The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey

Abstract

We present a study of the multi-wavelength properties, from the mid-infrared
to the hard X-rays, of a sample of 255 spectroscopically identified X-ray
selected Type-2 AGN from the XMM-COSMOS survey. Most of them are obscured the
X-ray absorbing column density is determined by either X-ray spectral analyses
(for the 45% of the sample), or from hardness ratios. Spectral Energy
Distributions (SEDs) are computed for all sources in the sample. The average
SEDs in the optical band is dominated by the host-galaxy light, especially at
low X-ray luminosities and redshifts. There is also a trend between X-ray and
mid-infrared luminosity: the AGN contribution in the infrared is higher at
higher X-ray luminosities. We calculate bolometric luminosities, bolometric
corrections, stellar masses and star formation rates (SFRs) for these sources
using a multi-component modeling to properly disentangle the emission
associated to stellar light from that due to black hole accretion. For 90% of
the sample we also have the morphological classifications obtained with an
upgraded version of the Zurich Estimator of Structural Types (ZEST+). We find
that on average Type-2 AGN have lower bolometric corrections than Type-1 AGN.
Moreover, we confirm that the morphologies of AGN host-galaxies indicate that
there is a preference for these Type-2 AGN to be hosted in bulge-dominated
galaxies with stellar masses greater than 10^10 solar masses.

Full text

Page 1

arXiv:1108.4925v1 [astro-ph.CO] 24 Aug 2011
Astronomy & Astrophysics manuscript no. 17175
August 26, 2011
c ? ESO 2011
The bolometric output and host-galaxy properties of obscured
AGN in the XMM-COSMOS survey
E. Lusso1,2⋆, A. Comastri2, C. Vignali1,2, G. Zamorani2, E. Treister3,9, D. Sanders3, M. Bolzonella2, A. Bongiorno4,
M. Brusa4, F. Civano6, R. Gilli2, V. Mainieri7, P. Nair2, M. C. Aller8, M. Carollo8, A. M. Koekemoer11, A. Merloni4,5,
and J. R. Trump10.
1Dipartimento di Astronomia, Universit` a di Bologna, via Ranzani 1, I-40127 Bologna, Italy.
2INAF–Osservatorio Astronomico di Bologna, via Ranzani 1, I-40127 Bologna, Italy.
3Institute for Astronomy, 2680 Woodlawn Drive, University of Hawaii, Honolulu, HI 96822, USA.
4Max Planck Institut f¨ ur extraterrestische Physik, Giessenbachstrasse 1, 85748 Garching, Germany.
5Excellence Cluster Universe, TUM, Boltzmannstr. 2, D-85748, Garching bei M¨ unchen, Germany.
6Harvard-Smithsonian Center for Astrophysics, 60 Garden Street,Cambridge, MA 02138, USA.
7ESO, Karl-Schwarzschild-Strasse 2, 85748 Garching bei M¨ unchen, Germany.
8ETH Z¨ urich, Physics Department, CH-8093, Z¨ urich, Switzerland.
9Universidad de Concepci´ on, Departamento de Astronom` ıa, Casilla 160-C, Concepci´ on, Chile.
10University of California Observatories/Lick Observatory, University of California, Santa Cruz, CA 95064.
11Space Telescope Science Institute, Baltimore, Maryland 21218, USA.
Accepted version August 24, 2011
ABSTRACT
We present a study of the multi-wavelength properties, from the mid-infrared to the hard X–rays, of a sample of 255 spectroscop-
ically identified X–ray selected Type-2 AGN from the XMM-COSMOS survey. Most of them are obscured the X–ray absorbing
column density is determined by either X–ray spectral analyses (for the 45% of the sample), or from hardness ratios. Spectral Energy
Distributions(SEDs)arecomputed for allsources inthe sample. Theaverage SEDsintheoptical band isdominated by thehost-galaxy
light, especially at low X–ray luminosities and redshifts. There is also a trend between X–ray and mid-infrared luminosity: the AGN
contribution in the infrared is higher at higher X–ray luminosities. We calculate bolometric luminosities, bolometric corrections, stel-
lar masses and star formation rates (SFRs) for these sources using a multi-component modeling to properly disentangle the emission
associated to stellar light from that due to black hole accretion. For 90% of the sample we also have the morphological classifications
obtained with an upgraded version of the Zurich Estimator of Structural Types (ZEST+). We find that on average Type-2 AGN have
lower bolometric corrections than Type-1 AGN. Moreover, we confirm that the morphologies of AGN host-galaxies indicate that there
is a preference for these Type-2 AGN to be hosted in bulge-dominated galaxies with stellar masses greater than 1010solar masses.
Key words. galaxies: active – galaxies: evolution – quasars: general – methods: statistical
1. Introduction
The formation and growth of supermassive black holes
(SMBHs) and their host-galaxies are related processes. This
is supported by various observational signatures: the SMBH
mass correlates with the mass of the bulge of the host-
galaxy (Magorrian et al. 1998; Marconi & Hunt 2003), with
the velocity dispersion of the bulge (Ferrarese & Merritt
2000; Tremaine et al. 2002), and with the luminosity of the
bulge (Kormendy & Richstone 1995). From theoretical mod-
els, AGN seem to be able to switch off cooling flows in
clusters (Hoeft & Br¨ uggen 2004) and star formation in galax-
ies (Somerville et al. 2008), with the result that the SMBH
mass is related to the host-galaxy bulge mass (or vice-versa).
Feedback between an accreting SMBH and the host-galaxy
may play an important role in galaxy formation and evolu-
tion. Understand the role of feedback is a demanding prob-
lem for both observers and theorists. Semi-analytical models
andhydrodynamicalsimulationshavebeendevelopedtoattempt
to link the formation and evolution of SMBHs to the struc-
⋆elisabeta.lusso2@unibo.it
ture formation over cosmic time. These models invoke different
mechanisms to fuel the central SMBHs and to build the host-
galaxy bulges, such as major/minor mergers of galaxies (e.g.,
Corbin2000; Kauffmann & Haehnelt2000; Springel et al.2005;
Hopkins et al. 2006), smooth accretion of cold gas from fila-
mentary structures (e.g., Kereˇ s et al. 2009; Dekel et al. 2009),
or accretion of recycled gas from dying stars (e.g., Ciotti et al.
2010). Several works also consider radiative feedback which
can reproducetwo importantphases of galaxyevolution,namely
an obscured-cold-phase, when the bulk of star formation and
black hole accretion occurs, and the following quiescent hot
phase in which accretion remains highly sub-Eddington and
unobscured (e.g., Sazonov et al. 2005; Lusso & Ciotti 2011).
In some of these models, the obscured/unobscured AGN di-
chotomy is more related to two different phases of galaxy evo-
lution (Hopkins et al. 2008), rather than to an orientation effect
(i.e., unified model scheme).
The obscured/unobscured time dependent AGN dichotomy
could be related to the bimodality in the rest-frame color dis-
tribution of host-galaxies (Rovilos & Georgantopoulos 2007;
Nandra et al. 2007; Brusa et al. 2009; Silverman et al. 2009),

