A Complete Multiwavelength Characterization of Faint Chandra X-Ray Sources Seen in the Spitzer Wide-Area Infrared Extragalactic (SWIRE) Survey

We exploit deep combined observations with Spitzer and Chandra of the Spitzer Wide-Area Infrared Extragalactic Survey (SWIRE) in the ELAIS N1 region to investigate the nature of the faint X-ray and IR sources in common, to identify active galactic nucleus (AGN)/starburst diagnostics, and to study the sources of the X-ray and IR cosmic backgrounds (XRB and CIRB). In the 17′ × 17′ area of the Chandra ACIS-I image there are approximately 3400 SWIRE near-IR sources with 4 σ detections in at least two Infrared Array Camera (IRAC) bands and 988 sources detected at 24 μm with the Multiband Imaging Photometer (MIPS) brighter than S24 ≃ 0.1 mJy. Of these, 102 IRAC and 59 MIPS sources have Chandra counterparts, out of a total of 122 X-ray sources present in the area with S0.5–8 keV > 10-15 ergs cm-2 s-1. We have constructed spectral energy distributions (SEDs) for each source using data from the four IRAC wavebands, Chandra fluxes in the hard (2–8 keV) and soft (0.5–2 keV) X-rays, and optical follow-up data in the wavebands U, g′, r′, i′, Z, and H. We fit a number of spectral templates to the SEDs at optical and IR wavelengths to determine photometric redshifts and spectral categories and also make use of diagnostics based on the X-ray luminosities, hardness ratios, X-ray to IR spectral slopes, and optical morphologies. Although we have spectroscopic redshifts for only a minority of the Chandra sources (10 type 1 QSOs or Seyfert sources and three galaxies), the available SEDs constrain the redshifts for most of the sample sources, which turn out to be typically at 0.5 < z < 2. We find that 39% of the Chandra sources are dominated by type 1 AGN emission (QSOs or Seyfert 1), 23% display optical/IR spectra typical of type 2 AGNs, while the remaining 38% show starburst-like or even normal galaxy spectra (including five passively evolving early-type galaxies). Since we prove that all these galaxies are dominated by AGN emission in X-rays (considering their large 0.5–8 keV rest-frame X-ray luminosities and their high X-ray to IR flux ratios), this brings the fraction of type 1 AGNs to 80% of the type 2 AGNs; even assuming that all the Chandra sources undetected by Spitzer are type 2 AGNs, the type 1 fraction would exceed 1/3 of the total population. Our analysis of the mid-IR MIPS 24 μm–selected sources, making up ∼50% of the CIRB, shows that the fraction of those dominated by an AGN (either type 1 or type 2) is relatively constant with the IR flux and around 10%–15%. Our combined IR and hard X-ray observations allow us to verify that the dust covering fraction in type 1 AGNs is widely distributed between ∼10% and 100%. A significant fraction, from 15% to 30% or more, of the sources of the XRB are hosted in galaxies whose optical/IR spectra are dominated by starburst (or normal galaxy) emission and for which only the hard X-ray spectra reveal the presence of a moderately luminous hidden AGN.


INTRODUCTION
The Spitzer Space Telescope is providing a new sensitive tool for cosmological investigations over a wide spectral region from 3 to 160 m.The Spitzer Wide-Area Infrared Extragalactic Legacy Survey (SWIRE) in particular, with its large areal and IR spectral coverage, will not only allow us to quantify the role of the environment in structure formation and evolution but is also providing rich enough statistical samples for investigating relatively rare source populations (Lonsdale et al. 2003(Lonsdale et al. , 2004)).
The sources of the cosmic X-ray background (XRB), efficiently detected by Chandra and XMM-Newton, belong to this last category of relatively uncommon objects.Recently, the importance of combined hard X-ray and mid-and far-IR data for testing the active galactic nucleus (AGN) unification model and for verifying the standard obscured accretion paradigm for the origin of the XRB has been emphasized (Risaliti et al. 2000;Franceschini et al. 2002;Fadda et al. 2002;Alonso-Herrero et al. 2004;Rigby et al. 2004;Manners et al. 2004;Alexander et al. 2003).Indeed, a relationship has been suggested to hold between the XRB and the more recently discovered cosmological background in the far-IR, the CIRB (Puget et al. 1996;Hauser et al. 1998).The X-ray emission that is photoelectrically absorbed in the type 2 AGNs dominating the XRB is expected to be downgraded in energy by the dusty circumnuclear medium and to emerge thermally reprocessed in the IR between a few and a few hundred microns (e.g., Granato et al. 1997).An important test of this picture is then provided by detection in the IR of the downgraded energy absorbed in X-rays.
A related aspect is to investigate how the X-ray and UV/ optical absorbed flux from AGNs contributes to explaining the energetics of the IR sources and of the whole IR background itself.Considering that the CIRB contains more than half of the total radiant energy produced by cosmic sources (Hauser & Dwek 2001;Franceschini et al. 2001), it is of crucial importance to assess the physical origin of this component.The issue is still relatively unsettled, with the expected AGN fraction ranging from 50% of the CIRB being due to AGNs (under some extreme assumptions about the IR emissivity of type 2 objects and the percentage of Compton-thick sources; see Almaini et al. 1999;Fabian & Iwasawa 1999) to a few percent (Elbaz et al. 2002).
The first attempts to answer these questions were based on the pioneering mid-IR sky surveys with the Infrared Space Observatory (ISO).Fadda et al. (2002) detected 24 sources in common between ISO at 15 m and XMM-Newton in an area of the Lockman Hole.On that basis, Fadda et al. and Franceschini et al. (2002) concluded that the expectations of the obscured accretion model are basically met.These studies also suggested that only a minor fraction of the CIRB seems to be due to gravitational accretion.These conclusions were, however, limited by smallnumber statistics and by poor spectral coverage of the source spectral energy distributions (SEDs) (essentially limited to the 15 m data point).
In addition, deep long-wavelength surveys with SCUBA and MAMBO have been cross-correlated with ultradeep Chandra images, with the result that although many high-z submillimeter SCUBA sources show detectable hard X-ray emission, the large majority of the bolometric flux does not appear to be due to an AGN (Alexander et al. 2005).In any case, not much can be concluded from these data about the origin of the CIRB, because it is still uncertain how much the submillimeter sources contribute to the CIRB at its peak at around 100 m.This is also made uncertain by a quite complicated source selection function (including submillimeter and radio/optical identification) and, again, poor statistics.
We report in this paper a refined analysis of the relationship between X-ray and IR emission for faint cosmic sources over a wide redshift interval based on combined Spitzer and Chandra observations of an area in ELAIS N1.This region has been previously observed with ISO as part of the ELAIS survey (Oliver et al. 2000).ISO detected six extragalactic sources at 15 m within the 17 0 ; 17 0 area of the ELAIS N1 Chandra ACIS-I image, three of which have detectable counterparts in X-rays, according to Manners et al. (2004).
For comparison, the present SWIRE survey in the same small area detects $900 sources at 24 m above S 24 ¼ 100 Jy and 3500 Infrared Array Camera (IRAC) near-IR galaxies.We find that 102 (84%) of the 122 Chandra sources in the ACIS-I area now have mid-and far-IR counterparts from Spitzer.
The Spitzer Space Telescope, Chandra, and ancillary optical observations of the ELAIS N1 region are detailed in x 2. Particular attention is paid here to the estimation of photometric redshifts for our sample sources, given that only a fraction of them have optical spectroscopic data.Although our approach is similar to others recently published for X-ray-selected source populations (Franceschini et al. 2002;Zheng et al. 2004), our advantage here is to benefit from an accurate coverage of the near-and mid-IR spectrum with the four IRAC channels centered at 3.6, 4.5, 5.8, and 8 m and the Multiband Imaging Photometer for Spitzer (MIPS) 24 m channel.
The main results of our analysis are reported in x 3, in which we detail the X-ray and IR cross-correlation and source identification procedures, the analysis of the SEDs and classification of the sources, and the X-ray to IR flux correlations and color and hardness ratio plots.We discuss, in particular, new diagnostic diagrams for AGN-and galaxy-dominated emission in the optical/IR based on structural differences in the spectra.Photometric redshifts are used to infer important statistical information on redshift and luminosity distributions.The X-ray and IR identification statistics and some inferences about the AGN dust covering, as well as a comparison of Chandra-detected and Chandra-undetected SWIRE populations, are also addressed in x 3. The paper's conclusions are summarized in x 4.

