ALMA and Herschel reveal that X-ray selected AGN and main-sequence galaxies have different star formation rate distributions
J. R. Mullaney, D. M. Alexander, J. Aird, E. Bernhard, E. Daddi, A. Del Moro, M. Dickinson, D. Elbaz, C. M. Harrison, S. Juneau, D. Liu, M. Pannella, D. Rosario, P. Santini, M. Sargent, C. Schreiber, J. Simpson, F. Stanley
aa r X i v : . [ a s t r o - ph . GA ] J u l Mon. Not. R. Astron. Soc. , 000–000 (0000) Printed 30 July 2015 (MN L A TEX style file v2.2)
ALMA and
Herschel reveal that X-ray selected AGN andmain-sequence galaxies have different star formation ratedistributions
J. R. Mullaney ⋆ , D. M. Alexander , J. Aird , E. Bernhard , E. Daddi , A. Del Moro ,M. Dickinson , D. Elbaz , C. M. Harrison , S. Juneau , D. Liu , M. Pannella ,D. Rosario , P. Santini , M. Sargent , C. Schreiber , J. Simpson , F. Stanley Department of Physics and Astronomy, The University of Sheffield, Hounsfield Road, Sheffield, S3 7RH, UK Centre of Extragalactic Astronomy, Department of Physics, Durham University, South Road, Durham, DH1 3LE, UK Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge CB3 0HA, UK Laboratoire AIM, CEA/DSM-CNRS-Universit´e Paris Diderot, Irfu/Service dAstrophysique, CEA-Saclay, Orme des Merisiers, 91191 Gif-sur-Yvette, France National Optical Astronomy Observatories, 950 N Cherry Avenue, Tucson, AZ 85719, USA Universit¨ats-Sternwarte M¨unchen, Scheinerstr. 1, D-81679 M¨unchen Max-Planck-Institut fr Extraterrestrische Physik (MPE), Postfach 1312, 85741, Garching, Germany INAF-Osservatorio Astronomico di Roma, via di Frascati 33, I-00040 Monte Porzio Catone, Roma, Italy Astronomy Centre, Department of Physics and Astronomy, University of Sussex, Brighton, BN1 9QH, UK
Date Accepted
ABSTRACT
Using deep
Herschel and ALMA observations, we investigate the star formation rate(SFR) distributions of X-ray selected AGN host galaxies at 0 . < z < . . < z < ≈
15 per cent of all MS galaxies, suggesting significantlydifferent SFR distributions. Indeed, when both are modelled as log-normal distributions, themass and redshift-normalised SFR distributions of X-ray AGNs are roughly twice as broad,and peak ≈ . Herschel data. This apparent contradictionis due to the linear-mean SFR being biased by bright outliers, and thus does not necessarilyrepresent a true characterisation of the typical SFR of X-ray AGNs.
Key words: galaxies: active—galaxies: evolution—galaxies: statistics
Today’s most successful models of galaxy evolution predict thatthe energy released via accretion onto supermassive black holes(hereafter, BHs) has played an important role in dictating how to-day’s galaxies have grown and evolved (e.g., Schaye et al. 2015).As such, understanding the connection between galaxy growthvia star-formation and the growth of their resident BHs is oneof the key challenges facing current extragalactic research. Thereare now numerous lines of empirical evidence in support of time-averaged/integrated BH growth correlating with star-formation in ⋆ E-mail: j.mullaney@sheffield.ac.uk their host galaxies; for example, (a) the tight proportionality be-tween BH mass and galaxy bulge mass (e.g., Gebhardt et al. 2000);(b) the similar cosmic histories of the volume-averaged BH growthand star formation rates (hereafter, SFR; e.g., Silverman et al.2008; Aird et al. 2015); and, more directly, (c) the correlationbetween average BH growth and SFR among the star-forminggalaxy population (e.g., Mullaney et al. 2012b; Chen et al. 2013;Delvecchio et al. 2015; Rodighiero et al. 2015). However, it isstill far from clear what physical processes (e.g., feedback pro-cesses/common fuel supply/common triggering mechanism) con-nect BH growth to star-formation to produce these average trends.One of the primary means of making progress in this areahas been to measure the SFRs and specific SFRs (i.e., SFR per c (cid:13) J. R. Mullaney et al.