Page 2

2E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey
namely the red-sequence (or “red-cloud”) and blue-cloud galax-
ies. Broad-line AGN (if the morphology of the host-galaxy is
available) are likely to be associated to galaxies belonging to
the blue-cloud, while obscured objects to red passive galaxies.
The green valley should be populated by transition objects. The
picture above is probably a too crude approximation. Moreover,
one should note that red sequence galaxies may well be pas-
sively evolving galaxies without significant star formation (e.g.,
Rovilos & Georgantopoulos 2007; Nandra et al. 2007), rather
than dusty starforming objects (e.g., Brusa et al. 2009).
Disentangling the contribution of the nuclear AGN from the
host-galaxy properties in the broad band SED is fundamental
to constrain the physical evolution of AGN and to place them
into the context of galaxy evolution. In the standard picture the
AGN energy output is powered by accretion onto SMBHs. The
disk accretion emission is visible in the optical-UV as the blue-
bump feature. The X–ray emission is believed to be due to a
hot-electrons corona that surrounds the accretion disk, while
the infrared emission is likely due to the presence of a dusty
torus around the disk at few parsec from the center, which re-
processes the nuclear radiation. According to the unified model
of AGN (e.g., Antonucci 1993; Urry & Padovani 1995), hot
dust is located in the inner edge of the torus. However, recent
studies predict and observe exceptions to the unified model.
From the theoretical point of view, an alternative solution to
the torus is the disk-wind scenario (e.g., Emmering et al. 1992;
Elitzur & Shlosman 2006). From the observational side, AGN
without any detectable hot dust emission (e.g., Jiang et al. 2010)
and weak infrared emission (e.g., Hao et al. 2010) are predicted
and observed.The vast majority of studies performedso far con-
cern unobscured (Type-1) AGN for which their SED is well
known from low-z (?z? ∼ 0.206, see Elvis et al. 1994) to high-
z (?z? ∼ 1.525, see Richards et al. 2006). An obvious compli-
cation in the study of their host-galaxy properties is that the
emission of the central AGN outshines the galaxy light in UV,
optical and infrared bands; therefore it is extremely difficult to
derive constraints on the colors, stellar populations, and mor-
phologies of the host. On the other hand, for obscured (Type-
2) AGN the host-galaxy light is the dominant component in
the optical/near-infrared SED, while it is difficult to recover
the AGN intrinsic nuclear emission. The lack of a proper char-
acterization of the nuclear componentof the SED of obscured
Type-2 AGN is a major limitation. As a consequence, the re-
lations between stellar masses, SFR, morphologies and accre-
tion luminosity remain poorly known. Since the relative con-
tribution in the SED of the different components (AGN/host-
galaxy) varies with wavelength, a proper decomposition can be
obtained by an SED-fitting approach, complemented by a mor-
phological analysis. This will provide a robust estimate of the
nuclear emission (bolometric luminosities and bolometric cor-
rections, absorption column density distributions, etc) and its re-
lationwiththehost-galaxyproperties(mass,starformationrates,
morphological classification). The AGN structure is reflected in
the shape of the SED, specifically the big-blue bump and the
infrared-bumpare related to the accretion disk and the surround-
ing torus, respectively. Therefore, a densely sampled SED over
a broad wavelength interval is mandatory to extract useful in-
formation from SED fitting procedures, allowing to tightly con-
strain physical parametersfrom multi-componentmodelingand,
in particular, to properly disentangle the emission associated to
stellar light from that due to accretion.
The combination of sensitive X–ray and mid-IR observa-
tories allows us to model the obscuring gas that in Type-2
AGN hides the nuclear region from the near-IR to the UV.
As supported by previous investigations, the reprocessed IR
emission could be a good proxy of the intrinsic disk emission.
Gandhi et al. (2009) confirm the correlation between the X–ray
luminosity at [2-10]keV and the IR emission for a sample of
Seyfert galaxies (see also Lutz et al. 2004). Their data are the
best estimate of the nuclear (non stellar) IR flux in AGN to date.
Ahighlysignificantcorrelationbetween L[2−10]keVandtheintrin-
sic nuclearIR luminosityat 12.3µmis observedin the high qual-
ity near-IR and X–ray data discussed by Gandhi et al. (2009).
This reinforces the idea that “uncontaminated” mid-IR contin-
uum is an accurate proxy for the intrinsic AGN emission.
In this work we present the largest study of the multi-
wavelength properties of an X–ray selected sample of ob-
scured AGN using the XMM-Newton wide field survey in
the COSMOS field (XMM-COSMOS). Following a similar ap-
proach to that of Pozzi et al. (2007) and Vasudevan et al. (2010),
we use the infrared emission to evaluate the nuclear bolometric
luminosity from a multi-component fit. The paper is aimed at
a detailed characterization of a large sample of obscured AGN
over a wide range of frequencies. The SEDs, morphology of the
host-galaxies,stellarmasses,colors,bolometricluminositiesand
bolometric corrections for the sample of obscured AGN are pre-
sented.
This paper is organized as follows. In Sect. 2 we report the
selection criteria for the sample used in this work. Section 3
presents the multi-wavelength data-set, while in Sect. 4 the
method to compute average SED is described. Section 5 con-
cerns the multi-componentmodeling used to disentangle the nu-
clear emission from the stellar light. In Section 6 the method
used to compute intrinsic bolometric luminosites and bolomet-
ric corrections for Type-2 AGN is described, while in Sect. 7 we
have applied the same method to a sample of Type-1 AGN. The
discussion of our findings is given in Sect. 8, while in Sect. 9 we
summarize the most important results.
We adopted a flat model of the universe with a Hubble con-
stant H0 = 70kms−1Mpc−1, ΩM = 0.27, ΩΛ = 1 − ΩM
(Komatsu et al. 2009).
2. The Data Set
The XMM-COSMOS catalog comprises 1822 point–like X–ray
sources detected by XMM-Newton overan area of ∼ 2deg2. The
total exposure time was ∼ 1.5 Ms with a fairly homogeneous
depth of ∼ 50 ks over a large fraction of the area (Hasinger et al.
2007, Cappelluti et al. 2009). Following Brusa et al. (2010), we
excluded from our analysis 25 sources which turned out to
be a blend of two Chandra sources. This leads to a total of
1797 X–ray selected point-like sources. We restricted the anal-
ysis to 1078 X–ray sources detected in the [2-10] keV band
at a flux larger than 2.5 × 10−15erg s−1cm−2(see Table 2 in
Cappelluti et al. 2009). The objects for which no secure optical
counterpartcouldbe assignedare often affectedby severeblend-
ingproblems,so thatwe considerinthis analysisthe971sources
(hereafter971-XMM) for which a secure optical counterpartcan
be associated (see discussion in Brusa et al. 2010)1.
From the 971-XMM catalog we have selected 255 sources,
whichdonotshowbroad(FWHM< 2000kms−1)emission lines
in their optical spectra2(hereafter we will refer to them as the
1The multi-wavelength XMM-COSMOS catalog can be retrieved
from: http://www.mpe.mpg.de/XMMCosmos/xmm53 release/, version
1stApril 2010.
2The origin of spectroscopic redshifts for the 255 sources is as
follows: 11 objects from the SDSS archive, 2 from MMT observa-

Page 3

E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey3
Fig.1. Plot of the [2 − 10]keV flux versus the total i∗CFHT
magnitude for our sample of 255 Type-2 AGN. The red circles
representsourceswitha de-absorbed2–10keVluminositylower
than 1042erg s−1. The dashed lines represent a constant X–ray
to optical flux ratio of Log (X/O) = ±1.
Type-2 AGN sample): 223 are classified not-broad-line AGN,
while 32 are absorption-line galaxies. Not-broad-line AGN are
objects with unresolved, high-ionization emission lines, ex-
hibiting line ratios indicating AGN activity, and, when high-
ionization lines are not detected, or the observed spectral range
does not allow to construct line diagnostics, objects without
broad line in the optical spectra. Absorption-line galaxies are
sources consistent with a typical galaxy spectrum showing only
absorption line.
In Figure 1 we plot the 2–10 keV X–ray flux as a func-
tion of i∗CFHT magnitude. The dashed lines limit the region
typically occupied by AGN along the X–ray to optical flux ra-
tio Log (X/O) = ±13. Nine sources have a de-absorbed 2–
10 keV luminosity lower than 1042erg s−1, the conventional
threshold below which the X–ray sources that can plausibly
be explained by moderate-strength starbursts, hot gas in ellip-
tical galaxies, or other sources besides accretion onto a nuclear
SMBH(Hornschemeier et al.2001).Thethreesourcesinsidethe
dashedlines haveX–rayluminosities close to 1042ergs−1, while
six AGN (6/255, 2%) lie in the part of the diagram usually oc-
cupied by star-forming galaxies, and have X–ray luminosities
< 1042erg s−1. Their inclusion in the analysis does not affect the
main results.
The Type-2 AGN sample used in our analysis comprises 255
X–ray selected AGN, all of them with spectroscopic redshifts,
spanning a wide range of redshifts (0.045 < z < 3.524) and X–
ray luminosities (41.06 ≤ Log L[2−10]keV ≤ 45.0). The redshift
distribution of the total sample and the distribution of the de-
absorbed hard X–ray luminosities are presented in Figure 2. The
tions (Prescott et al. 2006), 70 from the IMACS observation campaign
(Trump et al. 2007), 156 from the zCOSMOS bright 20k sample (see
Lilly et al. 2007), 7 from the zCOSMOS faint catalog and 9 from the
Keck/DEIMOS campaign.
3Log (X/O) = Log fx+ i∗/2.5 + 5.6.
mean redshift is ?z? = 0.76, while the mean Log L[2−10]keVis
43.34 with a dispersion of 0.64.
2.1. Correction for absorption for the X–ray luminosity
For a sub-sample of 111 AGN we have an estimate of the col-
umndensity NHfromspectralanalysis(seeMainieri et al.2007),
while for 144 AGN absorption is estimated from hardness ra-
tios (HR; see Brusa et al. 2010). For 25 sources for which a
column density estimate is not available from HR we consider
the Galactic column density. Therefore, we can compute the
de-absorbed X–ray luminosity at 0.5–2 keV (soft band) and 2–
10 keV (hard band) for all sources in our sample. In Figure 2
(right panel) we show the distribution of column densities that
ranges from 3 × 1020cm−2to 1.5 × 1024cm−2. The mean NH
value is 8.5 × 1021cm−2with a dispersion of 0.72 dex. The in-
tegrated intrinsic un-absorbed luminosity is computed assuming
a power-law spectrum with slope, Γ = 2 and Γ = 1.7 for the
0.5–2 keV and 2–10 keV bands, respectively. The average shift
induced by the correction for absorption in the Type-2 sample is
?∆Log L[2−10]keV? = 0.04± 0.01.
3. Rest-frame monochromatic fluxes and
Spectral Energy Distributions
We used the catalogby Brusa et al. (2010) whichincludesmulti-
wavelength data from mid-infrared to hard X–rays: MIPS 160
µm, 70 µm and 24 µm GO3 data (Le Floc’h et al. 2009), IRAC
flux densities (Sanders et al. 2007), u∗and i∗CFHT bands and
near-infrared KS-band data (McCracken et al. 2008), J UKIRT
(Capak et al.2008),HST/ACSF814WimagingoftheCOSMOS
field (Koekemoer et al. 2007), optical multiband photometry
(SDSS, Subaru, Capak et al. 2007) and near- and far-ultraviolet
bands with GALEX (Zamojski et al. 2007).
The number of X–ray sources detected at 160µm and 70µm
is 18 and 42, respectively. For the undetected sources in these
bands we consider 5σ upper limits of 65mJy and 8.5mJy for
160µm and 70µm, respectively. At 24µm the number of de-
tected sources is 237. For the 18 undetected sources at 24µm,
we consider 5σ upper limits of 80µJy. All 255 sources are de-
tected in the infrared in all IRAC bands, and only a few objects
were not detected in the optical and near-IR bands: we have only
8 upper limits in the z+band; 1 upper limit in the BJ and i∗
bands; 2 upper limits in the u∗band; 4 upper limits in the KS
CFHT band and 2 in the J UKIRT band. The observations in
the various bands are not simultaneous, as they span a time in-
terval of about 5 years: 2001 (SDSS), 2004 (Subaru and CFHT)
and 2006 (IRAC). Variability for absorbed sources is likely to
be a negligible effect, but, in order to further reduce it, we se-
lected the bands closest in time to the IRAC observations (i.e.,
we excluded SDSS data, that in any case are less deep than other
data available in similar bands). GALEX bands are not taken
into account because, given the large aperture they can include
light from close companions. All the data for the SED compu-
tation were shifted to the rest frame, so that no K-corrections
were needed. Galactic reddening has been taken into account:
we used the selective attenuation of the stellar continuum k(λ)
taken from Table 11 of Capak et al. (2007). Galactic extinction
is estimated from Schlegel et al. (1998) for each object in the
971-XMM catalog. Count rates in the 0.5-2 keV and 2-10 keV
are converted into monochromatic X–ray fluxes in the observed
frame at 1 and 4 keV, respectively, using a Galactic column den-
sity NH = 2.5 × 1020cm−2(see Cappelluti et al. 2009), and as-