Spitzer Space Telescope Observv ations
The SWIRE ELAIS N1 field was imaged by the IRAC multiband camera on Spitzer in 2004 January and with MIPS in early 2004 February, following the strategy described in Lonsdale et al. (2004).The original observing plan was corrected after the Spitzer performance verification (PV) phase by increasing the depth of the MIPS coverage to account for a reduced efficiency of the 70 m channel.The observations were centered at the position (16 h 00 m , +59 01 0 ).
Data processing began with the Spitzer Basic Calibrated Data (BCD), corrected for bias offsets and pixel-to-pixel gain variations (flat-fielding) and flux-calibrated in surface brightness.Additional IRAC processing corrected for latent images and electronic offsets.For MIPS, scan-mirror-dependent flats were derived from the data and applied to the BCD images.Individual images, which have measurable spatial distortions, were reprojected onto a single common projection system on the sky and co-added after correction for cosmic rays and other transient artifacts.
Fluxes were extracted in 5B8 apertures for IRAC ($2-3 times the FWHM beam) and 12 00 apertures for MIPS 24 m using SExtractor.Sources are typically unresolved by the large Spitzer beams (>2 00 at the shortest wavelength).The absolute flux calibrations are believed to be correct within roughly 10% for the IRAC and for the MIPS 24 m channel data and were confirmed for IRAC by comparison to the Two Micron All Sky Survey (2MASS).The resulting catalogs were examined by eye, and remaining spurious sources (radiation, scattered light, etc.) were removed by hand.More details on the data processing are given in Surace et al. (2005) and Shupe et al. (2005).
The achieved sensitivities with the SWIRE survey in ELAIS N1 turned out to be deeper than preflight expectations in the two short-wavelength IRAC channels and in the MIPS 24 m one, while performing less well in the other bands.The integral number counts of Spitzer sources in the Chandra ACIS-I 286 square arcmin field are reported in Figure 1.Dotted vertical lines mark the flux limits corresponding to 90% completeness limits, as determined from the deviation of the observed number counts from a power law and from simulations (Shupe et al. 2005).

The X-Ray Data
The X-ray observations were taken as part of the ELAIS Deep X-ray Survey (EDXS) and are described in detail in Manners et al. (2003).For this analysis we use the Chandra Advanced CCD Imaging Spectrometer (ACIS) observation of 75 ks centered on (16 h 10 m 20.s 11, +54 33 0 22B3) (J2000.0) in the ELAIS N1 region.The aim point was focused on the ACIS-I chips, which consist of four CCDs arranged in a 2 ; 2 array covering an area of 16A9 ; 16A9 (286 square arcmin).Bad pixels and columns were removed, and data were filtered to eliminate high background times (due to strong solar flares), leaving 71.5 ks of good data after filtering.Source detection and characterization in three bands (0.5-8 keV [full band], 0.5-2 keV [soft band], and 2-8 keV [hard band]) were achieved using the wavdetect software package (Freeman et al. 2002).For each band, exposure maps were constructed to account for variations in effective exposure across an image and used to remove bias from the source detection and to calculate source fluxes.Counts-to-photon calibration assumed a standard power-law model spectrum with photon index À ¼ 1:7.
Sources were detected to flux levels of 2:3 ; 10 À15 ergs s À1 cm À2 in the 0.5-8 keV band, 9:4 ; 10 À16 ergs s À1 cm À2 in the 0.5-2 keV band, and 5:2 ; 10 À15 ergs s À1 cm À2 in the 2-8 keV band.Sources are detectable to these flux limits over 90% of the nominal survey area.For this analysis we used sources detected in the full band of ACIS-I only, of which there are 122 in the N1 region.Of the 102 sources in common between Chandra and SWIRE, 83 have significant detections in the separate soft X-ray band (0.5-2 keV) and 64 are detected in the hard (2-8 keV) band.
The smoothed 0.5-8 keV Chandra ACIS-I image is reported in Figure 2 with the associated SWIRE sources overlaid.The image has been adaptively smoothed using the flux-conserving algorithm csmooth from the Chandra Interactive Analysis of Observations (CIAO) package.We defer to Manners et al. (2003Manners et al. ( , 2004) ) for more details on the X-ray data reduction and analysis.

The Optical Data
Follow-up observations of the EDXS sources have been taken in the bands U, g, r, i, and H and are detailed in Gonzalez-Solares et al. (2004).The Wide-Field Camera (WFC) on the Isaac Newton Telescope (INT) was used to observe a 30 0 ; 30 0 region centered on the Chandra N1 pointing.Data were observed in the g, r, and i bands in 2000 July and in the U band in 2001 June using the Sloan Digital Sky Survey (SDSS) photometric system.Total integration times were 100 minutes in the g, r, and i bands and 120 minutes in the U band, reaching Vega magnitude limits for a pointlike source in an aperture of 3 00 of U ¼ 24:0, g ¼ 25:5, r ¼ 25:2, i ¼ 24:1, and H ¼ 20:5.Because of potential source variability and the different observing time, the U-band data have significant uncertainty that has been taken into account in the subsequent analysis.
H-band observations were carried out in 2000 June using the Cambridge Infrared Survey Instrument (CIRSI) at the INT.The H filter is centered at 1.65 m with a width (FWHM) of 0.3 m and a transmission peak of 75%.A total integration of 60 minutes was taken over the 30 0 ; 30 0 used for the previous observations.
For several of our sources we also used fluxes in the Z band taken with the INT by Babbedge et al. (2005, the Wide-Field Survey).
All the images have been registered to a common world coordinate system and resampled to a common pixel scale.This step allowed us to carry out the detection in one reference image and the flux measurements in the others.This alleviates the task of cross-correlating different catalogs and allows us to measure magnitudes using the same aperture in all the images, as appropriate for SED analysis and required for the calculation of photometric redshifts.Source detection was performed using SExtractor (Bertin & Arnouts 1996).The measured parameters include positions and positional errors, basic shape parameters as well as elongation and ellipticity, and different types of fluxes and magnitudes: isophotal, corrected isophotal, fixed aperture, and Kron-like automatic aperture, as well as errors.The automatic aperture magnitude is designed to give the most precise flux measure and is the one used as the magnitude in this work.For the above-mentioned reasons, we require the magnitudes to be measured in the same aperture in each band.The object detection and aperture estimation is performed in the r-band images, and then magnitudes are obtained on all the images using that aperture.Since there are some X-ray objects that fall in the gaps between r-band chips, a second catalog was created by taking as the reference the g-band image.The absolute photometry was tied to the Wide-Field Survey (optical) catalog already existent in that area and 2MASS (IR) using bright nonsaturated stars.It was found that 85% of the X-ray sources have an optical counterpart, a fraction comparable to the detection rate achieved with Spitzer IRAC.

