Skip to main content

Full text of ""Revealing a Population of Heavily Obscured Active Galactic Nuclei at z=0.5-1 in the Chandra Deep Field-South"

See other formats


Draft version July 19, 201 1 

Preprint typeset using I4TgX style emulateapj v. 1 1/10/09 



REVEALING A POPULATION OF HEAVILY OBSCURED ACTIVE GALACTIC NUCLEI AT Z s 

CHANDRA DEEP FIELD-SOUTH 



; 0.5-1 IN THE 



o 
in 



O 
U 

6 



> 

oo 

m 

o 



s 



B. Luo 



1X3 W. N. Brandt,'- 2 Y. Q. Xue,'- 2 D. M. Alexander, 4 M. Brusa, 5 F. E. Bauer, 6 A. Comastri, 7 A. C. Fabian, 8 



R. GlLLI, 7 B. D. LEHMER, 



' D. A. Rafferty," D. P. Schneider, 1 & C. Vignali 1 

Draft version July 19, 2011 



ABSTRACT 

Heavily obscured (Nh > 3 x 10 23 cm" 2 ) Active Galactic Nuclei (AGNs) not detected in even the deepest 
X-ray surveys are often considered to be comparably numerous to the unobscured and moderately obscured 
AGNs. Such sources are required to fit the cosmic X-ray background (XRB) emission in the 10-30 keV band. 
We identify a numerically significant population of heavily obscured AGNs at z ~ 0.5-1 in the Chandra Deep 
Field-South (CDF-S) and Extended Chandra Deep Field-South by selecting 242 X-ray undetected objects 
with infrared-based star formation rates (SFRs) substantially higher (a factor of 3.2 or more) than their SFRs 
determined from the UV after correcting for dust extinction. An X-ray stacking analysis of 23 candidates in 
the central CDF-S region using the 4 Ms Chandra data reveals a hard X-ray signal with an effective power-law 
photon index of T = 0.6^JJ 4, indicating a significant contribution from obscured AGNs. Based on Monte Carlo 
simulations, we conclude that 74 ± 25% of the selected galaxies host obscured AGNs, within which « 95% are 
heavily obscured and sw 80% are Compton-thick (CT; Nh > 1.5 x 10 24 cm" 2 ). The heavily obscured objects 
in our sample are of moderate intrinsic X-ray luminosity [w (0.9^-) x 10 42 erg s" 1 in the 2-10 keV band]. 
The space density of the CT AGNs is (1.6 ± 0.5) x 10" 4 Mpc" 3 . The z « 0.5-1 CT objects studied here are 
expected to contribute w 1 % of the total XRB flux in the 10-30 keV band, and they account for s=s 5-15% of the 
emission in this energy band expected from all CT AGNs according to population-synthesis models. In the 6-8 
keV band, the stacked signal of the 23 heavily obscured candidates accounts for < 5% of the unresolved XRB 
flux, while the unresolved w 25% of the XRB in this band can probably be explained by a stacking analysis of 
the X-ray undetected optical galaxies in the CDF-S (a 2.5 a stacked signal). We discuss prospects to identify 
such heavily obscured objects using future hard X-ray observatories. 

Subject headings: Cosmology: cosmic background radiation — Galaxies: active — Galaxies: photometry — 
Galaxies: starburst — Infrared: galaxies — X-rays: galaxies 



1. INTRODUCTION 

Deep X-ray surveys have provided the most effective 
method of identifying reliable and fairly complete sam- 
ples of_A£tiw_Galactic Nuclei (AGNs) out to z ~ 5 (e.g., 
see iBrandt & Hasingerl 120051 for a review). The observed 
AGN sky density reaches « 10000 deg" 2 in the deepest 
X-ray surveys, the Cha ndra Deep Fields (e.g., iBauer etail 
2004; Xue et all 1201 lh . These X-ray point sources are 
largely responsible for the observed cosmic X-ray back- 
ground (XRB). A significant portion (rj 70-90%) of the 

1 Department of Astronomy & Astrophysics, 525 Davey Lab, The Penn- 
sylvania State University, University Park, PA 16802, USA 

2 Institute for Gravitation and the Cosmos, The Pennsylvania State Uni- 
versity, University Park, PA 16802, USA 

3 Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, 
Cambridge, MA 02138, USA 

4 Department of Physics, Durham University, Durham, DH1 3LE, UK 

5 Max-Planck-Institut fur Extraterrestrische Physik, Giessenbach- 
strasse, D-85748 Garching b. Munchen, Germany 

6 Pontificia Universidad Catolica de Chile, Departamento de As- 
tronomia y Astrofisica, Casilla 306, Santiago 22, Chile 

7 INAF — Osservatorio Astronomico di Bologna, Via Ranzani 1, 
Bologna, Italy 

8 Institute of Astronomy, Madingley Road, Cambridge, CB3 0HA, UK 

9 The Johns Hopkins University, Homewood Campus, Baltimore, MD 
21218, USA 

10 NASA Goddard Space Flight Centre, Code 662, Greenbelt, MD 
20771, USA 

11 Leiden Observatory, Leiden University, Oort Gebouw, P.O. Box 9513 
RA, Leiden, The Netherlands 

11 Universita di Bologna, Via Ranzani 1, Bologna, Italy 



XRB in the 0.5-6 keV range has been resolved into discrete 
sources by Chandra and XMM-Newton (e.g ., Moretti et al. 
l2002tlBauer et al.l2004tlHickox & Mark evitch 2006), the ma- 
jority of which are AGNs with moderate X-ray obscuration 
(N» < 3 x 10 23 cm" 2 ; e.g., ISzokoly et al]l2004l iBarger et all 
l2005l;lTozzi et al.l2006l) . However, the resolved fraction of the 
XRB decreases toward higher energies, being sa 60% in the 6— 
8 keV band and « 50% in the 8-12 keV band (IWorslev et all 
2004, 2005). Population-synthesis models suggest that at 
S3 20-30 key, where the XRB reache s its peak flux (e.g., 
iGruber et aH Il999h iMoretti et all 120091) . the unobscured or 
moderately obscured AGNs discovered at lower energies can- 
not account for the XRB flux entirely, and an additional 
population of heavily obscured (Nh >3x 10 23 cm" 2 ), or 
even Compton-thick (% > 1.5 x 10 24 cm" 2 , hereafter CT), 
AGNs at mainly z ~ 0-1.5 is required. The number den- 
sity of these heavily obscured AGNs is estimated to be of 
the same order as that of moderately obscured AGNs (e.g., 
Gilli et al. 2007). However, as the X-ray emission (below 
at least 10 keV) of such sources is significantly suppressed, 
only a few distant heavily obscured AGNs have been clear! 
identified even in the de epest X-ray surveys (e.g..lTozzi et al 
2006; Alexander et al. 2008; Georgantopoulos et al. 2009 
Comastri et al. 201 1; Feruglio et al. 201 1), and thus a signifi- 
cant fraction of the AGN population probably remains unde- 
tected. 

Infrared (IR) selection is a powerful tool for detecting heav- 
ily obscured AGNs that cannot be identified in the X-ray band. 
X-ray and UV emission absorbed by the obscuring material 



is re processed and reemitted mainly at mid-IR wavelengths 
(e.g„ ISilva etal]|2004t IPrieto et alJIJOlOl) . The mid-IR emis- 
sion is less affected by dust extinction than at optical or near- 
IR wavelengths, rendering it more suitab le for identifying ob - 
scured AGNs. With the Spitzer mission dWerner et al.l l2004), 
deep mid-IR data in multiple bands can be obtained by the 
Infrared Array Camera (IRAC) and Multiband Imaging Pho- 
tometer (MIPS) instruments. 

Various IR selection methods for obscured AGNs have 
been proposed. Spitzer IRAC power-law selection chooses 
sources whose IRAC spectral energy distributions (SEDs) fol- 
low a power law with a slope of a < -0.5 (f v oc v" ; e.g., 
lAlonso-Herrero et alj|2006t IDonlev et al] 120071 l20iol) . This 
criterion was determined based on the average o ptical-to-IR 
spectral properties of X-ray selected QSOs (e.g. jElvis et alj 
1994). It can select a relatively pure sample of AGNs with 
little contamination from star-forming galaxies b ut is gener- 
ally li mited to the most-luminous sources (e.g., Polletta et al. 
2008). IRAC color-color selection applies color cuts in the 
IRAC color spac e to identify AGNs (e.g.. lLacv et al.1 l2004t 
IStern et alj|2005l) . This method is based on the IRAC color 
properties of optically selected QSOs and AGNs, and it tends 
to have significant contamination from star-forming galax- 
ies when applied to deep mid-IR da t a (e.g.,|C ardamon e et alJ 
120081: iGeorgantopoulos et al.ll2008t iBrusa et alj|2010l) . IR- 
excess selection uses a combined UV-optical-IR color cut 
to identify AGNs ( e.g., iDev et alj l2008t iFiore et al.1 12001 
iLanzuisi et al. 2009). This techniq ue is also c ontaminated by 
star-forming galaxies (e.g. JDonley et al.ll2008h . 

Compared to the above methods, which utilize observa- 
tional data directly and de termine the selection criteria em- 
pirically, Daddi etaTI d2007l) adopted a generally different ap- 
proach to find obscured AGNs. They selected candidates with 
significantly higher IR-based (also including the component 
corresponding to the transmitted UV emission; see §2. 1 below 
for details) star formation rates (SFRs) than their UV-based 
(after correcting for dust extinction) SFRs. The amount of ex- 
cess in the IR-based SFR is a measurement of the excess in the 
IR emission, which is likely caused by reprocessed emission 
from the dusty torus of a heavily obscured AGN. This relative 
IR SFR excess (ISX) selection method requires and utilizes 
redshift information to measure the IR excess effectively, and 
it appe ars to isolate a different IR sample from t he methods 
above (IDonlev et alj|2008t lAlexander et al.l l2008). A few of 
the candidates in lDaddi et al. (2007) have been reliably iden- 
tified as CT AGNs ( lAlexander et alj|2008l) . The ISX sample 
is also contaminated by star-forming galaxies due to uncer- 
tainties in the c alculation of SFRs; the heavily obscured AGN 
fraction in the iDaddi et al] d2007l) sample is estimated to be 
« 25% ( lAlexander et alj|201 ll) . and the rest of the objects are 
likely star-forming galaxies. 

In this paper, we improve the ISX selection method 
and utilize it to search for X-ray undetected, heavily ob- 
scured AG Ns at z ~ 0-5-1 in th e Chandra Deep Field-South 
(CDF -S; iGiacconi et ail l200l iLuo et alJ 120081: iXueetalJ 
20111) and Extende d Chandra Deep Field-South (E-CDF-S; 
Lehmer et al. 2005). We choose to apply the ISX method on 



the CDF-S and E-CDF-S galaxy sample in this redshift range 
for the following reasons: 

• IR-selection of heavily obscured AGNs at z ~ 0.5-1 
is a poorly explored territory. Most of the previous 
IR selectio ns were focused on samples at z ~ 2, in- 
cluding the IDaddi et al] d2007l) work. Since AGNs at 



Z « .5-1 contribute significantly to the res olved XRB 
(e.g., IWorslev et allHool iGilli et alj|2007l) . and heav- 
ily obscured AGNs in this redshift range are expected 
to play a s imilar role in the missing fraction of the 
XRB (e.g.. IGilli et all 120071: ITreTster et alJl2009l) . it is 
of interest to study the h eavily o bscured population 
at these lower redshifts. IFiore et al] (2009) selected 
a sample of IR-excess sources in the COSMOS field, 
which included some objects at z < 1. However, de- 
tailed studies (e.g., the AGN fraction) were not fo- 
cused on this low-redshift bin. Moreover, the X-ray 
and IR (24 ^m) flux limits in the CDF-S field are 
> 10 times more sensitive than those in the COSMOS 
field dFiore et al] 120091) . allowing identification of the 
population of heavily obscured AGNs with lower in- 
trinsic lu minosities (which a r e likel y more numerous). 
Recently, Georgakakis et al. (2010) selected a sample 
of 19 IR-excess sources at z ~ 1 in the AEGIS and 
GOODS-N fields, but they argued that most of those 
X-ray undetected candidates are not heavily obscured 
AGNs based on IR SED modeling. 

• The ISX method is best ap plicable at z : ~ .5-1. 
The ISX method proposed by Daddi et al. (2007) is a 
promising technique because of its simple physical mo- 
tivation and its success in selecting identified CT AGNs 
at z ~ 2. However, the ISX method has not been ap- 
plied to any other studies, partially due to its known 
limitations. There are two essential quantities in the 
ISX method, the IR-based SFR (SFR IR+UV ) and the UV- 
based SFR (SFRuvxoir)- The IR-based SFR is derived 
mainly from the observed Sp itzer MIPS 24 /xm flux us- 
ing the lCharv & Elbazld2001l) galaxy SED templates. It 
has been noted that at z > 1 .4, the IR luminosity (and 
thus SFRir+uv) derived this way is ov erestimated by 
an average factor of « 5 dMurphv e t al. 2009, 201 1)0 
leading to additional contamination from galaxies in the 
ISX sample. The UV-based SFR is calculated using the 
dust-extinction corrected UV luminosity. Daddi et al. 
(2007) used an empirical color-extinction relation to es- 
timate the dust extinction. However, this correlation 
cannot be applied to relatively old galaxies that are in- 
trinsically red (dominated by old stellar populations in- 
stead of being reddened ), and a ll thes e galaxies were 
removed from the Dad di et al.l ([2007 ) sample^ As 
AGNs tend to be hosted by massive, red galaxies (e.g., 
Xue et al. 2010), this limitation might significantly af- 
fect the completeness of the resulting ISX sample. Both 
these problems can be substantially alleviated if the ISX 
method is applied at lower redshifts Th e 24 /xm de- 
duced IR luminosity is robust at z < 1 (e.g., Elbaz et al. 
2010), while t he dust extinction can be derived via SED 
fitting (e.g.. iBrammer et al.1 120091; iReddv et al.ll20Tob . 