Page 4

4E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey
Fig.2. Left panel: Redshift distribution of the 255 Type-2 AGN considered in this work. Center panel: Intrinsic hard X–ray lumi-
nosity distribution of the 255 Type-2 AGN considered in this work. Right Panel: Column density distribution of the 255 Type-2
AGN (black histogram), of the 111 Type-2 AGN with an NHestimate from spectral analysis (grey filled histogram), and of the 144
Type-2 AGN with an NHestimate from hardness ratios (orange hatched histogram).
Fig.3.Mean(orangecrosses) andmedian(redpoints)SEDfrom
the total sample of 255 Type-2 AGN. The blue points represent
the rest-frame data, from infrared to X–ray, used to construct
the averageSED, while the black lines representthe interpolated
SED.
suming a photon index Γx = 2 and Γx = 1.7, for the soft and
hard band, respectively. We do not correct these X–ray fluxes
for the intrinsic column density. All sources are detected in the
2-10keV bandbydefinitionofthe sample, whilein the soft band
we have 70 upper limits.
4. Average SED
We have computed the individual rest-frame SEDs for all
sources in the sample, following the same approach as in L10,
and we have normalized all of them at 1µm. After this nor-
malization we divided the frequency interval from Log νminto
Log νmaxusing a fixed step ∆Log ν. The minimum and max-
imum frequency depends on both the data and the redshift of
the source considered to compute the SED, in our case we
have used Log νmin = 12 Hz and Log νmax = 20 Hz, with
a ∆Log ν = 0.02. We averaged data in each given interval
Log νl≤ Log ν ≤ Log νl+1. The mean and median SEDs are ob-
tained by taking the arithmetic mean and the median of logarith-
mic luminosities, Log L, in each bin. It is important to note that
sources at different redshift contribute to the same bin. Because
of the relatively wide range of redshifts, the lowest and the high-
est frequency bins are populated by a variable number of points.
This effect may introduce relatively high fluctuations in the av-
erage luminosity in those bins. In order to minimize these ef-
fects we select a minimumnumberof 200SEDs in each bin (this
number depends on the total number of sources in the sample).
Then, we select the mean reference frequency, Log ν, of the bin
and use a binary-searchalgorithmto findall luminosities that
correspond at Log ν (if a source does not have a frequency that
correspond to Log ν, we choose the luminosity with the closer
frequency to Log ν). Finally, all adjacent luminosities in each
bin are then connected to compute the final mean and median
SED. In Figure 3 the resulting mean and median SEDs are re-
ported with orange crosses and red points, respectively.The data
points are reported in order to show the dispersion with respect
to 1µm. The average SED is characterized by a flat X–ray slope,
?Γ = 1.12?, while in the optical-UV the observed emission ap-
pears to be consistent with the host-galaxy. The flattening of the
X–ray slope is likely due to the fact that we do not correct for
the intrinsic absorption the fluxes at 1 and 4 keV. The average
SED in the mid-infraredis probablya combinationof dust emis-
sion from star-forming region and AGN emission reprocessed
by the dust. Before trying to deconvolve each source using an
SED-fitting code, we binned the total sample in X–ray and in-
frared luminosites and redshift. We used the luminosity at 4 keV
and 8µm to divide the total sample in 6 bins with the same num-
ber of sources in each bin. The wavelength of the luminosity
used to bin the sample is chosen to minimize the host-galaxy
contribution. In the three panels of Figure 4 the resulting mean
SEDs are shown. The redshift trend is directly connected to the
luminosity trend, since at higher redshifts we are looking at the
more luminous sources. The shapes of the average SEDs in the
optical bands are approximately the same in all luminosity and
redshift bins. As expected, there is a stronger host-galaxy con-
tribution at lower luminosity/redshift bins, where the average
SEDs have a typical galaxy shape. Moreover, there is a trend
between X–ray and mid-infrared luminosity: the contribution in

Page 5

E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey5
Fig.4. Average SEDs in the rest-frame Log (νLν) − Log ν plane. Right panel: Mean SEDs computed binning in X–ray luminosity
at 4 keV. Center panel: Mean SEDs computed binning in infrared luminosity at 8µm. Left Panel: Mean SEDs computed binning in
redshift. The color code refers to the different bins as labeled.
the infrared is higher at higher X–ray luminosities. This effect
is already known for both Type-1 and Type-2 AGN using the
intrinsic (non-stellar) emission from the AGN (e.g., Lutz et al.
2004, Gandhi et al. 2009). The observed average SED for our
sample is a combination of both host-galaxy and AGN light, but
the change in the average shape in the mid-infraredas a function
of the X–rayluminosity suggests that most luminoussources are
probablydominated by the AGN emission at those wavelengths.
5. SED-fitting
ThepurposeofperformingSED fitting is to properlydisentangle
the emission associated to stellar light from that due to accretion
and constrain physical parameters. Since the relative contribu-
tionofthedifferentcomponentsvarieswithwavelength,aproper
decompositioncan be obtainedthroughan SED-fitting approach
providing a robust estimate of the nuclear emission (bolometric
luminositiesand bolometriccorrections)and its relationwith the
host-galaxy properties (mass, star formation rates, morphologi-
cal classification). A well sampled SED is mandatory; in par-
ticular, far-infrared observations are fundamental to sample the
star-formation activity, while mid-infrared observations are nec-
essary to sample the region of the SED where most of the bolo-
metricluminosityof obscuredAGN is expectedto bere-emitted.
In Fig. 5 the broad-band SEDs of four XMM-Newton Type-2
AGN are plotted as examples. The lower two panels are repre-
sentative of a full SED with all detections from the far-infrared
to theoptical.Unfortunately,thereare veryfew detectionsat 160
and 70µm, so that the more representative situation is shown in
the upper left panel of Fig. 5. The three components adopted
in the SED-fitting code, starburst, AGN torus and host-galaxy
templates, are shown as a long-dashedline, solid line and dotted
line, respectively. All the templates used in the SED-fitting code
will be described in the following Sections. The red line repre-
sents the best-fit, while the black points represent the photomet-
ric data used in the code, from low to high frequency: MIPS-
Spitzer (160µm, 70µm and 24µm if available), 4 IRAC bands, K
CFHT, J UKIRT, optical Subaru and CFHT bands.
The observed data points from infrared to optical are fitted
with a combination of various SED templates (see Sect. 5.1) us-
ing a standard χ2minimization procedure
χ2=
nfilters
?
i=1
?Fobs,i− A × Fgal,i− B × Fagn,i−C × Fir,i
σi
?2
(1)
where Fobs,iand σiare the monochromatic observed flux and
its error in the band i; Fgal,i, Fagn,iand Fir,iare the monochro-
matic template fluxes for the host-galaxy, the AGN and the star-
burst component, respectively; A, B and C are the normaliza-
tion constants for the host-galaxy, AGN and starbust compo-
nent, respectively. The starburst component is used only when
the source is detected at 160µm and 70µm. Otherwise, a two
components SED-fit is used. Sixteen is the maximum number of
bandsadoptedintheSED-fitting(onlydetectionareconsidered),
namely: 160µm, 70µm, 24µm, 8.0µm, 5.8µm, 4.5µm, 3.6µm, KS,
J, z+, i∗, r+, g+, VJ, BJand u∗.
5.1. Template libraries
5.1.1. Optical template library
We used a set of 75 galaxy templates built from the
Bruzual & Charlot(2003, BC03 hereafter)codeforspectralsyn-
thesis models, using the version with the “Padova 1994” tracks,
solar metallicity and Chabrier IMF (Chabrier2003). For the pur-
poses of this analysis a set of galaxy templates representative of
the entire galaxy population from ellipticals to starbursts is se-
lected.To thisaim,10exponentiallydecayingstarformationhis-
tories with characteristic times ranging from τ = 0.1 to 30Gyr
and a model with constant star formation are included. For each
SFH, a subsample of ages available in BC03 models is selected,
to avoid both degeneracy among parameters and speed up the
computation. In particular, early-type galaxies, characterised by
a small amount of ongoing star formation, are represented by
models with values of τ smaller than 1 Gyr and ages larger than
2Gyr, whereas more actively star forming galaxies are repre-
sented by models with longer values of τ and a wider range of
ages from 0.1 to 10Gyr. An additional constraint on the age is
that, for each sources, the age has to be smaller than the age of
the Universe at the redshift of the source. Each template is red-
dened according to the Calzetti et al. (2000) reddening law. The
input value is E(B − V), corresponding to a dust-screen model,
with Fo(λ) = Fi(λ)10−0.4E(B−V)k(λ), where Foand Fiare the ob-
served and the intrinsic fluxes, respectively. The extinction at a
wavelength λ is related to the colour excess E(B− V) and to the
reddening curve k(λ) by Aλ= k(λ)E(B − V) = k(λ)AV/RV, with
RV = 4.05 for the Calzetti’s law. The E(B − V) values range
between 0 and 1 with a step of 0.05.