Redshift Measurements
Spectroscopic redshifts are available for only 13 of the 99 extragalactic sources in common between Chandra and Spitzer.These include six type 1 QSOs and two low-redshift galaxies observed by Gonzalez-Solares et al. (2005, including details on this optical spectroscopic run).Additional optical spectroscopy has been obtained at Palomar with the COSMIC spectrograph on 2004 June 10 and 11, providing the redshift for five further sources (two type 1 and two type 2 AGNs and a galaxy).
For the remaining objects, and thanks to the good coverage of the near-and mid-IR spectrum with the four IRAC channels, we estimated photometric redshifts with a tool based on the Hyper-z code (Bolzonella et al. 2000), as discussed in detail by Polletta et al. (2005).Hyper-z allowed a large database of galaxy spectra to be synthesized, including the effects of dust extinction.The following parameter ranges have been explored with Hyper-z: A V between 0 and 0.7, redshifts between 0 and 3.5, and absolute magnitudes in the g 0 band between À19 and À26.As expected, for the typical SEDs of our sample objects, the redshift estimates turned out to be rather insensitive to the amount of extinction.
Similar photometric redshift approaches have been successfully attempted for X-ray-selected source populations (Franceschini et al. 2002;Zheng et al. 2004), in which cases a comparison with spectroscopic measurements has revealed fair accuracy in the procedure (errors z $ 0:1).Independent photometric redshift estimates for SWIRE galaxy samples are reported in Rowan-Robinson et al. (2005; see also Rowan-Robinson 2003) and by Babbedge et al. (2004).Fair agreement between these codes is found by Babbedge et al., although in the current case the z-estimate is made more uncertain by the more complex nature of SEDs of AGN host galaxies.
Given the typical optical/IR SEDs of our sources, we expect that the photometric redshifts are more accurate for galaxydominated spectra and type 2 AGNs, which together make up the majority in our sample.Obviously, for the typically flat and featureless SEDs of type 1 QSOs, they are expected to be much less reliable.This expectation is confirmed by a comparison of photometric and spectroscopic redshifts for our 13 spectroscopically identified sources.

Cross-Correlation Analysis and Source Identification
A simple near-neighbor search has been performed to crosscorrelate the Spitzer and Chandra source catalogs within the Chandra ACIS-I chip image, using a d ¼ 5 00 search radius (roughly the quadratic sum of the astrometric errors).The distribution of separations between the Chandra and Spitzer centroids is reported in Figure 3.
The Spitzer source catalog used for the cross-correlation was a subset of the full band-merged catalog for the IRAC and MIPS 24 m bands.It contained only those sources with a signal-tonoise ratio greater than 4 in at least two of the Spitzer bands.This catalog contained 3420 objects within the region of the Chandra ACIS-I image, compared with the 122 Chandra fullband detections.
All together, we find reliably associated counterparts for 102 of the 122 Chandra sources (84% in total).The vast majority of these are detected with the IRAC channels 1 and 2 (3.6 and 4.5 m): 100 of the 122 Chandra sources in each case.Such an identification fraction decreases when considering the longer wavelength IRAC channels or MIPS.As many as 59 Chandra objects are reliably associated with MIPS 24 m sources (all of them having IRAC counterparts), and just 1 had a MIPS 70 m counterpart.More details on the source identification statistics are reported in Table 1, where the first column indicates the Spitzer band, the second indicates the number of Spitzer sources detected above the 4 limiting flux S lim (reported in the third column), the fourth column gives the number of Chandra sources identified in the various channels, and the fifth shows their percentage over the total number of Spitzer sources.
To estimate the reality of the associations, we have calculated the probability of random matches between the X-ray sources and the possible Spitzer counterparts.As discussed in Fadda  2002), we assume the IR population to follow a Poisson spatial distribution, such that the probability P of a random association is where d is the offset distance between the X-ray and Spitzer source and N is the areal number density of possible Spitzer counterparts derived from Figure 1.The probability P turned out to be sufficiently small to guarantee that almost all our associations are real.Approximately 90% have P < 0:03.Summing the probabilities, we expect two false associations in the 102 matched sources.
Only three Chandra sources have multiple associations within 5 00 in the IRAC images.Of these, two (N1 _ 8 and N1 _ 81) have been unambiguously identified, as they lie near to the center of the Chandra image, where the Chandra point-spread function (PSF) is small.Since the positional accuracy is dominated by the size of the Chandra PSF, we can reliably reject any associations with offsets greater than the 3 size of the Chandra PSF for these sources, leaving only one association in each case.The third source (N1 _ 104) is near the edge of the Chandra image and therefore has a large Chandra PSF that contains two Spitzer sources (as well as two optical sources) within the 3 PSF extent.We assume the association to be the Spitzer source closest to the center of the Chandra PSF and closest to the central optical source.This appears justified by the observed Seyfert-type spectrum found for this object (see Fig. 5h below).
Of the 102 Spitzer-identified Chandra sources, three turned out to correspond to Galactic stars on the basis of their position on color-magnitude plots and optical morphology and are excluded from our subsequent analysis.
We report in Figure 4 histograms of the ratio of the X-ray flux (in the total 0.5-8 keV band) to the 3.6 and 24 m fluxes for sources detected by SWIRE in the Chandra area brighter than 5 and 100 Jy, respectively.The figure compares the flux ratio distributions for sources with Chandra counterparts (solid-line histograms) and those without Chandra associations (shaded histograms).The latter are the distribution of the upper limits on the flux ratios computed using the Chandra 3 X-ray sensitivity limits, as in x 2.2.
The arrows in the top panel of Figure 4 indicate the observed values of the flux ratios for representative active galaxies: type 1 (A: NGC 4151) and type 2 AGNs (B: NGC 6240; C: IRAS 19254s), an ultraluminous IR quasar with high X-ray absorption (E: Mrk 231), and a low-luminosity starburst (D: M82).For these sources, the 24 m flux ( 24 S 24 ) is estimated from the 25 m IRAS data, while the X-ray data are taken from George et al. ( 1998), Cappi et al. (1999), Franceschini et al. (2003), and Braito et al. ( 2004).X-ray to IR flux ratios in our Chandra sample are then typical of X-ray-luminous AGNs and definitely larger than those of starburst galaxies or even heavily absorbed type 2 QSOs.