13 Thi s is largely due t o the fact that the local high-luminosity SED tem- 
plates in Chary & Elbaz 12001) do not accurately account for the aromatic 
features around 24 fim and the IR SEDs at z > 1 -4; they show weaker PAH 
emission than that in high-redshift galaxies by an average factor of pb 5. 

14 Daddi et al. (2007) removed galaxies with SSFR<median(SSFR)/3, 
where SSFR is the specific SFR, defined as SSFR = SFRi R+uv /M, with M» 
being the stellar mass. About 15% of the sources were excluded this way, the 
median stellar mass of which is f» 5 times larger than that of the remaining 
sources. 

15 We note that applying the ISX method at z ~ 2 has the advantage of 
being able to get a better contrast between star-formation and AGN emission 
at the observed 24 /im wavelength than at lower redshifts. 



The SED-fitting technique is reliable with the high- 
quality multiwavelength photometric data achievable at 
z<l. 

• The CDF-S and E-CDF-S are excellent fields in which 
to perform such a study. They have been covered by ex- 
tensive multiwavelength photometric and spectroscopic 
surveys, including very d eep X-ray and 24 /xm expo- 
sures. The COMBO- 17 dWolfet all 12004 120081) and 
Subaru (Cardamone etal.1 120101) surveys covering the 
entire CDF-S and E-CDF-S are particularly valuable 
for the determination of dust extinction via SED fitting. 
The 17-filter coverage of COMBO-17 between 3600 A 
and 9200 A and 18 medium-band coverage of Subaru 
between 4200 A and 8600 A are useful for distinguish- 
ing between a red dust-free galaxy and a blue dusty 
galaxy. Reliable photometric redshifts (photo-z's) can 
also be obtained with the high-quality multiwavelength 
data available. 

• We can assess the possibility of directly detecting the 
missing population of heavily obscured AGNs with fu- 
ture hard X-ray observatories. One of the science 
goals of several future hard X-ray missions, such as 
the Nuclear Sp ectroscopic Telescope Array (NuSTAR; 

2010) and ASTRO-H ( Takahashi et al. 
10-30 keV via 



Harri son et al.1 ._ _.. _. 

20101) . is to" resolve better the XRB at 



detecting heavily obscured and CT AGNs directly in the 
distant universe. While IR-selected heavily obscured 
AGN candidates at z ~ 2 are likely below their sensi- 
tivity limitsHj candidates at z ~ 0.5-1 should be more 
easily detectable, and we will critically assess such pos- 
sibilities. 

This paper is organized as follows. We describe the mul- 
tiwavelength data and ISX sample selection in §2. In §3 we 
present an X-ray stacking analysis of the ISX sample and es- 
timate the heavily obscured AGN fraction among the sample 
via simulations. In §4, we calculate the space density of the 
selected heavily obscured AGNs and their contribution to the 
XRB. We also discuss the feasibility of detecting these ISX 
sources with future hard X-ray observatories. We summarize 
in §5. Throughout this paper, we adopt the latest cosmology 
with H n = 70.4 km s" 1 Mpc -1 , fl M = 0.272, and fl A = 0.728 
dKomatsuet al.ll2011 | ). All give n magnitudes are in the AB 
system (e.g. JOke & Gunnlll983l) unless otherwise stated. 

2. THE RELATIVE IR SFR EXCESS SAMPLE 

2.1. Multiwavelength Data and Source Properties 

The CDF-S is the deepest X-ray survey ever performed, 
having a total Chandra exposure of « 4 Ms and covering a 
solid angle of sa 460 arcmin 2 dXue et all 120 lib . X-ray im- 
ages in three standard bands, 0.5-8.0 keV (full band; FB), 
0.5-2.0 keV (soft band; SB), and 2-8 keV (hard band; HB), 
along with other relevant products such as the main X-ray 
source catalog (740 sources) and sensitivity maps, are used 
in the sample creation and X-ray stacking analysis, which are 

16 Assuming an absorption column density of Nu = 10 24 cm" 2 , the 10- 
30 keV flux of the intrinsically luminous (2-10 keV luminosity of 2 X 
10 44 erg s _1 ) CT AGN HDF-oMD49 at z = 2.21 ( Alex ander et all2008D will 
still be below the detection limit of NuSTAR with a 1 Ms exposure (the NuS- 
TAR sensitivity limit is from http : //www.nusta r . caltech.edu /I. 
The average luminosity of the CT AGN candidates in Daddi et al. ( 2007) is 
about an order of magnitude fainter than that of HDF-oMD49. 



discussed below and in §3. The CDF-S is flanked by the E- 
CDF-S, which consists of four contiguous sa 250 ks Chan- 
dra observations with a total solid angle of « 1100 arcmin 2 
(Lehmeret al. 2005). The E-CDF-S main X-ray source cata- 
log (762 sources) is used only in the sample creation. 

Mid-IR-to-optical multiwavelength data are required to cal- 
culate the IR-based and UV-based SFRs in the ISX method. 
The CDF-S and E-CDF-S have been covered by extensive 
photometric and spectroscopic surveys, and we constructed a 
sample of mid-IR and optically selected sources in the CDF- 
S and E-CDF-S region. For the mid-IR data, we used the 
Spitzer MIPS 24 fim source catalog from the Spitzer Far In- 
frared Deep Extragalactic Legacy Survey (FIDEL; M. Dick- 
inson et al. 2011, in preparation). The FIDEL survey covers 
the entire CDF-S and E-CDF-S, and it has a 5 a limiting flux 
of m 20 /iJy. For the IR-to-optical data, we used the mas- 
ter source catalog compiled by iRafferty et al.l (120111) . which 
consists of m 100000 optically selected galaxies covering the 
entire CDF-S and E-CDF-S. The mast er source catal og was 
created based on the MUSYC catalog (Gawiser et al J 120061) . 
the COMBO-17 catalog (Wolfetal. 2004, 2008), and the 
GOODS-S MUSIC catalog (Grazianet al. 2006), and it was 
also cross-matched to several other photometric catalogs such 
as the SIMPLE IRAC catalog (Damen et al]|] 2011l) and the 

■ et al. 2005) 1 



GALEXUV catalog (e.g.,|Morrisseyj 



i to include up 



to 42 bands of IR-to-UV data. Additionally, we included the 
18 medium-band Subaru photometric data that have become 
available recently (Cardamone et al. 2010); about 80% of the 
sources in the master catalog have Subaru data. For our pur- 
pose of computing reliably the UV-based SFRs (requiring ro- 
bust IR-to-UV SED fitting), we chose sources with relatively 
bright /?-band magnitudes {R < 25); there are « 40000 such 
objects. These sources were matched to the 24 /xm sources 
using the likelihood-ratio matching technique described in 
ILuo et al. (2010), resulting in 5237 matches and a false-match 
probability of « 4%. For sources in the X-ray stacking sam- 
ples discussed in §3, we have visually examined their IR and 
optical images and removed sources that are probably affected 
by source blending, and thus the false-match probability is 
negligible for those samples. 

Reliable spectroscopic redshifts (spec-z's; 979/5237 
sources) for sources in the CDF-S and E-CDF-S were 
collected from the fo l lowing catalog s: |Ee Fevre et al. (2004), 
Szoko lv et al.1 d20 04), Mignoli et all (120051). iR avikum ar et all 
(2007),' Vanzella et al. " (2008), Popesso et al. (2009), 
iBalestra et all (l2010t) , and [Silverman et al.l ilOWtl . If a 



spec-z is not available for a given source, we calculated its 
photo-z using the Zurich Extragalactic Bayesian Redshift 
Analyzer (ZEBRA: iFeldmann et al1l2006l) . ZEBRA utilizes a 
maximum-likelihood approach to find the best-fit SED tem- 
plate; more details about the ZEBRA SED-fitting procedure 
are discussed in §3.2 of Luo et al. (2010). For the purpose of 
our study here, we used ZEBRA to derive photo-z's and dust 
extinction (in terms of the extinction in the V band, Ay) si- 
multaneously using up to 60 photometric band s. We selected 
99 typical PEGASE galaxy templates from Grazi an et al.l 
(2006) that span a wide range of star- formation history, 
and we applied intrinsic extinction in the range Ay = 0-4 
with an increment o f 0.1 to these templates employing the 
ICalzetti et all ([2000) extinction law. The resulting SED 
templates were used to fit the IR-to-UV SED data (excluding 
the 24 /im data) of the sources. For sources with spec-z's, 
the redshifts were fixed at the spec-z values during the SED 



fitting, and the values of Ay were obtained from the best-fit 
templates. For the other sources, both the photo-z and Ay 
values were determined based on the best-fit templates. To 
check the quality of the photo-z's, we performed another 
ZEBRA run of the spec-z sources, setting the redshift as a 
free parameter, and then compared the resulting photo-z's 
with the spec-z's. The photo-z accuracy was estimated using 
the outlier fraction and the normalized median absolute 
devia tion (ctnmad; e.g., iBrammer et al.l 120081; iLuo et al.l 
2010) parameters. Outliers are defined as sources having 



I Az|/(1 +z spec ) > 0.15, where Az = z, 
is defined as 



photo " 



<-spec 



, and Ctnmad 



ctnmad = 1 -48 x median 



j Az - median( Az) | 



l+Zs 



spec 



(i) 



For all the spec-z sources, we found an outlier fraction of 6% 
and ctnmad = 0.020. For sources in the redshift range of 0.5-1, 
which are of primary interest for this study, the outlier fraction 
is 3% and ctnmad = 0.017. These photo-z's have compara- 
bly high quality to those recently obtained for galaxies in the 
CDF-S region ( Cardamone et alJl2O10HDahlen et al.ll2010h . 

We calculated the IR-based and UV-based SFRs following 
lBelletal.ld2005h . 

SFR IR+UV = 9.8 x IO-^Lir/Lq + Luv/Lq) M yr" 1 , (2) 



and 



SFR, 



= 9.8 x 10- 11 (L uv ,con-/io) M Q yr" 



(3) 



where L Q = 3.8 x 10 erg s and Lir, Luv, and Luv.con 
are the IR, UV, and dust -extinc t ion co rrected UV luminosi- 
ties, respectively. The Kroupa| (1200 ll) initial mass function 
was adopted here. The IR luminosity was estimated by find- 
ing an IR SED template that produces the observed 24 fim 
luminosity via interpolation of a library of 105 IR SEDs 
and then calculat ing the integrated 8- 1000 /im luminosity 
for this template dCharv & Elbazll2001l) . It has been demon- 
strated that the IR luminosity determined this way is similar 
to that derived with additional longer wavelength photometric 
data and mid-IR spectroscopic for sources with z < 1 .4 and 
Lir < 3 x 10 12 L Q (Mur phy et al.ll2009l l20llh . The UV lu- 
minosity was computed following Luv = 3.3z4„ : 2800 (see §3.2 
of iBell et al.ll2005l) . where the rest-frame 2800 A monochro- 
matic luminosity /,,.2800 was interpolated from the multiwave- 
length data. We calculated the dust-extinction correction for 
the UV luminosity employing the Calzetti et al. (2000) extinc- 
tion law, Luvxoit = 10 a724 "Luv. 

As AGNs tend to reside in massive galaxies (e.g ., 
iKauffmann etaLl l200l iBrusa et all 120091; IXue et all [2010) . 
we estimated stellar masses for the 5237 galaxies and used 
these masses to filter o ut low-mass obje cts. The stellar mass 
is calculated following lXue et aTl J2010I) : 

log(M*/M ) = \og(L K /L (I) . K ) + b K (M B -M v ) + a K -0.1Q , 

(4) 
where Lq^ is the monochromatic /if -band luminosity of the 

Sun {L &iK = 3.6 x 10 18 erg s" 1 Hz" 1 ), t he coefficients a K = 
-1.390 and b K =1.176 are from IZibetti et al.l (2009), and the 
rest-frame monochromatic luminosity Lk and color (Ms-My) 
in the Vega system are from the SED-fitting results. All 
the sources have IRAC detections, and thus they all have 
rest-frame /f-band coverage in the SED fitting. The normal- 
ization has been adjusted by -0.10 dex to account for our 



adopted Kroupa (2001) initial mass function. The A'-band lu- 
minosity was used because it is sa 5-10 times less sensitive 
to dust and stellar-population effects than optical luminosities 
(e.g.. lBell&deJongll2000h . 

2.2. Sample Selection 

We define our parent sample from the 5237 optically and 
24 /im selected galaxies above with the following criteria: (1) 
The redshift of the source is between 0.5 and 1; there are 2037 
such sources. (2) The source must have COMBO-17 or Sub- 
aru detections (in > 10 photometric bands) to ensure robust 
SED fitting; there are 24 sources removed by this criterion. 
(3) The source has a stellar mass M»>5x 10 9 Mq, as AGNs 
tend to be hosted by ma s sive galaxies (e.g ., Ikauffman n et al.l 
120031: IBrusa et all £0091 IXue et al.l Hoiof ; this criterion re- 
moves ps 25% of the sources. (4) The source is not X-ray 
detected or overlapping with the 90% encircled-energy aper- 
ture (see §3.2 of IXue et al. 201 1) of any known X-ray source; 
the latter criterion is to avoid any contamination from nearby 
X-ray sources in our X-ray stacking analysis. We used the 
4 Ms CDF-S and 250 ks E-CDF-S X-ray source catalogs for 
this purpose dLehmer et al.l2005l;IXue et al.l20lTh . About 200 
sources are removed by this criterion. We note that the basic 
source properties, such as the dust extinction, stellar mass, 
and UV-based SFR, were calculated based on the assumption 
that the optical SED is dominated by the host galaxy; this as- 
sumption is valid after we exclude the X-ray sources from the 
sample (e.g., see §4.6.3 of Xue et al. 2010 for further discus- 
sion). 

