0708.3678/ms.tex
1: %\documentclass[12pt,preprint]{aastex}
2: \documentclass[apj]{emulateapj}
3: 
4: %\usepackage{fancyheadings}
5: %\usepackage{lscape}
6: %\usepackage{natbib}
7: \usepackage{apjfonts}
8: %\usepackage{psfig}
9: 
10: \newcommand{\ts}{\textstyle}
11: \newcommand{\tnm}{\tablenotemark}
12: \newcommand{\tnt}{\tablenotetext}
13: 
14: \newcommand{\ergs}{ergs s$^{-1}$}
15: \newcommand{\flux}{ergs cm$^{-2}$ s$^{-1}$}
16: \newcommand{\intens}{ergs cm$^{-2}$ s$^{-1}$ deg$^{-2}$}
17: \newcommand{\fluxhz}{ergs cm$^{-2}$ s$^{-1}$ Hz$^{-1}$}
18: \newcommand{\fluxlam}{ergs cm$^{-2}$ s$^{-1}$ ${\rm \AA^{-1}}$}
19: \newcommand{\pers}{s$^{-1}$}
20: \newcommand{\cdens}{cm$^{-2}$}
21: \newcommand{\vdens}{cm$^{-3}$}
22: 
23: \newcommand{\xmm}{{\it XMM}}
24: \newcommand{\xmmnewton}{{\it XMM-Newton}}
25: \newcommand{\chandra}{{\it Chandra}}
26: \newcommand{\asca}{{\it ASCA}}
27: \newcommand{\rxte}{{\it RXTE}}
28: \newcommand{\rosat}{{\it ROSAT}}
29: \newcommand{\bepposax}{{\it BeppoSAX}}
30: \newcommand{\ginga}{{\it Ginga}}
31: \newcommand{\exosat}{{\it EXOSAT}}
32: \newcommand{\einstein}{{\it Einstein}}
33: 
34: \newcommand{\mekal}{{\it MEKAL}}
35: 
36: \newcommand{\rxj}{RX J0059.2--7138}
37: \newcommand{\xtej}{XTE J0111.2--7317}
38: \newcommand{\fouru}{4U 1626--67}
39: \newcommand{\aofive}{A 0538--66}
40: \newcommand{\exo}{EXO 053109--6609.2}
41: 
42: \newcommand{\phip}{$\phi_{35}$}
43: \newcommand{\tout}{\theta_{\rm out}}
44: \newcommand{\tin}{\theta_{\rm in}}
45: \newcommand{\rout}{r_{\rm out}}
46: \newcommand{\rin}{r_{\rm in}}
47: \newcommand{\im}{\item}
48: \newcommand{\al}{\alpha}
49: \newcommand{\be}{\beta}
50: \newcommand{\ta}{\theta}
51: 
52: %\newcommand{\xiiu}{$\xi_{\rm iuv}$}
53: \newcommand{\xiiu}{$\log{(F_{R}/F_{\rm 4.5\mu m})}$}
54: \newcommand{\xiiuk}{$\log{(L_{\rm 2500\; \AA}/L_{\rm 2\mu m})}$}
55: \newcommand{\iiu}{$L_{R}/L_{\rm 4.5\mu m}$}
56: \newcommand{\liiu}{$\log{(L_{R}/L_{\rm 4.5\mu m})}$}
57: \newcommand{\iiuk}{$\log{(L_{\rm 2500\; \AA}/L_{\rm 2\mu m})}$}
58: \newcommand{\lumtwo}{$L_{\rm 2\mu m}$}
59: \newcommand{\alfven}{$\rm{Alfv\grave{e}n}$}
60: 
61: \newcommand{\alm}{$\alpha$}
62: \newcommand{\bem}{$\beta$}
63: \newcommand{\tam}{$\theta$}
64: \newcommand{\dgr}{$^{\circ}$}
65: \newcommand{\average}[1]{\ensuremath{\langle#1\rangle} }
66: 
67: 
68: \newcommand{\rstar}{R_{\star}}
69: \newcommand{\msun}{M_{\sun}}
70: \newcommand{\lsun}{$L_{\sun}$}
71: 
72: \newcommand{\W}{\hphantom{0}}
73: \newcommand{\Wm}{\hphantom{-}}
74: \newcommand{\Wp}{\hphantom{.}}
75: 
76: \newcommand{\pso}{$\phi_{\rm SO}$}
77: 
78: \newcommand{\nh}{$N_{\rm H}$}
79: \newcommand{\av}{$A_{\rm V}$}
80: 
81: \newcommand{\luv}{$L_{2500\; \rm \AA}$}
82: \newcommand{\lr}{$L_{R}$}
83: \newcommand{\lbol}{$L_{\rm bol}$}
84: 
85: \newcommand{\lirac}{$L_{\rm 4.5\mu m}$}
86: \newcommand{\lx}{$L_{\rm X}$}
87: 
88: \newcommand{\fhard}{$F_{\rm 2-7\; keV}$}
89: \newcommand{\fsoft}{$F_{\rm 0.5-2\; keV}$}
90: \newcommand{\lhard}{$L_{\rm 2-7\; keV}$}
91: \newcommand{\lsoft}{$L_{\rm 0.5-2\; keV}$}
92: 
93: \newcommand{\bootes}{Bo\"{o}tes}
94: \newcommand{\spitzer}{{\it Spitzer}}
95: 
96: \newcommand{\cstar}{${\tt CLASS\_STAR}$}
97: 
98: %\newcommand{\micron}{$\mu$m}
99: 
100: %\received{2004 April 8}
101: \begin{document}
102: \slugcomment{Accepted for publication in The Astrophysical Journal}
103: 
104: 
105: 
106: \title{A large population of mid-infrared selected,
107: obscured \\ active galaxies in the Bo\"{o}tes field}
108: \shorttitle{OBSCURED INFRARED AGNS}
109: \shortauthors{HICKOX ET AL.}
110: \author{R.~C. Hickox\altaffilmark{1}}
111: \author{C. Jones\altaffilmark{1}}
112: \author{W.~R. Forman\altaffilmark{1}}
113: \author{S.~S. Murray\altaffilmark{1}}
114: \author{M. Brodwin\altaffilmark{2,3}}
115: \author{M.~J.~I. Brown\altaffilmark{4}}
116: \author{P.~R. Eisenhardt\altaffilmark{2}}
117: \author{D. Stern\altaffilmark{2}}
118: \author{C.~S. Kochanek\altaffilmark{5}}
119: \author{D. Eisenstein\altaffilmark{6}}
120: \author{R.~J. Cool\altaffilmark{6}}
121: \author{B.~T. Jannuzi\altaffilmark{3}}
122: \author{A. Dey\altaffilmark{3}}
123: \author{K. Brand\altaffilmark{3,7}}
124: \author{V. Gorjian\altaffilmark{2}}
125: \author{N. Caldwell\altaffilmark{1}}
126: 
127: 
128: \altaffiltext{1}{Harvard-Smithsonian Center for Astrophysics, 60 Garden Street,
129:  Cambridge, MA 02138; rhickox@cfa.harvard.edu.}
130: \altaffiltext{2}{Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109.}
131: \altaffiltext{3}{National Optical Astronomy Observatory, Tucson, AZ
132:  85726-6732.}
133: \altaffiltext{4}{School of Physics, Monash
134: University, Clayton 3800, Victoria, Australia.}
135: \altaffiltext{5}{Department of Astronomy, The Ohio State University, 140 West 18th Avenue, Columbus, OH 43210.}
136: \altaffiltext{6}{Steward Observatory, 933 North Cherry Avenue, Tucson, AZ 85721.}
137: \altaffiltext{7}{Space Telescope Science Institute, 3700 San Martin
138:  Drive, Baltimore, MD 21218.}
139: 
140: 
141: %\slugcomment{To be submitted to The Astrophysical Journal}
142: 
143: \begin{abstract}
144: We identify a population of 640 obscured and 839 unobscured AGNs at redshifts $0.7<z\lesssim3$ using
145: multiwavelength observations of the 9 deg$^2$ NOAO Deep Wide-Field
146: Survey (NDWFS) region in \bootes.  We select AGNs on the basis of
147: \spitzer\ IRAC colors obtained by the IRAC
148: Shallow Survey.  Redshifts are obtained from optical spectroscopy or
149: photometric redshift estimators.  We classify the IR-selected AGNs as
150: IRAGN 1 (unobscured) and IRAGN 2 (obscured) using a simple criterion
151: based on the observed optical to mid-IR color, with a selection
152: boundary of $R-[4.5]=6.1$, where $R$ and [4.5] are the Vega magnitudes
153: in the $R$ and IRAC 4.5 $\mu$m bands, respectively.  We verify this
154: selection using X-ray stacking analyses with data from the \chandra\
155: X\bootes\ survey, as well as optical photometry from NDWFS and
156: spectroscopy from  MMT/AGES.  We show that (1) these sources
157: are indeed AGNs, and (2) the optical/IR color selection separates
158: obscured sources (with average $N_{\rm H}\sim3\times10^{22}$ \cdens\
159: obtained from X-ray hardness ratios, and optical colors and
160: morphologies typical of galaxies) and unobscured sources (with no
161: X-ray absorption, and quasar colors and morphologies), with a
162: reliability of $\gtrsim80\%$.  The observed numbers of IRAGNs are
163: comparable to predictions from previous X-ray, optical, and IR
164: luminosity functions, for the given redshifts and IRAC flux limits.
165: We observe a bimodal distribution in $R-[4.5]$ color, suggesting that
166: luminous IR-selected AGNs have either low or significant dust
167: extinction, which may have implications for models of AGN obscuration.
168: 
169: 
170: 
171: \end{abstract}
172: 
173: \keywords{galaxies: active ---  infrared: galaxies --- quasars: general
174:   --- surveys --- X-rays: galaxies}
175: 
176: \section{Introduction}
177: \label{intro}
178: 
179: In unified models of active galactic nuclei (AGNs), a significant
180: number of objects are expected to be obscured by a torus of gas and
181: dust that surrounds the central engine and blocks the optical emission
182: along some lines of sight \citep[see reviews by ][]{urry95, anto93}.
183: In addition, some models of merger-driven quasar activity predict a
184: prolonged phase in which the central engine is entirely obscured,
185: followed by a ``blowout'' of the absorbing material and a relatively
186: short unobscured phase \citep[e.g.,][]{silk98, spri05,hopk06apjs}.
187: While some obscured AGNs have been identified, the existence of a
188: large absorbed population ($N_{\rm H}>10^{22}$
189: \cdens) has been invoked to explain the slope of the cosmic X-ray
190: background (CXB) at $E>2$ keV, which is believed to be integrated
191: emission from active galaxies \citep[e.g.,][]{sett89, coma95, bran05}. 
192: 
193: \subsection{Obscured AGNs in the optical, X-ray, and radio}
194: 
195: There are three well-established methods for identifying obscured
196: AGNs.  The first is the existence of narrow, high-excitation emission
197: lines in the optical spectrum, along with the absence of a power-law
198: continuum and broad emission lines that are characteristic of
199: unobscured sources.  The lack of broad lines and continuum is
200: attributed to dust that obscures the broad-line region around
201: the central engine, but leaves visible the larger narrow-line region
202: \citep{urry95, anto93}.
203: 
204: In the standard nomenclature, AGNs with a power-law optical continuum
205: and broad emission lines are referred to as type 1 objects, and those
206: with only narrow lines as type 2 \citep{seyf43,khac74}.  In the Seyfert
207: galaxies, the optical luminosity of the nucleus is comparable to that
208: of the host galaxy, while in the quasars, the nuclear luminosity
209: dominates that of the host galaxy.  Many type 2 Seyfert galaxies are
210: known, and the ratio in number density between type 2 and type 1
211: Seyferts in the local Universe has been estimated to be $\gtrsim$3:1
212: \citep[e.g.,][]{oste88, maio95}, although there is evidence from the
213: X-rays \citep[e.g.,][]{ueda03, barg05, gill07} and optical
214: \citep[e.g.,][]{lawr91,hao_l05b}, that the ratio of type 2 to type 1
215: AGNs decreases with increasing luminosity, and may also change with
216: redshift \citep{lafr05, ball06b}.  While type 2 quasars have been
217: challenging to detect in the optical, $\approx$300 type 2 quasars at
218: redshifts $0.3<z<0.83$ have recently been identified in the Sloan
219: Digital Sky Survey \citep[SDSS,][]{zaka03,zaka04,zaka05}.
220: 
221: X-ray observations also can identify obscured AGNs, by the presence of
222: absorption in the spectrum due to intervening neutral gas that
223: preferentially absorbs soft X-rays
224: \citep[e.g.,][]{awak91,cacc04,guai05,alex05}.  X-ray detection of
225: obscuration is complementary to that in the optical, because it is
226: caused by absorbing neutral gas rather than dust.  Some authors have
227: classified X-ray AGNs similarly to optical AGNs, based on the absence
228: (type 1) or presence (type 2) of X-ray absorption
229: \citep[e.g.,][]{ster02, ueda03, zhen04}.  Typically, an X-ray AGN is defined
230: to be absorbed (type 2) if its spectrum implies a neutral hydrogen
231: column density $N_{\rm H}\gtrsim10^{22}$ \cdens\
232: \citep[e.g.,][]{ueda03}.
233: 
234: 
235: Finally, radio observations can detect the population of obscured AGNs
236: that are radio-loud.  Such radio galaxies were some of the first
237: objects detected at high redshifts \citep[for a review see][]{mcca93},
238: and have been identified out to $z=5.19$ \citep{vanbreu99}.
239: Radio-loud AGNs make up $\sim$10\% of the total AGN population, and
240: many are known to be obscured \citep[e.g.,][]{webs95}, however they
241: may represent a different mode of accretion from the radio-quiet AGNs
242: \citep{best05}.  For this study we concentrate on an
243: infrared-selected sample that is mostly radio-quiet.
244: 
245: Identification of obscured AGNs from their optical and X-ray
246: properties is complicated by the fact that these two classifications
247: do not always agree.  Some type 2 optical AGNs, which show no broad
248: emission lines, also show no absorption in their X-ray spectra and so
249: would be classified as type 1 X-ray AGNs \citep[e.g.,][]{mate05}.
250: Conversely, some type 1 optical AGNs show X-ray absorption
251: \citep[e.g.,][]{matt02}.  The observations of these anomalous objects
252: are quite robust and are not simply due to measurement errors.  We
253: do not expect a perfect correlation between dust extinction and gas
254: attenuation, but geometric or physical explanations for these
255: observed properties are not yet clear.
256: 
257: However, for $\gtrsim$70\%--80\% of AGNs, the optical and X-ray
258: classifications correspond \citep{tozz06,cacc04}, suggesting that in
259: most cases the absorption of X-rays emitted close to the central
260: engine is related to larger-scale obscuration of the broad-line
261: region.  In this paper, our classifications of IR-selected AGNs as
262: type 1 and type 2 are initially based on optical and IR colors
263: but are verified by measuring absorption in the X-rays.  We will
264: generally use the term ``obscuration'' to refer to dust extinction
265: observed in the ultraviolet (UV), optical, and IR, and ``absorption'' to refer to neutral
266: gas absorption in the X-rays.
267: 
268: \subsection{Obscured AGNs in the infrared}
269: 
270: The optical and X-ray selection techniques described above generally
271: require bright sources or long integrations to observe optical narrow
272: lines or X-ray absorption.  With the launch of the {\it Spitzer Space
273: Telescope}, the IR provides a new, highly sensitive window
274: to identify obscured AGNs, using new techniques to select AGNs based
275: on mid-IR colors \citep[e.g.,][hereafter S05]{lacy04, ster05}.  IR emission
276: is produced by the reprocessing of nuclear luminosity by surrounding
277: dust, and is not as strongly affected as optical or UV light by dust
278: extinction.  Therefore IR criteria can identify many AGNs that
279: are not detected in the optical or X-rays, because the optical lines
280: are highly extincted, or the X-ray emission is too faint to observe
281: without long exposures.
282: 
283: \defcitealias{ster05}{S05}
284: 
285: Recent works have identified populations of obscured AGNs among
286: IR-selected samples.  Using near-IR data from the Two Micron All Sky
287: Survey, \citet{cutr02} identified 210 red AGNs at $z<0.7$, and
288: \citet{wilk02} showed that most of these objects have X-ray properties
289: consistent with absorption of $N_{\rm H}=$(0.1--1)$\times10^{23}$
290: \cdens.  In the mid-infrared, \citet{lacy04} used the $\spitzer$ First
291: Look Survey to select $\sim$2000 candidate AGNs based on their
292: \spitzer\ Infrared Array Camera (IRAC) colors. Of these,
293: \citet{lacy04} identified 16 objects from their optical and mid-IR
294: properties that are likely to be luminous obscured AGNs at
295: $z\lesssim0.7$.  \citet{lacy07} obtained optical spectra for a sample
296: of 77 IR-selected AGNs and found that 47\% had broad emission
297: lines and 44\% had high-ionization narrow emission lines, while 9\%
298: had no AGN spectral signatures.  Similarly, \citet{mart06} used
299: the Multiband Imaging Photometer (MIPS) for \spitzer\ 24 $\mu$m and radio data
300: to select 21 luminous, obscured quasars at $z\sim2$, for which
301: follow-up optical spectroscopy showed that 10 of these objects had
302: narrow emission lines characteristic of type 2 optical AGNs, while the
303: remainder had no emission lines.  These optical spectra are consistent
304: with obscuration of the nucleus, although it is important to note that
305: such objects may still have broad lines in the rest-frame optical that
306: are redshifted out of the observed spectrum.  \citet{alon06} used
307: mid-IR colors to select 55 candidate obscured AGNs in the extremely
308: deep Great Observatories Origins Deep Survey (GOODS) fields.  Also
309: using GOODS data, \citet{dadd07comp} identified $\sim$100 AGNs based
310: on excess 24 \micron\ emission above that expected for star formation,
311: and used X-ray stacking to infer the presence of a significant
312: population in the sample of highly obscured ($N_{\rm H}>10^{24}$ \cdens) AGNs .  A
313: large IR-selected sample of obscured AGNs comes from \citet{poll06},
314: who used IRAC observations from the SWIRE survey in the 0.6 deg$^2$
315: Lockman Hole to select 120 obscured AGN candidates based on their
316: optical to IR spectral energy distributions (SEDs).  In the \bootes\
317: field, \citet{brow06} identified several hundred candidate $z>1$ type
318: 2 quasars, by selecting 24 $\mu$m MIPS sources with faint, extended
319: optical counterparts.
320: 
321: In the X-rays, a few hundred type 2 AGNs have been found in the
322: extremely deep, pencil-beam Chandra Deep Fields (CDFs)
323: \citep[e.g.,][]{trei04, zhen04, trei05, dwel05, dwel06, tozz06}, and
324: some have been identified with AGN counterparts selected in the IR
325: \citep{alon06} or submillimeter \citep{alex05}.  However, these narrow
326: fields miss rarer, more luminous objects.  Wide-field surveys offer
327: the best opportunity to select a large sample of AGNs with moderate to
328: high luminosity ($10^{45}\lesssim L_{\rm bol}\lesssim 10^{47}$ \ergs), moderate
329: obscuration ($10^{22}<N_{\rm H}<10^{23}$ \cdens), and high redshifts
330: ($0.7<z<3$), which is the goal of the present study.
331: 
332: The 9 deg$^2$ multiwavelength survey in the NOAO Deep Wide-Field
333: Survey region in \bootes\ is uniquely suited for identifying large
334: numbers of such obscured AGNs.  In this study, we develop IRAC and
335: optical selection criteria for finding obscured AGNs, and then use the
336: available multiwavelength data, principally X-rays, to confirm the
337: selection and to measure properties such as accretion luminosity and
338: absorbing column density.  To this end, we analyze a sample of 1479
339: IR-selected AGNs at $0.7<z\lesssim3$ for which we have spectroscopic
340: and/or photometric redshift estimates, and we select 640 candidate
341: luminous, obscured AGNs.
342: 
343: This paper is organized as follows.  In \S\ \ref{obs} we describe the
344: \bootes\ multiwavelength observations, and in \S\ \ref{sample} we
345: discuss the sample of IR-selected AGNs.  In \S\ \ref{anal} we develop
346: criteria based on optical-IR colors for selecting obscured AGNs.  In
347: \S\ \ref{tests} we confirm these selection criteria using the X-ray
348: and optical properties of these objects, and measure X-ray
349: luminosities and absorbing column densities.  In \S\ \ref{photoz} we
350: verify the photometric redshift estimates, and in \S\ \ref{caveats} we
351: discuss contamination and incompleteness in the IR-selected AGN
352: samples.  In \S\ \ref{discussion} we place the population of
353: IR-selected AGNs in the context of the known and expected populations
354: of obscured and unobscured objects, and in \S\ \ref{summary} we
355: summarize our results. Throughout this paper we use a cosmology with
356: $\Omega_{\rm m}=0.3$, $\Omega_{\Lambda}=0.7$, and $H_0=70$ km$^{-1}$
357: s$^{-1}$ Mpc$^{-1}$.  Unless otherwise noted, we use the Vega system
358: for optical and infrared magnitudes.
359: 
360: 
361: \section{Bo\"{o}tes data set}
362: \label{obs}
363: The 9 deg$^{2}$ survey region in \bootes\ of the NOAO Deep Wide-Field
364: Survey \citep[NDWFS;][]{jann99} is unique among extragalactic
365: multiwavelength surveys, in its wide field and uniform coverage using
366: space- and ground-based observatories, including the {\it Chandra
367: X-Ray Observatory} and \spitzer.  Extensive optical spectroscopy makes
368: this field especially well suited for studying the statistical properties
369: of a large number of AGNs (C.S. Kochanek et.~al. 2008, in preparation).
370: 
371: The \bootes\ field  was observed by the \spitzer\ IRAC Shallow
372: Survey \citep{eise04}.  Three or more 30 s exposures were taken per
373: position, in all four IRAC bands (3.6, 4.5, 5.8, and 8 $\mu$m), with
374: $5\sigma$ flux limits of 6.4, 8.8, 51, and 50 $\mu$Jy, respectively.
375: The sample includes $\approx$370,000 sources detected at 3.6 $\mu$m,
376: including $>80\%$ of the X-ray sources.  We limit our IRAC sample to
377: $\approx$15,500 objects that have 5$\sigma$ detections in all four
378: bands and at least three good exposures (for reliable rejection of cosmic
379: rays), which cover an area of 8.5 deg$^2$.
380: 
381: X-ray data are taken from the X\bootes\ survey, which is a mosaic
382: comprised of 126 5 ks \chandra\ ACIS-I exposures and is the largest
383: contiguous field observed to date with \chandra\ \citep{murr05}.  Due
384: to the shallow exposures and low background in the ACIS CCDs, X-ray
385: sources can be detected to high significance with as few as four
386: counts. In this field, 3293 X-ray point sources with four or more counts
387: are detected \citep{kent05}, of which 2960 lie within the area covered
388: by IRAC.  Optical identifications for the X-ray sources are presented
389: in \citet{brand06a}.  We also use radio data from the Very Large Array
390: (VLA) FIRST 20 cm radio survey \citep{beck95}, which detects 930
391: sources in the area covered by the IRAC, to a limiting flux of
392: $\approx$1 mJy.