Analysis of the Spectral Energg y Distributions
The positions, Spitzer IRAC and MIPS data, X-ray fluxes, spectral classifications, X-ray hardness ratios, and luminosities for our SWIRE-Chandra sample are reported in Table 2.The excellent UV/optical/IR spectral coverage available for our sources has prompted us to perform a very extensive analysis based on the Hyper-z code (Bolzonella et al. 2000) and a wide variety of additional spectral templates, aimed at a physical characterization of the sources.
Basically, starburst-dominated and quasar-dominated sources can be discriminated on the basis of the presence or absence of the classical UV excess and flat optical SED in type 1 QSOs or from the mid-IR excess characterizing the spectra of both type 1 and type 2 AGNs.The latter is due to emission of very hot dust with temperatures close to the grain sublimation temperature and is essentially missing in purely starburst regions.We defer to Berta et al. (2003) and Prouton et al. (2004) for further discussion about this point.The diagnostic power offered by full UV/optical/IR spectral coverage for disentangling AGNs and starbursts can be appreciated by considering that the typical peak around 1 m of the integrated emission of stellar populations in normal galaxies roughly (and coincidentally) corresponds to a minimum of the spectrum of quasar nuclei between the UV ''big bump'' and the near-IR hot dust component.We further discuss this point in subsequent sections.
While an exhaustive description of the spectral database used for our analysis can be found in Polletta et al. (2005), we summarize here its main features.It consists of 22 spectral templates that are derived from the observed SEDs of various galaxy       2 is also available in machine-readable form in the electronic edition of the Astronomical Journal.
classes, while in the optical/near-IR the observations are complemented with a spectrophotometric synthesis code presented in Berta et al. (2004).The library contains spectral templates of all kinds of AGNs, from optical quasars (Brotherton et al. 2001) to red quasars (Gregg et al. 2002), Seyfert 1 to 2 galaxies, and combined quasar/ultraluminous infrared galaxy (ULIRG) sources like the type 1 QSO Mrk 231 and the type 2 quasar/ULIRG Superantennae South (IRAS 19254s;Berta et al. 2003).More information on the spectral fits used with Hyper-z is summarized in Table 3.
Figures 5a-5j display the SEDs and spectral best fits for the 99 extragalactic Chandra/SWIRE sources in our sample (three stars are excluded).Data points are plotted for the four IRAC (3.6, 4.5, 5.8, and 8 m), MIPS 24 m, U, g 0 , r 0 , i 0 , Z, H, and Chandra X-ray bands.Where the sources are detected in the soft (0.5-2 keV) or hard (2-8 keV) Chandra images, their fluxes are converted to flux densities at 1 and 5 keV, respectively.Where the Chandra source is not detected in both bands, the full band (0.5-8 keV) flux is plotted, corrected to a flux density at 2 keV.
Spectral fits are also shown in the figures, where the bestfitting spectral type is given in the upper right corner of each plot, together with the photometric redshift (value given in parentheses).The spectroscopic redshifts available for 13 sources are reported outside parentheses.Also given in each panel are values for the bolometric luminosities within the observed optical (0.3-3 m), IR (3-1000 m), and IRAC (3-10 m) bands for the best-fitting spectral type.
The spectral fitting was performed only on the optical and IR data points.We have therefore used the best-fitting spectral types to make a prediction for the corresponding X-ray emission in each case.For those spectral types in which the IR emission is dominated by star formation processes, we have calculated the corresponding X-ray emission on the basis of our estimated star formation rate (SFR).Following Franceschini et al. (2003), the SFR is calculated from the bolometric IR luminosity.Given the SFR, the number of high-mass X-ray binaries (HMXBs) can be inferred along with their corresponding 2-10 keV emission.For the case of ULIRGs not dominated by an AGN, it is assumed that all of the X-ray emission comes from HMXBs.In the case of lower luminosity starbursts, a quarter of the 2-10 keV luminosity may be due to HMXBs (the rest being attributable to lowmass X-ray binaries and the quiescent disk).Therefore, our expected 2-10 keV luminosity due to star formation is given by where f ranges from 0.25 to 1 (Franceschini et al. 2003).These expected values of X-ray flux are plotted for each of the starburst-like and ''normal galaxy'' spectral types.In all cases they are found to underpredict the observed X-ray flux, providing clear evidence for the existence of a hidden AGN in these sources.For those spectral types in which the IR emission is dominated by the AGN (Seyfert types 1 and 2 and QSOs), we have predicted the associated X-ray emission on the basis of the mean radioquiet quasar spectrum of Elvis et al. (1994).This is achieved using the ratio of the flux density at 4.5 m to that at 2 keV; in roughly 2/3 of cases this provides a good prediction of the observed X-ray flux.In the remaining cases the X-ray emission is overpredicted, indicating a component of the IR emission not originating from the AGN.
For the spectral types best fitted by the Mrk 231 template (15 sources in total), the actual X-ray flux of Mrk 231 is plotted as observed by Braito et al. (2004).This strongly underpredicts the observed X-ray fluxes in all cases, consistent with the very low ratio of S X /S 3.6 of Mrk 231 indicated in Figure 4 (''D'' arrow).Our SWIRE/Chandra objects display typical X-ray energy fluxes that are within approximately a decade of those measured in the IR, which demonstrates an AGN dominance in these sources.

Source Classification
Our analysis has found good spectral solutions for a large majority of the sources.Figures 5a-5j illustrate the quality of the observational spectral coverage and that of our spectral best fits.For a few sources in the figure the photometric redshift is compared with the available spectroscopic measures, with a fairly good agreement between the two.
For seven sources the observational SED coverage is not enough for reliable source characterization.In these cases our adopted classification exploits additional constraints from the X-ray spectrum and the IR to X-ray flux ratios.