The parent sample defined above consists of 1313 sources, 
about 20% of which have spec-z's. The spectroscopic com- 
pleteness does not have any significant dependence on the IR 
or optical magnitude, and it is largely limited by the spec- 
troscopic coverage in the area; for example, w 60% of the 
sources in the GOODS-S region have spec-z's. We show 
in Figure [T] the dust extinction (Ay) against the ratio of the 
IR-based (SFR IR+UV ) and UV-uncorrected [SFR UV = 9.8 x 
10~ ll (L\jy/Lo) Mq yr" 1 ] SFRs, which is also a measure of 
extinction. The Ay values derived from SED fitting can well 
account for the extinction in general. In Figure 2, we plot the 
logarithmic ratios of the IR-based and UV-based SFRs, de- 
fined as Rsfr = log(SFRiR+uv/SFRuv,con-)- The distribution 
of Rsfr displays an exce ss on the positive side {Rs.fr > 0) 
similar to that observed in Da ddi et al.l J2007I) for z ~ 2 galax- 
ies. Various statistical errors are present in the calculations 
of SFRir + uv and SFRuv.corr, such as the uncertainties in Lir, 
Luv, and Ay. These tend to make the Rsfr distribution fol- 
low approximately a Gaussian function. We mirror the neg- 
ative half of the histogram to the positive side, and the com- 
bined distribution (blue histogram in Fig. 2) can be approx- 
imated by a Gaussian function with ct = 0.33. We thus con- 
sider that the excess in the distribution of Rsfr is caused by 
galaxies hosting heavily obscured or even CT AGNs. We de- 
fine IS X sources using the same criterion as in Daddi et al. 
J2007I) : Rsfr > 0.5. For comparison, we define IR SFR nor- 
mal (ISN) sources as those having Rsfr < 0.2. There are 242 
ISX and 736 ISN sources in the parent sample. Assuming 
that the intrinsic dispersion of Rsfr is Gaussian (blue his- 
togram in Fig. 2), and that the excess IR emission is powered 
by a heavily obscured AGN, about 73% of the ISX sources 
(176 objects out of the 242 ISX sources; shaded region in 
Fig. 2) host such AGNs. For simple comparison, we selected 
X-ray detected sources in the same way from the initial 5237 
sources (revising the fourth criterion above to require X-ray 




ik?- 



■■':■:'■&>> 



10 
5FR IUi 



FIG. 1. — The dust extinction (Ay) derived from SED fitting vs. the 
ratio of the IR-based (SFRir+uv; Eq. 2) and UV-uncorrected [SFRuv = 
9.8 X 10 (Luv/i<©) M@ yr~~ ] SFRs for sources in the parent sample. The 
stars indicate the mean Ay values in different SFRir+uv /SFRuv bins; each 
bin contains 60 sources. The errors on the mean values are smaller than or 
comparable to the symbol size. The solid line shows the ideal dust-extinction 
corrections which would lead to consistent IR-based and UV-based SFRs. 
The SED-derived Ay values scatter around this line; the deviation of the mean 
A v values from the ideal line at large ratios of the two SFRs is likely due to the 
contribution from ISX sources that still have IR excesses after dust-extinction 
corrections (see Fig. 2). 

detection in the CDF-S or E-CDF-S source catalogs); there 
are 198 X-ray sources selected. Following the AGN classifi- 
cation scheme in §4.4 of Xue et al. (201 1), including criteria 
for intrinsic X-ray luminosity, effective power-law photon in- 
dex, and X-ray-to-optical flux ratio, we identified 155 AGNs 
from the X-ray sources, mostly unobscured or moderately ob- 
scured. This number is comparable to that of expected heav- 
ily obscured AGNs (176 objects) in the ISX sample, consis- 
tent with predic tions from population-synthesis models (e.g., 
iGilli et alJl2007h . 

The ISX and ISN sources have different physical proper- 
ties. The median stellar mass, IR luminosity, and IR-based 
SFR of the ISX sample are 2.2, 1.4, and 1.2 times those for 
the ISN sample, respectively, while the median UV lumi- 
nosity, UV-based SFR, and Ay of the ISX sample are 50%, 
15%, and 30% those for the ISN sample. If the relative ex- 
cess in the IR luminosities is contributed by underlying heav- 
ily obscured AGNs, then such objects selected by the ISX 
method at z « 0.5-1 appear to be hosted by massive galax- 
ies with little dust. Since AGNs are preferentially hosted 
by massive galaxies ( e.g., Kauffm ann et al.1 12003 : Bru sa et all 
120091: IXue et all 12010ft . it is natural to find ISX sources in 
such galaxies. The dust extinction (in terms of Ay) applies 
to the entire galaxy, and a small value of Ay does not nec- 
essarily conflict with a high absorbing colum n density for a 
heavi ly 
l200lt TP 

ues in the ISX sources may be a selection effect. Heav- 
ily obscured AGNs at z ~ 0.5-1 are numerically dominated 
by moderate-luminosity objects (intrinsic X-ray luminosity 
ss 10 42 -10 43 erg s" 1 ) according to population-synthesis mod- 
els (e.g., Gilli et al. 2007), and sources hosted by galaxies 
with less dust (less IR emission from the host galaxy) will 
have more prominent IR excesses and are more likely to be 
identified by IR selection methods. 

As a test of whether ISX sources are preferential hosts 
of obscured AGNs, we also included X-ray detected CDF- 



heavi l y obscured A GN in the nucleus (e.g., iMaiolino et al.1 
" IPollettaetai] 120081) . The finding of smaller Ay val- 



$0 0.5 




fe -0.5 - ISI 



50 100150 
§ of Sources 

FIG. 2. — The logarithmic ratio of SFRir+uv and SFRuv.corr vs. the IR lu- 
minosity for sources in the parent sample. The filled and open dots represent 
sources with spec-z's and photo-z's, respectively. X-ray AGNs in the 4 Ms 
CDF-S main catalog are shown as red stars (obscured) and green squares 
(unobscured or weakly obscured). The classification of obscured/unobscured 
AGNs is based on X-ray data; see §2.2. The right panel shows the distribution 
of Rsfr ( n °t counting the X-ray sources). The dashed line is the reflection 
of the bottom half of the histogram around Ssfr = 0; the resulting distribu- 
tion (the blue histogram) can be approximated by a Gaussian function with 
a = 0.33. The horizontal dotted and dash-dotted lines indicate our criteria for 
defining the ISX and ISN samples, respectively. About 73% of the sources 
in the ISX sample (shaded area) belong to the excess population compared 
to the Gaussian distribution, and many of these are probably hosts of heavily 
obscured AGNs. The majority of the X-ray AGNs in the ISX region are ob- 
scured. [See the electronic edition of the Journal for a color version of this 
figure.] 



S AGNs (IXue et al.ll20Tlt) in Figure 2. There are 74 CDF- 
S AGNs selected from the initial 5237 sources (revising the 
fourth criterion above to require X-ray detection in the CDF- 
S source catalog). We further removed 6 luminous AGNs 
with Lfj.5-8 kev > 10 43 7 erg s _1 as the optical SEDs for such 
sources may be affected by AGN contamination (e.g., §4.6.3 
of IXue et al.H2.Q10h . The level of intrinsic absorption was es- 
timated by assuming an underlying X-ray power-la w photon 
index of V = 1.8 and using XSPEC (Version 12.5.1: lArnaudl 
1996) to derive the appropriate Nn value that produces the ob- 
served X-ray band ratio (defined as the ratio of count rates 
between the HB and SB). We consider a source to be ob- 
scured if A^h > 10 22 cm" 2 ; otherwise, it is unobscured or only 
weakly obscured. Figure 2 shows clearly that AGNs in the 
ISX (/?sfr > 0.5) region are mostly obscured (17 cases out 
19), while only half of the AGNs in the ISN region are ob- 
scured. We caution that the nature of the ISX population as 
a whole is likely different from these X-ray detected AGNs, 
given the apparent difference in the IR-to-X-ray flux ratios, 
and thus we perform detailed X-ray studies of the ISX sample 
in the following section. 

3. X-RAY STACKED PROPERTIES AND COMPTON-THICK AGN 
FRACTION 

3.1. X-ray Stacking Analysis 

X-ray stacking analysis is a useful technique to obtain the 
average X-ray properties and probe the nature of a sam- 
ple of X-ray undetected objects, and it has been used ex- 
tensi vely in previous studies of IR-selected AGN cand idates 
(e.g JPaddi et al.ll2007l:lDonlev et al.ll2008l;lFiore et alj2009ft . 
We performed X-ray stacking of the ISX and ISN samples in 
the three standard X-ray bands of the 4 Ms CDF-S, FB, SB, 
and HB. We did not stack the ISX and ISN sources in the 



250 ks E-CDF-S because the total X-ray exposure is low and 
no significant stacked signal can be achieved. Sources located 
further than 6' from the average aim point are not included in 
the stackin g due to sensitivity degradation at l arge off-axis an- 
gles (e.g.. iLehmer et alj|2007t Kue et al.11201 It) . We visually 
examined the IR and optical images of the remaining ISX and 
ISN sources, and excluded additional 10 sources (two ISX 
and eight ISN sources) whose 24 /im data appear to be af- 
fected by source blending. The final ISX and ISN samples 
used in the stacking contain 23 and 58 sources, respectively^! 
About 70% of these sources have spec-z's. We also visually 
checked the X-ray images and verified that these ISX and ISN 
sources are not close to any bright X-ray sources which could 
affect the stacking results. Basic properties of the ISX and 
ISN sources used in the X-ray stacking analysis are listed in 
Table 1. 

We followed a stacking procedure similar to that discussed 
in Steffen et al. (2007). For each source in each of the three 
standard bands, we calculated the total (source plus back- 
ground) counts within a 3"-diameter circular aperture cen- 
tered on its optical position. This extraction radius was 
found to p roduce the best signal-to-noise ratio (S/N) (e.g., 
Worsley et al. 2006) compared to other choices. The corre- 
sponding background counts within this aperture were deter- 
mined with a Monte Carlo approach which randomly (avoid- 
ing known X-ray sources) places 1000 apertures within a 1'- 
radius circle of the source position to measure the mean back- 
ground (e.g., Brandt et al. 2001). The total counts (S) and 
background counts (B) for the stacked sample were derived 
by summing the counts for individual sources. The net source 
counts are then given by (S-B), and the S/N is (S — B)/\/~B; 
we note that the numbers of source and background counts 
are large (> 100), and thus the S/N can be calculated using 
Gaussian statistics. The 3"-diameter aperture does not encir- 
cle the full point-spread function (PSF). Thus we determined 
an encircled-energy fraction (EEF) for each source in each 
band by interpolating an EEF table given the aperture radius, 
off-axis angle, and photon energy. The EEF table was de- 
rived from the PSF images of the main-catalog source s in the 

4 Ms CDF-S generated by ACIS EXTRACT (AE; Bro os et all 
2010) that uses the MARX ray-trace simulator; PSF images 
at five different photon energies (« 0.3-8.5 keV) for each 
main-catalog source in ea ch Chandra observation were pro- 
vided by AE (see §3.2 of IXue et all 120 111) . We created the 
EEF table by calculating the EEFs at different aperture radii, 
off-axis angles, and photon energies, averaged over all the 
observations weighted by exposure time. The EEFs calcu- 
lated this way are the best representative of the real CDF-S 
data. The EEF in a 3"-diameter aperture has a strong off- 
axis angle dependence, being w 95% for on-axis sources, and 
s=s 50% for sources at a 6' off-axis angle. The aperture correc- 
tion for the stacked counts was calculated using the exposure- 
weighted average of the EEFs for all the sources. The ef- 
fective power-law photon indices and fluxes were calculated 
based on the band ratios and aperture-corrected count rates us- 
ing the CXC's Portable, Interactive, Multi-Mission Simulator 
(PIMMS; see details in §3.3.1 of lLuo et a l. 2008). 

The stacking results are listed in Table 2, which show statis- 
tically significant differences between the ISX and ISN sam- 

17 Note that the entire sample of ISX sources was selected in the E-CDF- 

5 region which covers a total solid angle of Ri 1100 arcmin 2 , and thus the 
number of ISX sources within the inner 6' -radius area is only PS 10% of the 
total. 