393: 
394: Optical photometry in the \bootes\ field comes from the NDWFS, which
395: used the Mosaic-1 camera on the 4-m Mayall Telescope at Kitt Peak
396: National Observatory.  Deep optical imaging was performed over the
397: entire 9.3 deg$^{2}$ in the $B_{W}$, $R$, and $I$ bands with 50\%
398: completeness limits of 26.7, 25.5, and 24.9 mag, respectively
399: \citep{jann99}.  Optical spectroscopy in the \bootes\ field comes from
400: the AGN and Galaxy Evolution Survey (AGES), which uses the Hectospec
401: multifiber spectrograph on the MMT.  We use AGES Data Release 1 (DR 1)
402: and Internal Release 2 (IR 2), which consist of all the AGES spectra
403: taken in 2004--2005.  In AGES DR 1, targets include (1) all extended
404: sources with $R\le 19.2$ (2) a randomly selected sample of 20\% of all
405: extended sources with $19<R\le 20$, and (3) all extended sources with
406: $R\le 20$ and IRAC 3.6, 4.5, 5.8, and 8.0 $\mu$m magnitudes $\le$15.2,
407: 15.2, 14.7, and 13.2, respectively.  In addition, (4) fainter sources
408: were observed, selected mainly from objects with counterparts of
409: \chandra\ X-ray sources \citep{murr05, brand06a, kent05}, radio
410: sources from the VLA FIRST survey, and objects selected from 24 $\mu$m observations with
411: MIPS (E.\ Le Floc'h et al. 2008, in preparation).  AGES IR 2 contains
412: $I$-selected targets with $I\le 21.5$ for point sources and $I\le
413: 20.5$ for extended sources.  Because X-ray sources were preferentially
414: targeted, the survey contains a large number of spectral
415: identifications for distant AGNs.  Galaxy spectra are classified by
416: template fits into three categories: optically normal galaxies,
417: broad-line AGNs (BLAGNs), and narrow-line AGNs (NLAGNs).
418: 
419: We use the optical and IRAC photometry described in \citet{brod06},
420: for which optical and IRAC sources are matched using a 1\arcsec\
421: radius.  We then match the \chandra\ X-ray sources, AGES optical
422: spectra, and VLA FIRST 20 cm sources to the IRAC sources, using radii of
423: 3.5\arcsec, 2\arcsec, and 2\arcsec, respectively.  There were 1298
424: matches to X-ray sources, 6450 matches to AGES spectra, and 196
425: matches to radio sources. There were no sources with multiple
426: matches (owing to the $\sim$2\arcsec\ point-spread function [PSF] of the IRAC images, no two
427: sources in the $5\sigma$ catalog are closer than 3\arcsec).
428: 
429: To estimate the number of spurious matches, we offset the positions of
430: the IRAC sources by 16\arcsec\ and re-perform the source matching.
431: This places the IRAC sources at ``random'' positions away from the
432: X-ray or AGES sources but retains their surface density distribution
433: on larger scales.  We re-perform the matching with offsets in eight
434: directions and derive the median number of matches from these eight
435: trials.  For the full sample of $\approx$15,500 IRAC sources detected
436: at $5\sigma$ in all four bands, we expect spurious matches to 20 X-ray
437: sources, 45 AGES spectra, and 3 radio sources.  In this paper we focus
438: on a sample of 1479 IR-selected AGNs (\S\ \ref{sample}), for which we
439: expect spurious matches to only 2 X-ray sources, 4 AGES spectra, and no
440: radio sources.  Details of the IRAC $5\sigma$ sample and matches to
441: the optical and X-ray catalogs are given in Table \ref{tblsample}.
442: 
443: 
444: 
445: 
446: 
447: 
448: 
449: To calculate luminosities and to fit models to SEDs for the objects in
450: our sample, we require estimates of redshift.  For all objects with
451: AGES spectra, which have $17.5<R<22$, we have reliable spectroscopic
452: redshifts with uncertainties of $\sigma_{z} < 0.001$.  However, 51\%
453: of our IR-selected AGNs (as defined in \S\ \ref{sample}) do not have
454: optical spectra, either because they were not spectroscopically
455: targeted, or because they are fainter than the AGES spectroscopic
456: limits.  For these, we use photometric redshifts from the catalog of
457: \citet{brod06}, who use fluxes from the four IRAC bands, as well as
458: $B_{W}$, $R$, and $I$ in the optical. photo-$z$'s are obtained through
459: a hybrid technique; for objects with strong spectral features such as
460: most optically normal galaxies, redshifts are estimated using template
461: fitting.  For objects (such as AGNs) that have more featureless SEDs, an
462: artificial neural net is used.  Uncertainties in the photo-$z$ are
463: $\sigma_{z}=0.06(1+z)$ for galaxies at $z<1$ and
464: $\sigma_{z}=0.12(1+z)$ for optically bright AGNs.  Photo-$z$
465: uncertainties increase for fainter sources due to larger photometric
466: errors.  In \S\ \ref{photoz} we address possible systematic errors in
467: the photo-$z$'s and show that there are no large biases in the
468: photo-$z$ estimates that would significantly affect our conclusions.
469: However, because of the limited accuracy of the photo-$z$'s, in this
470: paper we do not use them to measure precise quantities such as the
471: evolution of the obscured AGN fraction with luminosity or redshift.
472: 
473: 
474: 
475: \begin{figure}
476: \epsscale{1.2}
477: \label{figsel}
478: \plotone{f1a.eps}
479: \plotone{f1b.eps}
480: \plotone{f1c.eps}
481: \caption{Illustration of mid-IR AGN color selection.  On the left, the
482:   solid lines show the rest-frame spectrum consisting of the sum of a
483:   starburst \citep[][dot-dashed lines]{sieb07} plus AGN power-law
484:   ($\alpha_\nu=-1$) templates.  The three panels are for objects at
485:   $z=0.3$, 0.7, and 2, and AGN contribution to the rest-frame mid-IR
486:   (3--8 \micron) emission $f_{\rm AGN}=0.2$, 0.8, and 0.95,
487:   respectively.  Fluxes (shown by the filled circles) are determined
488:   by convolving the spectra with the responses of the four IRAC bands
489:   (shown).  The dotted spectra and open squares in the left panels
490:   show the same model with the AGN power law extincted for $A_V=20$,
491:   with a Galactic extinction curve.  The panels at right show
492:   corresponding locations on the color-color diagram for these models
493:   along with the \citetalias{ster05} AGN color selection region.  The
494:   solid lines in the color-color diagrams represent the colors as a
495:   function of increasing $f_{\rm AGN}$ from 0 to 0.95 for each
496:   redshift (the blue arrows show $f_{\rm AGN}=0$, 0.5 and 0.8).  Note
497:   how increasing the contribution of the red AGN power law brings the
498:   objects into the \citetalias{ster05} color selection region for all
499:   three redshifts, with the objects entering the selection region at
500:   $f_{\rm AGN}=0.3-0.5$.  The dotted line in the right panel shows the
501:   effects of obscuration of the nuclear component (for $A_V=0-100$),
502:   for the spectrum with $f_{\rm AGN}=0.8$, with the open arrows
503:   representing $A_V=10$, 20, and 50.  Very high dust extinction
504:   $A_V\sim30$--50 will move the object out of the \citetalias{ster05}
505:   selection region.
506: \label{figcol}
507: \vskip0.2cm
508: }
509: 
510: \end{figure}
511: 
512: 
513: \begin{figure}
514: \epsscale{1.2}
515: \plotone{f2a.eps}
516: \plotone{f2b.eps}
517: \plotone{f2c.eps}
518: \caption{Distributions of (a) redshift, (b) \lirac, and (c) $L_R$ for
519:   the 1929 infrared-selected AGNs in the sample.  \lirac\ and $L_R$
520:   are $\nu L_\nu$ in the observed $R$ and IRAC 4.5 $\mu$m bands,
521:   respectively (see \S\ \ref{lum}).  Many of the objects at $z<0.7$
522:   (dotted line) are not AGNs but ``normal'' galaxies; above this
523:   redshift these objects are typically too faint to be detected in all
524:   four bands of the IRAC Shallow Survey.  We restrict our analysis to
525:   the 1479 objects with $z>0.7$, shown by the solid line.
526: \label{figdist}}
527: \vskip0.2cm
528: \end{figure}
529: 
530: 
531: \section{Infrared-selected AGN sample}
532: 
533: 
534: \label{sample}
535: The AGN sample used in this paper is selected in the mid-IR, which is
536: less affected by obscuration than optical or soft X-ray emission.  In
537: the (rest-frame) near- to mid-IR from 1 to 10 $\mu$m, AGNs have
538: markedly different SEDs from normal or starburst galaxies. AGNs
539: typically have a roughly power law continuum in the near- to mid-IR,
540: $S_\nu\propto \nu^{\alpha_{\nu}}$, where $\alpha_{\nu} \simeq -1$
541: \citep[e.g., S05;][]{glik06}.  In contrast, normal and starburst
542: galaxies have bluer continua in the rest-frame mid-IR, due to the fact
543: that the spectrum from the stellar population of the galaxy peaks at
544: $\simeq$1.6 $\mu$m, and falls at longer wavelengths.  In addition,
545: star-forming galaxies have prominent emission features at 3--10 $\mu$m
546: due mainly to lines from polycyclic aromatic hydrocarbons (PAHs) in dust
547: \citep{puge89}.  This difference in SEDs allows us to effectively
548: distinguish AGN-dominated objects from normal and starburst galaxies
549: using observed colors in the mid-IR.
550: 
551: \citetalias{ster05} developed a set of IRAC color-color selection
552: criteria based on the IRAC Shallow Survey photometry and AGES spectra,
553: described in \S\ \ref{obs}.  In this paper we use those criteria
554: to select AGNs.  To illustrate the \citetalias{ster05} color-color
555: selection, we show in Fig.\ \ref{figcol} the IRAC $[3.6]-[4.5]$ and
556: $[5.8]-[8.0]$ colors for a two-component template spectrum consisting
557: of a starburst galaxy \citep{sieb07} plus AGN power law with
558: $\alpha_\nu=-1$.  We show these colors for three redshifts and for
559: various values of the fraction $f_{\rm AGN}$ of the rest-frame 3-10
560: $\mu$m luminosity that is emitted by the AGN.  Because the colors of
561: the power law AGN spectrum are constant with redshift, increasing
562: $f_{\rm AGN}$ moves the colors into the \citetalias{ster05} AGN
563: selection region, regardless of the redshift of the source.  Fig.\
564: \ref{figcol} also shows the effect of dust extinction of the nuclear
565: component, for a Galactic extinction curve \citep{pei92}.  For
566: $A_V\gtrsim30$--50 (depending on redshift), extinction can cause the
567: IRAC colors to move out of the \citetalias{ster05} selection region,
568: even for $f_{\rm AGN}$ as high as 0.8.
569: 
570: 
571: \input{tab1.tex}
572: 
573: 
574: We stress that this color-color technique does not select all AGNs.
575: In the \bootes\ data \citep{gorj07}, as well as the extended Groth
576: strip \citep[EGS;][]{barm06}, only half of X-ray detected AGNs were identified
577: using the \citetalias{ster05} IRAC color-color criteria.  This is likely
578: due to the fact that some X-ray sources are too faint to be detected
579: in all four IRAC bands, while others might not have red power-law
580: mid-IR spectra.  Recent mid-IR spectroscopy of type 2 quasars with the
581: Infrared Spectrograph on \spitzer\ has shown that most luminous ($L_{\rm
582: X}>10^{44}$ \ergs) X-ray selected type 2 quasars have relatively
583: featureless mid-IR spectra \citep{stur06, weed06c}.  Still, many AGNs in
584: ultraluminous infrared galaxies (ULIRGs) show a variety of spectral
585: shapes including PAH emission and deep silicate absorption features
586: \citep{spoo05, buch06, bran07spec_aph}, which may
587: indicate deep obscuration of the nuclear IR emission.  Therefore, the
588: completeness of AGN color-color selection is still unclear.  The key
589: point for this study is that while color-color selection may miss
590: many AGNs, there should be little contamination in the
591: AGN color-color region from starburst-powered objects, particularly
592: for objects with $z>0.7$ (see \S\ \ref{lum}).  Sample completeness and
593: contamination are discussed in more detail in \S\ \ref{caveats}.
594: 
595: Our sample of IR-selected AGNs contains objects that have: (1)
596: $5\sigma$ detections in all four IRAC bands as well as the $R$ band of
597: the NOAO DWFS catalog, which we use to calculate optical luminosities;
598: (2) IRAC colors that fall in the \citetalias{ster05} AGN selection
599: region; and (3) spectroscopic redshifts from AGES or
600: photometric redshifts from the \citet{brod06} catalog, with $z_{\rm
601: phot}>0$.  These criteria select 1929 objects.  Only 13 additional
602: objects are not detected in the $R$ band but meet all the other
603: criteria, so this requirement has little effect on our results.
604: Excluding all objects with $z<0.7$ to minimize contamination by normal
605: galaxies (see \S\ \ref{lum}) leaves a sample of 1479 IR-selected AGNs,
606: of which 1469 have detections in all three NDWFS optical bands.
607: Details of AGES spectra and X-ray matches to the objects are shown in
608: Table \ref{tblsample}.
609: 
610: 
611: \section{Optical/IR SEDs and obscured AGN selection}
612: \label{anal}
613: 
614: In this section we calculate optical and IR luminosities for the
615: IR-selected AGNs, and perform template fits to the optical and IR SEDs
616: that provide evidence that roughly half of the sample has significant
617: nuclear extinction.  We then develop a simple optical-IR color
618: criterion for selecting obscured AGNs.  We show that the obscured AGN
619: candidates display absorption in their average X-ray spectra and have
620: the optical characteristics of normal galaxies, while the unobscured
621: candidates are on average X-ray unabsorbed and have optical colors
622: and morphologies typical of unobscured AGNs.
623: 
624: 
625: \begin{figure}
626: \epsscale{1.2}
627: \plotone{f3.eps}
628: \caption{Template spectra used for SED fits, normalized at 0.8
629:  \micron. The top panel shows the unabsorbed AGN template, with the
630:  mean SED and dispersion from \citet{rich06} for comparison.  The
631:  hatched region at right shows the allowed values of
632:  $\alpha_{\nu}$. The bottom panel shows elliptical (red solid line),
633:  Sb (blue dashed line) and starburst (green dotted line) galaxy
634:  templates.  See \S~\ref{lum} for details of the models.
635: \label{figtemp}
636: }
637: \vskip0.5cm
638: \end{figure}
639:  
640: 
641: 
642: 
643: \subsection{Luminosities and model fits}
644: \label{lum}
645: 
646: 
647: For each of the 1479 AGNs in our sample, we calculate the observed mid-IR
648: and optical luminosity densities using
649: \begin{equation}
650: L_\nu(\nu_{\rm rest})=\frac{4\pi d_{\rm L}^2}{1+z} S_\nu(\nu_{\rm obs}),
651: \end{equation}
652: where $d_{\rm L}$ is the luminosity distance for a given redshift in
653: our adopted cosmology \citep{hogg99}, $S_\nu$ is the flux density in \fluxhz,
654: and $\nu_{\rm obs}$ and $\nu_{\rm rest}$ are the observed and
655: rest-frame frequencies, respectively, where $\nu_{\rm
656: rest}=(1+z)\nu_{\rm obs}$.  Throughout the paper
657: we present optical and IR luminosities in terms of the bolometric
658: luminosity of the Sun, $L_{\sun}=3.83\times10^{33}$ \ergs.
659: 
660: We generally define luminosities and colors in terms of
661: the observed (rather than rest-frame) photometric bands; the
662: relationship between  rest-frame luminosity density
663: $L_\nu(\nu_{\rm rest})$ and the observed-frame luminosity density
664: $L_\nu(\nu_{\rm obs})$ is 
665: \begin{equation}
666: L_\nu(\nu_{\rm obs})=\frac{L_\nu(\nu_{\rm rest})}{(1+z)^{\alpha_\nu}},
667: \label{eqnrestobs}
668: \end{equation}
669: where $\alpha_\nu$ is the power-law index ($S_\nu\propto
670: \nu^{\alpha_{\nu}}$) for the spectrum between $\nu_{\rm obs}$ and
671: $\nu_{\rm rest}$.  Redshift estimates and detailed spectral shapes are
672: uncertain for many of the AGNs in our sample, so framing the selection
673: in terms of observed luminosities and colors makes our results less
674: subject to the details of $K$-corrections.
675: 
676: We define luminosities in each photometric band in terms of $\nu
677: L_{\nu}$, which unlike the luminosity density $L_\nu$, is not
678: strongly affected by corrections for redshift, at least for unobscured
679: quasars.  Eqn.~\ref{eqnrestobs} shows that for a typical broadband
680: quasar SED with IR power law with $\alpha_{\nu}=-1$, $\nu L_\nu$
681: remains constant with redshift.  For a typical optical continuum
682: (which is not exactly a power law, as described below), in the
683: redshift range we consider ($0.7<z<3$), the observed $\nu L_{\nu}$
684: varies by at most 0.25 dex.  In \S~\ref{kcor}, we estimate $K$-corrections and
685: show that they have no significant effect on our selection criteria.
686: 
687: For the mid-IR luminosity we use $L_{\rm 4.5\mu m}$, defined to be
688: $\nu L_{\nu}$ in the observed 4.5 $\mu$m IRAC band.  Because the
689: color-selected AGNs in our sample have similar IRAC SEDs, the \lirac\
690: is a simple and sufficiently accurate proxy for the total luminosity in the IRAC
691: bands.  In the optical, we use $L_{R}$, defined as $\nu L_{\nu}$
692: observed in the $R$ band centered on 6514 \AA.  The
693: distributions of the 1929 IR-selected AGNs in redshift, \lirac, and
694: $L_{R}$ are shown in Fig.\ \ref{figdist}.
695: 
696: We restrict our sample to AGNs at $z>0.7$.  At lower redshifts the
697: IRAC source counts are dominated by normal or star-forming galaxies
698: with relatively low luminosities ($L_{\rm 4.5\mu m}<10^{11}$
699: \lsun). Some of these objects may have red IRAC SEDs, for example, due
700: to heavy dust obscuration.  The model SED from \citet{sieb07} for the
701: heavily extincted starburst Arp 220 has IRAC colors that lie within
702: the \citetalias{ster05} AGN region, and less obscured sources can lie
703: close to  this region. Combined with photometric errors, this
704: results in a significant number of $z<0.7$ objects selected with the
705: \citetalias{ster05} criterion being either normal or starforming
706: galaxies.  By cutting our sample at $z>0.7$, however, we exclude most
707: of these ``normal'' galaxies as they are generally fainter than the
708: flux density limits in the 5.8 \micron, 8.0 $\mu$m, or $R$ bands
709: (heavily extincted starbursts, for example, are very faint in the
710: optical).  In addition, limiting the sample to $z>0.7$ allows for more
711: straightforward color selection of obscured AGNs, as shown in \S\
712: \ref{identify}.  Our final IR-selected AGN sample includes only the
713: 1479 IR-selected AGNs with $z>0.7$.
714: 
715: 
716: \begin{figure}
717: \epsscale{1.2}
718: \plotone{f4.eps}
719: \caption{Examples of fits to optical and IRAC photometry, for galaxy
720:    (red dot-dashed lines) and AGN (green dashed lines)
721:   spectral templates (see \S\ \ref{lum} for a description of the
722:   models).  These three example objects show a range
723: of best-fit values to the AGN power-law slope $\alpha_\nu$ (where $S_\nu
724: \propto \nu^{\alpha_\nu}$) and AGN template extinction $A_V$.
725: \label{figtempall}}
726: \end{figure}
727: 
728: 
729:  
730: \begin{figure}
731: \epsscale{1.2}
732: \plotone{f5.eps}
733: \caption{Distribution of best-fit $A_V$ for the IR-selected AGNs.  The
734:   far left bin corresponds to objects with $A_V\leq0.2$.  The three
735:   lines are for Galactic, LMC, and SMC extinction curves.
736: \label{figav}}
737: \end{figure}
738:  
739: To model the SEDs of the IR-selected AGNs, we fit the optical and IRAC
740: photometry of each source with spectral templates including AGN and
741: host galaxy components.    For the nuclear emission in the rest-frame
742: optical/UV, we use the AGN template of \citet{hopk07qlf}, which
743: consists of the composite SED of \citet{rich06}, with optical lines
744: taken from the SDSS composite quasar
745: template \citep{vand01}.  For $\lambda > 0.8$ $\mu$m, we use a power
746: law component.  Our grid of models includes 14 values of the slope
747: from $-2.2\leq \alpha_{\nu}\leq 0.4$.  We also include extinction of
748: the nuclear component, with a Galactic extinction curve \citep{pei92},
749: and $E_{B-V}/A_V=3.1$ for 12 logarithmically spaced values of
750: $0<A_V<32$.  This corresponds to a total of 168 separate AGN models.
751: 
752: For the host galaxy emission, we use two model galaxy templates
753: calculated using the PEGASE population synthesis code \citep{fioc97}.
754: The models are chosen so that at age 13 Gyr, they correspond to
755: observed low-redshift ellipticals and spirals.  The models differ in
756: their initial specific star formation rates ($5\times10^{-3}$ vs.\
757: $3.5\times10^{-4}$ $M_{\sun}$ Myr$^{-1}$ per unit gas mass in
758: $M_{\sun}$, for elliptical and Sb, respectively), the fraction of
759: stellar ejecta available for new star formation (0.5 vs. 1), and
760: extinction (none for the elliptical galaxy, disk extinction for the
761: Sb).  For simplicity, we use non-evolving spectra corresponding to an
762: age of 3 Gyr after formation.  Assuming that massive galaxies form at
763: $z>6$, this age roughly corresponds to the age of such a galaxy at
764: $z\sim1$--2 for our adopted cosmology.  At $\lambda > 0.8$ $\mu$m, the
765: models include either the quiescent galaxy spectrum, or the spectrum
766: of the starburst galaxy NGC 7714 \citep{sieb07}.  This gives a total
767: of four separate host galaxy models (E, Sb, E plus starburst, and Sb
768: plus starburst).  The quasar and galaxy template spectra are shown in
769: Fig.~\ref{figtemp}.
770: 
771: 
772: \begin{figure*}
773: \epsscale{1.1}
774: \plottwo{f6a.eps}{f6b.eps}
775: \plottwo{f6c.eps}{f6d.eps}
776: \caption{ (a) $L_{R}$ vs.\ \lirac\ and (b) \iiu\ vs. \lirac\ for
777:   IR-selected AGNs.  Objects with optical spectroscopic
778:   classifications as BLAGNs are shown in blue, and X-ray sources are
779:   shown as stars.  For clarity, only one out of every two objects is
780:   shown.  Empirical selection criteria to separate IRAGN 1s and IRAGN
781:   2s are shown by the dashed line.  Error bars show the median
782:   uncertainties for objects lacking spectroscopic redshifts (objects
783:   with spectroscopic redshifts have much smaller uncertainties in
784:   luminosity).  Luminosity uncertainties include both redshift and
785:   flux uncertainties; note that the error bars in (a) are not
786:   independent as the luminosity errors are dominated by uncertainty in
787:   $z$.  (c) shows the same points as (b), but includes contours of
788:   source density that clearly show the bimodal distribution in \iiu.