A New Diagg nostic Diagg ram Based on Optical/IR SEDs
We have first attempted to identify a diagnostic diagram able to disentangle various categories of high-redshift active galaxies based on optical/IR colors, which are easily achievable for huge a Same as QSO1 in the optical, IR based on average SED of SWIRE type 1 AGNs ( Polletta et al. 2005).
References.-( 1) Silva et al. 1998; (2) PAH modified using ISO PHT-S spectra; (3) Polletta et al. 2005; (4) Berta et al. 2004;(5) Composite quasar spectra from the Large Bright Quasar Survey ( Brotherton et al. 2001); ( 6) Composite quasar spectra from the FIRST Bright Quasar Survey, Brotherton et al. 2001;(7) FIRST J013435.7-093102 in the optical (Gregg et al. 2002).Fig. 5aFig.5bFig.5cFig.5dFig.5eFig.5fFig.5gFig.5hFig.5iFig.5j Fig. 5.-Broadband SEDs and spectral best fits for all 99 Chandra / SWIRE sources.The scale on the x-axis has been artificially broken between the UV and X-ray bands (dotted line) for plot optimization.See text for a more detailed explanation.We have found that a maximum segregation of our four categories is obtained by combining the H-band to 4.5 m color against the H-band (k eA ¼ 1:7 m) to R-band magnitudes.As illustrated in Figure 6, our three classes tend to populate different regions of the color-color diagram, which are delimited by the two thin solid lines.In this plot, type 1 AGNs occupy the lower left corner, corresponding to sources with excess 4.5 m emission (of quasar origin) but blue or very blue optical to H-band spectra.With respect to these, type 2 AGNs appear to be shifted toward an excess in the 4.5 m and H-band fluxes.Finally, galaxies classified as normal (i.e., starbursts or quiescent disks) tend to occupy a quite complementary domain of the graph with respect to the previous two: if such galaxies are relatively blue in the H to optical color, they show no mid-IR excess (upper left corner); if they are red in the optical, they are also red in the nearto mid-IR.
This behavior is easily understandable on the basis of what we expect the typical source SEDs to be and what we anticipated in x 3.3: type 1 quasars have ''concave'' SEDs with a minimum in the rest frame around 1 m.Normal galaxies have a maximum around1m, being redshifted toward longer wavelengths, which is a spectral behavior quite complementary to that of quasars.Type 2 AGNs have spectra similar to those of type 1 AGNs, but with an excess in the H band corresponding to the emergent contribution of the host galaxy.Another feature of type 2 AGNs is that they lack the UV excess of type 1 objects and avoid the region of very blue H/R colors (low values of the S H /S R ratio).
The fluxes in the H band and IRAC 4.5 m band are particularly critical for separating the contributions of the host galaxy from that of the nuclear AGNs.However, our result here indicates that only a two-color diagnostic can work, whereas a single-color criterion (e.g., that based on the S H /S 4.5 color) is subject to high degeneracy.
Note that the diagram of Figure 6 seems to work for sources within our properly sampled redshift interval, i.e., up to z ' 2:5, but with poor statistics above z ¼ 1:5.A more extensive data set will be needed to prove whether similar diagnostics might work on a wider redshift interval.
We have attempted to verify the effect of the source redshift in this plot in a modelistic way by overplotting lines corresponding to the expected evolution of colors as a function of redshift for various different spectral templates.For each of the overplotted lines in Figure 6, the color for redshift z ¼ 0 corresponds to the start point marked with ''0.''Then lines are drawn from here to the points corresponding to z ¼ 0:5, 1, 1.5, 2, 2.5, 3, and finally 3.5.The dot-dashed line is for an elliptical galaxy template, the long-dashed line is for an Sb spiral/starburst, the short-dashed line is for a Seyfert 2 galaxy, the solid line is for a type 1 Seyfert galaxy or quasar, and finally the dotted line is for the Mrk 231 spectral template.It is evident that while for typical type 1 AGNs (the latter two) the evolution of colors with redshift is very modest, and they keep well within the lower left quadrant for any z (because of the featureless broadband spectra), strong color evolution is expected for all other categories.In particular, the H/R/4.5 m colors migrate from the upper left to lower right at increasing redshifts.
In any case, the expected colors remain within the boundaries for the various categories up to z ¼ 3:5, except for type 2 AGNs and starbursts, for which they tend to become confused at z > 1:5 2.

Flux Correlations and Color-Color Plots
The relation between X-ray and mid-IR fluxes is an important one for constraining the distribution of the circumnuclear absorbing medium and eventually for testing the unification scheme.Various correlation analyses have been published (Maiolino et al. 1995;Krabbe et al. 2001;Lutz et al. 2004), not all of them finding unequivocal results, however.
We have performed an X-ray to IR flux correlation study by exploiting our large sample and extensive IR coverage.We report in Figure 7 plots of the X-ray flux densities at 4 keV (based on Chandra fluxes in the total 0.5-8 keV band) versus those in the IRAC and MIPS 24 m bands for our various classes of sources.
X-ray and mid-IR fluxes appear to be significantly correlated only for type 1 quasars and AGNs.The Spearman correlation coefficient for the 3.6, 4.5, and 8 m IRAC fluxes and the MIPS 24 m flux are C:C: ¼ 0:70, 0.69, 0.81, and 0.62, respectively; the tightest correlation is then with the 8 m flux, while the one between the X-ray and far-IR 24 m fluxes is less significant.
For the type 2 AGNs and for normal galaxies and starbursts (our sources in categories 3 and 4) the correlations are much poorer (Spearman C:C: ¼ 0:02, 0.16, 0.06, and 0.2, in the same order, and C:C: ¼ 0:17, 0.16, 0.17, and 0.40, respectively).Perhaps surprisingly, none of these results depend significantly on the X-ray bandwidth, soft or hard, adopted for the comparison.Although not statistically significant, it is interesting to note that the scatter in the X-ray to IR correlation for type 1 AGNs, and even more so for other sources, increases at the lower fluxes.This seems to be an intrinsic effect, not due to the Fig. 6.-Color-color plot with S H /S 4.5 vs. S H /S R .Here S H and S R are the flux densities in the near-IR H band and the optical R band.Four source categories are reported here: type 1 quasars and AGNs ( filled squares), type 2 AGNs (open squares), normal spirals and starbursts (three-point stars), and early-type galaxies ( five-point stars).These four categories preferentially populate different regions of the diagram, which are indicated by the two thin solid lines.The lower line is defined by the two points on the graph at (10 0.75 , 10 À0.1 ) and (10 1.15 , 10 À0.45 ).The upper line is defined by the two points at (10 0.92 , 10 0.07 ) and (10 1.22 , 10 À0.18 ).The other thicker lines show the expected evolution of colors as a function of redshift for various spectral templates: dot-dashed line, elliptical galaxy; long-dashed line, Sb spiral/starburst; short-dashed line, Seyfert 2; solid line, Seyfert 1 or quasar; dotted line, Mrk 231.The color for z ¼ 0 is the point marked with ''0''; then lines are drawn from here to the points corresponding to z ¼ 0:5, 1, 1.5, 2, 2.5, 3, and 3.5.
increased photometric uncertainties at the lower fluxes, which are lower than the observed scatter.
This may support the case that for type 1 AGNs, the IR and X-ray emissions originate from a single dominant physical process, i.e., energy production by gravitational accretion.Furthermore, absorption effects in the dust torus due to atomic photoelectric effects and dust extinction should be unimportant.This may be in keeping with the notions that (1) in type 1 AGNs the contribution of star formation to the energy production is irrelevant compared with the QSO emission itself, (2) the line of sight is not intersecting large column densities of absorbing material, and (3) the dust distribution covers only a fraction of the ionizing photons emitted by the central power source (see x 3.9 about this point).
All this is evidently not true for the other classes of sources.As recently discussed in Lutz et al. (2004), the large contribution of star formation, particularly to the mid-and far-IR emissions, is likely to be responsible for increasing the scatter going from the type 1 to type 2 AGN population.Eventually, the lack of apparent correlation between X-ray and IR emission in the X-ray subsample identified with normal or purely starburst galaxies in the optical/IR shows that the two emissions may come from independent physical processes (quasar nuclear activity and stellar emission, either direct or dust reprocessed).
Finally, an attempt to combine an optical/IR color criterion with the ratio of the IR to X-ray flux is reported in Figure 8.Here the ratio of the X-ray to IR flux is compared with the IRAC channel 2 to H-band color.Type 1 quasars cluster around the upper right corner (high X-ray and mid-IR emissions), while type 2 AGNs and sources with starburst optical/IR colors scatter over a much wider region.
In the same figure we report the predicted broadband colors for spectral templates as a function of the source redshift.As in Figure 6, for each line the color for redshift z ¼ 0 corresponds to the point marked with ''0,'' and lines are drawn from here to the points corresponding to z ¼ 0:5, 1, 1.5, 2, 2.5, 3, and 3.5.Going from the right to the left, the two solid lines are for a type1 quasar and a Seyfert 1 template, the short-dashed line is for a Seyfert 2, and the long-dashed line is for an Sb spiral/starburst.The X-ray to IR flux ratios of the templates come from observations of local sources (see xx 3.1 and 3.2).This confirms that for type 1 objects the evolution of flux ratios is very moderate, while it is more pronounced for type 2 objects and starbursts.
An important point emerging from Figure 8 is that the X-ray to IRAC flux ratios S X / 3.6 S 3.6 of our source sample are, almost without exception, inconsistent with X-ray emission due to a starburst (whose z-dependent path in this color-color diagram should be close to the long-dashed line in the far left).This agrees well with the results of Figure 4.