Figure 1.
Host galaxy star-forming properties of our low- z (i.e., 0 . < z < .
5; not observed by ALMA) and high- z (i.e., z > .
5) samples of AGNs (samplesseparated by the vertical dashed line). In all plots, grey circles indicate pre-ALMA (specific) star formation rates ( [ s ] SFRs) from
Herschel which are connectedto their ALMA-measured (s)SFRs by dotted lines. (s)SFRs from ALMA are indicated by small white circles. Red and blue circles represent AGNs with L X = − ergs s − and L X > ergs s − , respectively, with lighter colours used for 3 s upper limits. Top:
SFR vs. redshift. Despite our ALMA observationsprobing SFRs up to a factor of ≈
10 lower than
Herschel , only ≈
29 per cent of our ALMA-targeted AGNs are detected.
Middle: sSFR vs. redshift. In thispanel, the shaded region represents the average sSFR of main-sequence (MS) galaxies (SFR MS ) as described by Eqn. 9 of S15 for the stellar mass range of oursample. Bottom: R MS vs. redshift. By definition, the horizontal line represents the average R MS of MS galaxies. Shading indicates where R MS < .
5. Between34 and 55 per cent (dependent on upper limits) of AGNs in our combined (i.e., low- z + high- z ) sample lie within this shaded region, compared to ≈
15 per centof MS galaxies. unit stellar mass, or sSFR) of galaxies hosting growing BHs (wit-nessed as active galactic nuclei, or AGN) and search for correla-tions or differences (vs. the non-AGN population) that may sig-nify a causal connection. The
Herschel Space Observatory (here-after,
Herschel ) has played a major role in progressing this scienceby providing an obscuration-independent view of star-formationthat is largely uncontaminated by emission from the AGN. How-ever, with even the deepest
Herschel surveys detecting .
50 percent of the AGN population, most studies have resorted to aver-aging (often via stacking analysis, but see Stanley et al. 2015) tocharacterise the (s)SFRs of the AGN population. These studieshave typically reported that the average SFRs of AGNs trace thatof star-forming “main-sequence” (hereafter, MS) galaxies (e.g.,Mullaney et al. 2012a; Santini et al. 2012; Harrison et al. 2012;Rosario et al. 2013; Stanley et al. 2015), i.e., the dominant popula-tion of star-forming galaxies whose SFRs are roughly proportionalto their stellar mass (i.e., sSFR ≈ constant), with a constant of pro-portionality that increases with redshift (e.g., Noeske et al. 2007;Daddi et al. 2007). However, as averages can be biased by brightoutliers, it is feasible that these findings are being driven upwardsby a few bright sources (e.g., Fig. 14 of Rosario et al. 2015). Here,we test this by combining deep Herschel and ALMA observationsto instead constrain the distribution of host galaxy SFRs of a sam-ple of X-ray selected AGNs and comparing it to that of MS galax- ies. We adopt H =
71 km s − Mpc − , W L = . W M = .
27 anda Chabrier initial mass function (IMF).
To investigate any redshift evolution of the AGN (s)SFR distribu-tion, we use two samples of X-ray selected AGNs: a low- z samplespanning 0 . z < . z sample spanning 1 . z < z sample is dominated by AGNs at 1 . z < . z = . > . z < . Herschel in the deepest fields and thus already have obscuration-independentSFR measures and (b) the negative k -correction at sub-mm wave-lengths would call for prohibitive ALMA integration times.The high- z sample were all selected from the 4 Ms Chandra
Deep Field South (hereafter, CDF-S) survey catalogue described inXue et al. (2011) with updated redshifts from Hsu et al. (2014); werecalculate the rest-frame 2-10 keV luminosities ( L X ) of the sourcesusing these new redshifts. To ensure reliable AGN selection, weonly consider those sources with L X > ergs s − and reliableredshifts (spec- z , or phot- z with D z / ( + z ) < .