789:   (d) show the same distribution as (b), but includes sources at
790:   $z<0.7$ with various optical spectroscopic classifications.
791: \label{figuvir}}
792: \vskip0.3cm
793: \end{figure*}
794: 
795: For all AGN and galaxy models, we account for neutral hydrogen
796: absorption in the intergalactic medium by setting the templates equal
797: to zero blueward of the Lyman limit (912 \AA), which is probed by the
798: shortest-wavelength ($B_W$) band only for redshifts $z>2.7$.  Additional
799: absorption by Ly$\alpha$ becomes significant at $z\gtrsim3$ and
800: depends strongly on redshift; this absorption can be as strong as 50\%
801: at $\lesssim4$. \citep[e.g.,][]{beck07}. However, only 42 (3\%) of the
802: objects in our sample lie at $z>3$, so for simplicity we ignore
803: redshift-dependent Ly$\alpha$ absorption in our templates.
804: 
805: 
806: For each IR-selected AGN in our sample, we perform $\chi^2$ fits to
807: the optical and IRAC photometry with the redshifted sum for each
808: combination of starburst template and AGN power law (for this analysis
809: we omit the 10 sources that do not have detections in all three NDWFS
810: bands).  We leave the normalizations of the AGN and galaxy components
811: as free parameters, and we convolve the template spectra with the
812: appropriate Mosaic-1 and IRAC response functions\footnotemark.  From
813: the template with the lowest $\chi^2$, we derive the best fit
814: $\alpha_\nu$ and $A_V$ for the AGN.  Example fits are shown in Fig.\
815: \ref{figtempall}.  From the best-fit template, we also calculate
816: $K$-corrected luminosities $L_{2500\; \AA}$ and $L_{2\mu m}$,
817: corresponding to the rest-frame $\nu L_{\nu}$ at 2500 \AA, and 2
818: $\mu$m, respectively.  These wavelengths are probed by the optical and
819: IRAC photometry for all sources with $0.7<z<2.7$. The effects of
820: $K$-corrections are discussed in \S~\ref{kcor}.
821: 
822: There is some evidence that the extinction in AGNs is best
823: described by curves observed for the LMC and SMC, which have greater
824: extinction in the UV than is observed in the Galaxy.  To check the
825: dependence of the fit parameters on the choice of extinction curve, we
826: re-perform the SED fits using LMC and SMC curves \citep{pei92}.  These
827: do not significantly alter the quality of the fits, although the SMC
828: curve gives somewhat lower $A_V$ estimates for some objects with
829: $A_V\sim 1$.
830: 
831: 
832: \footnotetext{
833: {\tt http://www.noao.edu/kpno/mosaic/filters/filters.html} and  {\tt http://ssc.spitzer.caltech.edu/irac/\\spectral\_response.html}.}
834: 
835: 
836: 
837: 
838: 
839: 
840: 
841: 
842: \subsection{Color selection of obscured AGNs}
843: \label{identify}
844: 
845: The distribution in the best-fit $A_V$ from the optical/IR SED fits is
846: shown in Fig.~\ref{figav}.  The AGN extinctions are
847: bimodal, with a large fraction of sources having $A_V\gtrsim1$,
848: suggesting that the IRAC selection includes many obscured AGNs.  We do
849: not expect moderate extinction to strongly affect the IRAC color-color
850: selection because the (rest-frame) near- and mid-IR emission that is
851: probed with IRAC suffers relatively little obscuration by gas or dust
852: compared to the optical, UV, or soft X-ray bands (although the near-IR
853: can be extincted for sufficiently large $A_V$).  As shown by the
854: models in the lower two panels of Fig.~\ref{figtempall}, nuclear
855: emission with significant extinction in the optical can still dominate
856: over emission from the host galaxy in the IRAC bands.  Because the
857: extinction curve for the mid-IR is relatively flat
858: \citep[e.g.,][]{pei92, inde05}, extinction does not significantly
859: affect the shape of the observed IRAC spectrum.  Therefore, IRAC
860: color-color selection can identify AGNs even for $A_V\sim30$, as shown
861: in Fig.~\ref{figcol}.
862: 
863: 
864: 
865: \begin{figure}
866: \epsscale{1.2}
867: \plotone{f7.eps}
868: \caption{Distribution in $R$ for the IRAGN 1 and IRAGN 2 subsets.  Gray lines show the distribution for sources with AGES spectroscopy.  
869: \label{figr}}
870: \end{figure}
871:  
872: \begin{figure}
873: \epsscale{1.2}
874: \plotone{f8.eps}
875: \caption{Optical/mid-IR color ($R-[4.5]$) vs. redshift of IRAGN 1s,
876: compared to type 1 AGN selected from other works. Note that blue
877: colors are the top and red at the bottom, to correspond to the plots
878: in Fig.~\ref{figuvir}.  Squares and density contours are for IRAGN 1s.
879: Red circles and blue stars show type 1 AGN from \citet{rich06} and
880: \citet{hatz05}, respectively.  The figure shows that most the
881: color-selected IRAGN 1s show a similar distribution in color versus
882: redshift as other samples, although our sample includes more
883: moderately reddened AGNs (with $R-[4.5] > 5.5$).
884: \label{figcol_noabs}}
885: \vskip0.3cm
886: \end{figure}
887:  
888: \begin{figure}
889: \epsscale{1.2}
890: \plotone{f9.eps}
891: \caption{ $L_{R}/L_{\rm 4.5 \mu m}$ vs. redshift for a template of AGN
892:   plus host galaxy (as described in \S\ \ref{lum}), with different
893:   values of the extinction on the AGN template.  Because at higher $z$
894:   the optical observations probe shorter wavelengths, where the
895:   extinction curve is steeper, $L_{R}/L_{\rm 4.5 \mu m}$ varies more
896:   strongly with $A_V$.  Therefore candidate obscured AGN can be more easily
897:   selected on the basis of optical-IR color at $z>0.7$, to which we
898:   restrict our sample.  We also show, for comparison, the colors of a
899:   template for the starburst M82 \citep{sieb07}, and the obscured AGN
900:   selection cut shown in Fig.~\ref{figuvir}.
901: \label{figtempav}}
902: \vskip0.3cm
903: \end{figure}
904:  
905: 
906: \begin{figure}
907: \epsscale{1.2}
908: \plotone{f10a.eps}
909: \plotone{f10b.eps}
910: \caption{ Similar to Fig.\ \ref{figuvir}, with $K$-corrected
911:   luminosities $L_{\rm 2500\; \rm \AA}$ and $L_{\rm 2 \mu m}$
912:   calculated from template fits (\S\ \ref{lum}).  IRAGN 1s
913:   (blue squares) and IRAGN 2s (red circles) are classified as defined
914:   in \S\ \ref{identify} and shown in Fig. \ref{figuvir}.  For clarity,
915:   only one out of every two objects is shown.  Filled symbols show objects
916:   with optical spectroscopic redshifts, while empty symbols have only
917:   photo-$z$'s.  Errors are as in Fig.\ \ref{figuvir}.
918: \label{figuvirk}}
919: \vskip0.5cm
920: \end{figure}
921:  
922: 
923: 
924: Since extinction by dust is much stronger in the optical than the IR,
925: simple optical and IR color criteria (rather than detailed SED fits)
926: can be used to select obscured objects.  In Fig.~\ref{figuvir} (a) we
927: plot \lr\ versus \lirac\ for the IR-selected AGNs.  This plot shows
928: two separate distributions of sources.  The first has \lr\ values that
929: rise along with the \lirac\ and contains nearly all (96\%) of the
930: IR-selected AGNs that have optical spectra of BLAGNs.  A total of 79\%
931: of these objects have relatively low extinction $(A_V<1)$ from the SED
932: fits, so we associate them with candidate unobscured AGNs and classify
933: them as type 1 IR-selected AGNs (IRAGN 1s).  The second population has
934: lower values of \iiu, and 98\% with $A_V\geq 1$, so we associate these
935: with candidate obscured AGNs (IRAGN 2s).  Unfortunately, spectroscopic
936: classification is of little help with the IRAGN 2s, since most of them
937: are fainter than the spectroscopic limits of AGES (Fig.\
938: \ref{figr}). The 4\% of the BLAGNs that lie in this ``obscured'' region
939: have red colors in the optical, with $R-I\sim0.7$--1.2, compared to
940: $R-I\sim0.3$ for a typical unreddened BLAGN.
941: 
942: 
943: We elucidate the distinction between the subsets by plotting the
944: quantity \iiu\ (or equivalently in magnitudes, $R-[4.5]$) versus
945: \lirac\ in Fig.\ \ref{figuvir} (b) and (c).  The contours in
946: Fig.~\ref{figuvir}(c) show that the distribution in \iiu\ is bimodal,
947: so that there are two distinct populations.  We empirically define
948: the boundary between these two populations to be $\log{(L_{R}/L_{\rm
949: 4.5 \mu m})}=-0.4$, corresponding to $R-[4.5]=6.1$ (Vega) or $R-[4.5]=3.1$
950: (AB), as shown in all four plots in Fig.\ \ref{figuvir}.  We select
951: this boundary (1) to divide the region populated by AGES BLAGNs from
952: the region with few BLAGNs, and (2) to bisect the bimodal distribution
953: in \iiu\ shown in Fig.~\ref{figuvir}(c).  Because this boundary is
954: based in part on the AGES spectral classifications, it is possible that our
955: selection may be biased by the fact that AGES did not target optically
956: fainter sources.  However, as we show in \S\ \ref{xrayres}, X-ray
957: analysis independently confirms the division at
958: $\log{(L_{R}/L_{\rm4.5\mu m})}=-0.4$.  This criterion selects 839
959: IRAGN 1s and 640 IRAGN 2s.
960: 
961:  The IRAGN 1s have mid-IR/optical colors similar to those found for
962: other samples of type 1 AGNs.  Fig.~\ref{figcol_noabs} shows the
963: distribution in $R-[4.5]$ for the IRAGN 1s, with comparisons to
964: samples from \citet{rich06} and \citet{hatz05}.  Most of the IRAGN 1s
965: show the same trend in redshift and color as these previous samples,
966: although the IRAGN 1s include more moderately reddened AGNs (with
967: $R-[4.5]>5.5$), which make up 24\% of the total number of IRAGN
968: 1s.
969: 
970: The color distribution in Fig.~\ref{figuvir} can be interpreted in
971: terms of how the observed $R$ and IRAC fluxes for AGNs change with
972: extinction and redshift.  In Fig.~\ref{figtempav} we show \iiu\
973: versus $z$ for a template including an elliptical host galaxy plus AGN
974: (with $\alpha_\nu=-1$ and various $A_V$) as described in \S\
975: \ref{lum}.  The model AGN has an unabsorbed, rest-frame $R$-band
976: luminosity 5 times that of the host galaxy.  Fig.~\ref{figtempav}
977: shows that extinction of the AGN component decreases the observed
978: \iiu. This decrease becomes larger at higher $z$ because the $R$ band
979: probes shorter wavelengths in the rest-frame UV, where dust extinction
980: is greater.  Because obscured AGNs at higher redshift tend to have higher \lirac\ (owing to the flux limits of the survey), more
981: luminous objects appear redder in the observed \iiu\ for the same
982: $A_V$.  This explains the decrease in \iiu\ with \lirac\ observed in
983: Figs.~\ref{figuvir}(b) and (c).  
984: 
985: For starburst galaxies (shown by the M82 template), \iiu\ changes even
986: more strongly with redshift; at low $z$, starbursts, obscured AGNs,
987: and unobscured AGNs can have similar values of \iiu.  However, for most
988: redshifts, all but the most extincted starbursts are not selected by
989: the \citetalias{ster05} IRAC color-color criteria; the colors for M82
990: and NGC 7714 fall in the \citetalias{ster05} region only at $z\gtrsim
991: 3$.  We discuss possible contamination from these objects in
992: \S~\ref{contamination}.
993: 
994: 
995: 
996: \subsection{$K$-corrected colors}
997: \label{kcor}
998: In Fig.\ \ref{figuvirk} we include the $K$-corrections to the
999: IR-selected AGN luminosities and plot \luv\ versus \lumtwo.  The
1000: $K$-corrections have negligible effect on the color classification; if
1001: we apply an equivalent empirical boundary to separate IRAGN 1s and 2s
1002: using the $K$-corrected luminosities [$\log{(L_{\rm 2500\; \AA}/L_{\rm
1003: 2\mu m})}=-0.55$, shown by the dashed line in Fig.~\ref{figuvirk}],
1004: only 90 of the 1479 objects (6\%) change their
1005: classification. Therefore, almost all the IRAGNs can be empirically
1006: classified by their observed colors, independent of $K$-corrections,
1007: which allows this criterion to be used for samples that do not include
1008: accurate redshifts.
1009: 
1010: \subsection{Dependence of color selection on luminosity and redshift}
1011: The template used in Fig.~\ref{figtempav} represents a luminous AGN
1012: that dominates the optical emission from the host galaxy.  For
1013: lower-luminosity AGNs, whose unobscured optical flux is smaller than
1014: that of their hosts, extinction of the nucleus will have a
1015: relatively small effect on \iiu.  Therefore, our selection criterion
1016: is not applicable for samples of sources at lower luminosities and
1017: redshifts.  In Fig.~\ref{figuvir}(d), we show \iiu\ versus \lirac\ for
1018: subsets of objects at $z<0.7$, for comparison to the $z\geq0.7$ IRAGN
1019: sample.  At low \lirac, optical BLAGNs and NLAGNs have \iiu\ typical of
1020: normal galaxies, indicating that their total emission is dominated by
1021: the hosts, and this simple color criterion cannot distinguish obscured sources.
1022: 
1023: By cutting our IRAGN sample at $z=0.7$, we include only sources with
1024: $L_{\rm 4.5\mu m} \gtrsim 10^{11}$ \lsun\ (owing to the flux limits of
1025: the IRAC Shallow Survey). These AGNs are luminous enough that if
1026: unobscured, their nuclear optical luminosity is comparable to all but
1027: the most luminous host galaxies.  Therefore, our redshift cut at
1028: $z\geq0.7$ enables obscured AGN color selection, (1) by probing
1029: shorter rest-frame wavelengths in the optical and (2) by selecting
1030: luminous AGNs for which the intrinsic optical luminosity is larger
1031: than the host.
1032: 
1033: 
1034: 
1035: 
1036: \subsection{Are the IRAGN 2s intrinsically optically faint?}
1037: We consider the possibility that the IRAGN 2s are not obscured, but
1038: intrinsically faint in the observed optical band.  For example, there
1039: exist modes of accretion that lack a luminous accretion disk and
1040: therefore do not radiate strongly in the optical and UV; these are
1041: known as radiatively inefficient accretion flows
1042: \citep[e.g.,][]{nara95}.  However, in such a scenario it is difficult
1043: to explain the observed properties of IRAGN 2s in the mid-IR.  The red
1044: IRAC colors and high mid-IR luminosities of these objects are
1045: characteristic of dust that has been heated to high temperatures by
1046: high UV fluxes, and so imply some luminous UV emission from the
1047: nucleus.  Such emission would not be present in radiatively
1048: inefficient flows.  Therefore, we hypothesize that all the IRAGN 2s are
1049: intrinsically luminous enough in the UV to power the observed mid-IR
1050: emission, but they are optically faint because the nuclear emission is
1051: obscured.
1052: 
1053: 
1054: 
1055:  
1056: 
1057: \subsection{Bolometric luminosities}
1058: \label{bolometric}
1059: A fundamental property of AGNs is the bolometric accretion luminosity
1060: $L_{\rm bol}$.  For the IRAGN 2s, the nuclear optical light is
1061: extincted, and most objects are not individually detected in X-rays,
1062: so we cannot use optical or X-ray luminosities to estimate $L_{\rm
1063: bol}$.  Instead, we derive $L_{\rm bol}$ by scaling from the
1064: $K$-corrected luminosity of the AGN at 2 \micron, $L_{\rm 2 \mu
1065: m}^{\rm AGN}$, taken from the SED fits (\S\ \ref{lum}).  $L_{\rm bol}$
1066: is given by $L_{\rm bol}=BC_{2 \rm \mu m}L_{\rm 2 \mu m}^{\rm
1067: AGN}$, where $BC_{2 \rm \mu m}$ is the bolometric correction. We
1068: derive $BC_{2 \rm \mu m}$ from the luminosity-dependent quasar SED
1069: model of \citet{hopk07qlf}, for which the correction is in the range
1070: $BC_{2 \rm \mu m}=10$--15 for the luminosities of the sample (we note
1071: that the luminosity-independent model of \citet{rich06} gives a
1072: similar  $BC_{2 \rm \mu m}=12$).
1073: 
1074: The distributions in \lbol\ are shown in Fig.\ \ref{figlbol}.  The
1075: $L_{\rm bol}$ values of (0.1--10)$\times10^{46}$ \ergs\ are similar
1076: for IRAGN 1s and 2s and are typical of the accretion luminosities of
1077: bright Seyferts and quasars.  We see no systematic difference between
1078: the distributions for the two types of IRAGNs, indicating that at
1079: these high luminosities, the fraction of obscured to unobscured
1080: sources is relatively constant with luminosity.  However, these
1081: results give only approximate distributions in $L_{\rm bol}$ because
1082: of uncertainties in the photo-$z$'s for individual objects,
1083: particularly IRAGN 2s (see \S~\ref{photoz}).  While it would be very
1084: interesting to use this sample to study quantities such as the
1085: evolution of the obscured AGN fraction with $z$ or $L_{\rm bol}$, to
1086: confidently perform such measurements requires better calibration of
1087: the IRAGN 2 redshifts.
1088: 
1089: 
1090: 
1091: 
1092: 
1093: \begin{figure}
1094: \epsscale{1.2}
1095: \plotone{f11.eps}
1096: \caption{Distribution in bolometric luminosity $L_{\rm bol}$
1097: estimated as described in \S~\ref{bolometric}, for the
1098: two types of IR-selected AGN. \label{figlbol}}
1099: \vskip0.3cm
1100: \end{figure}
1101: 
1102: 
1103: \section{Multiwavelength tests of obscured AGN selection}
1104: 
1105: \label{tests}
1106: 
1107: Our classification of candidate AGNs as unobscured (IRAGN 1) or obscured
1108: (IRAGN 2) is based solely on the ratio of their observed optical
1109: to IR color.  This classification makes several predictions for the
1110: observed emission from these sources at X-ray, optical, and infrared
1111: wavelengths:
1112: 
1113: \begin{enumerate}
1114: 
1115: \im The average X-ray properties of the two populations should be
1116: consistent with unabsorbed and absorbed AGNs, respectively.  Both
1117: types should have high X-ray luminosities typical of Seyferts and
1118: quasars.  The IRAGN 1s should have X-ray spectral shapes consistent
1119: with unabsorbed AGNs, while $>$70\%--80\% of the IRAGN 2s should have
1120: harder X-ray spectra due to absorption by neutral gas \citep[e.g.,][]{tozz06}.
1121: 
1122: \im For IRAGN 2s, the observed X-ray absorption should be consistent
1123: with the extinction derived from the optical/UV colors, for a
1124: reasonable gas-to-dust ratio.
1125: 
1126: \im The IRAGN 1s and 2s  should have optical morphologies and optical colors
1127: characteristic of BLAGNs and galaxies, respectively.
1128: 
1129: \end{enumerate}
1130: 
1131: 
1132: In \S\S\ \ref{xray}--\ref{galedd} we test each of these predictions
1133: using the available data from \chandra, \spitzer, and optical
1134: photometry and spectroscopy.  In each case we show that the data are
1135: consistent with the above classification of IR-selected AGNs as
1136: unobscured (IRAGN 1) and obscured (IRAGN 2).  
1137: 
1138: \subsection{X-ray properties}
1139: \label{xray}
1140: 
1141: \input{tab2}
1142: 
1143: 
1144: 
1145: X-ray emission is an efficient and largely unbiased way of detecting
1146: AGN activity for objects with $N_{\rm H}\lesssim10^{24}$ \cdens\
1147: \citep[for reviews see][]{mush04book, bran05}.  Thus, the contiguous
1148: \chandra\ coverage of the \bootes\ field provides a useful diagnostic
1149: for confirming our classifications of IR-selected AGNs
1150: \citep[S05;][]{gorj07}, allowing us to estimate both the X-ray
1151: luminosity \lx\ and the absorbing neutral hydrogen column density \nh.
1152: 
1153: The main limitation of the wide-field X\bootes\ observations is that
1154: they are shallow, with exposures of only 5 ks yielding a 0.5--7 keV
1155: source flux limit of $(4-8)\times10^{-15}$ \flux.  Most IR-selected
1156: AGNs do not have firm X-ray detections, and most detected sources have
1157: fewer than 10 counts, so we do not have X-ray spectral information for
1158: most individual sources.  We therefore perform a stacking analysis,
1159: which compensates for the shallowness of the X-ray observations by
1160: averaging over the large number of IR-selected AGNs in the field.  By
1161: summing X-ray images around the known IR positions, we determine the
1162: average X-ray fluxes, luminosities and spectral shapes of various
1163: subsets of these sources.
1164: 
1165: \subsubsection{X-ray stacking}
1166: \label{xraystack}
1167: Around the position of each object in a given sample, we extract
1168: $40\times40$-pixel (19.7\arcsec) X-ray images in the soft band (0.5--2 keV)
1169: and hard band (2--7 keV).  Because the \chandra\ telescope PSF varies with angle $\theta$ from the optical axis, the
1170: aperture from which we extract source photons varies from source to
1171: source.  We take this aperture to be the 90\% energy encircled radius
1172: at 1.5 keV:\footnotemark
1173: \begin{equation}
1174: r_{90}\simeq 1\arcsec+10\arcsec(\theta/10\arcmin)^2.
1175: \end{equation} 
1176: We include in the stacking analysis only objects that lie within
1177: 6\arcmin\ of the optical axis of a \chandra\ pointing, for which
1178: $r_{90}<4\farcs6$.  This excludes over half the available sources
1179: but minimizes source confusion and maximizes signal-to-noise ratio.  Using a
1180: model of the \chandra\ PSF (from the \chandra\ CALDB) and sources with
1181: random positions inside the 6\arcmin\ radius, an aperture of $r_{90}$
1182: includes 90\%--92\% of the source counts in both the 0.5--2 keV and 2--7
1183: keV bands.  Accordingly, in our stacking analysis, we multiply the
1184: observed source counts by 1.1 to obtain the total counts from the
1185: source.
1186: 
1187:  \footnotetext{\chandra\ Proposer's Observatory Guide (POG), available at
1188:  {\tt http://cxc.harvard.edu/proposer/POG}.} 
1189: 
1190: 
1191: 
1192: Of the 126 pointings in the X\bootes\ data set, there are
1193: eleven\footnotemark\ that have significantly higher background
1194: intensities, due to background flares \citep[for a detailed discussion
1195: of ACIS backgrounds see][]{hick06a}.  In our stacking analysis, we do
1196: not include any source positions that lie within these eleven ``bad''
1197: exposures.  The total area over which we perform the stacking, which consists
1198: of the region covered by IRAC that lies within the central 6\arcmin\
1199: radii of these 115 pointings, is 2.9 deg$^2$.