■r 

" o 

n eM inn mm 



Fig. 3. — The adaptively smoothed stacked images for the ISX and ISN 
samples. The left panels are for the SB, and the right panels are for the HB. 
The images were adaptively smoothed with a minimal significance (S/N) of 
2.5 a, and have been scaled linearly with the same scaling. The black circles 
represent the 3"-diameter aperture used to extract photometry. It is evident 
that the ISX sample has a much harder stacked spectrum than the ISN sample. 

pies. For the ISN sample, there is a strong detection (10.6 a) 
in the SB, while the stacked signal is weak (1.6 crl3 m the 
HB. The corresponding band ratio is 0.29 ±0.19 (1 a er- 
rors), which indicates an effective power-law photon index of 
r = 2.0 ±0.6, consistent with X-ray emission from starburst 
galaxies (e.g., Ptak et al. 1999). The ISX sample has a « 5 a 
detection in the SB and a w 4 a detection in the HB (cor- 
responding to rest-frame s=a 3-14 keV), with a band ratio of 
1.5^Q5, corresponding to T = 0.6^4. X-ray sources with very 
flat spectral slopes (r < 1) are almost exclusively identified 
as heavily obscured AGNs (e.g., Bauer et al. 2004), and thus 
a significant contribution from obscured AGNs is required 
to produce this kind of hard X-ray signal for the ISX sam- 
ple. We note that the ISN sample may still contain a fraction 
of low-luminosity AGNs with soft X-ray spectra comparable 
to those of s tar-forming galaxies (e.g., Gonzalez-Martin et al. 
2006, 2009). These AGNs are generally unobscured and are 
not of primary interest to this study. In Figure 3, we show 
the adaptively smoothed stacked images for the ISX and ISN 
samples. It is clear that the ISX sample has a much harder X- 
ray signal than the ISN sample. The difference between the 
ISX and ISN samples is also apparent via comparison of their 
fluxes; the ISX SB flux is k, 40% smaller than the ISN SB 
flux, while the ISX HB flux is « 5 times higher than the ISN 
HB flux. 

3.2. Heavily Obscured AGN Fraction 

The stacked X-ray signal (T w 0.6) for the ISX sample indi- 
cates the existence of embedded AGNs and is consistent with 
X-ray emission from heavily obscu red or CT AGNs mod- 
eled with a reflection spectrum (e.g., George & Fabianl fl 99 lb 

18 This is a marginal detection. The chance of producing such a weak 
stacked signal by Poisson fluctuations is Ri 5%. Treating this kind of weak 
signal as a detection does not affect our later analyses. 



Maiolino et al. 1998) or a spectrum from a toroidal repro- 
cessor (Murphy & Yaqoob 2009). However, we cannot de- 
rive the average AGN properties using the stacked signal di- 
rectly, as the ISX sources have a range of redshifts and the 
sample is likely contaminated by star-forming galaxies (see, 
e.g., Fig. O. We thus performed Monte Carlo simulations 
to assess the fraction of heavily obscured AGNs/CT AGNs 
in the sample, using a procedure refined from that discussed 
in iFiore et ail (J2009). For a given AGN fraction, we ran 
10000 simulations to estimate the expected X-ray emission 
from both the star-forming and AGN activities of the sample, 
and we then compared the average simulated results to the 
observed stacked signal. 

3.2.1. Simulation for the IR SFR Normal Sample 

We first used the ISN sample (58 star-forming galaxies) 
to test our simulation method. For star-forming galaxies, 
a number of studies have found an approximately linear 
correlation between the SF R and rest- frame 2-10 keV X- 
ray lu minosity L2-io,gai (e-g-, iBauer et ai]|2002t iRanalli et al.l 
2003). However, later studies have pointed out that the X- 
ray-SFR correlation is likely to have significant scatter due 
to the fact that SFR only relates to the population of rel- 
atively young high-mass X-ray binaries (HMXBs), and the 
older population of low-mass X-ray bina ries (LMXBs) likely 
relates to the galaxy st ellar mass (e.g.. IColbert et al I l2004t 
Persic & Rephaeli 2007). Therefore, we adopted the most re- 
cent relation presented in[Lehmer et al. (2010), which consid- 
ers the correlation of L2-io,gai with both the SFR and stellar 
mass, 



^2-lO.sal 



ai = oM*+/3SFR, 



(5) 



where a = 2.33 x 10 29 erg s" 1 M Q : and ji = 1.62 x 
10 39 erg s" 1 (Mo yr" 1 )" 1 ; here a has been increased by a fac- 
tor of 2.6 to account for the average di fference between stel - 
lar masses calculated in this work and Leh mer et al.l (I2010I) . 
Ideally, the two SFRs for a given ISN galaxy, SFRir+uv and 
SFRuv.con, should be equal to each other. We consider that 
the Gaussian spread in /?sfr for ISN galaxies shown in Fig- 
ure 2 is caused by uncertainties in the two SFRs, and the 
intrinsic SFR is estimated as log(SFR) = [log(SFR IR+uv ) + 
log(SFRuv.corr)]/2. The corresponding 2-10 keV X-ray lu- 
minosity was then derived from Equation [5] also including a 
random Gaussian dispersion of 0.34 dex dLehmer et alj|2010h 
to account for the scatter of that relation. 

We assumed an X-ray power-law photon index of T = 2.0 
(given the stacking results) and no intrinsic absorption for 
the star-forming galaxies, and then converted the predicted 
2-10 keV X-ray luminosity to the observed SB and HB fluxes 
using the individual redshifts of the sources. As none of 
the ISN sources is individually X-ray detected, we required 
that the simulated X-ray fluxes in the SB and HB of each 
source do not exceed the sensitivity limits at the source po- 
sition. The X-r ay sensitivity limits are derived following 
IXue et all (120111) . by calculating the minimum flux at each 
pixel (converted from the minimum number of counts) re- 
quired for detection under the catalog source-detection cri- 
teria. We added a small Gaussian dispersion of 0. 1 dex to the 
sensitivity limits to account for statistical fluctuations (esti- 
mated using the few CDF-S X-ray sources with fluxes below 
the nominal sensitivity limits). If the simulated X-ray fluxes 
are greater than the sensitivity limits, we made another ran- 
dom generation of the source properties (in this case, only 
Eq. and the sensitivity limits have random scatters). The 



expected SB and HB counts were calculated given the expo- 
sure time, flux- to-count-rate conversion (using PIMMS; see 
iLuo et aJ1l2008l) . and EEF of the source. For each simula- 
tion, net counts of every source were summed in the SB and 
HB as the output counts. We performed 10000 simulations, 
and the average numbers of output counts are 204.2 ± 0.3 for 
the SB and 57.4 ±0.3 for the HB (the errors are the stan- 
dard errors of the mean; the 1 a dispersions for the average 
numbers of counts are both « 30), matching well with the ob- 
served stacked counts (194.8 ±23.1 and 52.4 ±33.0). There- 
fore we can successfully reproduce the X-ray emission from 
star-forming activity with this simulation approach. 

3.2.2. Simulations for the IR SFR Excess Sample 

We then performed simulations for the 23 sources in the 
ISX sample adopting a similar approach, calculating X-ray 
emission from both the star-forming and AGN activities of 
the sample. Assuming an AGN fraction in the range of sa 10- 
90%, the X-ray emission from the star-forming galaxies in the 
sample was derived following the same procedure as above. 

For a source selected as an AGN, its X-ray emission and 
IR emission consist of both a star-formation component (host 
galaxy) and an AGN component. For the star-formation 
component, we estimated its IR-based SFR to be SFRir s f = 
k x SFRuv.con, where the factor k is randomly drawn from a 
Gaussian function with a = 0.33 (blue histogram in Fig. |2]l 
to account for the scatter of the calculated SFR. The in- 
trinsic SFR of the star-formation component was then esti- 
mated as log(SFR) = [log(SFR IR , sf ) + log(SFR uv ,con-)]/2, and 
the corresponding X-ray emission was calculated. For the 
AGN component, the intrinsic rest-frame 2-10 keV luminos- 
ity, L2-io,agn, can be estimated fro m the rest -frame 6 /im lu- 
minosity ^ofJhe_AGN i _^L ( , i 6^m (e.g jLutz et al.l2004h . The re- 
lation in lLutz et al.l (120041) was derived using a sample of local 
AGNs with the 6 /im AGN continua decomposed from the IR 
spectra. Here we used the luminosity-dependent version of 
the relation which takes into account the possible luminosity 
dependence (Maiolino et al. 2007; Alexander et al. 2008), 



L 2 - 



10,AGN : 



10 



,4.57 



X {vL„ filim ) 



0.88 



(6) 



A Gaussian dispersion of 0.4 dex was added to account for 
the scatter of the relation, estimated based on the range of 
intrinsic X-ray-t o-mid-IR luminosity ratios for local AGNs 
dLutz et al.l 120041) . We note that this relation is consistent 
with the X-ray-to-mid-IR luminosity ratios of a sample of X- 
ray selected obscured quasars (Sturm et al. 2006), suggest- 
ing that it is applicable for deriving the intrinsic luminosity 
of heavily obscured or CT AGNs. To derive the rest-frame 
6 /im monochromatic luminosity coming from the AGN, we 
first computed the AGN 24 /im flux by subtracting the star- 
formation contribution from the observed 24 /im flux; the star- 
formation contribution to the 24 /im flux wa s estimated us- 
ing the IR-based SFR (SFR IR . sf ) above and the lCharv & Elbazl 
(2001) IR SED templates. The residual 24 /im flux was then 
converted to the rest-frame 6 /im flux using the IR SED of the 
local CT AGN NGC 10 68 (SE P data from lRigopoulouet all 
119991 and iGalliano et all 120031) . The observed 24 /im wave- 
length corresponds to rest-frame 12-16 /im for our sources, 
and the conversion factor to the rest-frame 6 /im flux is ss 0.2- 
0.3. 

We adopted an intrinsic photon index of T = 1.8 for the 
AGN and an Nh value randomly drawn from the Nh dis- 
tribution shown in Figure 7 of iGilli et al.l (120071) . in which 



8 



about half of the AGNs are CT. The adopted N H distribu- 
tion has only a small impact on the final A^h distribution, as 
will be discussed later. Given the intrinsic 2-10 keV lumi- 
nosity, absorption, and photon index, we estimated the ob- 
served absorbed SB and HB fluxes using the MYTORUS 
model (Murphy & Yaqoob 2009f!| implemented in XSPEC. 
The M YTORUS model is a recently developed spectral-fitting 
suite for modeling the transmitted and scattered X-ray spectra 
from a toroidal-shaped reprocessor, and it is especially de- 
signed to treat the spectra of CT AGNs. Compared to the 
commonly used disk-reflectio n model for CT AGNs (e.g., th e 
pexrav model in XSPEC; see Magdziarz & Zdziarski 1995), 
the MYTORUS model is more physically motivated and 
takes into account absorption and scattering simultaneously. 
We only considered the continuum output from MYTORUS 
for simplicity, neglecting the fluorescent iron emission lines 
which contribute less than 8% of the HB continuum flux. Two 
basic parameters describing the geometry of the model are the 
half-opening angle of the torus and the inclination angle (0° 
corresponds to a face-on observing angle). The half-opening 
angle was set at 37°, corresponding to a scenario where the 
number of obscured AGNs is four times the number of un- 
obscured AGNs; such an obscured AGN fraction (s=a 80%) 
for moderate-luminosit y AGNs at z ~ 1 has been reporte d 
in several studies (e.g.,[Hasinger 2008; JTreister et all 12008). 
We randomly chose an inclination angle between 37° and 90° 
(probability weighted by solid angle) for each source. Other 
parameters, such as the relative cross-normalizations of differ- 
ent components, were set at the default values (see §8.2 of the 
MYTORUS manual). The N u value in the MYTORUS model 
has an upper limit of 10 25 cm" 2 . For column densities beyond 
this limit, the observed spectrum is highly geometry depen- 
dent, and the scattered flux is likely dominated by scattering in 
any optically thin parts of the torus, even if the effective solid 
angle of the optically thin gas is tiny (T Yaqoob and K. D. 
Murphy 201 1, private communication). The above parameter- 
ization of the model cannot reproduce the exact environment 
of real obs cured AGNs, which could be much more c ompli- 
cated (e.g.. iGuainazzi et al.ll2005t IComastri et all 120101) . but 
we consider it to be the best available approximation of the 
average properties. We then combined the emission emerging 
from this obscured AGN component with that from the star- 
formation component to derive the expected observed emis- 
sion. 

As for the simulations of star-forming galaxies, we required 
that the observed SB and HB fluxes do not exceed the sensi- 
tivity limits at the source position. We would continue ran- 
domly regenerating the source properties (e.g., A^h value) un- 
til this requirement is satisfied. This is a strong constraint on 
the simulated source properties, and it generally requires the 
AGN to be heavily obscured or CT. Finally, we converted the 
fluxes into counts in the extraction aperture, and added the 
SB and HB counts from every source as the output of a single 
simulation. We again performed 10000 simulations, and the 
average output counts were compared to the stacking results 
in Table 2. 