Page 6

6E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey
5.1.2. AGN template library
The nuclear SED templates are taken from Silva et al. (2004).
They were constructed from a large sample of Seyfert galaxies
selected from the literature for which clear signatures of non-
stellarnuclearemissionweredetectedinthenear-IRandmid-IR.
After a proper subtraction of the stellar contribution, the nuclear
infrared data were interpolated with a radiative transfer code for
dust heated by a nuclear source with a typical AGN spectrum,
and including different geometries, dust distribution, variation
of the radii, density and dust grain sizes to account for possible
deviations from a standard ISM extinction curve (see for more
details Granato & Danese 1994; Maiolino et al. 2001).
The infrared SEDs were then normalized by the intrinsic,
unabsorbed X-ray flux in the 2–10 keV band, and are divided
into 4 intervals of absorption: NH < 1022cm−2for Sy1, while
1022< NH < 1023cm−2, 1023< NH < 1024cm−2and NH >
1024cm−2for Sy2 (see Fig. 1 in Silva et al. 2004). The main
differences between the SEDs of Sy1s and Sy2s with 1022<
NH< 1023cm−2are the absorption in the near-IR at about λ <
2µm and the silicate absorption at λ = 9.7µm, which are present
in the Sy2 template. The shape of the SED in the mid-infrared
with 1023< NH< 1024cm−2is quite similar to that with 1022<
NH < 1023cm−2. The Compton-thick SED (NH > 1024cm−2)
shows conspicuousabsorptionalso at λ ∼ 1.3µm. If a source has
the NHvalue available, this is used as a prior in the selection of
the best-fit AGN template.
5.1.3. Starburst template library
We used two different starburst template libraries for the SED-
fitting: Chary & Elbaz (2001) and Dale & Helou (2002). These
template libraries represent a wide range of SED shapes and lu-
minosities and are widely used in the literature. Here, we briefly
describe how each of these libraries was derived and discuss the
main differences between them.
The Chary & Elbaz (2001) template library consists of 105
templatesbasedontheSEDsoffourprototypicalstarburstgalax-
ies (Arp220 (ULIRG); NGC 6090 (LIRG); M82 (starburst); and
M51(normalstar-forminggalaxy)).Theywerederivedusingthe
Silva et al. (1998) models with the mid-infrared region replaced
with ISOCAM observationsbetween 3 and 18µm (verifyingthat
the observed values of these four galaxies were reproduced by
the templates). These templates were then divided into two por-
tions (4–20µm and 20–1000µm) and interpolated between the
four to generate a set of libraries of varying shapes and lumi-
nosities. The Dale et al. (2001) templates are also included in
this set to extend the range of shapes.
The Dale & Helou (2002) templates are updated versions of
the Dale et al. (2001) templates. This model involves three com-
ponents, large dust grains in thermal equilibrium, small grains
semistochastically heated,and stochastically heated PAHs. They
are based on IRAS/ISO observations of 69 normal star-forming
galaxies in the wavelength range 3–100µm. Dale & Helou
(2002) improveduponthese models at longer wavelengthsusing
SCUBA observations of 114 galaxies from the Bright Galaxy
Sample (BGS, see Soifer et al. 1989), 228 galaxies observed
with ISOLWS (52–170µm; Brauher 2002), and 170µm obser-
vations for 115 galaxies from the ISOPHOT Serendipity Survey
(Stickel et al. 2000). All together, these 64 templates span the
IR luminosity range 108− 1012L⊙. The total infrared template
sample used in our analysis is composed of 168 templates.
6. Bolometric luminosities and bolometric
corrections
The nuclear bolometric luminosities and bolometric corrections
are estimated, using an approach similar to Pozzi et al. (2007,
see also Vasudevan et al. 2010; Pozzi et al. 2010), whereas the
infrared luminosity is used as a proxy of the intrinsic nuclear
luminosity. The appropriate nuclear template from Silva et al.
(2004) is selected based on the absorbing column density NH,
when available, or from the best-fit nuclear infrared template. In
order to compute the hard X–ray bolometric correction we used
the standard definition
kbol=
Lbol
L[2−10]keV
(2)
where the L[2−10]keV is the intrinsic X–ray luminosity and the
bolometric luminosity is computed as the sum of the total in-
frared and X–ray luminosity
Lbol= LIR+ LX.
(3)
Afterperformingthe SED-fitting,onlythenuclearcomponentof
thebest-fit is integrated.Hence,the totalIR luminosity LIRis ob-
tained integrating the nuclear template between 1 and 1000µm.
To convert this IR luminosity into the nuclear accretion disk
luminosity, we applied the correction factors to account for
the torus geometry and the anisotropy (see Pozzi et al. 2007).
The first correction is parameterized by the covering factor f.
The covering factor is related to the geometry of the torus that
obscures the accretion disk emission in the optical-UV along
the line of sight, and its value is estimated from the ratio of
obscured/unobscured quasars found by the X–ray background
synthesis models (Gilli et al. 2007). This correction factor is
∼ 1.5. This value correspond to a typical covering factor of
f ∼ 0.67,consistentwith theresults basedonclumpytorusmod-
els (Nenkova et al. 2008).
Theanisotropyfactor is definedas the ratio of the luminosity
offace-onversusedge-onAGN, wheretheobscurationis a func-
tion of the column density NH. Therefore, SEDs in Silva et al.
(2004) have been integrated in the 1–30µm range, after normal-
izing these SEDs to the same luminosityin the 30–100µmrange.
The derived anisotropy values are 1.2–1.3 for 1022< NH< 1024
and 3–4 for NH> 1024. The same values as in Vasudevan et al.
(2010) are adopted: 1.3 for 1022< NH < 1024and 3.5 for
NH> 1024.
The total X–ray luminosity LXis estimated integratingin the
0.5-100 keV range the X–ray SED. We have interpolated the
de-absorbed soft and hard X–ray luminosities. Since we are in-
tegrating at the rest-frame frequencies, the X–ray SED is ex-
trapolated at higher and lower energies using the estimated X–
ray slope, and introducing an exponential cut-off at 200 keV
(Gilli et al. 2007, see also Sect. 3.1 in L10).
7. Robustness of the method
The robustness of the method used to estimate nuclear bolomet-
ric luminositiesandbolometriccorrectionsfromSED-fitting,for
the sample of Type-2 AGN, has been tested against the updated
soft X–ray selected sample of Type-1 AGN discussed in L10.
The Type-1 AGN sample in the L10 work was composed of 361
spectroscopicallyclassifiedbroad-lineAGN.Therecentworkby
Brusa et al. (2010) has updated the spectroscopic classification
and increased the number of Type-1 AGN with spectroscopic
redshift, so that the final sample is composed of 395 Type-1

Page 7

E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey7
Fig.5. Examples of SED decompositions. Black circles are the observed photometry in the rest-frame (from the far-infrared to the
optical-UV).The long-dashed,solid and dottedlines correspondrespectivelyto the starburst, AGN and host-galaxytemplates found
as the best fit solution. The red line represents the best-fit SED. The stellar mass and the SFR derived from the galaxy template are
reported.
AGN in the redshift range 0.103 ≤ z ≤ 4.255 with X–ray lu-
minosities 42.20 ≤ Log L[2−10]keV≤ 45.23. We have computed
bolometric and X–ray luminosities, and bolometric corrections
using the same approach as in L10 for the Type-1 sample: bolo-
metric luminosites are computed by integrating the rest-frame
SEDs from 1µm up to the UV-bump. In order to compare these
estimates with the results from the SED-fitting code, we have
applied to the same sample the method described in Sect. 5 and
6 to estimate bolometric parameters. To be consistent with the
selection criteria of the sample discussed in this paper, we have
considered only AGN with X–ray detection in the hard band, re-
moving from the main sample 87 Type-1 AGN with an upper
limit at 2–10 keV. Moreover, for 2 sources the best-fit does not
consider an AGN component, so we cannot compute the bolo-
metric luminosities for them. The final test sample is composed
of 306 Type-1 AGN in the redshift range 0.103 ≤ z ≤ 3.626 and
X–ray luminosities 42.20 ≤ Log L[2−10]keV≤ 45.04.
In order to select the appropriate nuclear template from
Silva et al. (2004), we consider the SED for Sy1 AGN (no cor-
rection for anisotropyis necessary in this case) and the Sy2 SED
with 1022< NH< 1023for 20 AGN that have NHin this range.
We present the comparison between the values of Lboland
kbolfrom L10 and this work in Fig. 6. The outlier in the bottom
side of the plot, XID=357 at redshift 2.151 has Log kbol= 1.95
from L10 and Log kbol = 1.04 using the new approach, and
presents large error bars in the 24µm detection, so that the total
bolometric luminosity, computed using the infrared luminosity,
is probably underestimated. The outlier in the right end of the
distribution, XID=5114 at redshift 0.212 (Log kbol= 2.82 from
L10 and Log kbol= 1.95 using the new approach) has detections
at 160, 70 and 24µm, Log NH = 22.68 and Log L[2−10]keV =
42.89. Probably this source is a star-forming galaxy, so that us-
ing the L10 approach we included stellar emission in the esti-
mate of the nuclear bolometric luminosity, thus overestimating
the nuclear bolometric luminosity and, therefore, the bolometric
correction. The last notable outlier in the top/left side of the dis-
tribution,XID=2152at redshift0.627(Logkbol= 1.35fromL10
and Log kbol= 2.27 using the new approach) presents a signif-
icant host-galaxy contribution in the optical-UV and, therefore,
the bolometric luminosity is likely to be underestimated in the
L10 approach.
Although the two methods are very different, the bolometric
luminosity estimates agree remarkably well, with a 1σ disper-
sion of 0.20 dex after performing a 3.5σ clipping method in or-
der to avoid outliers. Bolometric luminosities from SED-fitting
are on average slightly larger than those computed integrating
the rest-frame SED from 1µm to the X–ray (see the lower left
side in Fig. 6). This effect is also present in the Vasudevan et al.