1200: 
1201: 
1202: 
1203: 
1204: \footnotetext{ObsIDs 3657, 3641, 3625, 3617, 3601, 3607, 3612, 3623,
1205:   3639, 3645, and 4228.}
1206: 
1207:  
1208: 
1209: An accurate measure of the stacked source flux requires subtraction of
1210: the background, which we estimate by stacking X-ray images on random
1211: positions around the \bootes\ field, at least 20\arcsec\ away from any
1212: X-ray sources included in the X\bootes\ catalog \citep{kent05}.  We
1213: performed 20 trials, stacking $\sim$30,000 positions in each trial.
1214: As a check, we also calculate the surface brightness using the ACIS
1215: blank-sky data sets,\footnotemark\ which are obtained using deep
1216: exposures at high Galactic latitude and removing all detected sources.
1217: 
1218: \footnotetext{{\tt http://cxc.harvard.edu/contrib/maxim/acisbg/}}
1219: 
1220: 
1221:  
1222: 
1223: 
1224: 
1225: 
1226: Both estimates of the diffuse background give identical
1227: surface brightnesses of 3.0 counts s$^{-1}$ deg$^{-2}$ in the 0.5--2 keV
1228: band and 5.0 counts s$^{-1}$ deg$^{-2}$ in the 2--7 keV band.  We use
1229: these values to calculate the expected background counts within a
1230: circle of radius $r_{90}$ for each source position.  For a typical
1231: $r_{90}=3$\arcsec\ and an exposure time of 4686 s (see below), this
1232: corresponds to 0.03 and 0.05 background counts for each IRAC source in the
1233: 0.5--2 keV (soft) and 2--7 keV (hard) bands, respectively. 
1234: 
1235: Subtracting this background, we obtain the average X-ray flux in
1236: counts source$^{-1}$.  We assume that all source positions have an X-ray
1237: exposure time of 4686 s, which is the mean for all the X\bootes\
1238: observations excluding the ``bad'' exposures.  For simplicity we
1239: ignore variations in exposure time between pointings, as well as
1240: variations in effective exposure time within each single ACIS-I field
1241: of view due to mirror vignetting.  These variations are at most
1242: $\sim$10\% and do not significantly affect our results.  We convert
1243: count rates (in counts s$^{-1}$) to flux (in \flux) using the conversion
1244: factors $6.0\times10^{-12}$ ergs cm$^{-2}$ count$^{-1}$ in the 0.5--2 keV band and
1245: $1.9\times10^{-11}$  ergs cm$^{-2}$ count$^{-1}$ in the 2--7 keV band.  In addition to fluxes, we
1246: obtain rough X-ray spectral information by calculating the hardness
1247: ratio, defined as
1248: \begin{equation}
1249: {\rm HR}=\frac{H-S}{H+S},
1250: \end{equation}
1251: where $H$ and $S$ are the count rates in the hard and soft bands,
1252: respectively.  Errors in count rates are calculated using the
1253: approximation $\sigma_{X}=\sqrt{X+0.75}+1$, where $X$ is the number of
1254: counts in a given band \citep{gehr86}.  Uncertainties in HR are
1255: derived by propagating these count rate errors.  In the following
1256: analysis, we use these hardness ratios and fluxes to determine the
1257: typical absorption and X-ray luminosities from the stacking analysis.
1258: 
1259: 
1260: 
1261: \begin{figure}
1262: \epsscale{1.2}
1263: \plotone{f12.eps}
1264: \caption{Verification of IRAGN 2 selection using X-ray data.
1265:   Shown is the average hardness ratio from X-ray stacking analysis, for
1266:   objects with $L_{\rm 4.5 \mu m}<3\times10^{11}$ \lsun, in bins of \liiu. The
1267:   dashed line shows the boundary between IRAGN 1 and 2 as defined in
1268:   Fig.\ \ref{figuvir}.  \label{figstackxi}}
1269: \vskip0.3cm
1270: \end{figure}
1271: 
1272: 
1273: \begin{figure}
1274: \epsscale{1.2}
1275: \plotone{f13.eps}
1276: \caption{Comparison of average 2--7 keV flux for X-ray undetected
1277:   IRAGN 2s, versus X-ray fluxes expected for star formation.  Vertical
1278:   lines show observed average 2--7 keV X-ray flux from stacking
1279:   analysis of $\average{F_{\rm 2-7\;
1280:   keV}}=(1.83\pm0.20)\times10^{-15}$ \flux\ (Table \ref{tblxray}).
1281:   Histograms show the distribution of predicted X-ray fluxes for star
1282:   formation, given the measured $S_{\rm 5.8 \mu m}$ for each object
1283:   and the ratio $F_{\rm 2-7\; keV}/S_{\rm 5.8 \mu m}$ for star
1284:   formation (see Eqn.~\ref{eqnf27}), derived from the $F_{\rm 2-10\;
1285:   keV}/F_{\rm FIR}$ relation of \citet{rana03}, and typical starburst
1286:   SEDs.  The three histograms represent $F_{\rm 2-7\; keV}/S_{\rm 5.8
1287:   \mu m}=(0.5-4)\times10^{18}$ \flux\ $\mu$Jy$^{-1}$, for which the
1288:   average values of $F_{\rm 2-7\; keV}$ are $(0.8-6)\times10^{-16}$
1289:   \flux.  Even for the largest typical ratio of X-ray to 5.8 $\mu$m flux, the
1290:   observed average flux is $>3$ times larger than that expected for star
1291:   formation.   This indicates that the X-ray emission from these objects is
1292:   dominated by nuclear accretion.  \label{figsblx}}
1293: \vskip0.3cm
1294: \end{figure}
1295: 
1296: 
1297: \input{tab3}
1298: 
1299: 
1300: 
1301: \begin{figure}
1302: \epsscale{1.2}
1303: \plotone{f14a.eps}
1304: \plotone{f14b.eps}
1305: \plotone{f14c.eps}
1306: \caption{Results from X-ray stacking analysis, in bins of redshift, for
1307:   IRAGN 1s and 2s.  Here all sources are included, including those with X-ray detections.  Shown are the average $L_{\rm X}$ in the (a) 2--7
1308:   keV, and (b) 0.5--2 keV bands, and (c) average hardness ratio.
1309:   Note that the IRAGN 2s are consistently harder (larger HR) than
1310:   the IRAGN 1s. \label{figlxhr}}
1311: \end{figure}
1312:  
1313: \begin{figure}
1314: \epsscale{1.2}
1315: \plotone{f15.eps}
1316: \caption{X-ray hardness ratio HR vs.\ redshift for various \nh\
1317:   (given in \cdens), given
1318:   an intrinsic power-law photon index $\Gamma=1.8$ and the on-axis ACIS
1319:   response function.  Hatched regions show the observed $1\sigma$ limits in
1320:   HR for the IRAGN 2s in bins of redshift, as shown in Fig.\
1321:   \ref{figlxhr}.  Assuming the intrinsic $\Gamma=1.8$, these HR values
1322:   for IRAGN 2s are consistent with a constant $N_{\rm
1323:   H}=$(2--5)$\times10^{22}$ \cdens, shown by the shaded
1324:   region. \label{fignhz}}
1325: \end{figure}
1326: 
1327: 
1328:  
1329: \begin{figure}
1330: \epsscale{1.2}
1331: \plotone{f16a.eps}
1332: \plotone{f16b.eps}
1333: \plotone{f16c.eps}
1334: \caption{Same as Fig. \ref{figlxhr}, only including sources with no
1335:   detections in the X\bootes\ catalog \citep{kent05}. X-ray
1336:   undetected IRAGNs have average 
1337:   $L_{\rm 0.5-7\; keV}\sim10^{43}$ \ergs\ characteristic of luminous Seyfert
1338:   galaxies.  The X-ray undetected IRAGN 2s have systematically larger average hardness
1339:   ratios than the X-ray undetected IRAGN 1s. \label{figlxhr_max0}}
1340: \end{figure}
1341: 
1342: 
1343: \subsubsection{Calculation of \nh\ and \lx}
1344: \label{xraycalc}
1345:  Measuring \nh\ and \lx\ from X-ray fluxes requires an assumption for
1346: the X-ray spectrum of the source, which for most AGNs can be modeled
1347: by a simple power law with photon index $\Gamma$, such that the photon
1348: flux density (in photons cm$^{-2}$ s$^{-1}$ keV$^{-1}$) $F\propto
1349: E^{-{\Gamma}}$.  For all the AGNs in our sample, we assume an
1350: intrinsic X-ray spectrum with $\Gamma=1.8$, typical for unabsorbed
1351: AGNs \citep{tozz06}; as we show in \S\ \ref{xrayres}, this $\Gamma$
1352: corresponds to the average spectral shape of the IRAGN 1s.  Although
1353: X-ray AGNs do not all have the same intrinsic $\Gamma$, the typical
1354: intrinsic spectrum does not vary significantly with luminosity or \nh\
1355: \citep{tozz06}.  Therefore, it is reasonable to assume a constant
1356: $\Gamma$ when estimating \nh\ and \lx\ for an ensemble of sources.
1357: 
1358: Given a constant intrinsic X-ray spectrum, \nh\ is directly related to
1359: the observed hardness ratio.  Absorption by neutral gas preferentially
1360: obscures lower energy X-rays, and so a larger \nh\ corresponds to a
1361: larger (in this case, less negative) HR.  The conversion between
1362: HR and \nh\ depends on the response function of the X-ray detector and the
1363: redshift of the source.  With increasing $z$, the low-energy turnover
1364: due to absorption by neutral hydrogen is increasingly redshifted out
1365: of the 0.5--2 keV bandpass.  Therefore, for a power-law
1366: spectrum attenuated by a fixed column density of gas, the observed
1367: spectrum will become softer with increasing redshift, so that objects
1368: with higher $z$, but equal HR, correspond to greater absorption.
1369: Assuming $\Gamma=1.8$, we have calculated HR for a grid of
1370: absorptions ($10^{20}\leq N_{\rm H}\leq10^{23}$ \cdens) and redshifts
1371: ($0\leq z\leq4$), and will use these to convert observed hardness ratios to
1372: column densities.  Note that the Galactic $N_{\rm H}$ toward this
1373: field is very small ($\sim$$10^{20}$ \cdens), so we neglect it in our
1374: estimates of column density.
1375: 
1376: To derive the average \lx\ and \nh\ from stacking requires an estimate of redshift, so we perform the stacking in bins of $z$.  For each
1377: bin we calculate the rms value of $d_{\rm L}$ for the
1378: objects in the bin.  Using this distance, along with the average
1379: fluxes from stacking, we calculate $\average{L_{\rm X}}$ in each bin.  We
1380: also include a $K$-correction to $\average{L_{\rm X}}$, assuming
1381: $\Gamma=1.8$, to account for the fact that the X-ray bands we observe
1382: probe higher energies in the rest-frame spectrum.  This
1383: $K$-correction varies from 0.9 at $z=0.7$ to 0.8 at $z=2$, so it has
1384: a relatively small effect on $\average{L_{\rm X}}$.
1385: 
1386: 
1387: 
1388: 
1389: \subsubsection{Average X-ray fluxes}
1390: \label{xrayres}
1391: 
1392: 
1393: 
1394: 
1395: As a first step in the stacking analysis, we compare the average X-ray
1396: fluxes of different subsets of sources in the \bootes\ catalog.  We
1397: divide the sources into (1) IRAGN 1s and (2) IRAGN 2s, defined by
1398: \iiu\ as shown in Fig.\ \ref{figuvir},
1399: and for comparison (3) objects with detections in all four IRAC bands
1400: that are identified as optically normal galaxies in the optical from
1401: their AGES spectra\footnotemark.  The IRAGNs are all selected to have
1402: $z>0.7$, while the optically normal galaxies mainly lie at $z<0.7$.
1403: In interpreting X-ray stacking results, it is a concern that the X-ray
1404: brightest objects may dominate the average flux.  Therefore, we
1405: perform the stacking analysis twice, first using all objects in each
1406: subsample and then using only those objects that are not detected
1407: with 4 or more counts in the X\bootes\ source catalog \citep{kent05}.
1408: 
1409: \footnotetext{For a detailed X-ray stacking analysis of normal
1410:   galaxies in the AGES survey, see \citet{brand05}.}
1411: 
1412: Average fluxes in the soft and hard bands are given in Table
1413: \ref{tblxray}.  Including all sources, the IRAGN 1s have a larger
1414: average 2--7 keV flux than the IRAGN 2s, with $\average{F_{\rm 2-7\;
1415: keV}}=10.9\times10^{-15}$ and $6.5\times10^{-15}$ \flux\ source$^{-1}$,
1416: respectively.  However, when we exclude those objects that are detected in the
1417: X\bootes\ catalog, the hard X-ray fluxes of the two subsets
1418: closely agree.  This suggests that the IRAGN 1 sample contains more
1419: bright X-ray sources than the IRAGN 2 sample, but for faint sources ($<$4
1420: counts), the two IRAGN types have similar average fluxes.
1421: 
1422: While the hard X-ray fluxes are comparable between the IRAGN 1s and
1423: 2s, the soft X-ray fluxes are significantly smaller for the IRAGN 2s,
1424: indicating that they are more absorbed.  The IRAGN 1s have an average
1425: $HR=-0.47$, which is close to that expected for an unabsorbed AGN with
1426: $\Gamma=1.8$.  In contrast, the IRAGN 2s have $HR=-0.20$.  The
1427: optically normal galaxies have $HR=-0.19$, similar to the IRAGN 2s, but
1428: with $\sim$5 times smaller average flux.
1429: 
1430: As mentioned in \S\ \ref{identify}, the X-ray data can be used to
1431: verify our selection criterion for IRAGN 1s and 2s, by looking for
1432: systematic differences in HR (and thus absorption) on either side of
1433: our selection boundary.  We can therefore address concerns that the
1434: selection may be biased by the AGES spectroscopic flux limits,
1435: especially for $L_{\rm 4.5\mu m}<3\times10^{11}$ \lsun, where the
1436: distribution in \iiu\ is not as clearly bimodal (Fig.\ \ref{figuvir}).
1437: We perform the stacking analysis in bins of \iiu\ for the IRAGNs with
1438: $L_{\rm 4.5 \mu m}<3\times10^{11}$ \lsun\ and plot HR versus \iiu\
1439: in Fig.\ \ref{figstackxi}.  There is a significant increase in HR
1440: across our IRAGN 1/IRAGN 2 boundary of $\log{L_R/L_{\rm 4.5\mu m
1441: }}=-0.4$, verifying that this criterion effectively separates objects
1442: with unabsorbed (IRAGN 1) and absorbed (IRAGN 2) X-ray emission and
1443: is not significantly affected by the AGES flux limits.
1444: 
1445: Next, to confirm that the X-ray flux for these objects comes from
1446: nuclear emission, we show that the average X-ray flux is significantly
1447: larger than that expected for star formation.  The X-ray flux $F_X$
1448: from star formation is related to the far-IR flux $F_{\rm FIR}$
1449: \citep[see Eqn.~12 of][]{rana03}.  $F_{\rm FIR}$ is
1450: defined as \citep{helo85}
1451: \begin{equation}
1452: F_{\rm  FIR}=1.26\times10^{-11}(2.58 S_{60 \mu {\rm m}} + S_{100 \mu {\rm
1453:     m}}) {\rm \: ergs\: cm^{-2}\: s^{-1}},
1454: \label{eqnfir}
1455: \end{equation}
1456:  where $S_{60 \mu {\rm m}}$ and $S_{100 \mu {\rm m}}$ are in Jy.  For
1457:     the suite of starburst model SEDs given in \citet{sieb07}, we
1458:     calculate the ratio of rest-frame $F_{\rm FIR}$ (in \flux) to the
1459:     observed flux at 5.8 $\mu$m, $S_{\rm 5.8\mu m}$ (in $\mu$Jy).
1460:     Excluding Arp 220 (which has an extreme star formation rate and
1461:     is highly extincted in the optical, so it would not be detected
1462:     in our survey), we find that $F_{\rm FIR}/S_{\rm 5.8\mu
1463:     m}\sim(0.3-2)\times10^{-14}$ \flux\ $\mu {\rm Jy}^{-1}$ for the
1464:     redshift range $0.7<z<3$.  Combining with the \citet{rana03}
1465:     relation and converting from the rest-frame 2--10 keV band
1466:     luminosity to our observed 2--7 keV band flux using a $\Gamma=1.8$
1467:     power law spectrum (with the appropriate small $K$-correction), we
1468:     have
1469: \begin{equation}
1470: \label{eqnf27}
1471: F_{\rm 2-7\; keV}=S_{\rm 5.8\mu m }\times(0.5-4)\times10^{-18}{\rm \; ergs \; cm^{-2}\:
1472:     s^{-1}\: \mu Jy^{-1}}
1473: \end{equation}
1474: 
1475: Making the conservative assumption that the observed 5.8 \micron\ flux
1476: for every IRAGN 2 is due entirely to stars and star formation, we can
1477: put an upper limit on the \fhard\ we expect from star formation for
1478: each object.  The distribution in these \fhard\ values, for every
1479: IRAGN 2 that is not detected in X-rays, is shown in Fig.\
1480: \ref{figsblx}.  By comparison, the average \fhard\ observed for these
1481: objects is $>3$ times larger than that typically expected for star
1482: formation, even for the largest typical ratio of X-ray to 5.8 $\mu$m flux.
1483: This confirms that for most of the IRAGN 2s, the X-ray emission is not
1484: powered by star formation, but nuclear accretion.
1485: 
1486: 
1487: 
1488: 
1489: \subsubsection{Average \lx\ and \nh}
1490: In order to use the X-ray stacking analysis to measure physical
1491: parameters such as the accretion luminosity (\lx) or the gas
1492: attenuation (\nh), we must include redshift information.  Therefore,
1493: we have repeated the stacking for both types of IRAGN in bins of
1494: redshift from $z=0.7$--2.5.  We do not include sources at $z>2.5$
1495: because we do not have enough objects with best-fit $z>2.5$ to obtain
1496: well-constrained fluxes.  We stress here that although there may be
1497: significant uncertainties in photometric redshift estimates,
1498: particularly for IRAGN 2s, there is no large bias in the photo-$z$'s,
1499: as we show in \S\ \ref{photoz}.  Therefore, our stacking analysis
1500: using large bins in redshift should not be strongly affected by
1501: photo-$z$ uncertainties.
1502: 
1503: The stacking results as a function of $z$ are listed in Table
1504: \ref{tblxray2}, and we plot $\average{L_{\rm X}}$ versus $z$ in
1505: Fig.\ \ref{figlxhr}.  For both IRAGN types, $\average{L_{\rm X}}$ increases
1506: by a factor of $\sim$2 between $z=0.5$ and $z=2.5$, due to the evolution
1507: in the quasar luminosity function with redshift
1508: \citep[e.g.,][]{ueda03,barg05,hasi05} and the IR and optical flux limits that
1509: restrict us to selecting only the most luminous objects at high $z$.
1510: The $\average{L_{\rm 2-7\; keV}}$ range
1511: (0.3--3)$\times10^{44}$ \ergs\ is typical for Seyfert galaxies and
1512:  quasars and much larger than the typical \lx\ of
1513: starburst or normal galaxies.  Although the IRAGN 2s have
1514: $\average{L_{\rm 0.5-2\; keV}}$ and $\average{L_{\rm 2-7\; keV}}$ that
1515: are 3--5 and 2--3 times lower than the IRAGN 1s, respectively, these
1516: \lx\ values are still typical of AGNs and not starburst galaxies.  
1517: 
1518: Plotting HR in redshift bins (Fig.\ \ref{figlxhr}), the IRAGN 2s are
1519: significantly harder at all $z$.  The IRAGN 1s have $HR\simeq-0.45$
1520: for all $z$, which corresponds to an intrinsic $\Gamma=1.8$ with no
1521: absorption.  The IRAGN 2s are significantly harder, with
1522: $HR\simeq-0.3$--0.1.  Assuming that these have the same intrinsic
1523: $\Gamma$ as the IRAGN 1s, we estimate the corresponding \nh.  In Fig.\
1524: \ref{fignhz} we plot HR versus $z$ for several values of \nh\
1525: assuming $\Gamma=1.8$.  The hatched regions correspond to the
1526: $1\sigma$ errors in HR for the IRAGN 2s in each of our redshift
1527: bins.  For all redshifts, the IRAGN 2 HR values are consistent
1528: with a column density of $N_{\rm H}=$(2--5)$\times10^{22}$ \cdens,
1529: marked by the shaded region in Fig.\ \ref{fignhz}.
1530: 
1531: As mentioned in the previous section, it is a concern that the average
1532: \lx\ and \nh\ we measure may be dominated by a few bright sources.  To
1533: address this, we repeat the stacking as a function of $z$ but exclude
1534: those objects that are detected in the X\bootes\ catalog.  The results
1535: are shown in Fig.\ \ref{figlxhr_max0} and indicate that even those
1536: objects that are not detected in X-rays have
1537: $\average{L_{\rm X}}$ values consistent with AGNs.  In addition,
1538: the IRAGN 2s have harder spectral shapes than the IRAGN 1s, even
1539: at these fainter fluxes.
1540: 
1541: To summarize the X-ray stacking results, both IRAGN types have average
1542: X-ray fluxes that are too large to be due to star formation and thus
1543: strongly indicate AGN activity.  Performing the stacking as a function
1544: of redshift, we find that both IRAGN 1s and 2s have average \lx\
1545: values consistent with Seyferts and quasars, and the IRAGN 1s have
1546: hardness ratios consistent with unabsorbed AGNs ($\Gamma=1.8$). The
1547: IRAGN 2s, assuming the same intrinsic spectrum, correspond to absorbed
1548: sources with $N_{\rm H}\sim 3\times10^{22}$ \cdens.
1549: 
1550: 
1551: 
1552: 
1553: \subsection{Gas absorption and dust extinction}
1554: 
1555: \label{dust}
1556: 
1557:  In this section, we check that the dust extinction for the IRAGN 2s
1558:  that we inferred from the optical/UV data is consistent with the \nh\
1559:  we measure in X-rays, assuming that, in general, the X-ray-absorbing
1560:  gas is coincident with the extincting dust.  Fig.~\ref{figav} shows
1561:  that the SEDs of most of the IRAGN 2s are best fitted by templates with
1562:  $0.7<A_V<7$.  The ratio of gas to dust in the Galaxy is such that
1563:  $N_{\rm H}/A_V\simeq2\times10^{21} {\; \rm cm}^{-2}$, or $A_V\simeq15$
1564:  for the observed average $N_{\rm H} = 3 \times10^{22}$ \cdens.  This
1565:  extinction is more than enough to obscure the optical light from the
1566:  nucleus, although it is somewhat larger than the typical $A_V$
1567:  obtained from the SED fits. 