3.2.3. Simulation Results 

The simulation results are shown in Figure 4, with the sim- 
ulated SB and HB counts at different assumed AGN frac- 
tions. As the AGN fraction increases, the band ratio also 
increases and the simulated spectrum becomes harder. We 

" See http : / /www . my torus ■ com/| for details. 



note that the simulated SB photons almost exclusively orig- 
inate from star-forming galaxies and AGN hosts (their star- 
formation components), while the HB photons are mainly 
from the AGNs; the increase of band ratio with AGN frac- 
tion is largely due to the increase in the HB counts. How- 
ever, we did not include a soft-scattering component in the 
spectra of heavily obscured AGNs, which has been observed 
to provide significan t SB emission in most of the local CT 
AGNs (e.g ., Kinkha bwa¥etaLll2002l; IGuainazzi et al.M2Q09b 
iMarinucci et alj 1201 lb . The normalization of this compo- 
nent to the intrinsic power-law continuum is around 1% but 
has large object-to-object variations and uncertainties (e.g., 
IComastri et al"1l2007t lGmTeTaIll2007t lUeda et al.ll2007l) . and 
the resulting contribution to the SB emission relative to the 
host-galaxy contribution depends on both this normalization 
and the intrinsic X-ray luminosity. Therefore the SB simu- 
lated counts should be considered lower limits, and we use 
only the HB results to determine the AGN fraction. As shown 
in Figure 4, an AGN fraction of 74 ±25% can produce the 
observed HB counts (see Table 3)0 The error on the AGN 
fraction was determined given the error on the observed HB 
counts. The f=a 75% AGN fraction is in good agreement with 
the fraction of extra sources (73%, shaded area) in the /?sfr 
distribution plot in Figure 2. For this best-fit AGN fraction, 
we also computed the stacked and simulated X-ray signals 
in several sub-bands, and the comparisons are shown in Ta- 
ble 3. The simulations reproduced well the stacked counts 
in general. The cumulative A^h distribution for the 75% of 
AGNs in the 10000 simulations is displayed as an inset of 
Figure 4, where « 95% of the AGNs are heavily obscured, 
and « 80% are CT. The CT AGN fraction in the entire ISX 
sample is thus sa 60%. The ISX AGNs have moderate X-ray 
luminosities. The median intrinsic 2-10 keV luminosity de- 
rived from the simulations is L2-io,agn «2x 10 42 erg s" 1 ; this 
value would be ss 50% higher if the intrinsic luminosity was 
calculated (Equation 6) using the observed IR flux (without 
subtracting the star- formation contribution). The interquartile 
range of the X-ray luminosities is (0.9^1) x 10 42 erg s" 1 [(1- 
5)x 10 42 erg s _1 ] for the heavily obscured (CT) AGNs0 

We explored how the AGN fraction could change by vary- 
ing some of the assumptions in the simulations. First, we 
tried the commonly used disk-reflection model instead of the 
MYTORUS model to compute obscured AGN emission. An 
absorbed power-law {zwabs*zpow in XSPEC) plus reflection 
{pexrav in XSPEC) model was used. The absorbed power-law 
component typically dominates in the Compton-thin (A^h < 
1.5 x 10 24 cm" 2 ) regime, and the reflection component dom- 
inates in the CT regime. We note that the pexrav model has 
non-negligible uncertainties on some of its parameters, such 
as the reflection scaling f actor. For simplicity, we adopted the 
same parameters used in Gilli et al. (2007) to model the XRB; 
we assumed a cut-off energy of 200 keV, a reflection scaling 
factor of 0.37, and an inclination angle of 60° for the reflect- 
ing material. The hard X-ray flux produced with this model is 
about three times higher th an that from the MYTORUS model 
dMurphv & Yaqo ob 2009), and thus the required AGN frac- 
tion in the ISX sample is smaller. The fraction drops from 

20 We note that the ss 25% star-forming galaxies can still host low- 
luminosity AGNs, as long as the AGN X-ray luminosities are not comparable 
to their host galaxies; this kind of AGNs is not the focus of this study. 

21 For comparison to local studies, these intrinsic luminosities are compa- 
rable to those of some well-known CT AGNs, such as the Circinus galaxy 
and NGC 4945 (e.g., Comastri 2004 and references therein). 



60 




0.4 0.6 

AGN Fraction 

FIG. 4. — The average SB and HB counts of the ISX sample as a function 
of the AGN fraction, derived from Monte Carlo simulations. The statistical 
errors on the mean values are smaller than the symbol size. The SB simulated 
counts should be considered lower limits owing to the highly uncertain soft 
scattering component not included in the simulations. The two stars with 
error bars indicate the observed stacked SB (lower star) and HB (upper star) 
counts, with the best-fit AGN fraction as the .v-axis values. The HB (SB) data 
point is shifted rightward (leftward) by a small amount for clarity. An AGN 
fraction of 74 ± 25% can reproduce the observed signals. The inset panel 
shows the cumulative Nn distribution at this AGN fraction, with the dashed 
and dotted lines indicating the criteria for heavily obscured and CT AGNs. 
Approximately PS 95% of the AGNs in the ISX sample are heavily obscured, 
and « 80% are CT. 

74±25% to 55 ± 15%, still indicating a substantial AGN con- 
tribution. Again, almost all (sa 95%) of the AGNs are heavily 
obscured. Simply increasing the ^-lo.AGN-to-z/L^e/jm ratio in 
Equation[6]also reduces th e AGN frac t ion. F or example, if we 
adopt the relation used in iFiore et ail (|2009), which gives an 
X-ray luminosity that is s=s 0.2 dex more luminous than ours, 
the resulting AGN fraction is « 65%. The initial Nn distri- 
bution input into the simulation will affect the AGN fraction 
slightly. We adopted an alternative initial Nu distribution that 
has only 25% CT AGNs, which is simil ar to the Nu distri - 
bution for high excitation-line galaxies in lTozzi et alj (120061) . 
The derived AGN fraction is 60 ± 15% with « 90% being CT. 
The CT AGN fraction in the entire ISX sample (rs 55%) is 
thus comparable to that (i=s 60%) for the best-fit result above. 
Even in an extreme case where the initial TVh distribution 
has only 10% CT AGNs, an AGN fraction of « 60% and a 
CT AGN fraction of sa 55% are still required to produce the 
stacked X-ray emission and satisfy the requirement that the 
simulated sources cannot be individually detected in the SB or 
HB. We note that the adopted photon index for star- forming 
galaxies (T = 2.0) has little effect on the estimated AGN frac- 
tion as these galaxies produce mainly SB counts and we do 
not consider SB counts to be constraining. 

In general, the requirement that the simulated sources can- 
not be individually detected places a strong constraint on the 
final results, and an AGN fraction of > 50% (mostly heavily 
obscured) is always expected. We also caution that the simu- 
lation results were taken to be the average of 10000 tests, and 
thus they (as well as the following analyses based on these 
results) only represent the most probable scenario of the ISX 
source properties and might deviate from the real source prop- 
erties. We consider that this is the best available approach to 
probe the nature of ISX sources given the X-ray and multi- 
wavelength data available. 

4. DISCUSSION 



4.1. Space Density and Contribution to the 10-30 keV XRB 

Given the simulation results for the ISX sample, we can es- 
timate the space density of heavily obscured AGNs at z ~ 0.5- 
1 and calculate their contribution to the XRB at high energies 
(10-30 ke V) . There are 1 6 ± 5 ( 1 3 ± 4) heavily obscured (CT) 
AGNs expected in the X-ray stacking ISX sample. These 
objects are located within the central 6'-radius region of the 
CDF-S; the corresponding area is 0.029 deg 2 after exclud- 
ing the regions masked by X-ray sources. We further correct 
the number of sources by 10% to account for the fraction re- 
moved due to blended 24 /im photometry. Therefore the sky 
density of heavily obscured (CT) AGNs in our ISX sample is 
sw 600 deg" 2 (rj 500 deg" 2 )Qthe corresponding space density 
is (2.0 ±0.7) x 10" 4 Mpc" 3 [(1.6 ±0.5) x 10" 4 Mpc" 3 ] given 
the comoving volume between redshift 0.5 and 1. 

Note that there are several factors affecting the estimated 
space density above. The expected number of AGNs in the 
ISX sample may be up to « 30% lower if different assump- 
tions are made in the simulations (§3.2), and thus the space 
density will be reduced by the same fraction. On the other 
hand, there is probably a small fraction of sources excluded 
from the parent sample with stellar mass M* < 5 x 10 y M 
(§2.2) that also host heavily obscured AGNs. Also, the X-ray 
stacking ISX sample was selected in a relatively small area, 
and thus we have missed a few rare but intrinsically lumi- 
nous AGNs, which is also suggested by the moderate IR lu- 
minosities of all the ISX sources in Table 1 . Moreover, the X- 
ray spectra of heavily CT (Nu > 10 25 cm" 2 ) AGNs cannot be 
modeled by MYTORUS and an upper limit of 10 25 cm" 2 was 
used for the column density; if these heavily CT AGNs are 
present in the ISX sample, we could have overestimated their 
X-ray emission and thus underestimated the AGN fraction 
and space density. Most importantly, we define the ISX sam- 
ple conservatively (^sfr > 0.5) to avoid significant contami- 
nation from star-forming galaxies in the sample. There will 
be additional heavily obscured AGNs among those sources 
that are not in the ISX or ISN sample (0.2 < R SFR < 0.5; see 
jj4.3l below for X-ray stacking results); the expected number is 
nearly comparable to that in the ISX sample given the fraction 
of excess sources in the distribution of R$fr in Figure 2. Tak- 
ing into account the uncertainties and incompletenesses, we 
expect that the true space density of heavily obscured AGNs 
at z ~ 0.5-1 is w 1 .5-2 times that estimated above. In the fol- 
lowing analyses, we still use our conservative estimation as it 
is difficult to determine the exact correction factor. 

We show in Figure 5 the space densities of CT AGNs se- 
lected in this work and some previous studies ( mainly IR 
based) , as well as those from model predi ctions bvlGilli et al.l 
(12007b . The CT AGN space density for the lDaddietal .1(120071) 
CDF-S sample with z ~ 1 .4-2.5 and L2-10 ^ 10 43 erg s" 1 is es- 
timated to be «a 2.6 x 10~ 4 Mpc" 3 ; the actual value is expected 
to be lower (e.g.. lDonlev"etaT] l2008t iMurphvet al]|2009h 
and was recently revi sed downward to sa 2 x 10~ 5 Mpc" 3 
dAlexan der et al. 2011). At lower redshift, the space density 
for the IFiore et al.l (12009b COSMOS CT AGN sample with 
zw0.7-1.2is(3.7±l.l)xl0" 5 Mpc" 3 . We are probing a dif- 
ferent population of CT AGNs from the IFiore et al J (120091) 
COSMOS sample in terms of intrinsic X-ray luminosities. 
The interquartile range of the intrinsic 2-10 keV luminosities 

22 For simple comparison, the sky density of z = 0.5-1 X-ray AGNs 
from the original 5237 galaxies in the same region for the 4 Ms CDF-S is 
« 800 deg" 2 . 



10 




FIG. 5. — Space density of CT AGNs in the ISX sample. Also shown are 
the CT AGN space densities in some previous IR-based studies: Daddi et al. 
(2007; D07), Alexander et al. (2008; A08), Fiore et al. (2008; F08), Fiore 
et al. (2009; F09), and Alexander et al. (2011; All). We also include the 
local CT AGN density from Treister et al. (2009; T09). The redshift error 
bars indicate the redshift ranges of the samples, and the intrinsic 2-10 keV 
luminosity range of each sample is also shown. Note that the Alexa nder et al] 
12008) and Alexander et al. 12011) data points were derived from spectro- 
scopically (X-ray and/or IR) identified CT AGNs, the Treist er et al] (M®) 
data point was derived based on a sample of local CT AGNs, and the other 
studies are based on X-ray stacking analyses of X-ray undetected candidates. 
The dashed curves are the predicted space densities of CT AGNs for dif- 
ferent X-ray luminosity lower limits from the Gilli et al. (2007) population- 
synthesis model. Note that the various data points and model predictions have 
been derived with different underlying assumptions and are thus not strictly 
comparable. 

is (1-5) x 10 42 erg s" 1 for the CT AGNs in our sample, while it 
is (3-10) x 10 43 erg s" 1 for the iFiore et all (l2009t) sample. We 
thus selected mainly moderate-luminosity and more typical 
CT AGNs; this is largely attributed to the much higher sen- 
sitivities of the mid-IR and X-ray observations in the CDF-S. 
On the other hand, we expect that we have missed some rare, 
unrepresentative objects due to the smaller area of the CDF- 
S. We note that the data points in various studies and model 
predictions have been derived with different assumptions (see, 
e.g., §3.2.2) and are thus not strictly comparable. 

We estimated the expected XRB flux in the 10-30 keV band 
provided by the heavily obscured AGNs in the ISX sample. 
We used the M YTORUS model to calculate the emergent flux 
from these AGNs, given their redshifts, assumed power-law 
photon index (Y = 1.8), intrinsic X-ray luminosities, and col- 
umn densities (the latter two quantities are from the simu- 
lation results). On average, these AGNs produce a flux of 
« 0.6 keV cm" 2 s" 1 sr" 1 in the 10-30 keV band, « 85% of 
which is from CT AGNs. The total XRB flux in this b and 
is about 44 keV cm" 2 s" 1 sr" 1 (e.g., iMorettiet ail 120091 and 
references therein). Therefore, this population of z ~ 0.5-1 
heavily obscured (CT) AGNs selected using the ISX method 
is expected to contribute « 1 .4% (« 1 .2%) of the total XRB in 
the 10-30 keV band. Observations have not directly resolved 
the XRB at energies > 10 keV; however, theoretical predic- 
tions have been made by population-synthesis models about 
the contributions from unobscured, moderately obscured, and 
heavily obscured AGNs. For example, Gilli et al. (2007) con- 
cluded that « 24% of the XRB in the 10-30 keV band (the 
missing XRB) is produced by CT AGNs, while ITreister et all 
(2009) considered the total contribution from CT AGNs to 
be f=a 9%. The predicted contribution from the CT AGNs se- 
lected here (ss 1 .2%) is thus sa 5% (Gilli et al. 2007) or sa 15% 



( ITreister etal .1120091) of the missing XRB in the 10-30 keV 
band. Given the expected properties (lu minosity, r e dshift , 
column density) of the ISX sources, the iGilli et ail (2007) 
model predicts that sa 7% of the missing XRB comes from 
such objectsO agreeing well with our results. The remaining 
missing XRB is largely attributed to lum inous (Z_- io ~ 10 42 - 
10 44 erg s" 1 ) CT AGNs at z < 1.5 (IGilli et al.l2007l) : the X-ray 
stacking ISX sample selected here appears to have missed 
some of the high-luminosity objects due to the limited vol- 
ume surveyed. 