X-Ray Hardness Ratio Analysis
The spectral information contained in our deep Chandra X-ray survey provides additional value for source characterization.Given the limited depth and photon counts, the X-ray spectral data were summarized in the usual form of a broadband hardness ratio HR.This was defined as HR ¼ (H À S )=(H þ S ), where H and S are the background-subtracted X-ray counts in the hard (2.0-8.0 keV) and soft (0.5-2.0 keV) bands, respectively.Full details on the HR calculation can be found in Manners et al. (2003).
Figure 9 plots such a hardness ratio against the X-ray to IR flux ratio.A first obvious effect is the clustering of type 1 quasars and AGNs around values of HR ' À0:5 and values of the X-ray to IR flux ratio from 0.3 to 1.Such values of HR correspond to the classical X ¼ 0:7 unabsorbed spectrum.
Type 2 AGNs tend to spread out of the canonical AGN region and to display significantly harder X-ray spectra and a lower X-ray emissivity.Both these effects are easily interpreted as being due to X-ray photoelectric absorption.A similar depression of the X-ray emissivity is apparent in Figure 7, correlating the IRAC channel 1 and total X-ray fluxes.The population of optically classified normal galaxies and starbursts displays an even farther spread out of the type 1 region.
What is interesting to note about this and Figure 8, however, is that a hardening of the X-ray spectrum and a decrease of the mid-IR AGN excess do not correspond to a dramatic decrease of the source X-ray emissivity, as might be expected in the presence of strong absorption in either X-rays or the IR or because of a change in the source population: with few exceptions, the values of the X-ray to IR flux ratio span a range not larger than a factor of 30 all together.

Redshift and Luminosity Distributions
We now undertake the more uncertain task of exploiting the information on cosmic distances for our combined Chandra/ SWIRE sample, mostly based on the photometric redshift technique.Until we acquire a more systematic spectroscopic coverage for this sample, such information on redshifts should be taken in a statistical sense rather than as values for individual sources.This is particularly the case for the featureless type 1 AGNs.
We report in Figure 10 the photometric redshift distributions for our sample sources split into three different classes: type AGNs (shaded histogram), type 2 AGNs (dotted histogram), and normal galaxies and starbursts (our previous classes 3 and 4, dashed histogram).The total distribution is reported as a solid line.It is evident that our source sampling mainly detects galaxies Fig. 8.-IRAC 4.5 m to H-band colors vs. the ratios of the X-ray to IR fluxes.Here S X is the flux in the total 0.5-8 keV band, and 3.6 S 3.6 is the IRAC channel 1 band flux.Symbols are as in Fig. 6.The lines are the predicted broadband colors as a function of z; the color for z ¼ 0 corresponds to the point marked with ''0,'' and lines are drawn from here to points corresponding to z ¼ 0:5, 1, 1.5, 2, 2.5, 3, and 3.5.From right to left, the lines are type 1 quasar and Seyfert 1 (two solid lines), Seyfert 2 (short-dashed line), and Sb spiral / starburst (long-dashed line).Fig. 9.-X-ray hardness ratio vs. X-ray to IR flux ratio.The hardness ratio HR is defined as (H À S )=(H þ S ), where H and S are the net X-ray counts in the hard (2.0-8.0 keV ) and soft (0.5-2.0 keV ) bands.Here S X is the usual total 0.5-8 keV flux.Symbols are as in Fig. 6. and AGNs between z $ 0:5 and z $ 1:5, with a maximum at z ¼ 1 and a fairly sustained tail up to z ¼ 2:5.
This distribution is interestingly similar to that reported by Zheng et al. (2004), which comes from an optical spectroscopic survey of the deep Chandra survey in the Chandra Deep Field South (CDFS), complemented with a photometric redshift analysis similar to ours.In the CDFS the observed z-distribution shows a narrow peak at z ¼ 0:6, which we do not observe, probably due to a large clustering overdensity in their sample.Compared to that of Zheng et al., a substantial advantage of our analysis is in the very extensive coverage of the mid-and far-IR bands that contains important diagnostic information and brings improved capabilities for redshift determination.
It is important to note that our sample is very representative of the sources of the XRB, if we consider that it includes 84% of the Chandra X-ray sample complete to S 0:5 8 keV ' 2 ; 10 À15 ergs cm À2 s À1 ; sources detected at this limit contribute roughly 60% of the XRB (Manners et al. 2003; see also Hasinger et al. 2001;Brandt et al. 2001;Giacconi et al. 2001).Our present results confirm earlier reported evidence (Hasinger 2003;Barger et al. 2003;Szokoly et al. 2004;Gilli et al. 2003) that most of the X-ray sources of the XRB are at a redshift lower than 1.5.
As for the individual population contributions, type 1 AGNs are spread over a rather large z-interval to z > 2, while type 2 objects and starburst /normal galaxies appear to be more concentrated at z ¼ 0:5 1:5.Such evidence of a moderately low median redshift for the sources of the XRB, which we confirm here in a statistical sense, demands significant revision of the XRB synthesis models (Franceschini et al. 2002;Gandhi & Fabian 2003).
We report in Figure 11 the population of the Chandra/SWIRE sources in the X-ray luminosity-redshift plane.Here L 0.5-8 keV is the rest-frame X-ray luminosity , where d L is the usual luminosity distance computed for the H 0 ¼ 71, m ¼ 0:27, Ã ¼ 0:73 case.The spectral slope ( X ) used to derive the rest-frame luminosities was calculated on the basis of the X-ray hardness ratio (HR) for each source: The flux S X is calculated from the observed X-ray count rate adopting the X spectral slope, in a consistent way.After the analysis of Fadda et al. (2002) and Franceschini et al. (2003), the X-ray luminosity is by itself an important discriminant of the primary power sources.On statistical grounds, and also supported by synthetic modeling of the X-ray emission by young stellar populations (Persic et al. 2004), it turns out that an X-ray luminosity of L 0:5À8 keV ' 3 ; 10 42 ergs s À1 (L 2À10 keV ' 1:5 ; 10 42 ergs s À1 ) is the maximum achievable by an ultraluminous starburst.This X-ray luminosity corresponds to a rate of star formation of $1500 M yr À1 assuming a Salpeter stellar initial mass function between 0.1 and 100 M , again a limit achievable by a starburst in the most luminous ULIRGs.
An important conclusion then emerges from the comparison of this limit with the X-ray luminosities estimated for our sources: if we exclude a few among the most local sources, typically at z < 0:5, the vast majority exceed this X-ray luminosity value.Our conclusion is that, with few exceptions, our sources are dominated by quasar emission.
A further diagnostic is provided by the ratio of the X-ray to bolometric luminosity L X /L bol , where L bol is determined from our combined fits of the near-, mid-, and far-IR SEDs.We report in Figure 12 this L X /L bol luminosity ratio against the total 0.5-8 keV band X-ray luminosity L X .Our sources display a fairly large range of L X /L bol values, with the X-ray luminosity ranging from $0.5% to somewhat more than 10% of L bol .For comparison, in samples of local ULIRGs Braito (2003) and Franceschini et al. (2003) found that a discriminant between a starburstdominated and quasar-dominated ultraluminous source may be  Source symbols are as in Fig. 6.The dotted line marks the upper boundary of the X-ray luminosity produced by an ultraluminous starburst.The solid line shows the S X ¼ 10 À15 ergs cm À2 s À1 observational flux limit.6) with those of the SWIRE sources without Chandra counterparts (small circles).The top panel shows the i-band to IRAC channel 4 color vs. g /i color, and the bottom panel uses the i-band to IRAC channel 3 color.The region in the upper left corner, delimited by the vertices (À2.4,À1), (À0.5, À0.13), and (À0.5, 1), contains virtually all the Chandra type 1 AGNs in both panels.The majority of the type 2 AGNs and of the Chandra-undetected galaxies fall outside this region.
211 MIPS sources in the Chandra field, 27 of which are Chandra detections (13%).The fraction of X-ray counterparts decreases steadily at fainter IR fluxes, reaching 9% at 200 Jy (39 Chandra out of 433 MIPS sources) and 6% at 100 Jy (59 out of 988).Considering the X-ray to 24 m flux correlation ( Fig. 7), however, this last figure turns out to be already affected by incompleteness due to the X-ray flux limit.These numbers, in any case, have to be increased by a factor of 1.2-1.3 to account for the type 1 Chandra-undetected objects (see Fig. 13).
The above numbers appear to be quite consistent with the AGN fraction for faint 15 m sources estimated by Fadda et al. (2002) at comparable flux limits (>0.4 mJy, for which a 15% AE 5% AGN fraction was estimated).Our improved statistics and sensitivity below this limit show that the AGN fraction does not seem to increase at fainter fluxes.It remains to be seen, however, whether an additional population of X-ray-weak type 2 AGNs is hidden among SWIRE sources with normal (starburst) optical/IR colors.While an indirect argument against this is mentioned in x 4, quantifying it will require either much deeper X-ray imaging or, better, deep imaging in very hard (>10 keV) X-rays.