Page 8

8E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey
Fig.6. Upper panel: Comparison between the values of bolometric luminosity and bolometric correction from data presented in
L10 and from this work. The three red triangles mark the outliers discussed in Sect. 7. Lower panel: Distribution of the differences
between the values of bolometric luminosity and bolometric correction from data presented in L10 and from this work.
(2010)work. A possible explanationis that SED-fitting underes-
timates the host-galaxy contribution, or that the anisotropy and
geometry corrections are too large for some objects. The agree-
ment between the two methods is overall quite satisfactory and
in the followingwe will discuss ourfindings for the Type-2sam-
ple.
8. Results and discussion
8.1. Bolometric correction and luminosites for Type-2 AGN
Bolometric luminosities and bolometric corrections have been
computed for the Type-2 AGN sample. Intrinsic soft and hard
X–ray luminosities are estimated as described in Sect. 2.1. For
15 sources we do not have an estimate of the AGN compo-
nent from the SED-fitting, and we cannot compute the bolomet-
ric luminosity for them. In Fig. 7 the distribution of the bolo-
metric correction for the 240 Type-2 AGN sample and for the
306 Type-1 AGN (kbol for Type-1 AGN are computed using
the SED-fitting code) are presented. Figure 8 shows bolomet-
ric corrections for both the Type-1 and the Type-2 AGN sam-
ples as a function of the hard X–ray luminosity. For both sam-
ples, bolometric parameters are estimated from the SED-fitting
as discussed in Sect. 5. The green and orange curves represent
the bolometric corrections and their 1 σ dispersion as derived
by Hopkins et al. (2007) and Marconi et al. (2004), respectively.
Type-2 AGN have, on average, smaller bolometric corrections
than Type-1 AGN at comparable hard X–ray luminosity. For
example, at 43.30 ≤ Log L[2−10]keV ≤ 44.30 (vertical lines in
Fig. 8), where both AGN types are well represented, the me-
Fig.7. Distribution of the bolometric correction for the 240
Type-2 AGN sample (red hatched histogram) and for the 306
Type-1 AGN (blue hatched histogram).
dian bolometric correction for the Type-2 AGN (134 objects) is
?kbol? ∼ 13±1, to be comparedwith a medianbolometriccorrec-
tion ?kbol? ∼ 23 ± 1 for the Type-1 AGN (167 objects). The two
averages are statistically different at the ∼ 7 σ level and this is

Page 9

E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey9
Fig.8. Hard X–ray bolometric correction against the intrinsic
2–10 keV luminosity for 240 Type-2 AGN with AGN best-fit
(red data). The crosses represent the bolometric correction for
306 Type-1 AGN, computed with the approach described in
Sect. 6. The green and blue lines represent the bolometric cor-
rection and the 1σ dispersion obtained by Hopkins et al. (2007)
and Marconi et al. (2004), respectively. The red points and open
squares represent the 111 Type-2 AGN with NHfrom spectral
analyses and the 144 Type-2 AGN with NHfrom HR, respec-
tively.
consistent with the results in Vasudevan et al. (2010). The mean
L[2−10]keVfor the Type-1 and Type-2AGN within this luminosity
rangediffersbya factor1.8,andthis couldinprincipleexplainat
least part of the difference in the average bolometric corrections
for the two samples of AGN. However, the significance of the
difference is still present if we split this luminosity range in two
equal Log L[2−10]keVbins and perform a Kolmogorov–Smirnov
test for the Type-1 and Type-2 AGN luminosity distributions in
each bin.
Vasudevan & Fabian (2009) and Lusso et al. (2010) have
shown that hard X–ray bolometric corrections are correlated
with the Eddington ratios (λEdd = Lbol/LEdd) for Type-1 AGN
(see also Marconi et al. 2004; Kelly et al. 2008). The kbol− λEdd
relation suggests that there is a connection between the broad-
band emission, mostly in the optical-UV, and the Eddington
ratio, which is directly linked to the ratio between mass ac-
cretion rate and Eddington accretion rate. A high λEdd corre-
sponds to an enhanced optical-UV emission, which means a
prominent big-blue bump and therefore a higher kbol. The dif-
ference between the average bolometric corrections for Type-
1 and Type-2 AGN could be due to lower mass accretion rates
in Type-2 AGN, assuming the same black hole mass distribu-
tion for the two AGN populations (see Trump et al. 2011). The
current theoretical framework of AGN/host-galaxy co-evolution
predicts that obscured AGN are highly accreting objects and
their black hole is rapidly growing. However, we note that this
is true for z = 1 − 3 (see Hasinger 2008), while the majority
of Type-2 AGN in the present sample are relatively luminous
(42.37 ≤ Log Lbol ≤ 46.80), and at moderately low redshift
(0 < z < 1). Therefore, the Type-2 AGN sample is likely to
represent a later stage in AGN evolution history.
8.2. Infrared emission: indication of AGN activity
The re-processedinfraredemission can be used as a proxyof the
average disc emission, since the timescale for transfer of energy
from the disk to the outer edge of the torus into infrared emis-
sion is of the order of several years in standard AGN picture;
whereas optical, UV and X–ray variability in AGN is known to
occur on shorter timescales. The correlation between the 2–10
keV X–ray emission and IR emission at 12.3µm for a sample of
Seyfert nuclei has been discussed in Gandhi et al. (2009), and it
could be used to estimate the intrinsic AGN power. Using X–ray
data from the literature and new IR data from the Very Large
Telescope’s Imager and Spectrometer for mid-Infrared (VISIR),
taken specifically for addressing the issue of nuclear emission
in local Seyferts, they found a tight correlation between intrin-
sic, uncontaminated IR luminosity and X–ray luminosity in the
2–10 keV range
LogL12.3 µm
1043
= (0.19±0.05)+(1.11±0.07)LogL[2−10]keV
1043
. (4)
The relation is characterized by a small scatter with a standard
deviation of 0.23 dex. The expected nuclear mid-infrared lumi-
nosity is computed from Eq. (4) using the estimate of the intrin-
sic unabsorbed X–ray luminosity. From the observed rest-frame
SED (AGN+host-galaxy) the luminosity at 12.3 µm is com-
puted. A comparisonof the total observedluminosity at 12.3µm
and that predicted by Eq. (4) is plotted in Figure 9 for four rep-
resentative sources. In Figure 10 the distribution of the ratio
r = Log
?
AGN sample. The distribution of the ratio r has a mean which is
shifted from zero by ∼ 0.2. However, if we consider a gaussian
distributioncenteredat r = 0 with σ = 0.23,i.e. the samedisper-
sion observed by Gandhi et al. (2009) in their local sample, the
majority of the objects are found within 2σ of the r distribution.
The tail outside 2σ and extending to high r includes 73 sources
(with r ? 0.5) forwhich the predictedmid-infraredluminosityis
significantly lower than observed. The hard X–ray luminosities
of these 73 AGN are mainly in range Log L[2−10]keV∼ 42 − 44,
where the local correlation is well-sampled. There are two pos-
sible explanations for a significant (∼ 30% of the objects) tail
toward high-r values: either the Gandhi relation, which was de-
rived for a sample of local Seyfert galaxies, cannot be extended
to allthe sourcesinoursampleorthe SED-fittingproceduremay
overestimate, in a fraction of these objects, the nuclear contribu-
tion. In order to study the properties of these outliers, bolometric
corrections,morphologies,stellar masses and SFR are discussed
in following.We call “low-r”AGN all sourceswithin 2σ ofthe r
distribution, while the “high-r” AGN sample is populated by the
sources deviating more than 2σ (see Fig. 9 for some examples).
A clear separation in bolometric corrections for these two
sub-samples is found. This is shown in Figure 11 in which bolo-
metriccorrectionsareplottedasafunctionofthe2–10keVlumi-
nosity. At a given hard X–ray luminosity (43 ≤ Log L[2−10]keV≤
44) the low-r sample has a median bolometric correction of
?kbol? ∼ 11 ± 1 (110 objects), to be compared with a median
bolometric correction for the high-r sample of ?kbol? ∼ 26 ± 3
(44 objects). The two median values for kbolare statistically dif-
ferent at the ∼ 5σ level.
Furthermore,in the high-r sample 24 Type-2 AGN out of 73
havea detectionat 70µm(∼ 33%,significantly higherthan those
for the low-r sample, ∼ 4%) and 9 of these 24 have also a de-
tection at 160µm (∼ 12% considering the total high-r sample,
and only ∼ 1% for the low-r sample). This denotes that the dif-
ference in the average bolometric corrections between the low-r
L12.3 µm,obs/L12.3 µm,predicted
?
is plotted for the Type-2

Page 10

10E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey
Fig.9. Examples of SED decompositions. Black circles are the observed photometry in the rest-frame (from the far-infrared to the
optical-UV).The long-dashed,solid and dottedlines correspondrespectivelyto the starburst, AGN and host-galaxytemplates found
as the best fit solution. The red line represents the best-fit SED. The stellar mass and the SFR derived from the galaxy template are
reported. The green point represents the nuclear mid-infraredluminosity using Eq. (4), while the cross represents the total observed
luminosity at 12.3 µm computed from the rest-frame SED. XID=19 and 81 are examples of low-r AGN, while XID=172 and 117
represent high-r AGN.
and high-r samples is probably due to the fact that a significant
fraction of the infrared emission is attributable to an incorrect
modelingof the star-formationprocess, or the AGN contribution
is somehow overestimated by the SED-fitting procedure.
There is no significant difference in the average nuclear ab-
sorption between the low-r and the high-r sample, while there
is a possibly significant difference in SFR and stellar masses.
The median stellar mass in the high-r sample is ?Log M∗? ∼
10.93M⊙with a dispersion of 0.30, while for the low-r sam-
ple is ?Log M∗? ∼ 10.76M⊙ with σ = 0.30. The two av-
erages are statistically different at the ∼ 3 σ level. The me-
dian SFR, as derived from the SED-fit, for the high-r sample
is ?SFR? ∼ 17 ± 3M⊙/yrs with a σ = 0.30, while for the low-r
sample is ?SFR? ∼ 3 ± 1M⊙/yrs with a σ = 0.30 and the two
averages are statistically different at the 4.4 σ level.
Overall, the SED-fitting for the 73 Type-2 AGN is likely to
overestimate the AGN emission in the infrared, which is proba-
bly due to the infrared emission from star-forming regions. The
average bolometric correction for Type-2 AGN, excluding these
sources, would be even lower than what we have computed in
the previous Section. This reinforces the idea of lower bolomet-
ric corrections for Type-2 AGN with respect to Type-1 AGN.
The low bolometric corrections for Type-2 AGN could be also
explained if a fraction of the accretion disk bolometric output is
not re-emitted in the mid-infrared, but rather dissipated (e.g., by
AGN-feedback). This would not be accounted in the bolomet-
ric luminosity,and couldprovidea plausible explanationfor low
kbolespecially if the low-r sample is considered. At this stage,
this is just a speculation, and more work is needed to verify this
possibility.
8.3. Host-galaxy properties: M∗, SFR, colors and
morphologies
Galaxies show a colour bi-modality both in the local Universe
and at higher redshift (up to z ∼ 2; e.g., Strateva et al. 2001;
Bell et al. 2004). This bi-modality(red-sequenceand blue-cloud
galaxies) has been interpreted as an evidence for a dicothomy
in their star formation and merging histories (e.g., Menci et al.
2005, but see also Cardamone et al. 2010 for an alternative ex-
planation).Color-magnitudeandcolor-massdiagrams(e.g.,rest-
frame (U − V) versus stellar mass) have been used as tools in
galaxy evolution studies, and since many models invoke AGN
feedback as an important player in such evolution, it is inter-