4.2. Resolved Fraction of the XRB by Chandra 

Heavily obscured AGNs are expected also to contribute to 
the XRB at energies < 10 keV, though with a smaller con- 
tributed fra ction than that for the 10-30 keV band (see, e.g., 
Fig. 15b of lGilli et alj|2007h . The XRB in the 1-8 keV band, 
unlike that in the 10-30 ke V band, has been largely resolved 
into discrete sources (e.g ., Bauer et alJ l2004t IWorsley et al.1 
120051: Iffick ox & Markevitch 2006). With the deepest Chan- 
dra data available, the 4 Ms CDF-S, we expect to resolve 
the XRB further and improve our understanding of the nature 
of the X-ray source populations at low flux levels, including 
the heavily obscured population. We performed X-ray stack- 
ing analyses on the X-ray sources and optical galaxies in the 
CDF-S to explore the resolved XRB fraction and hidden AGN 
contribution to the unresolved fraction. 

Measurements of the normalization of the XRB spectrum 
have non-negligible uncertainties and field-to-field variations; 
the combined unce rtainty on the normalization is as 10-20% 
(e.g., iMoretti et aDl2003b Iffickox & Markevitch 2006) . Here 
we adopted the normalization from Hickox & Markevitch 
(2006), which was derived from the Chandra Deep Fields 
data inclu ding the 2 Ms Chandra Deep Fiel d-North (CDF- 
N: lAlexander et al] 120031) and 1 Ms CDF-S dGiacconi et all 
2002). The XRB has a power-law spectral slope with Y = 1.4 
and a normalization of 10.9 photons s" 1 keV" 1 sr" 1 at 1 keV. 
X-ray stacking analyses were performed in the following en- 
ergy bands: 0.5-1, 1-2, 2^1, 4-6, and 6-8 keV. For X-ray 
sources, we used the 740 sources in t he 4 Ms CDF-S main 
X-ray source catalog dXue et al.112.01 lh . The stacking proce- 
dure was the same as that described in §3.1; to maximize the 
S/N ratio, a 3"-diameter circular aperture was used and the 
stacking was performed for the 389 sources within the inner 
6'-radius area. To account properly for bright X-ray sources 
that have a rare occurrence in the narr ow CDF-S region, we 
adopted the bright-end correction in Hickox & Mar kevitch! 
(2006): sources brighter than 5 x 10" 15 erg cm" 2 s" 1 in the 
SB or 1.4 x 10~ 14 erg cm" 2 s" 1 in the HB are removed from 
the stacking, and the background intensity produced by such 
bright sources was calculated using the number counts of X- 
ray sources (Hickox & Markevitch 2006). Four sources in the 
SB and two in the HB are removed this way; the final stack- 
ing samples include 385 sources in the 0.5-1 and 1-2 keV 
bands, and 387 sources in the three > 2 keV bands. For the 
optical galaxies, we chose z-band sources i n the GOODS-S 
HST version r2.0z catalog (Giavali sco et al] |2004f 2 ^l with a 5 
a limiting AB magnitude of 28 .2. We stacked the X-ray unde- 
tected optical galaxies in the central 6'-radius region. Galax- 
ies within twice the 90% encircled-energy aperture radius of 
any known X-ray source are also removed from the stacking 

|http : //www, bo .astro . it/$\sim$gilli/xrb . html .| 
24 See http://archive.stsci.edu/pub/hlsp/goods/catalog_r2/ 



11 



to avoid X-ray source contamination. There are 18 272 optical 
sources included in the stacking. Note that this galaxy sam- 
ple contains all of the 23 sources in the X-ray stacking ISX 
sample. 

The stacking results are shown in Figure [6] The 1 a 
errors on the stacked flux es were calculated following 
iHickox & Markevitchl (|2006) including measurement errors 
and a 3% Chandra flux-calibration error. For the stacking of 
the X-ray sources, there is an additional Poisson error due to 
the limited number of sources below the bright-end flux cut 
(Hickox & Markevitch 2006). Taking into account both the 
X-ray source contribution and bright-end correction, the re- 
solved XRB fractions are m 75%-80% in all the energy bands, 
indicating that the average photon index of the X-ray sources 
is r w 1.4. This is consistent with the fact that the major- 
ity o f the X-ray sources in the CDF- S are obscured AGNs 
fe.g JLuo et alJl2010l:IXue et alJl2011l) . For the X-ray unde- 
tected optical galaxies, significant detections are found in all 
the bands except the 4-6 keV band, where there is only a 1 a 
signal (corresponding to a w 20% chance that the signal was 
created by Poisson noise). We thus calculated 3 a upper lim- 
its on the stacked counts and resolved XRB fraction in this 
band. In the 6-8 keV band, the optical galaxies produced a 
2.5 a stacked signal (wl% chance of being generated by Pois- 
son noise), responsible for 28 ± 11% of the XRB0 With all 
of the X-ray sources and galaxies considered, it appears that 
the XRB in the 6-8 keV band can be fully explained, though 
the uncertainties in the normalization and stacked signals are 
large (the slightly higher stacked flux than the XRB flux in this 
band could also be caused by cosmic variance in the CDF-S). 
At 1-6 keV, X-ray sources and galaxies can account for s=s 80- 
90% of the XRB. The remaining unresolved fraction is prob- 
ably due to cosmic variance in the narrow CDF-S region (e.g., 
iBauer et al.ll2004t iLuo et al.ll2008l) . It could also be partially 
contributed by extended X-ray sources (e.g., galaxy groups or 
clusters) in the CDF-S. 

The 2.5 a stacked signal from galaxies in the 6-8 keV 
band (28 ± 11% of the XRB; we refer to this as the galaxy 
CT XRB), combined with the weak signal in the 4-6 keV 
band (< 6% of the XRB), suggest an underlying population 
of heavily obscured AGNs among the X-ray undetected galax- 
ies; emission in the 4-6 keV band is more heavily absorbed 
than that in the 6-8 keV band. As a subsample of the optical 
galaxies, the 23 ISX sources produce a stacked signal that is 
0.2 ±0.1% of the total XRB flux in the 4-6 keV band; the 
stacked signal in the 6-8 keV band is weak (1.4 a), and we 
estimated a 3 a upper limit that is 1.2% of the total XRB. 
Therefore, in the 6-8 keV band, the ISX sources contribute 
< 5% of the galaxy CT XRB. It is expected that there are 
some additional heavily obscured AGNs not selected with the 
ISX method among the optical galaxies (see §4.1 and §4.3). 
The number density of such objects is of about the same order 
of magnitude as the ISX sources. 

Based on the population-synthesis model in iGilli et all 
(2007), we derived that heavily obscured AGNs below the 
4 Ms CDF-S sensitivity limro are responsible for f=a 8% of 
the XRB in the 6-8 keV band (we refer to this model pre- 
diction as the model CT XRB). This model CT XRB is only 

25 As a check of the stacking method, we stacked 1 8 272 random positions 
(excluding X-ray sources) in the central 6' -radius region, and the stacked 
signals in all the bands are consistent with zero (< 1 a significance). 

26 We adopted here the median 2 -8 keV sensitiv ity limit within the inner 
6' -radius region of the 4 Ms CDF-S IXueet all2011l) . 




FIG. 6. — Resolved fractions of the XRB in five energy bands between 0.5 
and 8 keV. The total XRB intensities are from Hickox & Markevitch ( 2006) 
with uncertainties indicated by the gray region. The stacked contributions 
from X-ray sources in the 4 Ms CDF-S, bright-end correction, GOODS-S 
optical galaxies, and the sum of the above are shown as blue, red, dark green, 
and magenta data points, respectively. Note that the stacked signal for the 
optical galaxies in the 4-6 keV band did not yield a significant detection and 
thus a 3 <t upper limit on the resolved fraction was calculated (triangle with 
a downward arrow). The upper limit was used to derive the upper error when 
calculating the total resolved fraction in this band. For the other data points, 
1 cr errors are shown. The contribution from the ISX objects is < 1.2% in the 
6-8 keV band and is not shown. [See the electronic edition of the Journal for 
a color version of this figure.] 

~ 30% of the galaxy CT XRB, due to possible cosmic vari- 
ance in the CDF-S and uncertainties in the model assumptions 
(e.g., assumptions about the spectral shape, CT AGN number 
density, and obscured fraction; see §9 of Gilli et al. 2007 for 
discussion). We also note that the IGilli et all (120071) model 
does not include low-luminosity (L2-10 S 10 42 erg s _1 ) AGNs. 
IR studies suggest that the AGN fraction in low-mass galax- 
ies may be significantly hi gher than previously reported using 
optical spectroscopy (e.g., Goulding & Alexander 2009). The 
obscured AGN fraction also increases as luminosity decreases 
(e.g., Hasinger 2008). It is thus probable that low-luminosity 
heavily obscured AGNs have a significant stacked contribu- 
tion to the galaxy CT XRB. These AGNs are faint in the X-ray 
band and challenging to identify even in deep Chandra obser- 
vations (e.g., Goulding et al. 2010). Their IR luminosities are 
more likely to be dominated by host-galaxy emission and thus 
are difficult to detect with the ISX method (or any IR-based 
method). 

4.3. Additional Samples and Subsamples 

We extracted three additional X-ray stacking samples with 
0.2 < /?sfr < 0.5 to test whether there are AGNs among these 
sources. We further break the ISX sample into two subsam- 
ples to explore how the stacking results depend on the /?sfr 
threshold value. The threshold cuts in /?sfr for these samples 
(samples A1-A5) are listed in Table 2, and they were chosen 
so that each sample has a similar number (w 10) of sources. 
We performed X-ray stacking analyses on the two subsamples 
and three additional samples following the same procedure as 
that for the ISX and ISN samples. The results are presented in 
Table 2. Due to limited sample sizes, these samples have less- 
significant detections in the HB than the ISX sample. How- 
ever, the band ratios and effective photon indices for samples 
A1-A4 all suggest heavily obscured AGN contributions to the 
stacked X-ray signals. A plot of the effective photon index 



12 



2.5 


















L _ 


2.0 


7 < « 


> 














- 




: isn 


















- 


1.5 


























'- 




*' 






1.0 








. 












- 












i 


, x 


• 


c 




N 


Z _ 








/ 


0.5 


• i 


^ 




0.0 






i 

ISX 










- 



0.4 

K. ci 



FIG. 7. — The effective photon index of the stacked X-ray signal vs. the 
average Rsfr of the sample. Besides the ISX and ISN samples (filled circles), 
several additional samples (open circles) are presented with different thresh- 
old cuts in i?sFR ( see Table 2). The error bars or arrows in the .v-axis direction 
indicate the intervals of the Rsfr cuts. In general, as the average Rsfr in- 
creases, the stacked X-ray spectrum gets flatter, suggesting more significant 
contributions from heavily obscured AGNs. 

as a function of the average /?sfr is shown in Figure 7; in 
general, the larger the Rsfr threshold value (more IR excess), 
the harder th e stacked signal. S imilar behavior has also been 
observed by Daddi et al. (2007), and it is likely due to less 
contamination from star-forming galaxies at larger threshold 
values (see Fig. 2). Given the stacking results for these ad- 
ditional samples, it is not likely that the hard X-ray stacked 
signal of the ISX sample was produced by coincidence. 

4.4. Observational Prospects for Distant Heavily Obscured 

AGNs 

A potentially straightforward way to detect distant heav- 
ily obscured or CT AGNs is via hard X-ray observations at 
sa 10-100 keV. Several future hard X-ray missions, suc h as 
NuSTAR (planned launch year 2012: lHarrison et al.l20l"oh an d 
ASTRO-H (planned launch year 2014-llakahashi et al .11201 01) . 
have as one science goal to detect hard X-ray emission from 
distant heavily obscured AGNs. For the three typical hard X- 
ray bands of NuSTAR, 6-10 keV, 10-30 keV, and 30-60 keV, 
the expected sensitivity limits are w 2 x 10~ 15 , 2 x 10~ 14 , and 
6 x 10~ 14 erg cm" 2 s" 1 for a 1 Ms exposure^ respectively. 
These flux limits are about two orders of magnitude more sen- 
sitive than those of previous missions. The ASTRO-H sensi- 
tivities are expected to be comparable to those for NuSTAR. 
The angular resolutions (half-power diameters) of NuSTAR 
and ASTRO-H are expected to be « 50" and sa 1.7', respec- 
tively. The NuSTAR positional accuracy is expected to be 
sa 1.5" for strong sources. To assess if the AGNs in our ISX 
samples at z « 0.5-1 will be detectable, we utilized the prop- 
erties for sources in the X-ray stacking ISX sample from the 
best-fit simulations (for the 74% AGN fraction) and the MY- 
Torus model to calculate their expected fluxes in the NuS- 
TAR bands. Only a tiny fraction (« 0.4%) of the simulated 
sources in our 10000 simulations have fluxes above the NuS- 
TAR sensitivity limit in its most sensitive band (10-30 keV); 
this corresponds to no detected sources expected within the 
16 ± 5 heavily obscured AGNs in the X-ray stacking ISX 
sample (or one detectable source among all the 242 ISX ob- 

27 See h ttp : //www.nustar . caltech.edu/l 



jects). The median simulated flux in the 10-30 keV band 
(sa 3 x 10~ 16 erg cm" 2 s" 1 ) is about two orders of magnitude 
below the sensitivity limit. This result is a natural conse- 
quence of the requirement that the sources are not individu- 
ally detected in the 2-8 keV band of the 4 Ms Chandra expo- 
sure; the extraordinarily high sensitivity of the 4 Ms CDF-S 
places a tight constraint on the intrinsic luminosities of the 
ISX sources and prevents them from being detected by NuS- 
TAR. Therefore, it is not likely that NuSTAR or ASTRO-H will 
detect any of the ISX sources presented here. However, they 
will probably detect some of the X-ray selected CT A GN can- 
didates (e.g.. ITozzi et al.ll2006t IComastri et al.ll201 ll) in hard 
X-rays; such detections will be useful for a clear determina- 
tion of the intrinsic spectral shape and power of these sources. 
One other approach to identify heavily obscured or CT 
AGNs is via X-ray spectroscopy at relatively low energies 
(< 10 keV) complement ed by multi wavelengfh data (e.g., 
Polletta et alJl20 06: Tozzi et ai1 l2006t I Alexander et "all 120081 
2011; Comastri et al. 2011). The X-ray emission of these 



objects is characterized by a flat continuum and often a 
strong rest-frame 6.4 keV iron Kg fluorescent line (e.g., 
iDella Ceca et"all 120081: iMurphv & Yaaoobl [2009). For the 
ISX sources presented here, spectroscopic analyses are prob- 
ably not feasible due to the small number of counts expected. 
However, it is worth performing a further study of th e X-ray 
selected CT AGN candidates in ITozzi et all d2006l) . which 
were previously studied using only the 1 Ms CDF-S data. 
With the current 4 Ms CDF-S data and the 3 Ms XMM-Newton 
observations on the CDF-S, we will be a ble to constrain bet- 
ter their nature (e.g.. IComastri et aill201 1[) . In the case of the 
CDF-S receiving further Chandra exposure (e.g., 10 Ms to- 
tal), some (« 15%) of the heavily obscured AGNs in the ISX 
sample could be detected in the HB, given the simulated prop- 
erties and expected 10 Ms sensitivity. 

