_{\rm bol}^{\rm AGN}/(\epsilon c^2)$, where c is the speed of light and is the efficiency by which accreting mass is converted into radiative energy; here we assume ≈ 0.1 (see, e.g., Marconi et al. 2004 for motivation). These assumptions yield mean mass accretion rates spanning $\dot{M}_{\rm BH} \approx$ 0.01–0.02 M_☉ yr⁻¹ for the C-COSMOS HAE samples and ≈0.2–0.4 M_☉ yr⁻¹ for the 2QZ Clus HAE samples (see Column 8 in Table 3).

The above X-ray stacking shows that the mean SMBH accretion rates of 2QZ Clus HAEs are significantly larger than those of C-COSMOS HAEs; however, we want to test whether the corresponding SFRs of the HAEs are on average consistent with that expected from the local M_BH/M_gal relation. Although the Hα emission line provides an immediate proxy for the SFR, Hα is often subject to extinction in starburst galaxies like those studied here (e.g., Calzetti 1997). Furthermore, in the case of QSOs, the Hα emission line fluxes from the AGN will overwhelm any signal related to star formation activity, making it difficult to infer directly the SFRs of these sources (see, e.g., Figure 3).

To remedy these issues, we have made use of Herschel observations at 250 μm using the Spectral and Photometric Imaging Receiver (SPIRE). Herschel SPIRE data are available in the C-COSMOS field via the HerMES campaign (Oliver et al. 2012) and in the 2QZ Clus field through a GO program (PI: Matsuda; Y. Matsuda et al. 2013, in preparation). These data allow us to probe emission from the rest-frame 80 μm, which is well beyond the peak of the expected dust-emission associated with QSOs (≈6–15 μm; e.g., Efstathiou & Rowan-Robinson 1995; Netzer et al. 2007; Hatziminaoglou et al. 2010; Mullaney et al. 2011). We therefore anticipate that SPIRE probes the cool dust emission associated with star formation activity, and the total SFR can be measured using these data. We performed SPIRE stacking of the three C-COSMOS and two 2QZ Clus samples described in Table 3 following the techniques detailed in Harrison et al. (2012). We obtained significant signal in the 250 μm stacks for all samples except for the highest overdensity C-COSMOS subsample (see Table 3). Errors on mean SPIRE flux represent the largest 1σ interval based on bootstrap resampling (see Harrison et al. 2012 for details). We converted 250 μm fluxes to L_IR (8–1000 μm) using the SED library of Chary & Elbaz (2001), selecting a SED (redshifted to z = 2.23) on the basis of the total monochromatic luminosity probed by the 250 μm emission. We converted the infrared luminosities to SFRs following SFR/(M_☉ yr⁻¹) ≈9.8 × 10⁻¹¹ × L_IR/L_☉. This conversion assumes negligible contributions from UV emission and is appropriate for a Kroupa (2001) initial mass function (IMF; see Equation (1) of Bell et al. 2005). The mean SFRs are reported in Column 10 of Table 3.

From the above X-ray and infrared stacking analyses, we obtain mean $\dot{M}_{\rm BH}$/SFR ratios for our samples. In Figure 4(b), we display $\dot{M}_{\rm BH}$/SFR versus overdensity (ρ/〈ρ〉) for these samples. We find that the 2QZ Clus HAEs have $\dot{M}_{\rm BH}$/SFR ≈ 3.2 × 10⁻³, similar to the mean M_BH/M_gal ratio at z = 0 (e.g., Häring & Rix 2004; Bennert et al. 2011). For the C-COSMOS HAE subsamples, we find much lower values of $\dot{M}_{\rm BH}$/SFR ≈ (0.2–0.4) × 10⁻³. Such a deficit in SMBH-to-galaxy growth rate ratio may be due to the SFR-biased selection of C-COSMOS HAEs, which has constrained our analysis to include only powerful star-forming galaxies with extinction uncorrected Hα-based SFR_Hα > 14 M_☉ yr⁻¹.

Recent Herschel SPIRE analyses of X-ray selected AGN in the CDF-N, CDF-S, and COSMOS fields have shown that on average the mean SFR increases with mean AGN luminosity; however, there is some debate as to whether the most powerful QSOs with L_X ≳ 10⁴⁴ erg s⁻¹ have declining SFRs (e.g., Harrison et al. 2012; Page et al. 2012). In the left panel of Figure 5, we compare the 2QZ Clus and C-COSMOS HAE $\dot{M}_{\rm BH}$/SFR values with those of X-ray selected AGN presented in Harrison et al. (2012) and z ≈ 1–3 AGN from Mullaney et al. (2012b). X-ray luminosities have been converted to $\dot{M}_{\rm BH}$ following the techniques discussed above and SFRs have been converted to be consistent with our choice of Kroupa (2001) IMF. It is apparent that the mean $\dot{M}_{\rm BH}$/SFR for X-ray selected AGN appears to be consistent with the local M_BH/M_gal relation (Häring & Rix 2004) for AGN with L_X ≲ 10⁴⁴ erg s⁻¹. At higher L_X, M_BH/M_gal increases dramatically. The 2QZ Clus HAE samples have $\dot{M}_{\rm BH}$/SFR versus L_X mean values in good agreement with the trends observed for the Harrison et al. (2012) X-ray selected AGN (see Figure 5). As discussed above, the C-COSMOS HAEs appear to have somewhat lower $\dot{M}_{\rm BH}$/SFR values below that expected from the local M_BH/M_gal relation; however, these values appear to be consistent with those of the low-luminosity X-ray selected AGN from Mullaney et al. (2012b). Given that the mean SFRs and masses of all the AGN samples shown in the left-panel of Figure 5 are relatively high (SFR ≈ 10–100 M_☉ yr⁻¹ and M_⋆ ≳ 5 × 10¹⁰), this trend suggests that SMBHs and galaxies do not grow simultaneously.