Page 11

E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey11
Table 1. Properties of the Type-2 AGN sample.
XIDRedshift Log L[2−10]keV
[ergs−1]
42.79
43.83
43.54
42.98
42.58
41.90
44.22
43.67
43.47
43.49
43.65
44.08
43.70
42.61
43.54
43.46
44.11
44.00
43.75
43.29
Log Lbol
[ergs−1]
43.80
44.85
44.52
44.28
43.38
42.91
45.28
44.76
45.59
44.50
44.65
45.11
44.68
43.54
44.78
44.84
45.05
45.60
44.82
44.25
kbol
Log M∗
[M⊙]
9.68
10.19
10.45
10.79
11.18
9.39
11.30
11.07
11.24
10.56
10.93
10.79
10.92
10.21
9.97
10.13
11.18
10.65
9.87
10.64
SFR
[M⊙/yrs]
1.86
38.60
32.82
4.94
0.11
0.00
1.47
35.88
199.40
3.77
0.62
0.45
0.61
0.00
116.26
38.52
0.00
542.60
92.88
0.55
MU
MV
MJ
Morphological classa
67
65
64
63
54
45
43
19
117
116
112
104
101
100
99
85
81
70
152
150
0.367
0.979
0.686
0.355
0.350
0.121
1.162
0.659
0.936
0.874
0.762
0.623
0.927
0.270
0.730
1.001
0.915
0.688
0.895
0.740
10.24
10.62
9.57
19.88
6.32
10.31
11.49
12.44
132.11
10.17
9.97
10.75
9.56
8.59
17.47
23.99
8.65
39.74
11.72
9.27
-18.63
-20.50
-20.17
-20.61
-20.57
-16.76
-21.55
-21.57
-21.81
-20.19
-20.00
-20.25
-21.19
-19.14
-21.61
-20.15
-20.35
-20.87
-21.06
-19.38
-19.80
-21.63
-21.56
-22.13
-22.62
-18.70
-23.48
-22.88
-23.34
-21.81
-22.13
-22.18
-22.93
-20.98
-22.26
-21.34
-22.59
-22.32
-21.82
-21.28
-20.72
-22.71
-23.05
-23.26
-23.78
-19.77
-24.61
-24.02
-24.80
-22.90
-23.52
-23.30
-23.78
-21.92
-23.15
-22.56
-24.07
-24.31
-22.85
-22.50
0
7
10
2
12
1
2
1
23
0
3
1
0
0
11
8
2
3
7
3
Notes—This table is presented entirely in the electronic edition; a portion is shown here for guidance.
aThe morphological classification of the Type-2 AGN hosts is coded from 0 to 23: 0 = elliptical, 1 = S0; 2 = bulge-dominated; 3 =
intermediate-bulge; 4 = disk-dominated; 5 = irregular; 6 = compact/irregular; 7 = compact; 8 = unresolved/compact; 9 = blended; 10 =
bulge-dominated/close-companion; 11 = intermediate-bulge/close-companion; 12 = S0/close-companion; 23 = possible mergers.
Fig.10. Histogram of the ratio between the total observed lumi-
nosity at 12.3 µm and the mid-infrared luminosity predicted by
Eq. (4). The red curve represents a gaussian with mean equal to
zero and standard deviation 0.23. The 1σ and 2σ standard devi-
ations of the correlation are also reported.
esting to locate the hosts of Type-2 AGN in those diagrams.
Using the galaxy component obtained from the best fit of the
Type-2 AGN, it is possible to derive rest-frame colors for the
host that, linked to the stellar mass and the morphology, can
provide hints on AGN feedback. Several studies found that
the hosts of obscured AGN tend to be redder than the over-
Fig.11. Hard X–ray bolometric correction against 2–10 keV lu-
minosity for 240Type-2 AGN with AGN best-fit. The 240 Type-
2 sample is divided into subsamples: low-r AGN sample (black
data) and high-r AGN sample (yellow data). The green and or-
ange lines represent the bolometriccorrectionand 1σ dispersion
obtained by Hopkins et al. (2007) and Marconi et al. (2004), re-
spectively.Inthede-absorbedhardX–rayluminosityrangehigh-
lighted by the solid lines, we have 167 low-r and 73 high-r (in
the infrared) sources.
all galaxy population in the rest-frame (U − V) color (e.g.,
Nandra et al. 2007). There are at least two possible and sig-

Page 12

12E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey
Fig.12. The morphology distribution (using the ZEST+ code)
of the 233 AGN host-galaxies on the (U − V) colour-mass di-
agram. We also plotted the 22 sources without morphological
information. The (U −V) color and stellar masses are computed
using the SED-fitting code. We overplot the contours of about
8700 galaxies in zCOSMOS (colours and stellar masses from
Hyperz). The morphology classification is labeled as follow: el-
liptical (Ell), S0, bulge-dominated galaxy (BD), intermediate-
bulge galaxy (IB), disk-dominated galaxy (DD), irregular (Irr),
Compact, possible mergers (PM) and unresolved compact (UC).
The red dashed line represents the red sequence cut defined by
Borch et al. (2006), while the green short dashed line defines an
approximategreenvalleyregion,bothlines arecalculatedat red-
shift ∼ 0.76, which is the average redshift of the main Type-2
sample.
nificantly different interpretations for this observational result:
the observed red colors are mainly due to dust extinction, so
that a significant fraction of obscured AGN would live in mas-
sive, dusty star-forming galaxies with red optical colors (e.g.,
Brusa et al. 2009); or red sources are linked with passive sys-
tems (e.g., Rovilos & Georgantopoulos 2007; Schawinski et al.
2009; Cardamone et al. 2010). Therefore, accurate stellar mass
and SFR estimates, together with detailed galaxy morphologies,
are of particular importance to discriminate between the two al-
ternative possibilities.
The very high resolution and sensitivity of ACS-HST imag-
ing in the COSMOS survey provides resolved morphologies
for several hundreds of thousands galaxies with iacs ≤ 24 (see
Scarlata et al 2007 for details). Galaxy morphologies were ob-
tained with an upgraded version of the Zurich Estimator of
Structural Types (ZEST; Scarlata et al. 2007), known as ZEST+
(Carollo et al. 2011, in prep). Relative to its predecessor,
ZEST+ includes additional measurements of non-parametric
morphological indices for characterising both structures and
substructures. For consistency with the earlier versions, ZEST+
uses a Principal Component Analysis (PCA) scheme in the
6-dimensional space of concentration, asymmetry, clumpiness,
M20(second-order moment of the brightest 20% of galaxy pix-
els), Gini coefficient, and ellipticity. ZEST+ classifies galax-
ies in seven morphological types located in specific regions
of the 6-dimensional space: elliptical, S0, bulge-dominated
disk, intermediate-bulge disk, disk-dominated, irregular, com-
pact. The different types were then visually inspected. For 19
objects ZEST+ is unable to give any information on morphol-
ogy because these sources lie off the edge of the ACS tiles and
4 sources are blended. As a result of the ZEST+ procedure and
visual inspectionofthe other233galaxiesin oursample,we find
that 16 are ellipticals (Ell), 53 are S0s, 74 are bulge-dominated
(BD) disks, 27 are intermediate-bulge (IB) disks, just 1 is disk-
dominated (DD), 19 are irregular galaxies (Irr), 15 are compact
galaxies(i.e. the structural parameterscomputedforthese galax-
ies from the HST-ACS images are highly affected by the in-
strumental PSF) and 18 are unresolved compact galaxies (UC,
i.e. essentially point-like sources). Ten galaxies show distortions
and potential signatures of ongoing or recent mergers (PM). At
the typical magnitudes of the objects in our sample, the ZEST+
classification is highly reliable for galaxies with redshift ? 1.
At higher redshifts, morphological k-correction and, to a lesser
extent, resolution effects can adversely affect measurements of
ZEST+ parameters (note that only 4% of the main Type-2 AGN
sample have z > 1.5). However, broad morphological bins (e.g.,
early/late type galaxies) should be relatively robust. For high-z
galaxies, resolution might also have an impact on the classifica-
tion for mergers, and ACS images could not be deep enough
to distinguish merger features (see also Mainieri et al. 2011).
Moreover,inclinationmightalso affectthemorphologicalclassi-
fication (e.g., Nair & Abraham 2010). However, a detailed study
of systematics and biases in the morphological classification is
beyond the purposes of the present paper. In Table 1 we list the
main properties of the sample.
The rest-frame (U −V) color encompasses the 4000Å break,
and it is particularly sentitive to age and metallicity variations
of stellar population in galaxies (e.g., Sandage & Visvanathan
1978; Bell et al. 2004; Borch et al. 2006; Cardamone et al.
2010).InFig. 12the distributionoftherest-frame(U−V)colors,
which are computed directly from the best-fit galaxy template,
and stellar masses (from the SED-fitting code) are reported for
the entire Type-2 AGN sample. In the same figure, the back-
ground contours for a sample of ∼ 8700 galaxies in zCOSMOS
(iacs < 22.5, 240 Type-2 are detected in the iacsband, 183/240
Type-2 AGN (76%) have iacs < 22.5) are also plotted, where
colours and stellar masses are computed using the Hyperz code
(Bolzonella et al. 2000).
AGN are known to reside in massive galaxies (e.g.,
Silverman et al. 2009; Brusa et al. 2009) and this is fully con-
firmed by the present analysis. The morphologies of the host-
galaxies and the stellar masses indicate that there is a prefer-
ence for these Type-2 AGN to be hosted in bulge-dominatedand
S0 galaxies (∼ 50%) with stellar masses greater than 1010M⊙.
This result is consistent with the previous studies on Type-2
AGN by Silverman et al. (2008, see also Kauffmann et al. 2003;
Bundy et al. 2008; Schawinski et al. 2011).
It should be noted that no correction for the internal extinc-
tion has been applied to the (U − V) colors of both background
galaxies in zCOSMOS and Type-2 AGN hosts. This correction
couldbeimportantas shownin Cowie & Barger(2008) (see also
Cardamone et al. 2010). In that work star formation and galac-
tic stellar mass assembly are analyzed using a very large and
highly spectroscopically complete sample selected in the rest-
frame NIR bolometric flux in the GOODS-N. They found that
applyingextinctioncorrectionsis critical whenanalyzinggalaxy
colors; nearly all of the galaxies in the green valley are 24µm
sources, but after correcting for extinction, the bulk of the 24µm
sources lie in the blue cloud. This correction introduces an av-
erage shift in color of ∼ 0.2 mag for the most extincted/star-