Given the above, the majority of the heavily obscured 
AGNs in the ISX sample will remain undetected in the X-ray, 
and the few percent of the XRB at sa 10-100 keV produced 
by these AGNs cannot be directly resolved in the near fu- 
ture. We note that X-ray absorption variability appears com- 
mon among local Seyfert 2 galaxies (e.g., Risaliti et al. 2002), 
and a few sourc es have ev en exhibited CT to Compton-thin 
transitions (e.g., Ma tt et al J[2003t iRisaliti et al.ll2005l) . There- 
fore, a fraction of the ISX AGNs may be detectable in the 
future if they become less obscured due to absorption vari- 
ability. Also, optical/mid-IR spectroscopy has been shown to 
be a promising technique for identifying distant heavily ob- 
scured AGNs, partic ularly when co mbined with sensi tive X - 
ray constraints (e.g., Alexander et al. 2008; Gilli et al. 2010). 
Future optical/IR instruments such as the James Webb Space 
Telescope (JWST), Extremely Large Telescopes (ELTs), and 
Atacama Large Millimeter Array (ALMA) will provide great 
opportunities for detecting the remaining sources. 

5. CONCLUSIONS AND FUTURE WORK 

We have identified a population of heavily obscured/CT 
AGNs at z sa 0.5-1 in the CDF-S and E-CDF-S utilizing the 
ISX selection method. The key points from this work are 
listed below: 



1. We ha ve improved the ISX method from Daddi "etal] 
(120071) by deriving the dust extinction via SED fitting 
and studying sources at lower redshifts with no bias in 
their IR-luminosity calculation. The ISX method can 
be well applied to the X-ray undetected galaxies in the 



13 



CDF-S region at z ~ 0.5-1, given the superb multi- 
wavelength data available. The parent sample of 1313 
galaxies were defined from an initial sample of 5237 
optically and 24 ^m selected sources in the E-CDF- 
S, with the requirements of selected redshift interval, 
excellent multiwavelength coverage, high stellar mass, 
and no X-ray detection (§2). 

2. We have identified 242 ISX sources in the CDF-S and 
E-CDF-S at z ~ 0.5-1; these sources tend to be hosted 
by evolved galaxies with high stellar mass and little 
dust (§2.2). An X-ray stacking analysis of 23 of the 
objects in the central CDF-S region resulted in a very 
hard X-ray signal with an effective photon index of 
r = 0.6^04, indicating a significant contribution from 
obscured AGNs (§3.1). 

3. We have performed Monte Carlo simulations to esti- 
mate the AGN fraction in the ISX sample and assess 
the intrinsic properties of the obscured AGNs. We 
modeled the observed X-ray flux considering both the 
star-formation and AGN contributions, and we utilized 
the MYTORUS model to treat the spectra of heav- 
ily obscured AGNs. The requirement that the sources 
are not individually detected in the 4 Ms CDF-S sets 
strong constraints on their intrinsic properties. We infer 
that 74 ± 25% of the ISX sources are obscured AGNs, 
within which ss 95% are heavily obscured and ca 80% 
are CT (§3.2). 

4. The heavily obscured (CT) AGNs discovered in our 
ISX sample have moderate intrinsic X-ray luminosities; 
the interquartile range of the intrinsic 2-10 keV lumi- 
nosities is (0.9^1)xl0 42 erg s" 1 [(l-5)xl0 42 erg s" 1 ]. 
The space density of the heavily obscured (CT) AGNs 
is(2.0±0.7)x 10" 4 Mpc- 3 [(1.6±0.5)x lO^Mpc" 3 ]. 
These heavily obscured (CT) AGNs are expected to 
contribute f=a 1.4% (w 1.2%) of the total XRB flux in 
the 10-30 keV band. In the 6-8 keV band, the 23 



ISX sources provide < 1.2% of the XRB flux. The X- 
ray undetected optical galaxies in the CDF-S produce a 
2.5 a stacked signal in the 6-8 keV band, accounting 
for 28 ± 1 1 % of the XRB flux, which is about the entire 
unresolved XRB fraction. The space density of the ISX 
AGNs and their contribution to the XRB could be in- 
creased by a factor of rj 1 .5-2 due to our conservative 
ISX definition (§4). 



5. These heavily obscured or CT AGNs will probably not 
be detected by hard X-ray observatories under develop- 
ment such as NuSTAR or ASTRO-H due to their moder- 
ate intrinsic X-ray luminosities and significant obscu- 
ration. Most of the hard X-ray sources that will be de- 
tected by these facilities are likely already detected in 
the 4 Ms CDF-S given the extremely high sensitivity of 
Chandra to point sources (§4.4). 

The ISX selection method presented in this paper can be ap- 
plied to other survey fields with good optical-to-IR coverage, 
and it can be expanded to higher redshifts if the IR luminos- 
ity can be estimated reliably, e.g., combining MIPS 24 /im 
data with data from deep Herschel surveys at 100 and 160 /im 
(e.g., Shao et al. 2010). We will explore these possibilities in 
future work. 

We acknowledge financial support from CXC grant SP1- 
12007A (BL, WNB, YQX), CXC grant G09-0134A (BL, 
WNB, YQX), NASA ADP grant NNX10AC99G (WNB), 
the Science and Technology Facilities Council (DMA), and 
ASI/INAF grant 1/009/10/0 (AC, CV, RG). We are grateful to 
T. Yaqoob and K. D. Murphy for providing support for the 
MYTORUS model and making available the data for our de- 
sired half-opening angle. We thank A. T Steffen for helpful 
discussions. We also thank the referee for carefully review- 
ing the manuscript and providing helpful comments that im- 
proved this work. 



REFERENCES 



Alexander, D. M., et al. 2003, AJ, 126, 539 

— . 2008, ApJ, 687, 835 

—.2011, ApJ, in press I arXiv: 1 10604431 

Alonso-Herrero, A., et al. 2006, ApJ, 640, 167 

Arnaud, K. A. 1996, in ASP Conf. Ser., Vol. 101, Astronomical Data 

Analysis Software and Systems V, ed. G. H. Jacoby & J. Barnes, 17 
Balestra, I., et al. 2010, A&A, 512, 12 
Barger, A. J., Cowie, L. L., Mushotzky, R. R, Yang, Y., Wang, W.-H., 

Steffen, A. T., & Capak, P. 2005, AJ, 129, 578 
Bauer, F. E., Alexander, D. M., Brandt, W. N., Hornschemeier, A. E., 

Vignali, C, Garmire, G. P., & Schneider, D. P. 2002, AJ, 124, 2351 
Bauer, F. E., Alexander, D. M., Brandt, W. N., Schneider, D. P., Treister, E., 

Hornschemeier, A. E., & Garmire, G. P. 2004, AJ, 128, 2048 
Bell, E. F, & de Jong, R. S. 2000, MNRAS, 312, 497 
Bell, E. F, et al. 2005, ApJ, 625, 23 

Brammer, G. B., van Dokkum, P. G., & Coppi, P. 2008, ApJ, 686, 1503 
Brammer, G. B., et al. 2009, ApJ, 706, L173 
Brandt, W. N., & Hasinger, G. 2005, ARA&A, 43, 827 
Brandt, W. N., Hornschemeier, A. E., Schneider, D. P., Alexander, D. M., 

Bauer, F E., Garmire, G. P., & Vignali, C. 2001, ApJ, 558, L5 
Broos, P. S., Townsley, L. K., Feigelson, E. D., Getman, K. V, Bauer, F. E., 

& Garmire, G. P. 2010, ApJ, 714, 1582 
Brusa, M., et al. 2009, A&A, 507, 1277 
—.2010, ApJ, 716, 348 
Calzetti, D., Armus, L., Bohlin, R. C, Kinney, A. L., Koornneef, J., & 

Storchi-Bergmann, T. 2000, ApJ, 533, 682 
Cardamone, C. N., et al. 2008, ApJ, 680, 130 



— . 2010, ApJS, 189,270 

Chary, R., & Elbaz, D. 2001, ApJ, 556, 562 

Colbert, E. J. M., Heckman, T. M., Ptak, A. F, Strickland, D. K., & Weaver, 

K. A. 2004, ApJ, 602, 231 
Comastri, A. 2004, in Astrophysics and Space Science Library, Vol. 308, 

Supermassive Black Holes in the Distant Universe, ed. A. J. Barger, 245 
Comastri, A., Gilli, R., Vignali, C, Matt, G., Fiore, E, & Iwasawa, K. 2007, 

Progress of Theoretical Physics Supplement, 169, 274 
Comastri, A., Iwasawa, K., Gilli, R., Vignali, C, Ranalli, P., Matt, G, & 

Fiore, F. 2010, ApJ, 717, 787 
Comastri, A., et al. 2011, A&A, 526, L9 
Daddi, E., et al. 2007, ApJ, 670, 173 
Dahlen, T., et al. 2010, ApJ, 724, 425 
Damen, M., et al. 201 1, ApJ, 727, 1 
Delia Ceca, R., et al. 2008, Memorie della Societa Astronomica Italiana, 79, 

65 
Dey, A., et al. 2008, ApJ, 677, 943 
Donley, J. L., Rieke, G H., Alexander, D. M., Egami, E., & Perez-Gonzalez, 

P. G. 2010, ApJ, 719, 1393 
Donley, J. L., Rieke, G H., Perez-Gonzalez, P. G, & Barro, G. 2008, ApJ, 

687, 111 
Donley, J. L., Rieke, G H., Perez-Gonzalez, P. G, Rigby, J. R., & 

Alonso-Herrero, A. 2007, ApJ, 660, 167 
Elbaz, D., et al. 2010, A&A, 518, L29 
Elvis, M., et al. 1994, ApJS, 95, 1 
Feldmann, R., et al. 2006, MNRAS, 372, 565 



14 



Feruglio, C, Daddi, E., Fiore, F., Alexander, D. M., Piconcelli, E., & 

Malacaria, C. 2011, ApJ, 729, L4 
Fiore, E, et al. 2008, ApJ, 672, 94 
— . 2009, ApJ, 693, 447 
Galliano, E., Alloin, D., Granato, G. L., & Villar-Martfn, M. 2003, A&A, 

412,615 
Gawiser, E., et al. 2006, ApJ, 642, L13 
Georgakakis, A., Rowan-Robinson, M., Nandra, K., Digby-North, J., 

Perez-Gonzalez, R G., & Barro, G. 2010, MNRAS, 406, 420 
Georgantopoulos, I., Akylas, A., Georgakakis, A., & Rowan-Robinson, M. 

2009, A&A, 507, 747 
Georgantopoulos, I., Georgakakis, A., Rowan-Robinson, M., & Rovilos, E. 