Estimatinggthe Dust Covv eringgFraction in AGNs
To constrain the dust covering fraction in our type 1 AGN sample, we compare in Figure 14 an estimate of the AGN torus luminosity due to dust reradiated emission, directly observed by Spitzer, with the AGN bolometric luminosity as inferred from the X-rays.We have estimated the former through the observed source luminosity in the MIPS 24 m and IRAC 8 m channels; for type 1 AGNs, these roughly correspond to the expected emission peak by the AGN dust torus for our typical source redshifts (the dust torus emission is expected to peak at k $ 5 10 m, as observationally seen, e.g., by Elvis et al. [1994], and expected from theoretical modeling; Efstathiou & Rowan-Robinson 1995;Granato et al. 1997).The basic assumption of our analysis is that the fluxes in the 8 and 24 m channels for our type 1 AGNs are dominated by the AGN torus emission rather than by interstellar medium-reprocessed stellar light, which is supported by the power-law shapes of their IR SEDs shown in Figures 5a-5j, very inconsistent with those of normal starburst galaxies.
These AGN IR luminosities are compared with the bolometric ones, L bol, X , that we estimated from our observed 0.5-8 keV flux of the Chandra sources by applying a bolometric correction based on Elvis et al. (1994).To obtain this correction factor, we have integrated their average quasar SED and found a value of $20 to go from L X (0.5-8 keV) to bolometric luminosity.
In the Figure 14a the bolometric emission based on the X-ray flux is compared with the AGN IR emission, L AGN, IR , estimated from the 8 m flux as L AGN; IR ¼ ln (10) Â L()j 8m Ã (corresponding to the energy in one decade in wavelength centered at 8 m).In Figure 14b the comparison is made with the 24 m-based estimate, L AGN; IR ¼ ln (10) Â L()j 24m Ã .The bolometric AGN luminosities inferred from the X-ray flux and the AGN IR luminosities turn out to be fairly well correlated in Figure 14a, while the correlation is poorer in Figure 14b (the 24 m; see also Fig. 7).The solid and dotted lines in the figure correspond to the predictions in which 100% and 10%, respectively, of the AGN bolometric emission is reprocessed by dust into the IR, assuming that both L AGN, IR and L bol,X are dominated by AGN emission.
Under the same hypothesis of an AGN dominance in the X-ray and 24-8 m fluxes (argued on the basis of the power-law IR spectra for type 1 AGNs), the position of each source compared with these two lines can be considered to bear an indication of the covering factor of the circumnuclear dust.This covering factor ranges between values of 10% and 100%.Occasionally, L AGN, IR exceeds L bol, X , particularly when the former is computed from the 24 m flux (Fig. 14b), which may indicate a possible starburst contribution to the IR emission in some of these sources.

DISCUSSION AND CONCLUSIONS
A clear limitation in our analysis can be seen in the shallow depth achieved with the IRAC channels 3 and 4, which would F i g .14aFig.14b Fig. 14.-Luminosity of the AGN dust emission component in type 1 AGNs estimated from the Spitzer IR luminosity vs. bolometric luminosity derived from the X-ray flux.In (a) the dust emission is estimated from the IRAC 8 m luminosities; in (b) it is from the MIPS 24 m luminosities.The solid line marks a covering fraction of 100% (i.e., L AGN; IR =L bol; X ¼ 1), and the dotted line is for a covering fraction of 10%.Only sources detected at 8 and 24 m are reported in the respective panels.