1568: 
1569:  However, there is evidence that AGNs have high gas-to-dust ratios,
1570:  similar to or perhaps even greater than that of the SMC
1571:  \citep[see][and references therein]{fall89,mart06}.  The SMC has
1572:  $N_{\rm H}/A_V\simeq2\times10^{22} {\; \rm cm}^{-2}$, which corresponds
1573:  to $A_V\simeq1.5$ for the observed average \nh\ and is close to the
1574:  typical $A_V$ obtained by the SED fits to the IRAGN 2s.  Therefore,
1575:  we conclude that the dust extinction implied by the optical and IR
1576:  observations is generally consistent with the average \nh\ we derive
1577:  from X-ray stacking.
1578: 
1579:  
1580: 
1581: 
1582: The bimodality in the $A_V$ distribution from SED fits
1583: (Fig.~\ref{figav}), as well as the clear separation of the two IRAGN
1584: types in optical-IR color (Fig.~\ref{figuvir}), suggests that there is
1585: a bimodal distribution in the dust column density to the IR-selected
1586: AGNs.  There is no obvious selection effect that could produce this
1587: bimodality, so we expect that it is real.
1588: This is broadly consistent with previous results on the distribution
1589: of \nh\ measured in X-rays.  These studies find many objects with
1590: $N_{\rm H}<3\times10^{20}$ \cdens\ or $N_{\rm H}>3\times10^{21}$
1591: \cdens, with relatively few at intermediate column densities
1592: \citep[e.g.][]{trei05,tozz06}.  However, such a bimodal distribution
1593: could simply be due to limitations of X-ray spectral fitting
1594: techniques, with which it is difficult to measure $N_{\rm H}$ as low
1595: as $\sim10^{21}$ \cdens, especially at high redshifts where X-ray
1596: telescopes probe energies higher than the photoelectric cutoff at
1597: $E\simeq 1$ keV \citep[e.g.,][]{akyl06}.  The colors we
1598: observe in IR-selected AGNs suggests that such a bimodal obscuration
1599: distribution does indeed exist, which has implications for models
1600: of AGN obscuration, as we discuss in \S\ \ref{impl}.
1601: 
1602: 
1603: 
1604: 
1605: 
1606: \begin{figure} 
1607: \epsscale{1.2} 
1608: \plotone{f17a.eps}
1609: \plotone{f17b.eps}
1610: \caption{Optical morphologies for the two types of IR-selected AGNs.
1611:   (a) Histograms of the $I$ band stellarity parameter \cstar\ output
1612:   by ${\tt SExtractor}$ (0 for extended sources, 1 for point sources).
1613:   Note that the vast majority of IRAGN 1s are point-like, while the
1614:   IRAGN 2s are extended, suggesting that their optical emission is
1615:   dominated by the nucleus and by the host galaxy, respectively.  (b)
1616:   Variation of $I$ band \cstar\ with $I$ magnitude, for the flux range
1617:   in which the two IRAGN subsets overlap.  The $y$ axis shows the
1618:   fraction of sources that have $I>0.5$ in each magnitude bin.  Errors
1619:   shown are counting statistics, and only bins with $>$5 sources are
1620:   shown.  The point-like fraction does not vary with $I$ and is
1621:   significantly different between the two subsets.
1622:   \label{figmorph} }
1623: \vskip0.2cm
1624: \end{figure}
1625:  
1626: 
1627: \subsection{Optical morphologies and colors}
1628: \label{galedd}
1629: 
1630: 
1631: 
1632: Since we expect the nuclear optical emission from the IRAGN 2s to be
1633: extincted, their optical light should be dominated by their host
1634: galaxies.  Normal galaxies differ from quasars in optical images in
1635: two principal ways: (1) galaxies have extended morphologies, while
1636: quasars are dominated by a small nucleus and so appear as point
1637: sources; and (2) normal galaxies have redder colors, characteristic of
1638: a composite stellar spectrum rather than a blue AGN continuum.  By
1639: examining the optical morphologies and colors of our IRAGN sample, we
1640: can confirm that optical emission is dominated by an AGN in IRAGN 1s
1641: and by the host galaxy in IRAGN 2s.
1642: 
1643: 
1644: 
1645: 
1646: 
1647: 
1648: To quantify morphologies, we use the \cstar\ parameter output by the ${\tt
1649: SExtractor}$ photometry code \citep{bert96}.  \cstar\ is a measure of how well an
1650: object can be approximated by a point source, with values ranging from
1651: 0 (extended) to 1 (point source).  In Fig.\ \ref{figmorph} (a), we
1652: plot the distribution in \cstar\ in the $I$ band (which best
1653: discriminates between the two IRAGN types) and find that 74\% of the
1654: IRAGN 1s have \cstar\ $>0.7$, indicating that the emission is point-like,
1655: while 85\% of the IRAGN 2s have \cstar\ $<0.5$, indicating mainly
1656: extended emission.
1657: 
1658: However, for very faint objects, it is possible to obtain low \cstar\
1659: values, even if the sources are point-like.  Therefore, we must confirm
1660: that the lower \cstar\ values for IRAGN 2s are not simply a
1661: result of their lower fluxes.  Fig.\ \ref{figmorph} (b) shows the
1662: fraction of objects with ${\tt CLASS\_STAR}>0.5$ for each IRAGN subset
1663: as a function of $I$ magnitude.  There is no clear trend in this
1664: fraction with $I$ for the IRAGN 2s, and for the magnitudes in which
1665: the subsets overlap, the IRAGN 2s have many fewer ``point-like''
1666: morphologies than the IRAGN 1s.  We conclude that the IRAGN 2s do have
1667: more extended morphologies than the IRAGN 1s, so that the color
1668: selection described in \S\ \ref{identify} can effectively distinguish
1669: between objects dominated by a nucleus and those dominated by extended
1670: emission.
1671: 
1672: 
1673: 
1674: 
1675: \input{tab4}
1676:  
1677: 
1678: 
1679: \begin{figure}
1680: \epsscale{1.2}
1681: \plotone{f18a.eps}
1682: \plotone{f18b.eps}
1683: \caption{Contours and points show observed $B_{W}-R$ and $R-I$ colors
1684:   for (a) IRAGN 1s and (b) IRAGN 2s, compared to colors for elliptical
1685:   galaxy (red solid line), Sb galaxy (blue dashed line), and quasar
1686:   (green dot-dashed line) templates.  The model tracks run from
1687:   $z=0-4$.  Stars show the color at $z=0$, and filled dots indicate
1688:   $z=1$, 2, 3, and 4.  Error bars show the median uncertainty in the
1689:   colors  for objects lacking
1690:   spectroscopic redshifts.  \label{figcolcol}}
1691: \vskip0.2cm
1692: \end{figure}
1693: 
1694: \begin{figure*}
1695: \epsscale{1.2}
1696: \plotone{f19.eps}
1697: \caption{Contours and points show observed $B_{W}-R$ and $R-I$ colors
1698:   for IRAGN 1s (top) and IRAGN 2s (bottom) versus redshift, compared
1699:   to model tracks as in Fig.~\ref{figcolcol}.  Error bars show the median uncertainty in the
1700:   colors  for objects lacking
1701:   spectroscopic redshifts. 
1702:   \label{figgalagncol}}
1703: \vskip0.2cm
1704: \end{figure*}
1705: 
1706: 
1707: 
1708: We also examine the observed $B_{W}-R$ and $R-I$ colors of the two IRAGN
1709: types and compare them to the colors of the galaxy and quasar templates
1710: described in \S~\ref{lum}.  In Fig.\ \ref{figcolcol} we plot the
1711: $B_{W}-R$ versus $R-I$ color tracks for the templates as a function of
1712: redshift and overplot the observed colors for the two IRAGN types.
1713: As expected, most IRAGN 1s have colors resembling a quasar spectrum,
1714: although there are a few objects (even those with optical BLAGN
1715: spectra) that have redder colors, owing to some optical extinction.  By comparison, the IRAGN 2s as a whole have colors that
1716: are redder than those of the IRAGN 1s and lie between the elliptical
1717: and spiral redshift tracks.  Fig.\ \ref{figgalagncol} shows a similar
1718: plot, but as a function of redshift.  Again we see that the IRAGN 2s
1719: have colors and trends with redshift that are closer to optically
1720: normal galaxies than to quasars.
1721: 
1722: To quantify this further, we fit the quasar, elliptical, and Sb
1723: templates (shown in Fig.\ \ref{figtemp}) to the $B_{W}$, $R$, and $I$
1724: photometry for the 1469 IR-selected AGNs with detections in all three
1725: optical bands.  For comparison, we perform the same fits for objects at all
1726: redshifts that have four-band $5\sigma$ IRAC detections and AGES
1727: classifications as BLAGN, NLAGN, and galaxies (as listed in the first
1728: column of Table \ref{tblsample}).  We fix the redshift of the template
1729: spectrum and leave only the normalization as a free parameter.  The
1730: distribution of templates that fit the photometry with the lowest
1731: $\chi^2$ is shown in Table \ref{tblgalphot}.  A total of 695 (83\%) of the IRAGN
1732: 1s are best fitted by the quasar or Sb templates (which have similar
1733: colors for $z>1.5$), similar to the fits for optical BLAGNs.  By
1734: contrast, 452 (71\%) of the IRAGN 2s are best fitted by the elliptical
1735: template, with almost all the rest fit by the Sb template, similar to
1736: the fits for optical NLAGNs and optically normal galaxies.  We
1737: conclude that the IRAGN 2s, as a population, do indeed have optical
1738: colors consistent with host galaxies and are markedly different from
1739: the IRAGN 1s.
1740: 
1741: 
1742: 
1743: 
1744: \section{Verification of photometric redshifts}
1745: \label{photoz}
1746: 
1747: The selection criteria developed here for IRAGN 2s depend only on
1748: observed color and so are independent of redshift.  However, our SED
1749: fits and luminosity calculations depend on redshift, so it is
1750: important to verify our redshift estimates.  Of the 1479 objects in
1751: our IR-selected AGN sample, 751 have no spectroscopic redshift, so for
1752: these we use photo-$z$'s calculated from IRAC and optical photometry.
1753: As described in \citet{brod06}, photo-$z$'s using template-fitting
1754: techniques generally fail for objects such as the IR-selected AGNs
1755: that have featureless, power law SEDs.  To overcome this difficulty,
1756: the technique of \citet{brod06} uses an artificial neural net to
1757: estimate the photo-$z$'s for such objects, using those objects that
1758: also have spectroscopic redshifts as a training set.
1759: 
1760: However, only 42 of the 640 IRAGN 2s have spectroscopic redshifts and
1761: thus are included in the training set.  As shown in Fig.\ \ref{figr},
1762: most of the IRAGN 2s are too faint to be spectroscopically targeted in
1763: AGES.  Therefore, it is not immediately clear that
1764: photo-$z$ estimates, which are calibrated against a training set of
1765: optically brighter objects (many of them optical BLAGNs), will also be
1766: valid for the IRAGN 2s that have significantly different mid-IR to
1767: optical SEDs.  It is encouraging that the average X-ray hardness ratio
1768: for the IRAGN 2s decreases with redshift as expected for a small range
1769: in \nh\ (Fig.~\ref{fignhz}).  However, the sharp cutoff at
1770: $z\simeq2.5$ in the redshift distribution of the IRAGN 2s (visible in
1771: Fig.\ \ref{figgalagncol}) suggests a possible systematic bias in the
1772: $z_{\rm phot}$.  It is important to verify that such errors do not
1773: significantly affect our results.
1774: 
1775: 
1776: \subsection{Comparison of spectroscopic and photometric redshifts}
1777: \label{specphotz}
1778: As a first check, we compare the photometric versus spectroscopic
1779: redshifts for the IRAGNs, as shown in Fig.\ \ref{figz}.  For
1780: completeness, this figure includes all objects selected by the
1781: \citetalias{ster05} IRAC criteria (including those with $z<0.7$), but it
1782: does not include 38 IRAGN 1s that have AGES spectroscopy but do not
1783: have well-constrained photo-$z$'s from the \citet{brod06} catalog.  For
1784: the IRAGNs, the distribution in $\delta z=(z_{\rm phot}-z_{\rm
1785: spec})/(1+z_{\rm spec})$ is roughly Gaussian, with mean, dispersion,
1786: and fraction of outliers (objects outside $2\sigma$ in the
1787: distribution) of -0.03, 0.16, and 0.06, respectively.  These values
1788: are (-0.03, 0.15, 0.05) for the 648 IRAGN 1s separately and (-0.06,
1789: 0.18, 0.10) for the 42 IRAGN 2s, indicating reasonably good agreement
1790: for both IRAGN types.
1791: 
1792: The distribution in $\delta z$ is skewed somewhat by $\sim20$ quasars
1793: at $z_{\rm spec}>2$ and $z_{\rm phot}<1$.  The presence of these
1794: sources suggests that for some high-$z$ IRAGN, our reliance on
1795: photo-$z$'s may give a large underestimate for the redshift (some such
1796: objects would be eliminated from the sample by our requirement that
1797: $z>0.7$).  Fig.~\ref{figz} also includes 73 sources with $z_{\rm
1798: phot}>0.7$ and $z_{\rm spec}<0.7$, indicating that there could
1799: be $\sim$10\% contamination from low-$z$ sources in the IRAGN sample.
1800: Still, 54 of these 73 sources have $z_{\rm spec}>0.5$, so the
1801: contamination from very low redshifts ($z<0.5$) is expected to be
1802: $\lesssim 3$\%.  In addition, the sample includes 33 sources
1803: ($\sim$5\%) with $z_{\rm phot}<0.7$ and $z_{\rm spec}>0.7$ that would
1804: not be included in the IRAGN sample.
1805: 
1806: 
1807: Other redshift estimates in the \bootes\ field come from
1808: \citet{houc05}, who obtained redshifts for 17 optically-faint sources
1809: using the Infrared Spectrograph on \spitzer.  Of these, 5 have $5
1810: \sigma$ detections in the IRAC bands, and 4 have IRAC colors inside
1811: the \citetalias{ster05} selection region.  These four sources have 
1812: photo-$z$ estimates from the \citet{brod06} catalog, although only two are in
1813: our IRAGN 2 sample (the other two  have no detection in the
1814: $R$ band).  Of the IRAGN 2s, one has $R=23.8$ and $(z_{\rm
1815: spec}, z_{\rm phot})=(1.95, 2.35)$  while the other has $R=24.7$
1816: and $(2.59, 3.96)$.  The two sources with no $R$ counterpart have
1817: $(0.70, 0.99)$ and $(1.75, 1.01)$.  Based on only these four
1818: objects it is difficult to make any conclusions about the whole
1819: sample, except that photo-$z$'s are more uncertain for fainter sources.
1820: 
1821: 
1822: 
1823: \begin{figure}
1824: \epsscale{1.2}
1825: \plotone{f20.eps}
1826: \caption{Photometric redshifts ($z_{\rm phot}$) from the
1827:   \citet{brod06} catalog versus spectroscopic redshifts ($z_{\rm
1828:   spec}$), for those objects with AGES optical spectra.  Squares show
1829:   IRAGN 1s, circles show IRAGN 2s.  BLAGNs are shown in blue, NLAGNs
1830:   are shown in orange, and optically normal galaxies are shown in
1831:   green.  The dashed line corresponds to $z_{\rm phot}=z_{\rm spec}$,
1832:   while the dotted lines show $z_{\rm spec}=0.7$ and $z_{\rm
1833:   phot}=0.7$.  Objects with no optical spectrum and with $z_{\rm
1834:   phot}<0.7$ would not be included in the IRAGN sample.
1835: \label{figz}}
1836: \end{figure}
1837: 
1838: 
1839:  
1840: \begin{figure}
1841: \epsscale{1.2}
1842: \plotone{f21.eps}
1843: \caption{Verification of \citet{brod06} photo-$z$'s ($z_{\rm phot}$)
1844: for IRAGN 2s using three-band optical galaxy template redshifts
1845: ($z_{\rm temp}$).  The dashed line corresponds to $z_{\rm temp}=z_{\rm
1846: phot}$.  Filled points have spectroscopic redshifts, while open points
1847: have only photo-$z$'s; note that we plot $z_{\rm phot}$ here even for
1848: objects with spectroscopic redshifts.
1849: \label{figtempz}}
1850: \end{figure}
1851:  
1852: 
1853: 
1854: 
1855: 
1856: 
1857: \subsection{Comparison to optical template redshifts}
1858: 
1859: 
1860: To test the photo-$z$'s for the entire IRAGN 2 sample, we note that
1861: most IRAGN 2s have galaxy-like optical colors (\S\
1862: \ref{galedd}). Therefore, for these sources we can perform a rough
1863: template photo-$z$ estimate by using only the optical photometry,
1864: fitting the $B_W$, $R$, and $I$ SED as in \S\ \ref{galedd}, but
1865: allowing the redshift to vary.  
1866: 
1867: The accuracy of the template fits is limited by the fact that we have
1868: only three optical photometric data points, so that the fits are
1869: underdetermined if they include too many free parameters.  We have
1870: tried fits using a wide range of galaxy and starburst templates with
1871: varying ages and extinctions, and consistently find that if we include
1872: two or more templates, the photo-$z$'s are poorly constrained.  We
1873: therefore use a single, non-evolving template, of which the elliptical
1874: galaxy model described in \S~\ref{lum} provides the best constraints
1875: over the wide range in redshift ($0.7<z\lesssim3$) covered by our
1876: sample.
1877: 
1878: The best-fit redshifts ($z_{\rm temp}$) from these template
1879: fits are shown in Fig.\ \ref{figtempz}.  The $z_{\rm temp}$ estimates
1880: follow the \citet{brod06} empirical photo-$z$'s reasonably well and
1881: cover the same range in redshift, except for a group of 80 objects
1882: that have very low $z_{\rm temp}<0.1$ (we note, however, that most of
1883: these sources have a second minimum in the $\chi^2$ function that lies
1884: within $\pm0.5$ of the $z_{\rm phot}$).   Excluding the sources with
1885: $z_{\rm temp}<0.1$, 79\% of the IRAGN 2s have $\left| z_{\rm
1886: temp}-z_{\rm phot} \right| < 0.25(1+z_{\rm phot})$, with a
1887: bias toward lower redshifts at $1.7<z<2.2$.  Objects
1888: with $z_{\rm phot}\sim2.5$ have a wide range in $z_{\rm temp}$, which
1889: may indicate that the real redshift distribution of the IRAGN 2s
1890: extends smoothly out to $z\gtrsim3$, similar to the IRAGN 1 sample.
1891: 
1892: We also obtain similar results with the ${\tt HyperZ}$ photometric
1893: redshift package \citep{bolz00}, using the same fixed, non-evolving
1894: template spectrum. These results give us confidence that the IRAGN 2s
1895: lie at redshifts $0.7\lesssim z \lesssim 3$ and that the photo-$z$'s have no
1896: systematic bias large enough to significantly affect the physical
1897: interpretation of our results.  
1898: 
1899: 
1900: 
1901: 
1902: \section{Sample contamination and completeness}
1903: \label{caveats}
1904: In the previous section we showed that our mid-IR and optical
1905: color classification for obscured AGNs is verified by the typical
1906: X-ray, IR, and optical properties of these objects.  Therefore, we are
1907: confident in the general technique of selecting obscured AGNs.
1908: However, to make estimates of how our IRAGN 2 sample relates to the
1909: total population of obscured AGNs, it is important to address issues
1910: of contamination and completeness.  
1911: 
1912: \subsection{Photometric uncertainties and color selection}
1913: \label{photoerr}
1914: 
1915: We first address the photometric uncertainty in the IRAC colors that
1916: are used to select the IRAGN.  Photometric error will lead some
1917: sources to move into or out of the \citetalias{ster05} selection
1918: region, causing contamination or incompleteness, respectively.  These
1919: will be dominated by the 5.8 and 8 \micron\ IRAC bands, which
1920: are less sensitive than the shorter wavelength bands; the $1 \sigma$
1921: uncertainty in the $[5.8]-[8.0]$ color is typically in the range
1922: $0.1-0.4$, compared to $0.02-0.08$ for $[3.6]-[4.5]$.
1923: 
1924: The color-color distribution indicates that incompleteness is a greater
1925: problem than contamination.  Fig.~\ref{figcol_stern_errors}(a) shows the
1926: IRAC color-color distribution, highlighting those objects with
1927: $S_{5.8}/\sigma_{5.8}>15$, where $\sigma_{5.8}$ is the error in the
1928: 5.8 \micron\ band flux.  This shows that the bright sources in the \citetalias{ster05} AGN region occupy a small locus in color-color space around a line defined by 
1929: \begin{equation}
1930: \label{eqstern}
1931: [3.6]-[4.5]=0.2([5.8]-[8.0])+1.8.
1932: \end{equation}
1933: The spread of points about this line is consistent with the
1934: photometric uncertainties.  For all sources lying above the lower
1935: boundary in the \citetalias{ster05} criteria (shown as black points in Fig.~\ref{figcol_stern_errors}(a), we derive the difference
1936: $\Delta C$ between the observed $[5.8]-[8.0]$ and the line defined
1937: above.  The distribution in $\Delta C/\sigma_C$, where $\sigma_C$ is
1938: the $1 \sigma$ uncertainty in the color, is shown in
1939: Fig.~\ref{figcol_stern_errors}(b), and is well fitted by a Gaussian
1940:  with mean $-0.04$ and $\sigma=1.05$.  This indicates that
1941: most of the objects with high $[3.6]-[4.5]$ can be associated with the
1942: \citetalias{ster05} region and in fact may occupy a remarkably tight
1943: locus in color-color space.  However, photometric errors cause
1944: $\sim$10\% to be observed outside the AGN selection region.
1945: Conversely, we only expect $\sim$100 sources to be scattered into this
1946: region, indicating that contamination due to photometric errors is
1947: $\lesssim$5\%.
1948: 
1949: 
1950: \begin{figure}
1951: \epsscale{1.2}
1952: %\plotone{oi_plot_ir.eps}
1953: \plotone{f22a.eps}
1954: \plotone{f22b.eps}
1955: \caption{(a) IRAC color-color distribution for the $\approx$15,500
1956:   IRAC sources with four-band $5\sigma$ detections.  We have excluded
1957:   from this figure 139 stars that are brighter than the saturation
1958:   limits given in \citet{eise04}; all have $[5.8]-[8.0]<0.4$ and so do
1959:   not lie in the \citetalias{ster05} AGN selection region (shown by
1960:   the red dashed line).  Sources with $[3.6]-[4.5]$ greater than the
1961:   lower boundary of the \citetalias{ster05} region are shown in black.
1962:   Objects with $S_{5.8}/\sigma_{5.8}>15$ are shown in red, and the
1963:   locus defined by these points (Eqn.~\ref{eqstern}) is shown by the
1964:   diagonal line.  (b) Distribution of the deviation  ($\Delta C$) of $[5.8]-[8.0]$ colors from the
1965:   diagonal line for the black points in (a), in units of
1966:   the $1 \sigma$ uncertainty in the $[5.8]-[8.0]$ color.  A Gaussian fit to the
1967:   distribution is shown in red, and is consistent with most points
1968:   having intrinsic colors defined by the line shown in (a).