2008, A&A, 484, 671 
George, I. M., & Fabian, A. C. 1991, MNRAS, 249, 352 
Giacconi, R., et al. 2002, ApJS, 139, 369 
Giavalisco, M., et al. 2004, ApJ, 600, L93 
Gilli, R., Comastri, A., & Hasinger, G. 2007, A&A, 463, 79 
Gilli, R., Vignali, C, Mignoli, M., Iwasawa, K., Comastri, A., & Zamorani, 

G. 2010, A&A, 519, 92 
Gonzalez-Martin, O., Masegosa, J., Marquez, I., Guainazzi, M., & 

Jimenez-Bailon, E. 2009, A&A, 506, 1107 
Gonzalez-Martin, O., Masegosa, J., Marquez, I., Guerrero, M. A., & 

Dultzin-Hacyan, D. 2006, A&A, 460, 45 
Goulding, A. D., & Alexander, D. M. 2009, MNRAS, 398, 1 165 
Goulding, A. D., Alexander, D. M., Lehmer, B. D., & Mullaney, J. R. 2010, 

MNRAS, 406, 597 
Grazian, A., et al. 2006, A&A, 449, 951 
Graber, D. E., Matteson, J. L., Peterson, L. E., & Jung, G V. 1999, ApJ, 

520, 124 
Guainazzi, M., Matt, G., & Perola, G. C. 2005, A&A, 444, 119 
Guainazzi, M., Risaliti, G., Nucita, A., Wang, J., Bianchi, S., Soria, R., & 

Zezas, A. 2009, A&A, 505, 589 
Harrison, F. A., et al. 2010, Proc. SPIE, 7732, 77320S 
Hasinger, G. 2008, A&A, 490, 905 
Hickox, R. C, & Markevitch, M. 2006, ApJ, 645, 95 
Kauffmann, G, et al. 2003, MNRAS, 346, 1055 
Kinkhabwala, A., et al. 2002, ApJ, 575, 732 
Komatsu, E., et al. 2011, ApJS, 192, 18 
Kroupa, P. 2001, MNRAS, 322, 231 
Lacy, M., et al. 2004, ApJS, 154, 166 
Lanzuisi, G., Piconcelli, E., Fiore, F, Feruglio, C, Vignali, C, Salvato, M., 

& Gruppioni, C. 2009, A&A, 498, 67 
Le Fevre, O., et al. 2004, A&A, 428, 1043 
Lehmer, B. D., Alexander, D. M., Bauer, F E., Brandt, W. N., Goulding, 

A. D., Jenkins, L. P., Ptak, A., & Roberts, T. P. 2010, ApJ, 724, 559 
Lehmer, B. D., et al. 2005, ApJS, 161, 21 
—.2007, ApJ, 657, 681 
Luo, B., et al. 2008, ApJS, 179, 19 
— . 2010, ApJS, 187, 560 
Lutz, D., Maiolino, R., Spoon, H. W. W., & Moorwood, A. F. M. 2004, 

A&A, 418, 465 
Lyons, L. 1991, Data Analysis for Physical Science Students (Cambridge: 

Cambridge Univ. Press) 
Magdziarz, P., & Zdziarski, A. A. 1995, MNRAS, 273, 837 
Maiolino, R., Marconi, A., Salvati, M., Risaliti, G., Severgnini, P., Oliva, E., 

La Franca, R, & Vanzi, L. 2001, A&A, 365, 28 
Maiolino, R., Salvati, M., Bassani, L., Dadina, M., della Ceca, R., Matt, G., 

Risaliti, G., & Zamorani, G. 1998, A&A, 338, 781 
Maiolino, R., Shemmer, O., Imanishi, M., Netzer, H, Oliva, E., Lutz, D., & 

Sturm, E. 2007, A&A, 468, 979 



Marinucci, A., Bianchi, S., Matt, G., Fabian, A. C, Iwasawa, K., Miniutti, 

G., & Piconcelli, E. 2011, A&A, 526, A36 
Matt, G., Guainazzi, M., & Maiolino, R. 2003, MNRAS, 342, 422 
Mignoli, M., et al. 2005, A&A, 437, 883 

Moretti, A., Campana, S., Lazzati, D., & Tagliaferri, G. 2003, ApJ, 588, 696 
Moretti, A., Lazzati, D., Campana, S., & Tagliaferri, G. 2002, ApJ, 570, 502 
Moretti, A., et al. 2009, A&A, 493, 501 
Morrissey, P., et al. 2005, ApJ, 619, L7 
Murphy, E. J., Chary, R.-R., Alexander, D. M., Dickinson, M., Magnelli, B., 

Morrison, G., Pope, A., & Teplitz, H. I. 2009, ApJ, 698, 1380 
Murphy, E. J., Chary, R.-R., Dickinson, M., Pope, A., Frayer, D. T., & Lin, 

L. 2011, ApJ, 732, 126 
Murphy, K. D., & Yaqoob, T. 2009, MNRAS, 397, 1549 
Oke, J. B., & Gunn, J. E. 1983, ApJ, 266, 713 
Persic, M., & Rephaeli, Y. 2007, A&A, 463, 481 
Polletta, M., Weedman, D., Honig, S., Lonsdale, C. J., Smith, H. E., & 

Houck, J. 2008, ApJ, 675, 960 
Polletta, M., et al. 2006, ApJ, 642, 673 
Popesso, P., et al. 2009, A&A, 494, 443 
Prieto, M. A., Reunanen, J., Tristram, K. R. W., Neumayer, N, 

Fernandez-Ontiveros, J. A., Orienti, M., & Meisenheimer, K. 2010, 

MNRAS, 402, 724 
Ptak, A., Serlemitsos, P., Yaqoob, T, & Mushotzky, R. 1999, ApJS, 120, 179 
Rafferty, D. A., et al. 2011, ApJ, submitted 
Ranalli, P., Comastri, A., & Setti, G. 2003, A&A, 399, 39 
Ravikumar, C. D., et al. 2007, A&A, 465, 1099 
Reddy, N. A., Erb, D. K., Pettini, M., Steidel, C. C, & Shapley, A. E. 2010, 

ApJ, 712, 1070 
Rigopoulou, D., Spoon, H. W. W., Genzel, R., Lutz, D., Moorwood, 

A. R M., & Tran, Q. D. 1999, AJ, 118, 2625 

Risaliti, G., Elvis, M., Fabbiano, G., Baldi, A., & Zezas, A. 2005, ApJ, 623, 

L93 
Risaliti, G., Elvis, M., & Nicastro, F 2002, ApJ, 571, 234 
Shao, L., et al. 2010, A&A, 518, L26 

Silva, L., Maiolino, R., & Granato, G. L. 2004, MNRAS, 355, 973 
Silverman, J. D., et al. 2010, ApJS, 191, 124 
Steffen, A. T., Brandt, W. N, Alexander, D. M., Gallagher, S. C, & Lehmer, 

B. D. 2007, ApJ, 667, L25 
Stern, D., et al. 2005, ApJ, 631, 163 
Szokoly, G. P., et al. 2004, ApJS, 155, 271 
Takahashi, T, et al. 2010, Proc. SPIE, 7732, 77320Z 
Tozzi, P., et al. 2006, A&A, 451, 457 

Treister, E., Krolik, J. H, & Dullemond, C. 2008, ApJ, 679, 140 

Treister, E., Urry, C. M., & Virani, S. 2009, ApJ, 696, 110 

Ueda, Y, et al. 2007, ApJ, 664, L79 

Vanzella, E., et al. 2008, A&A, 478, 83 

Werner, M. W., et al. 2004, ApJS, 154, 1 

Wolf, C, Hildebrandt, H, Taylor, E. N., & Meisenheimer, K. 2008, A&A, 

492, 933 
Wolf, C, et al. 2004, A&A, 421, 913 
Worsley, M. A., Fabian, A. C, Barcons, X., Mateos, S., Hasinger, G., & 

Brunner, H. 2004, MNRAS, 352, L28 
Worsley, M. A., Fabian, A. C, Bauer, F E., Alexander, D. M., Brandt, 

W. N, & Lehmer, B. D. 2006, MNRAS, 368, 1735 
Worsley, M. A., et al. 2005, MNRAS, 357, 1281 
Xue, Y Q., et al. 2010, ApJ, 720, 368 
—.2011, ApJS, 195, 10 
Zibetti, S., Chariot, S., & Rix, H. 2009, MNRAS, 400, 1181 



15 



TABLE 1 
List of ISX and ISN Sources in the X-ray Stacking Analysis 



RA 

(1) 


Dec 

(2) 


z 
(3) 


Z Lower 
(4) 


Z Upper Ay 
(5) (6) 


(V) 


logL IR 
(8) 


SFRir+uv 
(9) 


SFRuV.corr 
(10) 


#SFR 
(11) 


ISX Sample 


03 32 49.61 
03 32 45.58 


-27 49 00.1 
-27 49 36.4 


0.98 
0.680 


0.97 
-1.00 


1.02 0.2 
-1.00 0.9 


58.7 
472.3 


10.74 
11.46 


6.2 
29.5 


1.1 

5.0 


0.76 
0.77 


ISN Sample 


03 32 50.38 
03 32 48.58 


-27 47 07.1 
-27 45 04.9 


0.537 
0.89 


-1.00 
0.89 


-1.00 1.9 
0.92 1.2 


166.5 
121.8 


10.86 
11.03 


7.7 
12.2 


14.1 
12.2 


-0.26 
0.00 



NOTE. — Table 1 is presented in its entirety in the electronic edition. An abbreviated version of the table is shown 
here for guidance as to its form and content. The full table contains 1 1 columns of information for the 23 ISX sources 
and 58 ISN sources used in the X-ray stacking analysis. Cols. (1) and (2): The J2000 right ascension and declination of 
the ISX or ISN source. Cols. (3)-(5): The spec-z or photo-z of the source. Spec-z's are denoted by having three decimal 
places, while photo-z's with their 1 a confidence intervals (lower and upper bounds) were derived using ZEBRA. Col. 
(6): The V-band dust extinction derived from ZEBRA SED fitting. Col. (7): The MIPS 24 fim flux, in unites of fiiy. Col. 
(8): The logarithmic IR (8-1000 /im integrated) luminosity estimated based on the observed 24 /ira flux, in units of solar 
luminosity. Cols. (9) and (10): The IR-based and UV-based SFRs, in units of tfg yr . Col. (11): The logarithmic ratio 
of the IR-based and UV-based SFRs. 



OS 



TABLE 2 
Stacked X-ray Properties 



Net Source Counts Signal-to-Noise Ratio Flux 







%. 


Zmean 


fcxp (Ms) 














Band 
Ratio 


Effective 

r 








Sample 


FB 


SB 


HB 


FB 


SB 


HB 


SB 


HB 


L X 


(1) 




(2) 


(3) 


(4) 


(5) 


(6) 


(V) 


(8) 


(9) 


(10) 


(11) 


(12) 


(13) 


(14) 


(15) 


ISX (R SFR > 0.5) 




23 


0.69 


76.7 


138.8 + 26.1 


57.6+13.9 


78.1 + 22.1 


6.0 


4.9 


3.9 


i 4.Q+0.63 


o-er^i 


4.9 


36.6 


8.3 


ISN (i? SFR < 0.2) 




58 


0.77 


193.0 


256.8 + 40.3 


194.8 + 23.1 


52.4 + 33.0 


6.9 


10.6 


1.6 


0.29 + 0.19 


2.0 + 0.6 


7.7 


7.3 


4.2 


Al (R S FR > 0.6) 




14 


0.68 


46.9 


71.5 + 20.0 


31.4+10.6 


39.0+17.0 


4.0 


3.5 


2.6 


j 35+0.90 
1 64+^ 


0.7 4 * 5 
5 +0 ^ 

u 6 


4.2 


28.4 


6.3 


A2 (0.5 < i? SFR < 


0.6) 


9 


0.71 


29.8 


67.3 + 16.7 


26.1+9.0 


39.1 + 14.1 


4.5 


3.5 


3.2 


5.9 


50.7 


12.2 


A3 (0.4 < R SFR < 


0.5) 


10 


0.76 


33.3 


45.8+16.8 


25.1 + 9.1 


20.7+14.1 


3.1 


3.2 


1.8 


QQ1+0.81 


1 o +0 - 7 


5.6 


22.7 


7.1 


A4 (0.28 < Rsfr < 0.4) 


11 


0.69 


37.7 


81.7 + 18.5 


47.7+10.6 


33.0+15.2 


5.1 


5.9 


2.4 


76 +a41 
U -^ 4 -0.36 


1 1+04 
'■ -0.5 


8.8 


28.1 


7.4 


A5 (0.2 < R SFR < 


0.28) 


13 


0.74 


43.7 


69.4+19.4 


49.2+11.2 


19.7+15.9 


4.0 


5.6 


1.3 


1.7 + 0.8 


8.1 


12.7 


5.0 



NOTE. — Col. (1): The X-ray stacking sample. The threshold cut in ^sfr is indicated. Col. (2): Number of sources used in the stacking. Col. (3): Mean redshift of the stacked sample. Col. (4): Total FB 
exposure time. Cols. (5)-(7): Stacked net source counts in the FB, SB, and HB, with 1 a Gaussian statistical errors. Cols. (8)— (10): Stacked signal-to-noise ratios in the FB, SB, and HB. Note that there are a few 
marginal detections (< 2 a) in the HB. Treating this kind of weak signals as detections does not affect our analyses in the paper. Col. (11): Stacked band ratio for the stacked sample, defined as the ratio of count 
rates between the HB and SB. The 1 a errors were calculated following the "numerical method" described in §1.7.3 of Lyons f 1991). Col. (12): Effective photon index with 1 a errors for the stacked sample. Cols. 
(13)-(14) SB and HB fluxes for the stacked sample, in units of 10~ erg cm"" s~ . Col. (15): Rest-frame 0.5-8 keV X-ray luminosity calculated from the observed-frame 0.5-8 keV flux for the stacked sample, in 
units of 10 erg s~ . The mean redshift, observed flux, and effective power-law photon index were used in the calculation, and no assumption was made about the intrinsic absorption. 



17 



TABLE 3 

Comparison of Stacked and Simulated 

Counts for the ISX Sample 

Band (keV) Stacked Counts Simulated Counts 



0.5-2.0 


57.6 ±13.9 


57.5 ±0.2 


2.0-8.0 


78.1 ±22.1 


78.9 ±0.3 


0.5-1.0 


22.6 ±8.6 


19.9 ±0.2 


1.0-2.0 


33.2± 11.0 


37.6 ±0.3 


2.0-4.0 


31.6±13.3 


36.8 ±0.4 


4.0-8.0 


44.7 ±17.6 


42.1 ±0.4 



NOTE. — The uncertainties for the stacked counts are 1 a 
Gaussian statistical errors. The simulated counts are the aver- 
age values over the 10 000 simulations; see §3.2 for details.