Fig. 14b
Fig. 14a otherwise be important for source diagnostics.In many instances, even the upper limits of our IRAC survey have set useful constraints on the spectral fits.Deeper IRAC surveys will soon offer such improved capabilities.
We have found that 39% of the Chandra sources are dominated by type 1 AGN emission (QSOs or Seyfert 1), 23% are type 2 AGNs, and the remaining 38% are consistent with starburst-like or even normal galaxy spectra in the optical/IR.
On the basis of our estimated X-ray luminosities and the X-ray to IR flux ratios, we have shown that, with few exceptions, all the Chandra sources in our sample (including those with galaxylike optical/IR spectra) are dominated in X-rays by AGN emission.This allows us to estimate the global fraction of type 1 (our class 1) to type 2 AGNs (our classes 2, 3, and 4); assuming conservatively that all Chandra sources undetected by SWIRE are type 2 AGNs, the fraction of type 1 AGNs would still be half of the type 2 objects, and the type 1 fraction of the total AGNs would be at least one-third (39 objects out of 119).This type 1 AGN fraction is definitely larger than the canonical value 1/4 to 1/5, or even 1/10, needed to explain the XRB under the strict hypotheses of the standard XRB models (e.g., Comastri et al. 1995;Gilli et al. 2001).A solution is likely to come from Figure 10, showing that an important fraction of the sources of the XRB are found around z ¼ 1 (and 70% at z < 1:5); the XRB comes mostly from type 2 AGNs at moderate to low z, much lower than previously inferred from type 1 quasar surveys (typically found at z $ 2).
Our analysis of the mid-IR MIPS 24 m-selected sources, making up much of the CIRB, has shown that the fraction of those sources dominated by an AGN (either type 1 or type 2) is around $10% to 20% down to 0.3 mJy and may decrease at fainter fluxes.This is consistent with previous findings (Fadda et al. 2002).However, this information does not translate directly into a constraint on the AGN contribution to the CIRB, which is at the moment rather uncertain.Indeed, the above AGN fraction could be considered as an upper limit to the bolometric AGN contribution to the CIRB, because the AGN IR emission peaks at much shorter wavelengths (rest frame 10 m) than the CIRB.On the other hand, the available data, like those discussed in the present paper, cannot yet rule out the existence of an additional population of Compton-thick AGNs undetectable in X-rays that could contribute to the CIRB beyond our current 10% limit (e.g., Farrah et al. 2002;Wilman et al. 2003).
It may be worth noting that the locally observed mass density of supermassive black holes (BHs) in galaxies may set a constraint here, although indirect and somehow model dependent.Putting together the optical/X-ray type 1 quasar counts and the XRB spectral intensity, Fabian (2004) (see also Franceschini 2005) argues that, assuming a standard 10% efficiency of gas accretion in AGNs, the whole BH mass density can be explained by normal Compton-thin X-ray/optical AGNs.This sets an upper limit of 20% at most to the local BH density possibly coming from Compton-thick accretion, which implies a modest addition, if any, to the above numbers and a minor contribution altogether by Compton-thick AGNs to the CIRB.
To summarize, we have exploited combined Chandra and Spitzer multiwavelength imaging and optical/near-IR data to perform a systematic analysis of faint far-IR-and X-ray-selected sources.Our observations substantiate the concept that the two well-separated observational windows tend to select cosmic sources powered by fundamentally different processes, the far-IR probing preferentially stellar thermonuclear reactions, while the X-rays sample AGN accretion.Our results show that the IR observations would be capable of detecting with high efficiency (even at the moderate depths of the SWIRE Legacy survey) the XRB sources.On the other hand, we have shown that for a significant fraction (from 15% up to 40%) of the X-ray/IR sources in common, only deep X-ray observations have revealed the presence of moderately luminous and strongly absorbed AGNs.Support for this work, part of the Spitzer Space Telescope Legacy Science Program, was provided by NASA through an award issued by the Jet Propulsion Laboratory, California Institute of Technology, under NASA contract 1407.This work was also supported by the European Community RTN Network ''POE'' (grant HPRN-CT-2000-00138).

Fig. 1 .
Fig. 1.-Cumulative number counts of galaxies detected with signal-to-noise ratio >4 by Spitzer within the 286 square arcmin of the Chandra ACIS-I image.The vertical dotted lines mark the limiting fluxes for 90% completeness.

Fig. 3 .
Fig.3.-Distribution of positional distances between the centroids of the 3.6 m sources and the Chandra sources.The distribution includes a slight systematic offset in declination, also apparent in Fig.2, which, however, does not influence the source identification.

Fig. 4 .
Fig. 4.-Top: Histograms of the ratio of the 0.5-8 keV band X-ray flux (S X ) to the 24 m flux for sources detected by SWIRE in the area covered by Chandra.The solid-line histogram refers to sources with Chandra counterparts, while the shaded histogram is the distribution of upper limits for those without Chandra associations.Bottom: Same as top panel for the IRAC 3.6 m sources.Our adopted flux limits for the SWIRE samples are 5 and 100 Jy for 3.6 and 24 m, respectively.In the top panel the arrows indicate the values of representative active galaxies: A, NGC 4151; B, NGC 6240; C, IRAS 19254s; D, M82; and E, Mrk 231.

Fig. 7 .
Fig. 7.-Relation between the IRAC and MIPS 24 m flux densities (in Jy) and the X-ray flux densities at 4 keV ( based on Chandra fluxes in the total 0.5-8 keV band).In all panels filled squares are type 1 quasars, open squares are type 2 AGNs, three-point stars are spirals and starbursts, and five-point stars are early-type galaxies.The different panels show the correlation with the 3.6 m (top left), 4.5 m (top right), and 8 m (top right) IRAC and 24 m MIPS (bottom right) fluxes.

Fig. 10
Fig. 10.-Redshift distributions for 99 sources detected by Chandra in ELAIS N1 and identified with galaxies or AGNs (three stars are excluded).Three classes of sources are shown: type 1 AGNs (shaded histogram), type 2 AGNs (dotted histogram), and normal galaxies and starbursts (our previous classes 3 and 4, dashed histogram).

Fig. 11
Fig.11.-Distribution of rest-frame 0.5-8 keV X-ray luminosity and redshift for 99 sources detected by Chandra in ELAIS N1.The bulk of redshifts are photometric and are to be taken as only indicative for type 1 AGNs.Source symbols are as in Fig.6.The dotted line marks the upper boundary of the X-ray luminosity produced by an ultraluminous starburst.The solid line shows the S X ¼ 10 À15 ergs cm À2 s À1 observational flux limit.

Fig. 12
Fig.12.-Plot of the ratio of the total 0.5-8 keV X-ray to bolometric luminosity as a function of L X .Symbols are as in Fig.6.

Fig. 13
Fig. 13.-Comparison of the optical / IR colors of SWIRE /Chandra AGNs (large symbols, see legend of Fig.6) with those of the SWIRE sources without Chandra counterparts (small circles).The top panel shows the i-band to IRAC channel 4 color vs. g /i color, and the bottom panel uses the i-band to IRAC channel 3 color.The region in the upper left corner, delimited by the vertices (À2.4,À1), (À0.5, À0.13), and (À0.5, 1), contains virtually all the Chandra type 1 AGNs in both panels.The majority of the type 2 AGNs and of the Chandra-undetected galaxies fall outside this region.

TABLE 1
Identification Statistics of Spitzer and Chandra Sources