1) that lie within 6 ′ of the average aim point of the survey (the latter ensures reliable po- c (cid:13) , 000–000 GN SFR distributions Figure 2.
The posterior probability distributions (PDs) for the parametersdescribing the assumed log-normal R MS distribution for AGN host galaxies: µ is the mode of the log-normal, while s is its 1 s width (see Eq. 1). PDs forboth our low- z and high- z samples are shown (see key). Contours of 20, 68and 95 per cent confidence are shown. The best-fit parameters of the com-bined (i.e., redshift-averaged) R MS distribution of MS galaxies is indicatedby the solid black circle (from Schreiber et al. 2015). The bottom and right-most plots indicate the relative probability of µ and s values; the locationof the peak represent the most probable parameter values. When modelledas a log-normal, the R MS distribution of AGN host galaxies is significantlybroader, and shifted significantly lower than that of MS galaxies. sitions for matching to ALMA counterparts). Our primary sciencegoal of constraining the SFR distributions of AGN host galaxies inthe context of the MS requires knowledge of the host galaxy stellarmasses ( M ∗ ), which we derive following Santini et al. (2012). Werefer to that study for a description of the relative uncertainties on M ∗ , which is estimated to be 50 and 20 per cent ( s ) for opticallyobscured (Type 2) and unobscured (Type 1) AGN, respectively.Since the majority (i.e., >
70 per cent) of the AGNs in our samplesare optically obscured, this level of uncertainty has no significantimpact on our conclusions. We restrict our sample to AGNs with M ∗ > × M ⊙ ; below this threshold, it becomes prohibitive toreach low enough flux limits to probe to SFRs significantly belowthe mean SFR of MS galaxies (hereafter, SFR MS ) with ALMA. De-spite this M ∗ cut we still sample the vast majority of the luminousAGN population since the M ∗ distribution of L X > ergs s − AGNs peaks at ≈ × M ⊙ (e.g., Mullaney et al. 2012a).The above selection returned 49 AGNs (our high- z sample),with 20 and 29 having spec- z and phot- z , respectively. Of these49, 13 are detected in the GOODS- Herschel µ m maps of theCDF-S (Elbaz et al. 2011) from which SFRs are be derived. Ofthe remaining 36 AGNs, 24 were observed by ALMA. However,since making our original ALMA target list, a more sensitive Her-schel µ m map of the CDF-S has been generated by com-bining the PEP (Lutz et al. 2011) and GOODS- Herschel surveys(Magnelli et al. 2013) and four of our 24 ALMA targets are nowdetected in that new map. For these four, we adopt the mean (s)SFRderived from the two facilities (see § Herschel fluxesand 3 s upper limits (including for the twelve Herschel -undetectedAGNs not targeted by ALMA) are also taken from the combinedPEP + GOODS-
Herschel dataset.The low- z sample were selected from the regions of the Chan-dra
Deep Field North (from Alexander et al. 2003 and adoptingthe same redshifts and M ∗ as Mullaney et al. 2012a) and South(Xue et al. 2011, but using the updated redshifts and M ∗ ) sur- Table 1.