1969: \label{figcol_stern_errors}}
1970: \vskip0.4cm
1971: \end{figure}
1972:  
1973: 
1974: 
1975: \subsection{Reliability of obscured AGN selection}
1976: It is important to estimate the reliability of our classification of
1977: IRAGNs based on IR-optical colors; that is, how many IRAGN 1s are
1978: actually obscured, and how many IRAGN 2s are unobscured?  In the
1979: sample of 839 IRAGN 1s, 719 have BLAGN spectra from the AGES data set,
1980: or have point-like optical morphologies (${\tt CLASS\_STAR}>0.7$) and
1981: are best fitted by blue (quasar or Sb) optical templates.  These are
1982: strong indicators that a source is an unobscured, type 1 AGN, so the
1983: color selection is at least 85\% reliable for IRAGN 1s.  Of the 640
1984: IRAGN 2s, 517 have galaxy-like optical colors and ${\tt
1985: CLASS\_STAR}\leq0.5$, and do not have BLAGN optical spectra (only 29,
1986: or 3\%, of the IRAGNs with BLAGN spectra are classified as IRAGN 2s).  These criteria
1987: only indicate that an object is dominated by the host galaxy in the
1988: optical; however, as discussed in \S\ \ref{identify}, the high IRAC
1989: luminosities of these sources would suggest dominant nuclear emission
1990: in the optical, were they unobscured.  We conclude that our selection
1991: of obscured AGNs based on optical-IR color is at least 80\% reliable.
1992: 
1993: 
1994: 
1995: \subsection{Contamination from starburst and normal galaxies}
1996: \label{contamination}
1997: 
1998: We expect almost all of the IRAGN 1s to be AGNs rather than starburst
1999: or normal galaxies.  Most of these objects were targeted by AGES, so
2000: we can verify their classification with optical spectra.  Of the 686
2001: IRAGN 1s that have AGES spectroscopy (or 82\% of the IRAGN 1 sample),
2002: 668 (97\%) are BLAGNs, while 3 are NLAGNs and 15 are optically normal
2003: galaxies.  Keeping in mind that many AGES targets were selected to be
2004: X-ray sources and thus are biased toward bright AGNs, we consider
2005: separately the 233 IRAGN 1s that have optical spectra but no X-ray
2006: counterpart.  These should be a representative sample of the IRAGN 1s
2007: that are not detected in X-rays, and of these 226 (97\%) are BLAGNs, one
2008: is a NLAGN, and five are optically normal galaxies.  We conclude that
2009: there is little ($<$5\%) contamination in the IRAGN 1 sample.
2010: 
2011: It is more difficult to estimate contamination in the IRAGN 2s.  Only
2012: 42 IRAGN 2s have AGES spectra (29 BLAGNs, 1 NLAGN, 12 galaxies),
2013: because most IRAGN 2s are fainter than the AGES flux limits (Fig.\
2014: \ref{figr}).  A total of 155 of the IRAGN 2s have X-ray detections and
2015: thus \lx\ values that imply that they must be powered by accretion.
2016: Of the remaining 485 objects, some at high redshifts might not be AGNs
2017: but instead luminous starburst galaxies with IRAC colors that lie
2018: inside the \citetalias{ster05} AGN color-color selection region.  As
2019: mentioned in \S~\ref{lum}, heavily extincted starbursts (i.e., Arp
2020: 220) can have very red IRAC colors.  However, the \citet{sieb07} Arp
2021: 220 template is also very red in the optical ($B_W-I>3$ at $z>0.7$),
2022: which is much redder than observed for the IRAGN 2s
2023: (Fig.~\ref{figgalagncol}).  Still, it is possible that some high-$z$
2024: starbursts have similar IRAC colors to Arp 220 but are bluer in the
2025: optical, and these could contaminate the IRAGN 2 sample.  In addition,
2026: at $z\gtrsim 3$, the colors of less obscured starbursts (e.g., M82)
2027: would also lie in the \citetalias{ster05} region \citep[see Fig.~6
2028: of][]{barm06}.  However, to be detected to our IRAC flux limits at
2029: $z>3$, a source must have a very high $L_{\rm 5.8 \mu m}>10^{12}$
2030: \lsun\ (where $L_{\rm 5.8 \mu m}$ is the observed $\nu L_\nu$ in the
2031: 5.8 \micron\ band).  For a typical ratio of rest-frame far-infrared
2032: (FIR) to observed 5.8 \micron\ fluxes for starburst galaxies
2033: (\S~\ref{xrayres}), this implies $L_{\rm FIR}\gtrsim10^{13}$ \lsun.
2034: In most such ``hyperluminous infrared galaxies'', a significant (and
2035: often dominant) contribution to the IR emission comes from an AGN
2036: \citep[e.g.,][]{farr02}.  Also considering that our IRAGN sample
2037: contains only 27 objects at $z>3$ that do not have BLAGN optical
2038: classifications, contamination from such high-$z$ starbursts should be
2039: small.
2040: 
2041: 
2042: One empirical constraint on contamination comes from the X-ray
2043: stacking results, due to the fact that starburst galaxies tend to be
2044: significantly fainter in the X-rays than AGNs.  If we exclude
2045: sources that have X-ray counterparts, the IRAGN 1s and 2s have similar
2046: average X-ray fluxes in the 2--7 keV band of $0.47\pm0.06$ and
2047: $0.46\pm0.05$ counts source$^{-1}$, respectively (Table
2048: \ref{tblxray}).  Because there is little contamination in the IRAGN 1
2049: sample, 0.47 counts source$^{-1}$ should be typical for IR-selected
2050: AGNs that are fainter than the X\bootes\ detection limit.  We thus
2051: consider the possibility that the AGNs among the X-ray--undetected
2052: IRAGN 2s have the same average flux, but the sample is 40\%
2053: contaminated by starburst galaxies, which have 0.5--7 keV fluxes that
2054: are 5 times smaller.  The observed average flux from stacking would
2055: then be 68\% of that for the IRAGN 1s, or 0.32 counts source$^{-1}$,
2056: which is $\simeq3\sigma$ below the observed value.  We are therefore
2057: confident that $<40\%$ of the 485 X-ray--undetected IRAGN 2s are
2058: contaminating starbursts, implying a $3 \sigma$ upper limit of
2059: $\sim$30\% contamination for the total sample of IRAGN 2s.
2060: 
2061: Results from deeper surveys can help put more concrete limits on
2062: contamination.  \citet{alon06} examined a population of objects in the
2063: Chandra Deep Field-South (CDF-S) selected using the
2064: \citetalias{ster05} IRAC color-color criteria.  Based on X-ray
2065: luminosities and spectral shapes for the individual sources,
2066: \citet{alon06} find that at least 70\% of the IR-selected objects are
2067: AGNs.  We conclude that while it is difficult to accurately estimate
2068: the contamination by normal galaxies of the IRAGN 2 sample, we expect
2069: it to be no larger than $\sim$30\%.  In addition, a further $\sim$10\%
2070: contamination of the IRAGN 2 sample could come from objects at $z<0.7$
2071: (as discussed in \S~\ref{specphotz}), although many of these would
2072: likely be AGNs rather than galaxies.
2073: 
2074: 
2075: \subsection{Sample completeness}
2076: \label{incompleteness}
2077: 
2078: We next estimate our selection completeness; that is, of the AGNs
2079: brighter than the flux limits of the survey, how many are included in
2080: the IRAGN sample?  For AGNs with broad-line optical spectra at
2081: $z>0.7$, the IRAC color-color selection is highly complete to the IRAC
2082: flux limits.  The AGES sample contains 1306 BLAGNs at $z>0.7$ in the
2083: area observed by IRAC, of which 784 (60\%) have $5\sigma$ detections
2084: in all four IRAC bands.  Of these, 697 (89\%) have IRAC colors in the
2085: \citetalias{ster05} selection region.  Of four NLAGNs with $z>0.7$, all
2086: have four-band IRAC detections and are selected by the
2087: \citetalias{ster05} criteria.
2088: 
2089: For optically faint or obscured AGNs, however, the completeness is more
2090: difficult to estimate.  Of the 1298 X\bootes\ sources with four-band 
2091: $5\sigma$ IRAC counterparts (almost all of which are AGNs), 879 (68\%)
2092: are selected by the \citetalias{ster05} criteria.  Likewise, in the
2093: much deeper \spitzer\ and \chandra\ data from the EGS, \citet{barm06}
2094: find that only $\sim$50\% of X-ray AGNs are selected by the
2095: \citetalias{ster05} criteria.
2096: 
2097: This incompleteness can be caused by either obscuration or dilution.
2098: Heavy obscuration can absorb even mid-IR emission.  We consider an AGN
2099: with $N_{\rm H}=6\times10^{23}$ \cdens\ (roughly 20 times higher than
2100: the typical column for the IRAGN 2s), for which an SMC gas-to-dust ratio of
2101: $N_{\rm H}/A_V\simeq2\times10^{22}$ \cdens\  implies $A_V=30$.
2102: This corresponds to a rest-frame extinction at 2 \micron\ of 3.6
2103: mag (this is largely independent of the choice of extinction
2104: curve, which are very similar redward of the $V$ band); for a smaller
2105: Galactic dust-to-gas ratio, the IR extinction would be even higher.
2106: Therefore, high column densities can obscure the nucleus such that
2107: either the IRAC fluxes drop below our detection limits or the IRAC
2108: color-color selection criteria would not select such an object as an
2109: AGN (note that Fig.~\ref{figcol} shows that at $z\gtrsim0.7$, sources
2110: with $A_V\gtrsim30$ move out of the \citetalias{ster05} AGN color
2111: selection).  For these reasons, we expect our IRAGN sample to include
2112: very few highly absorbed objects ($N_{\rm H}\gg10^{23}$ \cdens\ in the
2113: X-ray).
2114: 
2115: AGNs can also be missed if their IR emission is diluted by starburst
2116: activity.  The luminosities of the IRAGNs in our sample, $L_{\rm 4.5
2117: \mu m}\sim(0.3-3)\times10^{12}$ \lsun, are comparable to that of
2118: luminous starburst galaxies \citep[e.g.,][]{rowa05}.  There is
2119: compelling evidence that starburst activity and AGN activity are often
2120: linked \citep{ho05,king05, farr03}, so we expect some sources with an
2121: AGN also to have a powerful starburst that dominates the mid-IR
2122: luminosity.  Such an object would have a starburst-like SED
2123: (corresponding to a low $f_{\rm AGN}$ as in Fig.~\ref{figcol}) and
2124: would not be selected using the AGN color-color technique.
2125: 
2126: One way to estimate this incompleteness is to examine bright radio
2127: sources with relatively faint IR counterparts.  These are likely to be
2128: AGNs and not starbursts, and the radio emission will not be strongly
2129: affected by extinction.  The VLA FIRST 20 cm radio survey
2130: \citep{beck95} detects 301 radio sources that are brighter than 5 mJy
2131: in the area covered by IRAC.  Of these, 24 are matched to the
2132: four-band IRAC catalog with 3.6 $\mu$m magnitude fainter than 15.  The
2133: \citetalias{ster05} color-color criteria select 14 of these objects as
2134: AGNs, of which 11 are in our IRAGN sample at $z>0.7$.  Of these 11
2135: sources, 7 are IRAGN 1s and 4 are IRAGN 2s.  These results suggest
2136: that the completeness of our IRAC color-color selection may be as low
2137: as $\sim$60\% for AGNs at our IRAC flux limits.  However, radio-loud
2138: AGNs may be different from the more numerous radio-quiet objects, so
2139: it is difficult to draw conclusions about the total AGN population,
2140: except to say that incompleteness effects may be significant.
2141: 
2142: 
2143: 
2144: \begin{figure}
2145: \epsscale{1.2}
2146: \plotone{f23.eps}
2147: \caption{Cumulative redshift distribution of IRAGNs, compared to
2148:   predictions using optical and X-ray luminosity functions, and considering the
2149:   \bootes\ flux limits.  See text in \S~\ref{compmodel} for discussion.
2150:   \label{figmodel}}
2151: \end{figure}
2152: 
2153: \subsection{Comparison to predictions from luminosity functions}
2154: \label{compmodel}
2155: Another check on the completeness of our selection is to compare
2156: the number of IRAGNs in our sample with the number that are predicted by
2157: optical, IR, and X-ray luminosity functions, accounting for the
2158: \bootes\ flux limits.  Our AGN detection is usually limited by the 5.8
2159: $\mu$m band (the 8 $\mu$m band has a similar flux limit, but since the
2160: quasars have red SEDs, they are usually brighter at 8 $\mu$m than at
2161: 5.8 $\mu$m).  To convert X-ray and optical luminosities to IRAC fluxes,
2162: we use the AGN model (SDSS optical spectrum plus mid-IR power law)
2163: described in \S\ \ref{lum}.  We include a distribution in IR slopes,
2164: modeled by a Gaussian with $\average{\alpha_\nu}=-1$ and
2165: $\sigma_{\alpha_\nu}=0.5$.  This distribution approximately reproduces
2166: the observed dispersion in the optical to IR colors for type 1 AGNs
2167: from \citet{rich06}, as shown in Fig.~\ref{figtemp}.
2168: 
2169: We first compare the IRAGN 1 sample to the predictions of the
2170: broad-line quasar luminosity function of \citet{rich05} from the 2QZ
2171: survey. Unlike more recent luminosity functions derived from SDSS
2172: data, this data set includes objects below the ``knee'' of the
2173: luminosity function.  This model SED, convolved with the
2174: \citet{rich05} luminosity function, predicts 840 type 1 quasars
2175: brighter than our flux limits in the 8.5 deg$^2$ field covered by IRAC
2176: (see the the green dashed line in Fig.~\ref{figmodel}).  We detect 839
2177: IRAGN 1s, very close to this total.  This indicates that the IRAC
2178: color selection is highly complete for broad-line quasars (although
2179: some highly-reddened IRAGN 1s might not be included in the
2180: \citet{rich05} sample).
2181: 
2182: We also evaluate predictions for X-ray luminosity functions.  We use
2183: the same model UV/IR spectrum described above, and take the
2184: relationship between UV luminosity and the UV/X-ray spectral slope
2185: $\alpha_{\rm ox}$ from \citet{stef06}.  For simplicity, we assume a
2186: constant unabsorbed X-ray spectrum with $\Gamma=2$.  The 0.5--2 keV
2187: luminosity function for unabsorbed (type 1) quasars of \citet{hasi05}
2188: predicts 970 AGNs above our flux limits at $z>0.7$ (see the red
2189: dashed curve in Fig.~\ref{figmodel}), or 16\% more than
2190: the number of IRAGN 1s.  
2191: 
2192: In contrast, the 2--10 keV luminosity function of \citet{ueda03}
2193: predicts $\sim$2300 total AGNs, 55\% more than we detect.
2194: However, although the \citet{ueda03} X-ray sample includes many AGNs
2195: that are X-ray absorbed, our model AGN SED used here does not
2196: include corresponding dust extinction.  The solid blue line in
2197: Fig.~\ref{figmodel} therefore represents the number of total AGNs that would
2198: be observed at the \bootes\ flux limits, in the absence of dust
2199: extinction.
2200: 
2201: The effects of dust extinction were included by \citet{hopk07qlf}, who
2202: determined a parametrization of the {\it bolometric} luminosity
2203: function by fitting observed X-ray, optical, and IR luminosity
2204: functions.  This work used the distribution in \nh\ observed by
2205: \citet{ueda03}, and a typical Galactic gas-to-dust ratio, in
2206: predicting the numbers of observed AGNs.  The total number of AGNs
2207: for the \bootes\ 4.5 $\mu$m flux limit is shown by the dot-dashed cyan line in
2208: Fig.~\ref{figmodel}.  The model predicts 1320 detectable AGNs at
2209: $z>0.7$, 10\% fewer than the total number of IRAGNs we observe.
2210: 
2211: We note that the \citet{hopk07qlf} predictions may provide only a
2212: lower limit on the number of detectable AGNs.  If AGNs typically have
2213: a gas-to-dust ratio that is higher than the Galactic value (see
2214: \S~\ref{dust}), this would tend to decrease the dust extinction for a
2215: given \nh\ distribution, and so would increase the number of
2216: detectable AGNs at a given flux limit.  Still, the agreement between
2217: the observed number of IRAGN 1s and 2s and the predictions from
2218: luminosity functions indicates that our selection is reasonably
2219: complete {\it to the flux limits of the survey}.  However, we are
2220: likely missing half or more of the total obscured AGN population
2221: because dust extinction causes them to fall below our IRAC flux limits
2222: or out of the \citetalias{ster05} AGN color selection.
2223: 
2224: 
2225: 
2226: 
2227: \section{Discussion}
2228: \label{discussion}
2229: 
2230: 
2231: \subsection{Comparison to other obscured AGN samples}
2232: 
2233: 
2234: 
2235:  
2236: \begin{figure}
2237: \epsscale{1.2}
2238: \plotone{f24.eps}
2239: \caption{$L_{\rm bol}$ versus redshift for four large samples
2240:   of obscured AGNs: optically selected objects from SDSS
2241:   \citep[green triangles]{zaka03}, X-ray selected objects from the CDF-S
2242:   \citep[blue stars]{zhen04}, IR-selected AGNs from the \chandra/SWIRE
2243:   survey \citep[orange squares]{poll06}, and IRAGNs from this work
2244:   (red circles).  $L_{\rm bol}$ is estimated using bolometric corrections
2245:   from \citet{hopk07qlf}, as described in \S\ \ref{othersamples}.
2246:   \label{figz_lum}}
2247: \vskip0.5cm
2248: \end{figure}
2249:  
2250: 
2251: 
2252: 
2253: 
2254: \label{othersamples}
2255: In order to perform a complete census of obscured accretion in the
2256: Universe, it is important to place the IRAGN 2 sample described
2257: in this paper in the context of obscured AGNs found in other surveys.
2258: We compare the distribution in $z$ and \lbol\ of our sample with three
2259: of the largest ($>$100 objects) samples of obscured AGNs with redshift
2260: estimates.
2261: 
2262: The first sample consists of 291 optically-selected, luminous type 2
2263: AGNs at $0.3<z<0.83$ from the SDSS \citep{zaka03}.  To estimate
2264: intrinsic \lbol\ from the observed spectroscopic properties, we assume
2265: that the unobscured SED of the type 2 quasars in this sample is
2266: similar to that for type 1 optical AGNs.  In this case we use Equation
2267: (8) of \citet{zaka03} to convert the observed [\ion{O}{3}]
2268: $\lambda$5007 luminosity to the intrinsic, unobscured $L_B$.  We then use the bolometric corrections from the model of \citet{hopk07qlf}, for which $BC_B=9-12$.
2269: 
2270: The second sample consists of 145 X-ray sources in the CDF-S that are
2271: classified as type 2 AGNs on the basis of their X-ray spectrum and/or
2272: optical properties \citep{zhen04}.  Of these objects, 68 have
2273: spectroscopic redshifts, while 78 have photo-$z$ estimates.  For these
2274: objects we estimate an unabsorbed $L_{\rm 0.5-8\; keV}$ on the basis
2275: of the redshift, observed flux, and HR, assuming an intrinsic
2276: power-law spectrum with $\Gamma=1.8$.  We then convert these values to
2277: $L_{\rm bol}$, again using the model of \citet{hopk07qlf}, for which
2278: $BC_{\rm 0.5-8\; keV}\simeq20-80$.
2279: 
2280: The third sample consists of the 120 obscured AGNs selected from
2281: optical and IR SEDs in the \chandra/SWIRE survey by \citet{poll06}.
2282: Of these objects, 11 have spectroscopic redshifts, and the remainder
2283: have photo-$z$'s.  We calculate \lbol\ from the mid-IR luminosity
2284: \lirac, as described in \S\ \ref{bolometric}.  As a further check of
2285: the robustness of these $L_{\rm bol}$ estimates, we estimate $L_{\rm
2286: bol}$ for the 41 objects in the survey with X-ray detections, using
2287: the unabsorbed X-ray luminosities calculated by \citet{poll06} and the
2288: \citet{hopk07qlf} SED model.  We find that while there is significant
2289: scatter between the IR and X-ray estimates of $L_{\rm bol}$, on
2290: average they agree to within a factor of $\lesssim 2$.  Nevertheless,
2291: these discrepancies between $L_{\rm bol}$ estimates require that our
2292: comparison between samples selected at different wavelengths is at best
2293: qualitative.
2294: 
2295: These samples of obscured AGNs selected in the optical \citep{zaka03},
2296: X-ray \citep{zhen04}, and IR \citep[this work]{poll06} are shown
2297: in Fig.\ \ref{figz_lum}.  The \bootes\ IRAGN 2s are more luminous than
2298: the X-ray selected AGNs from the CDF-S and at higher redshifts than
2299: those in the SDSS sample.  The SWIRE obscured AGNs and \bootes\ IRAGN
2300: 2s have similar distributions in $z$ and \lbol, but because of the
2301: much larger area in the \bootes\ field (8.5 deg$^2$ vs. 0.6
2302: deg$^2$), the \bootes\ IRAGN 2 sample contains $\sim$6 times more
2303: objects.  We note that even accounting for possible 30\% contamination (\S\
2304: \ref{contamination}), the \bootes\ IRAGN 2s contain the largest sample
2305: to date of luminous, moderately obscured AGNs at high redshift.
2306: 
2307: 
2308: \subsection{Contribution to the cosmic X-ray background}
2309: \label{cxb}
2310: Synthesis models require a population of obscured AGNs to produce the
2311: intensity and spectral shape of the CXB.  From our stacking analysis,
2312: we estimate the contribution to the total CXB of the two IRAGN types
2313: defined in this paper.  The extragalactic component of the CXB is
2314: well modeled by a power law with $\Gamma=1.4$ and normalization $10.9\pm0.5$
2315: photons cm$^{-1}$ s$^{-1}$ keV$^{-1}$ sr$^{-1}$ at 1 keV
2316: \citep[e.g.,][]{hick06a,delu04}.  This gives CXB intensities of
2317: $7.6\times10^{-12}$ \intens\ in the 0.5--2 keV band and
2318: $1.5\times10^{-11}$ \intens\ in the 2--7 keV band.
2319: 
2320: Over the 2.9 deg$^2$ for which we perform the X-ray stacking
2321: analysis, the total fluxes listed in Table \ref{tblxray} correspond to
2322: intensities for the IRAGN 1s of $1.2\times10^{-12}$ \intens\ in the
2323: 0.5--2 keV band and $1.3\times10^{-12}$ \intens\ in the 2--7 keV
2324: band, which represent 16\% and 9\%, respectively, of the total CXB.
2325: For the IRAGN 2s the intensities are $3.0\times10^{-13}$ \intens\ in
2326: the 0.5--2 keV band and $6.0\times10^{-13}$ \intens\ in the 2--7 keV
2327: band, or 4\% and 4\% of the CXB, respectively.  This indicates that
2328: selecting IR AGNs in shallow exposures and at $z>0.7$ only captures a
2329: small fraction of the sources that produce the X-ray background.  
2330: 
2331: \subsection{Implications of a bimodal distribution in obscuration}
2332: \label{impl}
2333: The bimodal distribution in \iiu\ (and accordingly $A_V$) observed in
2334: the IRAGNs may give clues to the distribution of material that
2335: obscures the central engine.  A detailed comparison of the
2336: distribution of dust extinction to AGN obscuration models is beyond
2337: the scope of this paper, but we qualitatively consider two
2338: explanations for the bimodal extinction: a hard-edged torus, or
2339: obscuration as an evolutionary phase.