Page 13

E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey13
forming galaxies. However, to be consistent with the colors of
the background galaxies no correction for intrinsic extinction is
considered.
AGN host-galaxies belong to the red-sequence if their (U −
V) color is above the threshold (Borch et al. 2006):
(U − V)AB,rest−frame> 0.277Log M∗− 0.352z− 0.39(5)
Sources in the green-valley are approximately defined shifting
this relation by 0.25 downward towards bluer colors. With these
definitions, ∼ 42% (108/255) and ∼ 25% (63/255) of the to-
tal sample are included in the red-cloud and the green-valley,
respectively. For all sources the Specific Star-Formation Rate
(SSFR) is estimated, defined as the ratio of the SFR per unit of
galaxy stellar mass (SSFR=SFR/M∗). The inverse of the SSFR,
SSFR−1, is called “growth time”
SSFR−1= M∗/˙ M∗,
(6)
and corresponds to the time required for the galaxy to double its
stellar mass, assuming its SFR remained constant. Actively star-
forminggalaxiesaredefinedas sourceswithgrowthtimesmaller
than the age of the Universe at their redshift (SSFR−1< tHubble),
while sources with SSFR−1larger than the age of the Universe
can be consideredpassive galaxies (see also Fontana et al. 2009;
Brusa et al. 2009). Figure 13 shows SSFR−1as a functionof the
stellar mass in three different redshift bins for the AGN host-
galaxies in the red-sequence,in the green-valleyand in the blue-
coud and for the zCOSMOS galaxies in same redshift ranges.
The horizontal lines mark the age of the Universe at the two
redshift boundaries of the chosen intervals. At face value, al-
most all the sources in the red-sequence have SSFR−1larger
than the age of the Universe at their redshift, which is consis-
tent with passive galaxies. However, the value of SSFR−1has
to be considered only as an approximate indication of the star-
formation activity; in fact, there is some possible evidence of
some residual star-formation, in red-cloud AGN host-galaxies,
as witnessed by their morphologies. In the red-sequence 8 and
28 sources are classified as ellipticals and S0s, respectively; all
together they represent 34% of the host-galaxy population in
the red-sequence. About 42% is represented by disk galaxies
(both bulge-dominated and intemediate-bulge), which are prob-
ably still forming stars but not at high rates. In fact, 15 over 108
sources (∼ 14%) have a detection at 70µm and 5 have also a
detection at 160µm (∼ 6%).
For these objects, the SFR inferred from the far-infrared de-
tections is significantly higher than the SFR derived from the
SED-fitting procedure. Indeed, an SED-fitting over the UV, op-
tical and near-infrared bands is not always able to discriminate
between the red continua of passive galaxies and those of dusty
star-forming galaxies. Therefore, we decided to include another
indicatorin the present analysis broadlyfollowing the procedure
described in Cardamone et al. (2010, i.e., the (U − V) − (V − J)
color diagram). Near-infrared emission can distinguish between
red-passive or dust-obscured galaxies: given a similar 0.5µm
flux, a star-forming galaxy has more emission near ∼ 1µm than
a passive galaxy. A sub-sample of galaxies is selected in the
same redshift range explored by Cardamone et al. (2010), we
find 92 AGN host-galaxies with 0.8 ≤ z ≤ 1.2. Fig. 14 shows
both inactive galaxies and AGN host-galaxies in the same red-
shift range and the thresholds considered to divide galaxies in
the red-sequence and in the green-valley (we consider an aver-
age redshift of 1 to define the threshold for the red-sequence
and the green-valley).Thirty-fiveout of 92 AGN hosts are found
to lie in the red-sequence (∼ 38%) and 23 in the green-valley
Fig.13. Inverseof the SSFR rate as a functionof the stellar mass
of the AGN host-galaxies in three different redshift bins for the
zCOSMOS galaxies and for the Type-2 AGN sample in the red-
sequence (red crosses), in the green-valley (green triangles) and
in the blue-cloud (blue open circles). The horizontal lines mark
the age of the Universe at the two redshift boundariesof the cho-
sen intervals.
(∼ 25%); while for inactive galaxies about 32% and 21% lie in
the red-sequenceand green-valley,respectively.In Figure 15 the
(U −V)−(V − J) color diagram for the 92 Type-2 AGN hosts is
presented. From a preliminary analysis of the rest frame (U −V)
against the rest-frame (V − J) color (see Fig. 15, but see also
Fig. 2 in Cardamone et al. 2010), only ∼ 9% of the AGN host-
galaxiesinthered-sequenceand∼ 30%ofAGNhost-galaxiesin
the green-valleyaremovedin theregionpopulatedbydusty star-
forming galaxies in the color-color diagram. To be compared
with 20% AGN host-galaxiesin the red-sequenceand75% AGN
host-galaxies in the green-valley found by Cardamone and col-
laborators. The fractions of dust-obscured galaxies among the
red-cloudand green-valleyAGN in our sample, at 0.8 ≤ z ≤ 1.2,
are lower than those in the Cardamone et al. (2010) sample.
However, the global fractions of AGN hosts, tentatively asso-
ciated to passive galaxies, are very similar (∼ 50%) in the two
samples. The fractions for both AGN host-galaxies and inactive
galaxies are reported in Table 2.
9. Summary and Conclusions
A detailed analysis of the SEDs of 255 spectroscopically identi-
fiedhardX–rayselectedType-2AGNfromtheXMM-COSMOS
survey is presented. In obscured AGN, the optical-UV nuclear
luminosity is intercepted along the line of sight by the dusty
torus and reprocessed in the infrared, so what we see in the
optical-UV is mostly the light from the host-galaxy. On the one
hand, this allows us to study the galaxy properties, on the other
hand it makes difficult to estimate the nuclear bolometric power.
An SED-fitting code has been developed with the main purpose
of disentagling the various contributions (starburst, AGN, host-
galaxy emission) in the observedSEDs using a standard χ2min-
imization procedure(the starburst component is only used in the
case of detection at 70µm). The code is based on a large set of

Page 14

14 E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey
Table 2. AGN hosts and galaxies properties.
SampleNRed-sequenceGreen-valleyBlue-cloud
0.045 ≤ z ≤ 3.452
96 (89%) P
12 (11%) D
2306 (91%) P
229 (9%) D
52 (82%) P
11 (18%) D
1596 (83%) P
327 (17%) D
Type-2 AGN255108 (42%)63 (25%) 84 (33%)
Galaxies 87422535 (29%) 1923 (22%)4284 (49%)
0.8 ≤ z ≤ 1.2
32 (91%) P
3 (9%) D
569 (97%) P
18 (3%) D
16 (70%) P
6 (30%) D
269 (70%) P
116 (30%) D
Type-2 AGN 9235 (38%)23 (25%) 34 (37%)
Galaxies 1836587 (32%)385 (21%) 864 (47%)
Note – P=Passive, D=Dusty.
Fig.14. Distribution of the stellar masses as a function of the
rest-frame (U − V) colors in the redshift range 0.8 ≤ z ≤ 1.2.
The red dashed line represents the red sequence cut defined by
Borch et al. (2006), while the green short dashed line defines an
approximategreenvalleyregion,bothlines arecalculatedat red-
shift ∼ 1. The points are color coded as in Fig. 12.
starburst templates from Chary & Elbaz (2001) and Dale et al.
(2001),andgalaxytemplates fromthe Bruzual & Charlot(2003)
code for spectral synthesis models, while AGN templates are
taken from Silva et al. (2004). These templates represent a wide
range of SED shapes and luminosities and are widely used in
the literature. The total (nuclear) AGN bolometric luminosities
are then estimated by adding the X–ray luminosities integrated
overthe 0.5-100keV energyrangeto the infrared luminositybe-
tween 1 and 1000µm. The total X–ray luminosity is computed
integrating the X–ray SED using the de-absorbed soft and hard
X–ray luminosities. The SED is extrapolated to higher energies
using the observed X–ray slope, and introducing an exponential
cut-off at 200 keV. The total infrared luminosity is evaluated in-
Fig.15. Distribution of Type-2 AGN hosts in the rest-frame
(U − V) against the rest-frame (V − J) color. Color coded as
in Fig. 13. Sources with the same best-fit galaxy template and
the same extinction lie in the same position in the color-color
diagram. Point size is keyed to the number of objects.
tegrating the infrared AGN best-fit and then converted into the
nuclear accretion disk luminosity applying the appropriate cor-
rection factors to account for the geometry and the anisotropy of
the torus emission. The reprocessed IR emission is considered
to be a good proxy of the intrinsic disk emission and this is sup-
ported by previous investigations (Pozzi et al. 2007; Gandhi et
al. 2009; Vasudevan et al. 2010). In the distribution of the ratio
r = Log
?
the objects are within 2σ of the r distribution. The tail outside
2σ and extending to high r includes 73 sources (with r ? 0.5)
for which the predicted mid-infrared luminosity is significantly
lower than observed. We call “low-r” AGN all sources within
2σ of the r distribution, while the “high-r” AGN sample is rep-
resented by the sources deviating more than 2 σ.
L12.3 µm,obs/L12.3 µm,predicted; see Eq. 4) the majority of