Best-fit parameters for the log-normal R MS ( = SFR / SFR MS ) dis-tributions (see Eqn. 1) of the samples of galaxies described in the main text.(1) (2) (3)Sample µ s MS galaxies (Schreiber et al. 2015) − . a . ± . z AGN sample − . + . − . . + . − . High- z AGN sample − . + . − . . + . − . Combined AGN sample − . + . − . . + . − . N OTES : Values given are the median of the posterior probability distribu-tions (PDs) and the 68 per cent confidence intervals. a This is slightly offsetfrom exactly zero as R MS is the SFR relative to the linear mean SFR of MSgalaxies, whereas µ is the mode of the R MS distribution. veys with Herschel coverage by the PEP + GOODS surveys. Wealso restrict this low- z sample to L X > ergs s − and M ∗ > × M ⊙ to allow meaningful comparison with the high- z sam-ple. This returned a sample of 110 AGNs (i.e., our low- z sample),94 of which have spec- z . Sixty five of these 110 are detected inthe Herschel µ m band, from which we derive (s)SFRs (see § s flux upper limits were measured for the 45 Herschel non-detections.
All 24 of our ALMA targets were observed with ALMA Band-7 (i.e., observed-frame ∼ µ m) during November, 2013, witha longest baseline of 1.3 km. To maximise observing efficiency,the ALMA-targeted sample was split into three groups accordingto the flux limit required to probe down to at least SFR MS at agiven redshift. This corresponds to RMS flux limits of 200 µ Jy,125 µ Jy and 90 µ Jy for the three groups. ALMA continuum fluxeswere measured using uv fit of GILDAS v.apr14c, adopting pointsource profiles for two unresolved sources and circular Gaussianprofiles for the other five detected targets.Measured ALMA and
Herschel fluxes and upper limits wereconverted to 8-1000 µ m infrared luminosities (hereafter, L IR ) us-ing our adopted redshifts (see §
2) and the average infrared SEDsof MS galaxies described in B´ethermin et al. (2015), which areconstructed using the theoretical templates of Draine & Li (2007).However, we note that our conclusions do not change if we insteaduse either the Chary & Elbaz (2001) SEDs or a starburst SED (i.e.,Arp220). At the redshifts of our high- z sample, Band-7 probes therest-frame 180–340 µ m, close to the peak of the far-infrared emis-sion due to star-formation. While these rest-frame wavelengths arealso sensitive to dust mass (e.g., Scoville et al. 2014), based on therange of Draine & Li (2007) SED templates we estimate that thecorresponding L IR are accurate to within ± . Herschel and ALMAfluxes and upper limits to reduce the uncertainties associated withthe adopted SED, but such detailed fitting is beyond the scope ofthis Letter. As a check, however, we note that the SFRs derivedfrom ALMA and
Herschel data for the four AGNs that are de-tected with both are consistent to within this tolerance. SFRs arederived from L IR using Eqn. 4 from Kennicutt (1998), but adoptinga Chabrier IMF. Finally, to explore the distributions of AGN hostSFRs relative to SFR MS , we define R MS ≡ SFR / SFR MS , the rela-tive offset from the MS, where SFR MS is computed using Eqn. 9 ofSchreiber et al. (2015; hereafter, S15). c (cid:13) , 000–000 J. R. Mullaney et al.
Despite our ALMA observations probing to SFRs up to a factor of ≈
10 below that achieved with
Herschel (Fig. 1, top) only seven(i.e., ≈
29 per cent) of the 24 ALMA-targeted AGNs in our high- z sample are detected at > s at 850 µ m. The fractions of ALMA-undetected AGN are roughly the same for targets with spec- z andphot- z , suggesting that redshift uncertainties are not the primarycause of the non-detections. Despite the high fraction of non-detections, the 3 s upper limits provided by the ALMA + Herschel data enable us to infer the level of consistency between the dis-tributions of R MS for AGN and MS galaxies (see § M ∗ and redshifts rangesconsidered here (e.g., Rodighiero et al. 2011; Sargent et al. 2012).To explore our AGN hosts’ star-forming properties in the con-text of the evolving MS, we plot their sSFRs and R MS values as afunction of redshift (Fig. 1, middle and lower panels, respectively).We find that 54 to 88 (range due to upper limits) of the 159 AGNs(i.e., ≈
34 to ≈
55 per cent) in our combined (i.e., low- z + high- z )sample have R MS < .