Zoom In Zoom Out Reset image size

In the right panel of Figure 5, we compare the 2QZ Clus and C-COSMOS HAE $\dot{M}_{\rm BH}$/SFR values with those of z ≈ 1–2 star-forming galaxies (Mullaney et al. 2012a). We find that the C-COSMOS HAEs have $\dot{M}_{\rm BH}$/SFR in good agreement with field z ≈ 1–2 star-forming galaxies, which appear to also agree with z ≈ 2 submillimeter galaxies studied by Alexander et al. (2008). These values appear to lie a factor of ≈3 below the local M_BH/M_gal. We note that the various assumptions (e.g., bolometric correction and accretion efficiency) and unknown uncertainties used in our calculations may contribute to systematic offsets in $\dot{M}_{\rm BH}$/SFR. However, if our findings are correct, typical star-forming galaxies must undergo vigorous SMBH growth in short duty cycles, where $\dot{M}_{\rm BH}$/SFR is above the local M_BH/M_gal ratio. If the QSO phase at L_X ≳ 10⁴⁴ erg s⁻¹ is responsible for such episodic growth, we can use our observations to estimate the QSO duty cycle required to bring star-forming galaxies onto the M_BH/M_gal relation:

$$\begin{equation} M_{\rm BH}/M_{\rm gal} \approx (\dot{M}_{\rm BH}/{\rm SFR})_{\rm QSO}f_{\rm QSO} + (\dot{M}_{\rm BH}/{\rm SFR})_{\rm gal}(1 - f_{\rm QSO}), \end{equation} \tag{ 1 }$$

where f_QSO is the fraction of time that a galaxy is in a QSO phase. Taking M_BH/M_gal = 10^−2.5–10^−3.0, ($\dot{M}_{\rm BH}$/SFR)_QSO ≈ 10^−1.8, and ($\dot{M}_{\rm BH}$/SFR)_gal ≈ 10^−3.3, we find f_QSO ≈ 2%–8%. For C-COSMOS HAEs, we estimate that the QSO fraction for L_X ≳ 10⁴⁴ erg s⁻¹ AGN is ≈1.0^+1.3_−0.6%. This value and our duty-cycle estimate of f_QSO are consistent within errors, indicating that the short-term SMBH growth in a QSO phase may be sufficient to bring the typical HAE up to the M_BH/M_gal relation. This is consistent with the observation that the majority of SMBH growth density (using X-ray emissivity as a proxy) in the universe at z ≈ 2 can be attributed to luminous AGN (L_X ≳ 10⁴⁴ erg s⁻¹; e.g., Hasinger et al. 2005); a significant fraction of this population is expected to be fueled by mergers (≈30%–40%; e.g., Treister et al. 2012). We note that none of the luminous AGN among the C-COSMOS HAEs fell within the observational footprint of the HST WFC3 IR-imaged region of the Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey (CANDELS; Grogin et al. 2011; Koekemoer et al. 2011), so it was not possible to examine directly the rest-frame optical morphologies of the HAE AGN subpopulation. An alternative possibility is that the apparent sequence of increasing values of $\dot{M}_{\rm BH}$/SFR with increasing L_X is in a steady state and is due primarily to an increase in black-hole mass with increasing L_X. This explanation is supported by more recent detailed studies of local galaxies and their central SMBHs indicate M_BH/M_gal may be increasing with galaxy and SMBH mass (e.g., Graham et al. 2012; Graham & Scott 2012). Disentangling these two possibilities would require direct measurements of the galaxy and SMBH masses of the HAEs themselves. We used the HST ACS optical (F814W) morphology catalogs from Tasca et al. (2009) from the broader COSMOS field (Scoville et al. 2007) to compare the concentrations and asymmetries of the rest-frame UV emission from the HAEs and their AGN subpopulation. These parameters were available for 270 C-COSMOS HAEs and all 10 AGN. Compared with the C-COSMOS HAE population, the AGN subpopulation morphology distribution is skewed toward higher optical-light concentrations (K-S test reveals the populations differ at the 98.1% confidence level). The AGN subpopulation asymmetry distribution is consistent with the broader HAE population, with no evidence for more luminous AGN having larger asymmetries, as might be expected from merging systems. However, the observed HST images reveal that the rest-frame UV light from the most luminous AGN is dominated by the QSO emission, skewing the light to higher concentrations and making merger signatures elusive. Observations of these sources with HST WFC3 IR (rest-frame optical) would improve our ability to address whether mergers play a significant role in triggering driving SMBH growth in the HAE population.