2340: 
2341: In the unified model, the obscuring material is in an extended
2342: distribution that surrounds the nucleus on scales of $\lesssim100$ pc,
2343: possibly in the shape of a torus, such that the level of obscuration
2344: depends on the observer's line of sight. These obscuring structures
2345: are well-established for local Seyfert galaxies \citep[see ][and
2346: references therein]{anto93}, and there is evidence that they also
2347: exist in more distant, luminous quasars, for example from the
2348: detection of broad emission lines in polarized light that is scattered
2349: from the nucleus \citep{zaka05}.  In some models of the torus
2350: \citep[e.g.,][]{trei04, ibar07}, the obscuring medium is not
2351: homogeneous but varies in density with radius and angle from the axis
2352: of symmetry, so that there is a slow increase in obscuring column as
2353: the torus is seen more edge-on.  However, this slow increase is
2354: inconsistent with the $A_V$ distribution we observe.  Instead, a
2355: bimodal $A_V$ distribution could indicate an abrupt edge to the
2356: obscuring material rather than a smooth distribution and so provides a
2357: constraint on the obscuring geometry.
2358: 
2359: Alternatively, the obscuring material could be in the form of
2360: irregular clouds that surround the nucleus on scales as large as
2361: kiloparsecs.  This material can be driven to the center of the galaxy
2362: by major galaxy mergers and can feed the AGNs (as well as nuclear
2363: starbursts) while also obscuring the central engine
2364: \citep[e.g.,][]{sand88,hopk06merge}.  In time, AGN winds may blow this
2365: material away from the nucleus, leading to an unobscured phase of AGN
2366: activity \citep[e.g.,][]{silk98, spri05,hopk06apjs}.  In this picture,
2367: obscured accretion is an evolutionary stage in the life of the quasar;
2368: as long as the blowout phase is short-lived, quasars will be seen with
2369: either significant or very little obscuration.  Therefore, the bimodal
2370: distribution we observe may place constraints on the timescale for AGN
2371: feedback.
2372:  
2373: \vspace{1cm}
2374: 
2375: 
2376: \section{Summary}
2377: \label{summary}
2378: 
2379: 
2380: In this paper we analyze a sample of 1479 AGNs at $0.7<z\lesssim3$
2381: from the wide-field multiwavelength \bootes\ survey, selected on the
2382: basis of their IRAC colors.   This work has two key elements
2383: that together make it unique among studies of IR-selected AGN: (1) the
2384: wide area and deep optical photometry in the \bootes\ field allow us
2385: to identify a large number of obscured sources, and (2) the contiguous
2386: X-ray coverage allows us to verify independently that the IRAGN 2s are
2387: obscured AGNs, and to measure their neutral gas column densities. 
2388: 
2389: Key results of this paper are as follows:
2390: \begin{enumerate}
2391: 
2392: \im The optical-IR color distribution of the IR-selected AGNs is
2393: bimodal, with a boundary of $R-[4.5]=6.1$ (Vega) between the two
2394: subsets.  Based on this color criterion, we divide our sample into 640
2395: obscured (IRAGN 2) and 839 unobscured (IRAGN 1) AGNs.  The optical-IR
2396: color distribution can be interpreted in terms of dust extinction of
2397: the nuclear optical emission for the IRAGN 2s.  The obscured AGN color
2398: selection is valid for AGNs at $z>0.7$ and with mid-IR luminosities
2399: $\gtrsim10^{11}$ \lsun.
2400: 
2401: \im X-ray and optical data confirm our selection of obscured AGNs.
2402: X-ray stacking shows that both subsets of IRAGNs have average X-ray
2403: luminosities characteristic of luminous AGNs.  The IRAGN 1s have
2404: average X-ray hardness ratios typical of unobscured sources, while the
2405: IRAGN 2s have harder X-ray spectra, corresponding to absorption with
2406: typical $N_{\rm H}\sim3\times10^{22}$ \cdens.  The optical colors and
2407: morphologies are typical of galaxies for most IRAGN 2s and of quasars
2408: for most IRAGN 1s, consistent with the optical emission from the IRAGN
2409: 2s being extincted.
2410: 
2411: \im For a typical range of AGN gas-to-dust ratios, the \nh\ for the
2412: IRAGN 2s (derived from X-ray stacking) corresponds to $1\lesssim A_V
2413: \lesssim10$, consistent with the $A_V$ values derived from optical-IR
2414: SED fits, and sufficient to completely extinct the nuclear optical
2415: emission.  This indicates that, on average, absorption by neutral gas
2416: and extinction by dust are correlated in these luminous AGNs.
2417: 
2418: \im The \citetalias{ster05} IRAC color-color AGN selection is
2419: reasonably complete to our survey  flux limits. The numbers
2420: of IRAGN 1s and 2s are within $\sim$15\% of predictions from
2421: optical and X-ray luminosity functions.  We expect the optical/IR
2422: color selection to be at least 80\% reliable in distinguishing between
2423: unobscured and obscured AGNs, while contamination from starburst
2424: galaxies in the IRAGN 2 sample should be at most $\sim$30\% and is likely
2425: much lower.
2426: 
2427: \im The bimodal distribution in optical-IR color for IRAGNs suggests
2428: that these objects have either low ($A_V\lesssim0.1$) or significant
2429: ($A_V\gtrsim0.7$) extinction.  This distribution may have implications
2430: for models of AGN obscuration.  In the context of the unified model,
2431: this may imply a hard edge to distribution of obscuring material.
2432: Alternatively, obscuration may be an evolutionary phase that is
2433: followed by rapid blowout of the obscuring dust, leading to a bimodal
2434: distribution in $A_V$.
2435: 
2436: \im The IRAGN 2s comprise the largest sample to date of AGNs with high
2437: redshifts ($0.7<z<3$), high bolometric luminosities
2438: ($10^{45}\lesssim L_{\rm bol}\lesssim 10^{47}$ \ergs), and moderate absorption
2439: ($10^{22}\lesssim N_{\rm H}\lesssim 10^{23}$ \cdens), even after accounting
2440: for possible sample contamination of at most $\sim$30\%.   This work
2441: shows that IRAC and
2442: optical selection is a powerful tool for identifying  large numbers
2443: of luminous,
2444: obscured AGNs for follow-up study.
2445: 
2446: \end{enumerate}
2447: 
2448: 
2449: 
2450: \acknowledgements We thank our colleagues on the AGES, IRAC Shallow
2451: Survey, NDWFS, and X\bootes\ teams, and Ramesh Narayan, Michael Pahre,
2452: Pauline Barmby, and Kamson Lai for productive discussions.  We are
2453: grateful to the referee for suggestions that significantly
2454: strengthened the paper.  This paper would not have been possible
2455: without the efforts of the \chandra, \spitzer, KPNO, and MMT support
2456: staffs.  This work is based in part on observations made by the {\it
2457: Spitzer Space Telescope}, which is operated by the Jet Propulsion
2458: Laboratory, California Institute of Technology under a contract with
2459: NASA.  This research was supported by the National Optical Astronomy
2460: Observatory, which is operated by the Association of Universities for
2461: Research in Astronomy (AURA), Inc., under a cooperative agreement with
2462: the National Science Foundation.  Optical spectroscopy discussed in
2463: this paper was obtained at the MMT Observatory, a joint facility of
2464: the Smithsonian Institution and the University of Arizona.  R.C.H. was
2465: supported by a NASA GSRP Fellowship and a Harvard Merit Fellowship.
2466: 
2467: \begin{thebibliography}{99}
2468: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi
2469: 
2470: \bibitem[{{Akylas} {et~al.}(2006){Akylas}, {Georgantopoulos}, {Georgakakis},
2471:   {Kitsionas}, \& {Hatziminaoglou}}]{akyl06}
2472: {Akylas}, A., {Georgantopoulos}, I., {Georgakakis}, A., {Kitsionas}, S., \&
2473:   {Hatziminaoglou}, E. 2006, \aap, 459, 693
2474: 
2475: \bibitem[{{Alexander} {et~al.}(2005){Alexander}, {Bauer}, {Chapman}, {Smail},
2476:   {Blain}, {Brandt}, \& {Ivison}}]{alex05}
2477: {Alexander}, D.~M., {Bauer}, F.~E., {Chapman}, S.~C., {Smail}, I., {Blain},
2478:   A.~W., {Brandt}, W.~N., \& {Ivison}, R.~J. 2005, \apj, 632, 736
2479: 
2480: \bibitem[{{Alonso-Herrero} {et~al.}(2006){Alonso-Herrero}, {P{\'e}rez-Gonz
2481:   {\'a}lez}, {Alexander}, {Rieke}, {Rigopoulou}, {Le Floc'h}, {Barmby},
2482:   {Papovich}, {Rigby}, {Bauer}, {Brandt}, {Egami}, {Willner}, {Dole}, \&
2483:   {Huang}}]{alon06}
2484: {Alonso-Herrero}, A., et~al.\ 2006, \apj, 640, 167
2485: 
2486: \bibitem[{{Antonucci}(1993)}]{anto93}
2487: {Antonucci}, R. 1993, \araa, 31, 473
2488: 
2489: \bibitem[{{Awaki} {et~al.}(1991){Awaki}, {Koyama}, {Inoue}, \&
2490:   {Halpern}}]{awak91}
2491: {Awaki}, H., {Koyama}, K., {Inoue}, H., \& {Halpern}, J.~P. 1991, \pasj, 43,
2492:   195
2493: 
2494: \bibitem[{{Ballantyne} {et~al.}(2006){Ballantyne}, {Shi}, {Rieke}, {Donley},
2495:   {Papovich}, \& {Rigby}}]{ball06b}
2496: {Ballantyne}, D.~R., {Shi}, Y., {Rieke}, G.~H., {Donley}, J.~L., {Papovich},
2497:   C., \& {Rigby}, J.~R. 2006, \apj, 653, 1070
2498: 
2499: \bibitem[{{Barger} {et~al.}(2005){Barger}, {Cowie}, {Mushotzky}, {Yang},
2500:   {Wang}, {Steffen}, \& {Capak}}]{barg05}
2501: {Barger}, A.~J., {Cowie}, L.~L., {Mushotzky}, R.~F., {Yang}, Y., {Wang}, W.-H.,
2502:   {Steffen}, A.~T., \& {Capak}, P. 2005, \aj, 129, 578
2503: 
2504: \bibitem[{{Barmby} {et~al.}(2006){Barmby}, {Alonso-Herrero}, {Donley}, {Egami},
2505:   {Fazio}, {Georgakakis}, {Huang}, {Laird}, {Miyazaki}, {Nandra}, {Park},
2506:   {P{\'e}rez-Gonz{\'a}lez}, {Rieke}, {Rigby}, \& {Willner}}]{barm06}
2507: {Barmby}, P., et~al.\ 2006, \apj, 642, 126
2508: 
2509: \bibitem[{{Becker} {et~al.}(2007){Becker}, {Rauch}, \& {Sargent}}]{beck07}
2510: {Becker}, G.~D., {Rauch}, M., \& {Sargent}, W.~L.~W. 2007, \apj, 662, 72
2511: 
2512: \bibitem[{{Becker} {et~al.}(1995){Becker}, {White}, \& {Helfand}}]{beck95}
2513: {Becker}, R.~H., {White}, R.~L., \& {Helfand}, D.~J. 1995, \apj, 450, 559
2514: 
2515: \bibitem[{{Bertin} \& {Arnouts}(1996)}]{bert96}
2516: {Bertin}, E. \& {Arnouts}, S. 1996, \aaps, 117, 393
2517: 
2518: \bibitem[{{Best} {et~al.}(2005){Best}, {Kauffmann}, {Heckman}, {Brinchmann},
2519:   {Charlot}, {Ivezi{\'c}}, \& {White}}]{best05}
2520: {Best}, P.~N., {Kauffmann}, G., {Heckman}, T.~M., {Brinchmann}, J., {Charlot},
2521:   S., {Ivezi{\'c}}, {\v Z}., \& {White}, S.~D.~M. 2005, \mnras, 362, 25
2522: 
2523: \bibitem[{{Bolzonella} {et~al.}(2000){Bolzonella}, {Miralles}, \&
2524:   {Pell{\'o}}}]{bolz00}
2525: {Bolzonella}, M., {Miralles}, J.-M., \& {Pell{\'o}}, R. 2000, \aap, 363, 476
2526: 
2527: \bibitem[{{Brand} {et~al.}(2006){Brand}, {Brown}, {Dey}, {Jannuzi}, {Kochanek},
2528:   {Kenter}, {Fabricant}, {Fazio}, {Forman}, {Green}, {Jones}, {McNamara},
2529:   {Murray}, {Najita}, {Rieke}, {Shields}, \& {Vikhlinin}}]{brand06a}
2530: {Brand}, K., et~al.\ 2006, \apj, 641, 140
2531: 
2532: \bibitem[{{Brand} {et~al.}(2005){Brand}, {Dey}, {Brown}, {Watson}, {Jannuzi},
2533:   {Najita}, {Kochanek}, {Shields}, {Fazio}, {Forman}, {Green}, {Jones},
2534:   {Kenter}, {McNamara}, {Murray}, {Rieke}, \& {Vikhlinin}}]{brand05}
2535: {Brand}, K., et~al.\ 2005, \apj, 626, 723
2536: 
2537: \bibitem[{{Brand} {et~al.}(2007){Brand}, {Weedman}, {Desai}, {Le Floc'h},
2538:   {Armus}, {Dey}, {Houck}, {Jannuzi}, {Smith}, \& {Soifer}}]{bran07spec_aph}
2539: {Brand}, K., et~al.\ 2007, \apj\ in press (arXiv:0709.3119), 709
2540: 
2541: \bibitem[{{Brandt} \& {Hasinger}(2005)}]{bran05}
2542: {Brandt}, W.~N. \& {Hasinger}, G. 2005, \araa, 43, 827
2543: 
2544: \bibitem[{{Brodwin} {et~al.}(2006){Brodwin}, {Brown}, {Ashby}, {Bian}, {Brand},
2545:   {Dey}, {Eisenhardt}, {Eisenstein}, {Gonzalez}, {Huang}, {Jannuzi},
2546:   {Kochanek}, {McKenzie}, {Murray}, {Pahre}, {Smith}, {Soifer}, {Stanford},
2547:   {Stern}, \& {Elston}}]{brod06}
2548: {Brodwin}, M., et~al.\ 2006, \apj, 651, 791
2549: 
2550: \bibitem[{{Brown} {et~al.}(2006){Brown}, {Brand}, {Dey}, {Jannuzi}, {Cool}, {Le
2551:   Floc'h}, {Kochanek}, {Armus}, {Bian}, {Higdon}, {Higdon}, {Papovich},
2552:   {Rieke}, {Rieke}, {Smith}, {Soifer}, \& {Weedman}}]{brow06}
2553: {Brown}, M.~J.~I., et~al.\ 2006, \apj, 638, 88
2554: 
2555: \bibitem[{{Buchanan} {et~al.}(2006){Buchanan}, {Gallimore}, {O'Dea}, {Baum},
2556:   {Axon}, {Robinson}, {Elitzur}, \& {Elvis}}]{buch06}
2557: {Buchanan}, C.~L., {Gallimore}, J.~F., {O'Dea}, C.~P., {Baum}, S.~A., {Axon},
2558:   D.~J., {Robinson}, A., {Elitzur}, M., \& {Elvis}, M. 2006, \aj, 132, 401
2559: 
2560: \bibitem[{{Caccianiga} {et~al.}(2004){Caccianiga}, {Severgnini}, {Braito},
2561:   {Della Ceca}, {Maccacaro}, {Wolter}, {Barcons}, {Carrera}, {Lehmann}, {Page},
2562:   {Saxton}, \& {Webb}}]{cacc04}
2563: {Caccianiga}, A., et~al.\ 2004, \aap, 416, 901
2564: 
2565: \bibitem[{{Comastri} {et~al.}(1995){Comastri}, {Setti}, {Zamorani}, \&
2566:   {Hasinger}}]{coma95}
2567: {Comastri}, A., {Setti}, G., {Zamorani}, G., \& {Hasinger}, G. 1995, \aap, 296,
2568:   1
2569: 
2570: \bibitem[{{Cutri} {et~al.}(2002){Cutri}, {Nelson}, {Francis}, \&
2571:   {Smith}}]{cutr02}
2572: {Cutri}, R.~M., {Nelson}, B.~O., {Francis}, P.~J., \& {Smith}, P.~S. 2002, in
2573:   ASP Conf. Ser. 284: IAU Colloq. 184: AGN Surveys, ed. R.~F. {Green}, E.~Y.