5, with significant overlap between the frac-tions in our low- z (i.e., ≈
43 per cent to ≈
54 per cent) and high- z (i.e., ≈
14 per cent to ≈
59 per cent) samples. Comparing thesefractions to the ≈
15 per cent of MS galaxies with R MS < . z sample, and possibly alsoour high- z sample, do not trace the same R MS distribution as MSgalaxies, instead displaying a strong bias toward lower R MS values.Finally, we note that only ≈ R MS > With the large fraction of AGNs with R MS < . z samples being inconsistent with the R MS distribution of MS galaxies, we now attempt to place constraintson the distribution of SFRs (relative to the MS; i.e., R MS ) of AGNhosts. We place particular emphasis on quantifying the level of con-sistency/discrepancy between the AGN and MS R MS distributions.Our relatively small sample sizes, combined with the largefraction of non-detections prevents us from determining the AGN R MS distribution directly. Since a key goal here is to quantitativelycompare the AGN and MS R MS distributions, we instead assume the same log-normal form for the AGN R MS distribution as foundfor MS galaxies (e.g., Rodighiero et al. 2011; Sargent et al. 2012,S15): N ( R MS ) (cid:181) exp (cid:18) − ( log ( R MS ) − µ ) s (cid:19) (1)and infer its parameters (i.e., similar to Shao et al. 2010 who in-ferred the AGN L IR distribution). This is done purely to ease com-parison between the AGN and MS R MS distributions by allowingus to compare like-for-like parameters (i.e., the mode, µ , and thevariance, s , of the log-normal R MS distribution), and is not to betaken as a literal description of the true AGN R MS distribution. We adopt a hierarchical Bayesian framework to determinethe best-fit parameters (i.e., µ and s ) for our assumed log-normaldistributions, using Gibbs sampling and the Metropolis-Hastings Investigating whether other forms better describe the R MS distribution ofAGN hosts will be the focus of a later study incorporating a larger set ofALMA observations from Cycle 2 (PI: Alexander; awaiting completion). Figure 3. R MS distributions for our high- z and low- z samples of X-ray se-lected AGNs ( Top ) and MS galaxies (
Bottom ; from S15). Here, we show thelog-normal distributions with best fitting parameters shown in Table 1 (solidand dotted curves; see key). The histograms in the top panel shows the rel-ative numbers of AGNs from our combined (i.e., low- z + high- z ) sample ineach R MS bin; the solid grey histogram represents those AGNs detected at > s with either Herschel or ALMA, whereas the empty histogram (withleft-pointing arrows) also includes upper limits. The solid points in the toppanel indicate the linear means of the log-normal distributions (equivalentto what would be obtained via, e.g., stacking analyses) and lie within 1 s ofthe linear mean R MS of MS galaxies (vertical dashed line). MCMC algorithm to randomly sample their posterior probabilitydistributions (hereafter, PDs; Gelman et al. 2014). The benefits oftaking this approach are that (a) upper limits and uncertainties on R MS can be readily taken into account and (b) the resulting poste-rior PDs provide us with meaningful parameter uncertainties. Weuse weak prior PDs, noting that the centring of these priors (withinreasonable limits) has no significant effect on our results.The posterior PDs on µ and s for our two samples are pre-sented in Fig. 2, while the best-fit parameters (median of the PDsand 68 per cent confidence intervals) are given in Table 1. For com-parison, we also include the best-fit parameters of the log-normal R MS distribution for non-AGN MS galaxies from S15. As expectedfrom the smaller size of our AGN sample and the high fractions ofnon-detections compared to the MS galaxy sample of S15, the un-certainties on the posterior parameter values for the assumed AGNlog-normal R MS distribution are considerably larger than those forMS galaxies. Despite this, our analysis shows that the R MS distribu-tions of our low- z and high- z AGNs are both significantly broaderand peak at significantly lower values (both at > . R MS distribution, with the modes and variances of the log-normal distributions describing our low- z and high- z samples beingconsistent to within 1 s . In light of this, we infer the R MS distri-bution of our combined sample, which we find is roughly twice asbroad as, and peaks ≈ . c (cid:13) , 000–000 GN SFR distributions In the previous section we used our combined ALMA+
Herschel data to demonstrate that, when modelled as a log-normal, the AGN R MS distribution is significantly broader and peaks at significantlylower values than that of MS galaxies. This appears to be at-oddswith recent findings based on mean-stacked Herschel data that theaverage star-forming properties of AGN hosts is consistent withthose of MS galaxies (e.g., Mullaney et al. 2012a; Santini et al.2012; Rosario et al. 2013). Here, we place our results in the contextof these studies to explore the root of these apparent discrepancies.When comparing to results derived from mean-stacked
Her-schel data, it is important to note that mean-stacking provides alinear mean which will not correspond to the mode, µ , of a log-normal distribution. Instead, the linear mean will always be higherthan the mode, with the discrepancy between the two increasingas function of both µ and s . Therefore, while results from mean-stacking still hold when interpreted as the linear mean, dependingon the underlying distribution this may not necessarily correspondto the mode.We compare our results against those from stacking by cal-culating the linear mean of our log-normal distributions, taking aMonte-Carlo approach to sample the µ and s PDs. This gives linear-mean AGN R MS values (i.e., h R MS i ) of 0 . + . − . and 1 . + . − . forour low- z and high- z samples, respectively (Fig. 3). These valuesare remarkably close to the linear mean R MS of MS galaxies (i.e., h R MS i≈
1) and are broadly consistent with the linear means calcu-lated by mean-stacking
Herschel µ m maps at the positions ofour AGN (i.e., h R MS i = . ± .
12 and 0 . ± .
15, respectively).We conclude that these linear-means are, indeed, influenced by thehigh tail of the broad R MS distribution and may not necessarily givea reliable indication of the modal SFR of AGN hosts.Despite finding that the R MS distribution of AGN hosts isshifted toward lower values compared to MS galaxies, our resultsremain consistent with AGNs preferentially residing in galaxieswith comparatively high (s)SFRs by z ∼ MS . Indeed, applying our analy-ses to sSFR (rather than R MS ) gives distributions peaking at ≈ . − and ≈ . − for our low- z and high- z samples, re-spectively. To put this in context, h sSFR MS i ≈ . − at z ≈ . − and 0 . − wouldbe classed as MS and starbursting galaxies, respectively.Our result compare favourably to those derived from AGN sur-veys conducted at other wavelengths. For example, using SFRs de-rived from optical SED fitting, Bongiorno et al. (2012) reported abroad sSFR distribution for X-ray selected AGNs that peaks be-low that of the MS at redshifts similar to those explored here (i.e.,0 . < z < . R MS distribution of X-ray selected AGNs (with a similar M ∗ selectionas here) peaks at ∼ . R MS distribution of M ∗ -matched galaxies (i.e., not just star-forming galaxies). As such,these studies and the results presented here support the view thatX-ray selected AGN hosts at moderate to high redshifts span thefull range of relative sSFRs of M ∗ & × M ⊙ galaxies (e.g.,Brusa et al. 2009; Georgakakis et al. 2014). However, with recentresults suggesting that X-ray absorbed AGN may have higher SFRsthan unabsorbed AGN (e.g., Juneau et al. 2013; Del Moro et al.2015), it is feasible that alternative AGN selections may bring theAGN R MS distribution closer to that of MS galaxies.We thank the anonymous referee. DMA, ADM, CMH acknowledgeSTFC grant ST/I001573/1. This paper makes use of ALMA data:ADS/JAO.ALMA REFERENCES
Aird, J., Coil, A. L., Georgakakis, A., et al. 2015, MNRAS, 451,1892Alexander, D. M., Bauer, F. E., Brandt, W. N., et al. 2003, AJ,126, 539Azadi, M., Aird, J., Coil, A., et al. 2014, ArXiv e-prints,1407.1975B´ethermin, M., Daddi, E., Magdis, G., et al. 2015, A&A, 573,A113Bongiorno, A., Merloni, A., Brusa, M., et al. 2012, MNRAS, 427,3103Brusa, M., Fiore, F., Santini, P., et al. 2009, A&A, 507, 1277Chary, R., & Elbaz, D. 2001, ApJ, 556, 562Chen, C.-T. J., Hickox, R. C., Alberts, S., et al. 2013, ApJ, 773, 3Daddi, E., Dickinson, M., Morrison, G., et al. 2007, ApJ, 670, 156Del Moro, A., Alexander, D. M., Bauer, F. E., et al. 2015, ArXive-printsDelvecchio, I., Lutz, D., Berta, S., et al. 2015, ArXiv e-prints,1501.07602Draine, B. T., & Li, A. 2007, ApJ, 657, 810Elbaz, D., Dickinson, M., Hwang, H. S., et al. 2011, A&A, 533,A119Gebhardt, K., Bender, R., Bower, G., et al. 2000, ApJL, 539, L13Gelman, A., Carlin, J. B., Stern, H. S., et al. 2014, Bayesian DataAnalysis, 3rd. edGeorgakakis, A., P´erez-Gonz´alez, P. G., Fanidakis, N., et al. 2014,MNRAS, 440, 339Harrison, C. M., Alexander, D. M., Mullaney, J. R., et al. 2012,ApJL, 760, L15Hsu, L.-T., Salvato, M., Nandra, K., et al. 2014, ApJ, 796, 60Juneau, S., Dickinson, M., Bournaud, F., et al. 2013, ApJ, 764,176Kennicutt, Jr., R. C. 1998, ARA&A, 36, 189Lutz, D., Poglitsch, A., Altieri, B., et al. 2011, A&A, 532, A90Magnelli, B., Popesso, P., Berta, S., et al. 2013, A&A, 553, A132Mullaney, J. R., Pannella, M., Daddi, E., et al. 2012a, MNRAS,419, 95Mullaney, J. R., Daddi, E., B´ethermin, M., et al. 2012b, ApJL,753, L30Noeske, K. G., Weiner, B. J., Faber, S. M., et al. 2007, ApJL, 660,L43Rodighiero, G., Daddi, E., Baronchelli, I., et al. 2011, ApJL, 739,L40Rodighiero, G., Brusa, M., Daddi, E., et al. 2015, ApJL, 800, L10Rosario, D. J., Trakhtenbrot, B., Lutz, D., et al. 2013, A&A, 560,A72Rosario, D. J., McIntosh, D. H., van der Wel, A., et al. 2015, A&A,573, A85Santini, P., Rosario, D. J., Shao, L., et al. 2012, A&A, 540, A109Sargent, M. T., B´ethermin, M., Daddi, E., & Elbaz, D. 2012,ApJL, 747, L31Schaye, J., Crain, R. A., Bower, R. G., et al. 2015, MNRAS, 446,521Schreiber, C., Pannella, M., Elbaz, D., et al. 2015, A&A, 575, A74Scoville, N., Aussel, H., Sheth, K., et al. 2014, ApJ, 783, 84Shao, L., Lutz, D., Nordon, R., et al. 2010, A&A, 518, L26Silverman, J. D., Green, P. J., Barkhouse, W. A., et al. 2008, ApJ,679, 118Stanley, F., Harrison, C. M., Alexander, D. M., et al. 2015, ArXive-prints, 1502.07756Xue, Y. Q., Luo, B., Brandt, W. N., et al. 2011, ApJS, 195, 10 c (cid:13)000