Page 15

E. Lusso et al: The bolometric output and host-galaxy properties of obscured AGN in the XMM-COSMOS survey 15
Our main results are the following:
1. The average observed SED is characterized by a flat X–ray
slope, ?Γ = 1.12?, as expected for obscured AGN (not cor-
rected for absorption), while in the optical-UV the observed
light appears to be consistent with the host-galaxy emission.
The average SED in the mid-infraredis more likely a combi-
nation of dust emission from star-forming region and AGN
emission reprocessed by the dust.
2. The full sample is split into four bins of different X–ray and
infrared luminosities and redshift. The shapes of the average
SEDs in the optical bands are approximately the same in all
luminosity and redshift bins. There is a stronger host-galaxy
contribution at lower luminosity/redshift bins, where the av-
erage SEDs have a typical galaxyshape. Moreover,there is a
trend between X–ray and mid-infrared luminosity: the con-
tribution of the AGN in the infrared (around 8 − 15µm) is
higher at higher X–ray luminosities.
3. Type-2 AGN appear to have smaller bolometric corrections
than Type-1 AGN. At the same hard X–ray luminosity,
43.30 ≤ Log L[2−10]keV≤ 44.30,wherebothsamplesarewell
represented, we find that the median bolometric correction
for Type-2 AGN (134 objects) is ?kbol? ∼ 13 ± 1, to be com-
pared with a medianbolometriccorrection?kbol? ∼ 23±1for
Type-1AGN(167objects).Thetwo averagesarestatistically
different at the ∼ 7 σ level.
4. A clear separation in bolometric corrections for the low-r
and the high-r samples is found. The relation provided by
Gandhi and collaborators is valid for the majority of objects,
while for 30% of the sample SED-fitting procedure may un-
derestimate the non-nuclearcontribution.At a givenhard X–
ray luminosity (43 ≤ Log L[2−10]keV≤ 44) the low-r sample
has a median bolometric correction of ?kbol? ∼ 11 ± 1 (110
objects), to be compared with a median bolometric correc-
tion for the high-r sample of ?kbol? ∼ 26 ± 3 (44 objects).
The two median values for kbolare statistically different at
the ∼ 5σ level.
5. Host-galaxies morphologies and the stellar masses indicate
that Type-2 AGN are preferentially hosted in galaxies which
have a bulge,irrespectiveofthe strengthofthe bulgeor if the
galaxy is on the red sequence or blue cloud, and with stellar
masses greater than 1010M⊙.
6. Almost all the sources in the red-sequence have SSFR−1
larger than the age of the Universe at their redshift, which
is consistent with passive galaxies. Following the same ap-
proach as in Cardamone and collaborators (i.e., combining
the rest-frame (U −V) vs Log M∗and the rest-frame (U −V)
vs (V−J) color diagrams),we find that, consistent with their
results, ∼ 50% of AGN hosts lie in the passive region of this
diagram. In contrast from Cardamone et al. (2010), only ∼
30% of AGN host-galaxies in the green-valleyin our sample
are consistent with dust-obscured sources in 0.8 ≤ z ≤ 1.2.
It is clear that the mid and far-infrared parts of the SED are
under-sampled with respect to the optical part. The ongoing
Herschel survey over various fields at different depths (100µm
and160µmintheCOSMOS field)andtheupcomingALMAsur-
veys will allow us to gain an optimal multiwavelength coverage
also in the far-infrared.
Acknowledgements. We gratefully thank B. Simmons for her useful comments
and suggestions. In Italy, the XMM-COSMOS project is supported by ASI-
INAFgrants I/009/10/0, I/088/06 and ASI/COFIS/WP3110 I/026/07/0. Elisabeta
Lusso gratefully acknowledges financial support from the Marco Polo program,
University of Bologna. In Germany the XMM-Newton project is supported by
the Bundesministerium f¨ ur Wirtshaft und Techologie/Deutsches Zentrum f¨ ur
Luft und Raumfahrt and the Max-Planck society. Support for the work of E.T.
was provided by NASA through Chandra Postdoctoral Fellowship Award grant
number PF8-90055, issued by the Chandra X-ray Observatory Center, which is
operated by the Smithsonian Astrophysical Observatory for and on behalf of
NASA under contract NAS8-03060. The entire COSMOS collaboration is grate-
fully acknowledged.
References
Antonucci, R. 1993, ARA&A, 31, 473
Bell, E. F., Wolf, C., Meisenheimer, K., et al. 2004, ApJ, 608, 752
Bolzonella, M., Miralles, J., & Pell´ o, R. 2000, A&A, 363, 476
Borch, A., Meisenheimer, K., Bell, E. F., et al. 2006, A&A, 453, 869
Brusa, M., Civano, F., Comastri, A., et al. 2010, ApJ, 716, 348
Brusa, M., Fiore, F., Santini, P., et al. 2009, A&A, 507, 1277
Bruzual, G. & Charlot, S. 2003, MNRAS, 344, 1000
Bundy, K., Georgakakis, A., Nandra, K., et al. 2008, ApJ, 681, 931
Calzetti, D., Armus, L., Bohlin, R. C., et al. 2000, ApJ, 533, 682
Capak, P., Aussel, H., Ajiki, M., et al. 2008, VizieR Online Data Catalog, 2284,
0
Capak, P., Aussel, H., Ajiki, M., et al. 2007, ApJS, 172, 99
Cappelluti, N., Brusa, M., Hasinger, G., et al. 2009, A&A, 497, 635
Cardamone, C. N., Urry, C. M., Schawinski, K., et al. 2010, ApJ, 721, L38
Chabrier, G. 2003, ApJ, 586, L133
Chary, R. & Elbaz, D. 2001, ApJ, 556, 562
Ciotti, L., Ostriker, J. P., & Proga, D. 2010, ApJ, 717, 708
Corbin, M. R. 2000, ApJ, 536, L73
Cowie, L. L. & Barger, A. J. 2008, ApJ, 686, 72
Dale, D. A. & Helou, G. 2002, ApJ, 576, 159
Dale, D. A., Helou, G., Contursi, A., Silbermann, N. A., & Kolhatkar, S. 2001,
ApJ, 549, 215
Dekel, A., Sari, R., & Ceverino, D. 2009, ApJ, 703, 785
Elitzur, M. & Shlosman, I. 2006, ApJ, 648, L101
Elvis, M., Wilkes, B. J., McDowell, J. C., et al. 1994, ApJS, 95, 1
Emmering, R. T., Blandford, R. D., & Shlosman, I. 1992, ApJ, 385, 460
Ferrarese, L. & Merritt, D. 2000, ApJ, 539, L9
Fontana, A., Santini, P., Grazian, A., et al. 2009, A&A, 501, 15
Gandhi, P., Horst, H., Smette, A., et al. 2009, A&A, 502, 457
Gilli, R., Comastri, A., & Hasinger, G. 2007, A&A, 463, 79
Granato, G. L. & Danese, L. 1994, MNRAS, 268, 235
Hao, H., Elvis, M., Civano, F., et al. 2010, ApJ, 724, L59
Hasinger, G. 2008, A&A, 490, 905
Hasinger, G., Cappelluti, N., Brunner, H., et al. 2007, ApJS, 172, 29
Hoeft, M. & Br¨ uggen, M. 2004, ApJ, 617, 896
Hopkins, P. F., Hernquist, L., Cox, T. J., et al. 2006, ApJS, 163, 1
Hopkins, P. F., Hernquist, L., Cox, T. J., & Kereˇ s, D. 2008, ApJS, 175, 356
Hopkins, P. F., Richards, G. T., & Hernquist, L. 2007, ApJ, 654, 731
Hornschemeier, A. E., Brandt, W. N., Garmire, G. P., et al. 2001, ApJ, 554, 742
Jiang, L., Fan, X., Brandt, W. N., et al. 2010, Nature, 464, 380
Kauffmann, G. & Haehnelt, M. 2000, MNRAS, 311, 576
Kauffmann, G., Heckman, T. M., Tremonti, C., et al. 2003, MNRAS, 346, 1055
Kelly, B. C., Bechtold, J., Trump, J. R., Vestergaard, M., & Siemiginowska, A.
2008, ApJS, 176, 355
Kereˇ s, D., Katz, N., Fardal, M., Dav´ e, R., & Weinberg, D. H. 2009, MNRAS,
395, 160
Koekemoer, A. M., Aussel, H., Calzetti, D., et al. 2007, ApJS, 172, 196
Komatsu, E., Dunkley, J., Nolta, M. R., et al. 2009, ApJS, 180, 330
Kormendy, J. & Richstone, D. 1995, ARA&A, 33, 581
Le Floc’h, E., Aussel, H., Ilbert, O., et al. 2009, ApJ, 703, 222
Lilly, S. J., Le F` evre, O., Renzini, A., et al. 2007, ApJS, 172, 70
Lusso, E. & Ciotti, L. 2011, A&A, 525, A115+
Lusso, E., Comastri, A., Vignali, C., et al. 2010, A&A, 512, A34
Lutz, D., Maiolino, R., Spoon, H. W. W., & Moorwood, A. F. M. 2004, A&A,
418, 465
Magorrian, J., Tremaine, S., Richstone, D., et al. 1998, AJ, 115, 2285
Mainieri, V., Bongiorno, A., Merloni, A., et al. 2011, ArXiv e-prints
Mainieri, V., Hasinger, G., Cappelluti, N., et al. 2007, ApJS, 172, 368
Maiolino, R., Marconi, A., & Oliva, E. 2001, A&A, 365, 37
Marconi, A. & Hunt, L. K. 2003, ApJ, 589, L21
Marconi, A., Risaliti, G., Gilli, R., et al. 2004, MNRAS, 351, 169
McCracken, H. J., Radovich, M., Iovino, A., et al. 2008, VizieR Online Data
Catalog, 2286, 0
Menci, N., Fontana, A., Giallongo, E., & Salimbeni, S. 2005, ApJ, 632, 49
Nair, P. B. & Abraham, R. G. 2010, ApJS, 186, 427
Nandra, K., Georgakakis, A., Willmer, C. N. A., et al. 2007, ApJ, 660, L11
Nenkova, M., Sirocky, M. M., Nikutta, R.,ˇZ. Ivezi´ c, & Elitzur, M. 2008, ApJ,
685, 160