2574:   {Khachikian}, \& D.~B. {Sanders} (San Francisco: ASP), 127
2575: 
2576: \bibitem[{{Daddi} {et~al.}(2007){Daddi}, {Alexander}, {Dickinson}, {Gilli},
2577:   {Renzini}, {Elbaz}, {Cimatti}, {Chary}, {Frayer}, {Bauer}, {Brandt},
2578:   {Giavalisco}, {Grogin}, {Huynh}, {Kurk}, {Mignoli}, {Morrison}, {Pope}, \&
2579:   {Ravindranath}}]{dadd07comp}
2580: {Daddi}, E., et~al.\ 2007, \apj\  in press, astro-ph/0705.2832
2581: 
2582: \bibitem[{{De Luca} \& {Molendi}(2004)}]{delu04}
2583: {De Luca}, A. \& {Molendi}, S. 2004, \aap, 419, 837
2584: 
2585: \bibitem[{{Dwelly} \& {Page}(2006)}]{dwel06}
2586: {Dwelly}, T. \& {Page}, M.~J. 2006, \mnras, 372, 1755
2587: 
2588: \bibitem[{{Dwelly} {et~al.}(2005){Dwelly}, {Page}, {Loaring}, {Mason},
2589:   {McHardy}, {Gunn}, \& {Sasseen}}]{dwel05}
2590: {Dwelly}, T., {Page}, M.~J., {Loaring}, N.~S., {Mason}, K.~O., {McHardy}, I.,
2591:   {Gunn}, K., \& {Sasseen}, T. 2005, \mnras, 360, 1426
2592: 
2593: \bibitem[{{Eisenhardt} {et~al.}(2004){Eisenhardt}, {Stern}, {Brodwin}, {Fazio},
2594:   {Rieke}, {Rieke}, {Werner}, {Wright}, {Allen}, {Arendt}, {Ashby}, {Barmby},
2595:   {Forrest}, {Hora}, {Huang}, {Huchra}, {Pahre}, {Pipher}, {Reach}, {Smith},
2596:   {Stauffer}, {Wang}, {Willner}, {Brown}, {Dey}, {Jannuzi}, \&
2597:   {Tiede}}]{eise04}
2598: {Eisenhardt}, P.~R., et~al.\ 2004, \apjs, 154, 48
2599: 
2600: \bibitem[{{Fall} \& {Pei}(1989)}]{fall89}
2601: {Fall}, S.~M. \& {Pei}, Y.~C. 1989, \apj, 337, 7
2602: 
2603: \bibitem[{{Farrah} {et~al.}(2003){Farrah}, {Afonso}, {Efstathiou},
2604:   {Rowan-Robinson}, {Fox}, \& {Clements}}]{farr03}
2605: {Farrah}, D., {Afonso}, J., {Efstathiou}, A., {Rowan-Robinson}, M., {Fox}, M.,
2606:   \& {Clements}, D. 2003, \mnras, 343, 585
2607: 
2608: \bibitem[{{Farrah} {et~al.}(2002){Farrah}, {Serjeant}, {Efstathiou},
2609:   {Rowan-Robinson}, \& {Verma}}]{farr02}
2610: {Farrah}, D., {Serjeant}, S., {Efstathiou}, A., {Rowan-Robinson}, M., \&
2611:   {Verma}, A. 2002, \mnras, 335, 1163
2612: 
2613: \bibitem[{{Fioc} \& {Rocca-Volmerange}(1997)}]{fioc97}
2614: {Fioc}, M. \& {Rocca-Volmerange}, B. 1997, \aap, 326, 950
2615: 
2616: \bibitem[{{Gehrels}(1986)}]{gehr86}
2617: {Gehrels}, N. 1986, \apj, 303, 336
2618: 
2619: \bibitem[{{Gilli} {et~al.}(2007){Gilli}, {Comastri}, \& {Hasinger}}]{gill07}
2620: {Gilli}, R., {Comastri}, A., \& {Hasinger}, G. 2007, \aap, 463, 79
2621: 
2622: \bibitem[{{Glikman} {et~al.}(2006){Glikman}, {Helfand}, \& {White}}]{glik06}
2623: {Glikman}, E., {Helfand}, D.~J., \& {White}, R.~L. 2006, \apj, 640, 579
2624: 
2625: \bibitem[{{Gorjian} {et~al.}(2007){Gorjian}, {Stern}, {Brodwin}, {Fazio},
2626:   {Rieke}, {Rieke}, {Werner}, {Wright}, {Allen}, {Arendt}, {Ashby}, {Barmby},
2627:   {Forrest}, {Hora}, {Huang}, {Huchra}, {Pahre}, {Pipher}, {Reach}, {Smith},
2628:   {Stauffer}, {Wang}, {Willner}, {Brown}, {Dey}, {Jannuzi}, \&
2629:   {Tiede}}]{gorj07}
2630: {Gorjian}, V., et~al.\ 2007,  submitted to ApJ
2631: 
2632: \bibitem[{{Guainazzi} {et~al.}(2005){Guainazzi}, {Matt}, \& {Perola}}]{guai05}
2633: {Guainazzi}, M., {Matt}, G., \& {Perola}, G.~C. 2005, \aap, 444, 119
2634: 
2635: \bibitem[{{Hao} {et~al.}(2005){Hao}, {Strauss}, {Fan}, {Tremonti}, {Schlegel},
2636:   {Heckman}, {Kauffmann}, {Blanton}, {Gunn}, {Hall}, {Ivezi{\'c}}, {Knapp},
2637:   {Krolik}, {Lupton}, {Richards}, {Schneider}, {Strateva}, {Zakamska},
2638:   {Brinkmann}, \& {Szokoly}}]{hao_l05b}
2639: {Hao}, L., et~al.\ 2005, \aj, 129, 1795
2640: 
2641: \bibitem[{{Hasinger} {et~al.}(2005){Hasinger}, {Miyaji}, \& {Schmidt}}]{hasi05}
2642: {Hasinger}, G., {Miyaji}, T., \& {Schmidt}, M. 2005, \aap, 441, 417
2643: 
2644: \bibitem[{{Hatziminaoglou} {et~al.}(2005){Hatziminaoglou}, {P{\'e}rez-Fournon},
2645:   {Polletta}, {Afonso-Luis}, {Hern{\'a}n-Caballero}, {Montenegro-Montes},
2646:   {Lonsdale}, {Xu}, {Franceschini}, {Rowan-Robinson}, {Babbedge}, {Smith},
2647:   {Surace}, {Shupe}, {Fang}, {Farrah}, {Oliver}, {Gonz{\'a}lez-Solares}, \&
2648:   {Serjeant}}]{hatz05}
2649: {Hatziminaoglou}, E., et~al.\ 2005, \aj,  129, 1198
2650: 
2651: \bibitem[{{Helou} {et~al.}(1985){Helou}, {Soifer}, \&
2652:   {Rowan-Robinson}}]{helo85}
2653: {Helou}, G., {Soifer}, B.~T., \& {Rowan-Robinson}, M. 1985, \apjl, 298, L7
2654: 
2655: \bibitem[{{Hickox} \& {Markevitch}(2006)}]{hick06a}
2656: {Hickox}, R.~C. \& {Markevitch}, M. 2006, \apj, 645, 95
2657: 
2658: \bibitem[{{Ho}(2005)}]{ho05}
2659: {Ho}, L.~C. 2005, astro-ph/0511157
2660: 
2661: \bibitem[{{Hogg}(1999)}]{hogg99}
2662: {Hogg}, D.~W. 1999, astro-ph/9905116
2663: 
2664: \bibitem[{{Hopkins} {et~al.}(2006{\natexlab{a}}){Hopkins}, {Hernquist}, {Cox},
2665:   {Di Matteo}, {Robertson}, \& {Springel}}]{hopk06apjs}
2666: {Hopkins}, P.~F., {Hernquist}, L., {Cox}, T.~J., {Di Matteo}, T., {Robertson},
2667:   B., \& {Springel}, V. 2006{\natexlab{a}}, \apjs, 163, 1
2668: 
2669: \bibitem[{{Hopkins} {et~al.}(2007){Hopkins}, {Richards}, \&
2670:   {Hernquist}}]{hopk07qlf}
2671: {Hopkins}, P.~F., {Richards}, G.~T., \& {Hernquist}, L. 2007, \apj, 654, 731
2672: 
2673: \bibitem[{{Hopkins} {et~al.}(2006{\natexlab{b}}){Hopkins}, {Somerville},
2674:   {Hernquist}, {Cox}, {Robertson}, \& {Li}}]{hopk06merge}
2675: {Hopkins}, P.~F., {Somerville}, R.~S., {Hernquist}, L., {Cox}, T.~J.,
2676:   {Robertson}, B., \& {Li}, Y. 2006{\natexlab{b}}, \apj, 652, 864
2677: 
2678: \bibitem[{{Houck} {et~al.}(2005){Houck}, {Soifer}, {Weedman}, {Higdon},
2679:   {Higdon}, {Herter}, {Brown}, {Dey}, {Jannuzi}, {Le Floc'h}, {Rieke}, {Armus},
2680:   {Charmandaris}, {Brandl}, \& {Teplitz}}]{houc05}
2681: {Houck}, J.~R., et~al.\ 2005, \apjl, 622, L105
2682: 
2683: \bibitem[{{Ibar} \& {Lira}(2007)}]{ibar07}
2684: {Ibar}, E. \& {Lira}, P. 2007, \aap, 466, 531
2685: 
2686: \bibitem[{{Indebetouw} {et~al.}(2005){Indebetouw}, {Mathis}, {Babler}, {Meade},
2687:   {Watson}, {Whitney}, {Wolff}, {Wolfire}, {Cohen}, {Bania}, {Benjamin},
2688:   {Clemens}, {Dickey}, {Jackson}, {Kobulnicky}, {Marston}, {Mercer},
2689:   {Stauffer}, {Stolovy}, \& {Churchwell}}]{inde05}
2690: {Indebetouw}, R., et~al.\ 2005, \apj, 619, 931
2691: 
2692: \bibitem[{{Jannuzi} \& {Dey}(1999)}]{jann99}
2693: {Jannuzi}, B.~T. \& {Dey}, A. 1999, in Astronomical Society of the Pacific
2694:   Conference Series, Vol. 191, Photometric Redshifts and the Detection of High
2695:   Redshift Galaxies, ed. R.~{Weymann}, L.~{Storrie-Lombardi}, M.~{Sawicki}, \&
2696:   R.~{Brunner} (San Francisco: ASP), 111
2697: 
2698: \bibitem[{{Kenter} {et~al.}(2005){Kenter}, {Murray}, {Forman}, {Jones},
2699:   {Green}, {Kochanek}, {Vikhlinin}, {Fabricant}, {Fazio}, {Brand}, {Brown},
2700:   {Dey}, {Jannuzi}, {Najita}, {McNamara}, {Shields}, \& {Rieke}}]{kent05}
2701: {Kenter}, A., et~al.\ 2005, \apjs, 161, 9
2702: 
2703: \bibitem[{{Khachikian} \& {Weedman}(1974)}]{khac74}
2704: {Khachikian}, E.~Y. \& {Weedman}, D.~W. 1974, \apj, 192, 581
2705: 
2706: \bibitem[{{King}(2005)}]{king05}
2707: {King}, A. 2005, \apjl, 635, L121
2708: 
2709: \bibitem[{{La Franca} {et~al.}(2005){La Franca}, {Fiore}, {Comastri}, {Perola},
2710:   {Sacchi}, {Brusa}, {Cocchia}, {Feruglio}, {Matt}, {Vignali}, {Carangelo},
2711:   {Ciliegi}, {Lamastra}, {Maiolino}, {Mignoli}, {Molendi}, \&
2712:   {Puccetti}}]{lafr05}
2713: {La Franca}, F., et~al.\ 2005, \apj, 635, 864
2714: 
2715: \bibitem[{{Lacy} {et~al.}(2007){Lacy}, {Petric}, {Sajina}, {Canalizo},
2716:   {Storrie-Lombardi}, {Armus}, {Fadda}, \& {Marleau}}]{lacy07}
2717: {Lacy}, M., {Petric}, A.~O., {Sajina}, A., {Canalizo}, G., {Storrie-Lombardi},
2718:   L.~J., {Armus}, L., {Fadda}, D., \& {Marleau}, F.~R. 2007, \aj, 133, 186
2719: 
2720: \bibitem[{{Lacy} {et~al.}(2004){Lacy}, {Storrie-Lombardi}, {Sajina},
2721:   {Appleton}, {Armus}, {Chapman}, {Choi}, {Fadda}, {Fang}, {Frayer},
2722:   {Heinrichsen}, {Helou}, {Im}, {Marleau}, {Masci}, {Shupe}, {Soifer},
2723:   {Surace}, {Teplitz}, {Wilson}, \& {Yan}}]{lacy04}
2724: {Lacy}, M., et~al.\ 2004, \apjs, 154, 166
2725: 
2726: \bibitem[{{Lawrence}(1991)}]{lawr91}
2727: {Lawrence}, A. 1991, \mnras, 252, 586
2728: 
2729: \bibitem[{{Maiolino} \& {Rieke}(1995)}]{maio95}
2730: {Maiolino}, R. \& {Rieke}, G.~H. 1995, \apj, 454, 95
2731: 
2732: \bibitem[{{Mart{\'{\i}}nez-Sansigre} {et~al.}(2006){Mart{\'{\i}}nez-Sansigre},
2733:   {Rawlings}, {Lacy}, {Fadda}, {Jarvis}, {Marleau}, {Simpson}, \&
2734:   {Willott}}]{mart06}
2735: {Mart{\'{\i}}nez-Sansigre}, A., {Rawlings}, S., {Lacy}, M., {Fadda}, D.,
2736:   {Jarvis}, M.~J., {Marleau}, F.~R., {Simpson}, C., \& {Willott}, C.~J. 2006,
2737:   \mnras, 370, 1479
2738: 
2739: \bibitem[{{Mateos} {et~al.}(2005){Mateos}, {Barcons}, {Carrera}, {Ceballos},
2740:   {Hasinger}, {Lehmann}, {Fabian}, \& {Streblyanska}}]{mate05}
2741: {Mateos}, S., {Barcons}, X., {Carrera}, F.~J., {Ceballos}, M.~T., {Hasinger},
2742:   G., {Lehmann}, I., {Fabian}, A.~C., \& {Streblyanska}, A. 2005, \aap, 444, 79
2743: 
2744: \bibitem[{{Matt}(2002)}]{matt02}
2745: {Matt}, G. 2002, in ASP Conf. Ser. 258: Issues in Unification of Active
2746:   Galactic Nuclei, ed. R.~{Maiolino}, A.~{Marconi}, \& N.~{Nagar} (San
2747:   Francisco: ASP), 3
2748: 
2749: \bibitem[{{McCarthy}(1993)}]{mcca93}
2750: {McCarthy}, P.~J. 1993, \araa, 31, 639
2751: 
2752: \bibitem[{{Murray} {et~al.}(2005){Murray}, {Kenter}, {Forman}, {Jones},
2753:   {Green}, {Kochanek}, {Vikhlinin}, {Fabricant}, {Fazio}, {Brand}, {Brown},
2754:   {Dey}, {Jannuzi}, {Najita}, {McNamara}, {Shields}, \& {Rieke}}]{murr05}
2755: {Murray}, S.~S., et~al.\ 2005, \apjs, 161, 1
2756: 
2757: \bibitem[{{Mushotzky}(2004)}]{mush04book}
2758: {Mushotzky}, R. 2004, in Astrophysics and Space Science Library, Vol. 308,
2759:   Supermassive Black Holes in the Distant Universe, ed. A.~J. {Barger}
2760:   (Dordrecht: Kluwer), 53
2761: 
2762: \bibitem[{{Narayan} \& {Yi}(1995)}]{nara95}
2763: {Narayan}, R. \& {Yi}, I. 1995, \apj, 452, 710
2764: 
2765: \bibitem[{{Osterbrock} \& {Shaw}(1988)}]{oste88}
2766: {Osterbrock}, D.~E. \& {Shaw}, R.~A. 1988, \apj, 327, 89
2767: 
2768: \bibitem[{{Pei}(1992)}]{pei92}
2769: {Pei}, Y.~C. 1992, \apj, 395, 130
2770: 
2771: \bibitem[{{Polletta} {et~al.}(2006){Polletta}, {Wilkes}, {Siana}, {Lonsdale},
2772:   {Kilgard}, {Smith}, {Kim}, {Owen}, {Efstathiou}, {Jarrett}, {Stacey},
2773:   {Franceschini}, {Rowan-Robinson}, {Babbedge}, {Berta}, {Fang}, {Farrah},
2774:   {Gonz{\'a}lez-Solares}, {Morrison}, {Surace}, \& {Shupe}}]{poll06}
2775: {Polletta}, M.~d.~C., et~al.\ 2006, \apj, 642, 673
2776: 
2777: \bibitem[{{Puget} \& {Leger}(1989)}]{puge89}
2778: {Puget}, J.~L. \& {Leger}, A. 1989, \araa, 27, 161
2779: 
2780: \bibitem[{{Ranalli} {et~al.}(2003){Ranalli}, {Comastri}, \& {Setti}}]{rana03}
2781: {Ranalli}, P., {Comastri}, A., \& {Setti}, G. 2003, \aap, 399, 39
2782: 
2783: \bibitem[{{Richards} {et~al.}(2005){Richards}, {Croom}, {Anderson},
2784:   {Bland-Hawthorn}, {Boyle}, {De Propris}, {Drinkwater}, {Fan}, {Gunn},
2785:   {Ivezi{\'c}}, {Jester}, {Loveday}, {Meiksin}, {Miller}, {Myers}, {Nichol},
2786:   {Outram}, {Pimbblet}, {Roseboom}, {Ross}, {Schneider}, {Shanks}, {Sharp},
2787:   {Stoughton}, {Strauss}, {Szalay}, {Vanden Berk}, \& {York}}]{rich05}
2788: {Richards}, G.~T., et~al.\ 2005, \mnras, 360, 839
2789: 
2790: \bibitem[{{Richards} {et~al.}(2006){Richards}, {Lacy}, {Storrie-Lombardi},
2791:   {Hall}, {Gallagher}, {Hines}, {Fan}, {Papovich}, {Vanden Berk}, {Trammell},
2792:   {Schneider}, {Vestergaard}, {York}, {Jester}, {Anderson}, {Budav{\'a}ri}, \&
2793:   {Szalay}}]{rich06}
2794: {Richards}, G.~T., et~al.\ 2006, \apjs, 166, 470
2795: 
2796: \bibitem[{{Rowan-Robinson} {et~al.}(2005){Rowan-Robinson}, {Babbedge},
2797:   {Surace}, {Shupe}, {Fang}, {Lonsdale}, {Smith}, {Polletta}, {Siana},
2798:   {Gonzalez-Solares}, {Xu}, {Owen}, {Davoodi}, {Dole}, {Domingue},
2799:   {Efstathiou}, {Farrah}, {Fox}, {Franceschini}, {Frayer}, {Hatziminaoglou},
2800:   {Masci}, {Morrison}, {Nandra}, {Oliver}, {Onyett}, {Padgett},
2801:   {Perez-Fournon}, {Serjeant}, {Stacey}, \& {Vaccari}}]{rowa05}
2802: {Rowan-Robinson}, M., et~al.\ 2005, \aj, 129, 1183
2803: 
2804: \bibitem[{{Sanders} {et~al.}(1988){Sanders}, {Soifer}, {Elias}, {Madore},
2805:   {Matthews}, {Neugebauer}, \& {Scoville}}]{sand88}
2806: {Sanders}, D.~B., {Soifer}, B.~T., {Elias}, J.~H., {Madore}, B.~F., {Matthews},
2807:   K., {Neugebauer}, G., \& {Scoville}, N.~Z. 1988, \apj, 325, 74
2808: 
2809: \bibitem[{{Setti} \& {Woltjer}(1989)}]{sett89}
2810: {Setti}, G. \& {Woltjer}, L. 1989, \aap, 224, L21
2811: 
2812: \bibitem[{{Seyfert}(1943)}]{seyf43}
2813: {Seyfert}, C.~K. 1943, \apj, 97, 28
2814: 
2815: \bibitem[{{Siebenmorgen} \& {Kr{\"u}gel}(2007)}]{sieb07}
2816: {Siebenmorgen}, R. \& {Kr{\"u}gel}, E. 2007, \aap, 461, 445
2817: 
2818: \bibitem[{{Silk} \& {Rees}(1998)}]{silk98}
2819: {Silk}, J. \& {Rees}, M.~J. 1998, \aap, 331, L1
2820: 
2821: \bibitem[{{Spoon} {et~al.}(2005){Spoon}, {Keane}, {Cami}, {Lahuis}, {Tielens},
2822:   {Armus}, \& {Charmandaris}}]{spoo05}
2823: {Spoon}, H.~W.~W., et~al.\ 2005, in IAU Symposium, ed.  D.~C. {Lis}, G.~A. {Blake}, \& E.~{Herbst} (Cambridge: Cambridge University  Press), 281--290
2824: 
2825: \bibitem[{{Springel} {et~al.}(2005){Springel}, {Di Matteo}, \&
2826:   {Hernquist}}]{spri05}
2827: {Springel}, V., {Di Matteo}, T., \& {Hernquist}, L. 2005, \mnras, 361, 776
2828: 
2829: \bibitem[{{Steffen} {et~al.}(2006){Steffen}, {Strateva}, {Brandt}, {Alexander},
2830:   {Koekemoer}, {Lehmer}, {Schneider}, \& {Vignali}}]{stef06}
2831: {Steffen}, A.~T., {Strateva}, I., {Brandt}, W.~N., {Alexander}, D.~M.,
2832:   {Koekemoer}, A.~M., {Lehmer}, B.~D., {Schneider}, D.~P., \& {Vignali}, C.
2833:   2006, \aj, 131, 2826
2834: 
2835: \bibitem[{{Stern} {et~al.}(2005){Stern}, {Eisenhardt}, {Gorjian}, {Kochanek},
2836:   {Caldwell}, {Eisenstein}, {Brodwin}, {Brown}, {Cool}, {Dey}, {Green},
2837:   {Jannuzi}, {Murray}, {Pahre}, \& {Willner}}]{ster05}
2838: {Stern}, D., et~al.\ 2005, \apj, 631, 163
2839: 
2840: \bibitem[{{Stern} {et~al.}(2002){Stern}, {Moran}, {Coil}, {Connolly}, {Davis},
2841:   {Dawson}, {Dey}, {Eisenhardt}, {Elston}, {Graham}, {Harrison}, {Helfand},
2842:   {Holden}, {Mao}, {Rosati}, {Spinrad}, {Stanford}, {Tozzi}, \& {Wu}}]{ster02}
2843: {Stern}, D., et~al.\ 2002, \apj,  568, 71
2844: 
2845: \bibitem[{{Sturm} {et~al.}(2006){Sturm}, {Hasinger}, {Lehmann}, {Mainieri},
2846:   {Genzel}, {Lehnert}, {Lutz}, \& {Tacconi}}]{stur06}
2847: {Sturm}, E., {Hasinger}, G., {Lehmann}, I., {Mainieri}, V., {Genzel}, R.,
2848:   {Lehnert}, M.~D., {Lutz}, D., \& {Tacconi}, L.~J. 2006, \apj, 642, 81
2849: 
2850: \bibitem[{{Tozzi} {et~al.}(2006){Tozzi}, {Gilli}, {Mainieri}, {Norman},
2851:   {Risaliti}, {Rosati}, {Bergeron}, {Borgani}, {Giacconi}, {Hasinger},
2852:   {Nonino}, {Streblyanska}, {Szokoly}, {Wang}, \& {Zheng}}]{tozz06}
2853: {Tozzi}, P., et~al.\ 2006, \aap, 451, 457
2854: 
2855: \bibitem[{{Treister} \& {Urry}(2005)}]{trei05}
2856: {Treister}, E. \& {Urry}, C.~M. 2005, \apj, 630, 115
2857: 
2858: \bibitem[{{Treister} {et~al.}(2004){Treister}, {Urry}, {Chatzichristou},
2859:   {Bauer}, {Alexander}, {Koekemoer}, {Van Duyne}, {Brandt}, {Bergeron},
2860:   {Stern}, {Moustakas}, {Chary}, {Conselice}, {Cristiani}, \&
2861:   {Grogin}}]{trei04}
2862: {Treister}, E., et~al.\ 2004, \apj, 616, 123
2863: 
2864: \bibitem[{{Ueda} {et~al.}(2003){Ueda}, {Akiyama}, {Ohta}, \& {Miyaji}}]{ueda03}
2865: {Ueda}, Y., {Akiyama}, M., {Ohta}, K., \& {Miyaji}, T. 2003, \apj, 598, 886
2866: 
2867: \bibitem[{{Urry} \& {Padovani}(1995)}]{urry95}
2868: {Urry}, C.~M. \& {Padovani}, P. 1995, \pasp, 107, 803
2869: 
2870: \bibitem[{{van Breugel} {et~al.}(1999){van Breugel}, {De Breuck}, {Stanford},
2871:   {Stern}, {R{\"o}ttgering}, \& {Miley}}]{vanbreu99}
2872: {van Breugel}, W., {De Breuck}, C., {Stanford}, S.~A., {Stern}, D.,
2873:   {R{\"o}ttgering}, H., \& {Miley}, G. 1999, \apjl, 518, L61
2874: 
2875: \bibitem[{{Vanden Berk} {et~al.}(2001){Vanden Berk}, {Richards}, {Bauer},
2876:   {Strauss}, {Schneider}, {Heckman}, {York}, {Hall}, {Fan}, {Knapp},
2877:   {Anderson}, {Annis}, {Bahcall}, {Bernardi}, {Briggs}, {Brinkmann}, {Brunner},
2878:   {Burles}, {Carey}, {Castander}, {Connolly}, {Crocker}, {Csabai}, {Doi},
2879:   {Finkbeiner}, {Friedman}, {Frieman}, {Fukugita}, {Gunn}, {Hennessy},
2880:   {Ivezi{\'c}}, {Kent}, {Kunszt}, {Lamb}, {Leger}, {Long}, {Loveday}, {Lupton},
2881:   {Meiksin}, {Merelli}, {Munn}, {Newberg}, {Newcomb}, {Nichol}, {Owen}, {Pier},
2882:   {Pope}, {Rockosi}, {Schlegel}, {Siegmund}, {Smee}, {Snir}, {Stoughton},
2883:   {Stubbs}, {SubbaRao}, {Szalay}, {Szokoly}, {Tremonti}, {Uomoto}, {Waddell},
2884:   {Yanny}, \& {Zheng}}]{vand01}
2885: {Vanden Berk}, D.~E., et~al.\ 2001, \aj, 122, 549
2886: 
2887: \bibitem[{{Webster} {et~al.}(1995){Webster}, {Francis}, {Peterson},
2888:   {Drinkwater}, \& {Masci}}]{webs95}
2889: {Webster}, R.~L., {Francis}, P.~J., {Peterson}, B.~A., {Drinkwater}, M.~J., \&
2890:   {Masci}, F.~J. 1995, \nat, 375, 469
2891: 
2892: \bibitem[{{Weedman} {et~al.}(2006){Weedman}, {Polletta}, {Lonsdale}, {Wilkes},
2893:   {Siana}, {Houck}, {Surace}, {Shupe}, {Farrah}, \& {Smith}}]{weed06c}
2894: {Weedman}, D., et~al.\ 2006, \apj, 653, 101
2895: 
2896: \bibitem[{{Wilkes} {et~al.}(2002){Wilkes}, {Schmidt}, {Cutri}, {Ghosh},
2897:   {Hines}, {Nelson}, \& {Smith}}]{wilk02}
2898: {Wilkes}, B.~J., {Schmidt}, G.~D., {Cutri}, R.~M., {Ghosh}, H., {Hines}, D.~C.,
2899:   {Nelson}, B., \& {Smith}, P.~S. 2002, \apjl, 564, L65
2900: 
2901: \bibitem[{{Zakamska} {et~al.}(2005){Zakamska}, {Schmidt}, {Smith}, {Strauss},
2902:   {Krolik}, {Hall}, {Richards}, {Schneider}, {Brinkmann}, \&
2903:   {Szokoly}}]{zaka05}
2904: {Zakamska}, N.~L., et~al.\ 2005, \aj, 129, 1212
2905: 
2906: \bibitem[{{Zakamska} {et~al.}(2004){Zakamska}, {Strauss}, {Heckman},
2907:   {Ivezi{\'c}}, \& {Krolik}}]{zaka04}
2908: {Zakamska}, N.~L., {Strauss}, M.~A., {Heckman}, T.~M., {Ivezi{\'c}}, {\v Z}.,
2909:   \& {Krolik}, J.~H. 2004, \aj, 128, 1002
2910: 
2911: \bibitem[{{Zakamska} {et~al.}(2003){Zakamska}, {Strauss}, {Krolik}, {Collinge},
2912:   {Hall}, {Hao}, {Heckman}, {Ivezi{\'c}}, {Richards}, {Schlegel}, {Schneider},
2913:   {Strateva}, {Vanden Berk}, {Anderson}, \& {Brinkmann}}]{zaka03}
2914: {Zakamska}, N.~L., et~al.\ 2003, \aj, 126,  2125
2915: 
2916: \bibitem[{{Zheng} {et~al.}(2004){Zheng}, {Mikles}, {Mainieri}, {Hasinger},
2917:   {Rosati}, {Wolf}, {Norman}, {Szokoly}, {Gilli}, {Tozzi}, {Wang}, {Zirm}, \&
2918:   {Giacconi}}]{zhen04}
2919: {Zheng}, W., et~al.\ 2004, \apjs, 155, 73
2920: 
2921: 
2922: \end{thebibliography}
2923: 
2924: 
2925: 
2926: 
2927: 
2928: 
2929: 
2930: \end{document}
2931: