Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://hea-www.harvard.edu/~kim/pap/2005_ChaMP_AGN_space_density.pdf Äàòà èçìåíåíèÿ: Mon Dec 6 20:20:01 2010 Äàòà èíäåêñèðîâàíèÿ: Tue Oct 2 05:09:53 2012 Êîäèðîâêà: koi8-r Ïîèñêîâûå ñëîâà: ï ï ï ï ï ï ï ï ï ï

The Astrophysical Journal, 624:630 ­ 637, 2005 May 10
# 2005. The American Astronomical Society. All rights reserved. Printed in U.S.A.

COMOVING SPACE DENSITY OF X-RAY­ SELECTED ACTIVE GALACTIC NUCLEI
J. D. Silverman,1, 2 P. J. Green,1 W. A. Barkhouse,1 R. A. Cameron,1 C. Foltz,3 B. T. Jannuzi,4 D.-W. Kim, M. Kim,1 A. Mossman,1 H. Tananbaum,1 B. J. Wilkes,1 M. G. Smith,5 R. C. Smith,5 and P. S. Smith6
Received 2004 June 14; accepted 2005 February 1
1

ABSTRACT For measurement of the active galactic nucleus (AGN ) luminosity function and its evolution, X-ray selection samples all types of AGNs and provides reduced obscuration bias in comparison with UV excess or optical surveys. The apparent decline in optically selected quasars above z $ 3 may be strongly affected by such a bias. The Chandra Multiwavelength Project (CHAMP) is characterizing serendipitously detected X-ray sources in a large number of fields with archival Chandra imaging. We present a preliminary measure of the comoving space density using a sample of 311 AGNs found in 23 CHAMP fields ($1.8 deg 2) supplemented with 57 X-ray ­ bright AGNs from the Chandra Deep Field ­ North and Chandra Deep Field ­ South. Within our X-ray flux ( f0:3 ­ 8:0 keV > 4 ; 10þ15 ergs cmþ2 sþ1 )and optical magnitude (r 0 < 22:5) limits, our sample includes 14 broad emission-line AGNs at z > 3. Using this X-ray ­ selected sample, we detect a turnover in the comoving space density of luminous type 1 AGNs ( log LX > 44:5ergs sþ1 , measured in the 0.3 ­ 8.0 keV band and corrected for Galactic absorption) at z > 2:5. Our X-ray sample is the first to show a behavior similar to the well-established evolution of the optical quasar luminosity function. A larger sample of high-redshift AGNs and with a greater fraction of identified sources, either spectroscopic or photometric, at faint optical magnitudes (r 0 > 22:5) are required to remove the remaining uncertainty in our measure of the X-ray luminosity function, particularly given the possibility that AGNs might be more easily obscured optically at high redshift. We confirm that for z < 1, lower luminosity AGNs ( log LX < 44:5) are more prevalent by more than an order of magnitude than those with high luminosity. We have combined the Chandra sample with AGNs from the ROSAT surveys to present a measure of the space density of luminous type 1 AGNs in the soft X-ray band (0.5 ­ 2.0 keV ) that confirms the broadband turnover described above. Subject headingg galaxies: active -- quasars: general -- surveys -- X-rays: galaxies s:

1. INTRODUCTION Optical surveys have measured the evolution of QSOs out to z $ 6 ( Fan et al. 2004). The most dramatic feature found is the rise and fall of the comoving space density with peak activity at z $ 2:5. A systematic decrease in luminosity from z $ 2tothe present is evident with very few intrinsically bright QSOs in the local universe (e.g., Croom et al. 2004). This fading of the QSO population is attributed to a decreased fuel supply and /or fueling rate (e.g., Cavaliere & Vittorini 2000; Kauffmann & Haehnelt 2000). The dropoff in the space density at z > 3 (e.g., Warren et al. 1994; Schmidt et al. 1995; Fan et al. 2001; Wolf et al. 2003) could represent the growth phase of supermassive black holes (SMBHs), possibly under a veil of obscuration ( Fabian 1999). It has become clear over the past decade that X-ray selection of active galactic nuclei (AGNs) offers many benefits over detection methods in other wavebands. X-rays can penetrate large absorbing columns of gas that can effectively hide an accreting black hole in the optical. The existence of a significant, missed population of obscured AGNs is required to explain the spectral shape of the cosmic X-ray background (CXRB; e.g., Gilli et al. 2001). The current generation of X-ray observatories (Chandra, XMMNewton) are probing the faint source population responsible for
Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138; jsilverman@mpe.mpg.de. 2 Astronomy Department, University of Virginia, P.O. Box 3818, Charlottesville, VA 22903-0818. 3 National Science Foundation, 4201 Wilson Boulevard , Arlington , VA 22230. 4 National Optical Astronomy Observatory, P.O. Box 26732, Tucson, AZ 85726-6732. 5 Cerro Tololo Inter-American Observatory, National Optical Astronomical Observatory, Casilla 603, La Serena, Chile. 6 Steward Observatory, University of Arizona, Tucson, AZ 85721.
1

the bulk of the CXRB (e.g., Alexander et al. 2003; Rosati et al. 2002; Hasinger et al. 2001). X-ray surveys with Chandra, XMM-Newton, and ASCA are refining our knowledge of the AGN X-ray luminosity function. The peak, in the comoving space density of low-luminosity (L2:0 ­ 8:0 keV < 1044 ergs sþ1 ) AGNs found in hard (2 ­ 10 keV ) X-ray surveys ( Barger et al. 2003; Cowie et al. 2003; Fiore et al. 2003; Steffan et al. 2003; Ueda et al. 2003), occurs at z $ 1. This contrasts starkly with the behavior of the luminous (L2:0 ­ 8:0keV > 1044 ergs sþ1 ) QSOs, which are most prevalent at z $ 2:5. Using a highly complete sample of 941 AGNs selected in the soft band, Hasinger et al. (2005) show the same luminosity dependence and extend the space density measurement of low-luminosity AGNs (LX < 1044 ergs sþ1 ) out to z $ 3. These latest results clearly require a luminosity-dependent density evolution model ( Miyaji et al. 2000) in contrast to a ``pure'' luminosity evolution model, descriptive of optically selected QSOs (Croom et al. 2004). With these results, semianalytic models ( Menci et al. 2004) posit that the more luminous QSOs illuminate early epochs when most of the massive galaxies are forming (z > 2), thereby inducing high accretion rates, whereas lower luminosity AGNs dominate at a later period (z < 2) when most of the galaxies have fully assembled. X-ray ­ selected QSOs from ROSAT have hinted at a constant space density in the range 1:5 < z < 4:5 ( Miyaji et al. 2000) for the most luminous AGNs ( log L0:3 ­ 8:0 keV > 44:5). The lack of a decline in the space density at z > 3, in contrast to the behavior seen in optical surveys, could be evidence for obscuration at early epochs. We can test this model ( Fabian 1999) by measuring the luminosity function and comoving space density of X-ray ­ selected AGNs over a wide area to compile a significant sample at z > 3. Unfortunately, the ROSAT sample includes only seven QSOs at these high redshifts. Of these, only five QSOs have 630


SPACE DENSITY OF X-RAY­ SELECTED AGNs luminosities high enough to be detected over a broad range of redshift (0 < z < 5) and therefore included in the measurement of the comoving space density ( Miyaji et al. 2000). While the Chandra and XMM-Newton Deep Field observations have great scientific merit, their narrow area provides only four highluminosity ( log L0:3 ­ 8:0keV > 44:5) AGNs at z > 3. Recent measurement of the hard X-ray luminosity function ( Barger et al. 2003; Fiore et al. 2003; Steffan et al. 2003; Ueda et al. 2003) have limited numbers of AGNs at these high redshifts. A wider area survey is needed to compile a significant sample of highly luminous QSOs with redshifts greater than 3. We present a preliminary measure of the comoving space density of X-ray ­ selected AGNs out to z $ 4 from the Chandra Multiwavelength Project (CHAMP). We supplement the CHAMP sample with AGNs of comparable X-ray and optical fluxes from the Chandra Deep Field ­ North (CDF-N ) and Chandra Deep Field ­ South (CDF-S) to boost our spectroscopic completeness at faint optical magnitudes and include one additional z > 3 QSO. We merge the subsample of these AGNs detected in the soft (0.5 ­ 2.0 keV ) band with those from ROSAT to directly compare our results with those of Miyaji et al. (2000, 2001). A full presentation of the CHAMP X-ray luminosity function is forthcoming (J. Silverman et al. 2005, in preparation). We assume H0 ¼ 70 km sþ1 Mpcþ1 , ö ¼ 0:7, and M ¼ 0:3 with the exception of those parameters used in x 6.2. 2. CHANDRA MULTIWAVELENGTH PROJECT The CHAMP ( Kim et al. 2004a; Green et al. 2004) is carrying out an extragalactic X-ray survey encompassing 10 deg 2 using serendipitous detections in archival Chandra fields. Chandra's small point-spread function ($100 resolution on-axis) and low background allow sources to be detected to fainter flux levels ($10þ15 ergs cmþ2 sþ1 ) than any X-ray observatory past or present, thus enabling the detection of high-redshift (z $ 5) AGNs ( Treister et al. 2004; Barger et al. 2003; Castander et al. 2003; Silverman et al. 2002). For this study, we use the full Chandra energy range (0.3­ 8.0 keV ) to detect the absorbed sources missed by previous optical, UV, or soft X-ray surveys, and we take advantage of the high collecting area at soft energies to detect the faint, high-redshift AGNs. At z > 3, our survey is sensitive to AGNs with absorbing columns up to $1023 cmþ2 owing to the favorable k-correction. With a sample of AGNs selected from a large number of noncontiguous Chandra fields that reach similar depths ($10þ15 ergs cmþ2 sþ1 ), we effectively smooth out any effects from large-scale structures such as those evident in the CDF-S (Gilli et al. 2003). We have chosen 23 Chandra fields (1.8 deg 2) for which we have acquired extensive follow-up optical imaging and spectroscopy. The deepest observations have exposure times that are sensitive to sources with f0:3 ­ 8:0 keV > 8 ; 10þ16 ergs cmþ2 sþ1 . A full description of the CHAMP image reduction and analysis pipeline XPIPE can be found in Kim et al. (2004b). With our 4 m MOSAIC optical imaging, we are able to identify counterparts to the Chandra sources down to r 0 $ 25 (Green et al. 2004). We acquire optical imaging in three ( g 0 ; r 0 ; and i 0 ) Sloan Digital Sky Survey (SDSS) filters ( Fukugita et al. 1996). Optical colors provide preliminary source classification and crude photometric redshifts. These diagnostics are required to characterize the optically faint X-ray sources that cannot be identified with spectroscopy and, in particular, their influence on the X-ray luminosity function. The use of the SDSS photometric system allows more direct comparison between the CHAMP and the SDSS AGN surveys (Silverman et al. 2004). Optical spectroscopic follow-up currently focuses on identifying counterparts with r 0 < 22:5 for which

631

Fig. 1.--Area coverage as a function (0.5 ­ 2.0 keV; bottom) band X-ray flux for is generated since small-scale variations limited number of simulated sources and

of broad (0.3 ­ 8.0 keV; top) and soft 23 CHAMP fields. A smoothed curve or sharp features are a result of the are not real.

spectra can be acquired on a 4 ­ 6 m (i.e., MMT, Magellan, WIYN, CTIO Blanco) class telescope. To date, we have spectroscopically classified a sample of 358 AGNs detected in the broad band (0.3­ 8.0 keV ) in these 23 CHAMP fields. 3. X-RAY SENSITIVITY AND AREA COVERAGE There are complications with measuring the X-ray luminosity function that must be put into context. The difficulty in focusing X-rays results in a point-source function ( PSF ) and flux sensitivity that varies across the field of view. The PSF degrades as a function of off-axis angle, decreasing the flux sensitivity. In terms of measuring the luminosity function, it is not a trivial task to determine the incompleteness at the faintest X-ray fluxes and the actual sky area over which a source of a specific flux would be detected. In addition, the exposure times of these Chandra fields range from 17 to 114 ks, generating a wide range of limiting fluxes even on-axis. To characterize the sensitivity, completeness, and sky area coverage as a function of X-ray flux, a series of simulations were performed. The full details will be presented in an upcoming CHAMP X-ray analysis paper ( M. Kim et al. 2005, in preparation). The simulations consists of three parts: (1) generating artificial X-ray sources with MARX ( MARX Technical Manual7)
7

See http://space.mit.edu /CXC/ MARX.


632

SILVERMAN ET AL.

Vol. 624

Fig. 2.-- Optical magnitude (r 0 ) vs. X-ray flux (0.3 ­ 8 keV ) of X-ray sources found in 23 CHAMP fields. Optical spectroscopic classifications are indicated with the sample size. X-ray sources with no optical counterparts are shown by an arrow placed at the magnitude for a 5 detection from our optical imaging.

and adding them to real X-ray images, (2) detecting these artificial sources using wavdetect and extracting source properties with XPIPE identically as performed for actual sources, and (3) estimating the sky area coverage by comparing the input and detected source properties as a function of flux. The simulations are restricted to specific CCDs for ACIS-I ( I0, I1, I2, and I3) and ACIS-S ( I2, I3, S2, and S3). These CCDs are closest to the aim point for each observation. Sources far off-axis (á > 12 0 ) are excluded since the flux sensitivity is low and the PSF is degraded. These simulations allow determination of corrections for the source detection incompleteness at faint flux levels quantified in the first CHAMP X-ray analysis paper ( Kim et al. 2004b). In Figure 1 we show the sky coverage determined from the simulations using the broad (0.3 ­ 8.0 keV ) and soft (0.5 ­ 2.0 keV ) band source detections for 23 CHAMP fields. We are surveying a sky area of 1.8 deg 2 for the brightest sources. The sky area falls below 0.1 deg 2 at the faintest flux levels. 4. AGN SELECTION In these CHAMP fields, we find a diversity of objects (AGNs, clusters, galaxies, and stars), although 85% of them are attributed to an AGN (Green et al. 2004; Silverman et al. 2004). We show the optical magnitude (r 0 ) as a function of 0.3 ­ 8.0 keV X-ray flux for the sources detected in 23 CHAMP fields ( Fig. 2). We only include sources with greater than 9.5 net counts. Although sources with counts as low as 2 may be significant detections as a result of Chandra's low background, we use a higher count limit to ensure that we have well-measured X-ray fluxes within our sample. In Figure 2 we label sources as classified from our optical spectroscopy. Extragalactic objects with strong emis-

sion lines (Wk > 5 8) are labeled as either broad-line AGNs ( BLAGNs; FWHM > 1000 km sþ1 ) or narrow emission-line galaxies ( NELGs; FWHM < 1000 km sþ1 ). The BLAGNs are equivalent to type 1 AGNs. We loosely refer to the NELGs with LX > 1042 ergs sþ1 as type 2 AGNs, since optical spectra may not cover all the emission lines needed to confirm photoionization from a nonthermal source. Extragalactic counterparts with weak emission-line (Wk < 5 8) or pure absorption-line spectra are classified as absorption-line galaxies (ALGs). In addition, a handful of stars are identified at faint X-ray fluxes. Six clusters have been found based primarily on their extended X-ray emission. We calculate the rest-frame (0.3 ­ 8.0 keV; ergs sþ1 ) X-ray luminosity for each extragalactic source, after correction for Galactic absorption. The conversion from X-ray count rate to flux units (ergs cmþ2 sþ1 ) is determined from simulated detections on each CCD of a source with a power-law spectrum8 [ fE / Eþ( þþ1) ;þ ¼ 1:7] and galactic absorption ( Dickey & Lockman 1990). The observed luminosity is converted to the rest frame assuming a power-law spectrum with the photon index ( þ ) set to the average value for that object type ( BLAGN, NELG, ALG) on the basis of our X-ray spectral fit results ( T. L. Aldcroft et al. 2005, in preparation). We find that the BLAGNs have a mean spectral index (hþi) of 1.9, while the NELGs have hþi ¼ 0:9 and ALGs have hþi ¼ 1:5. We do not apply a correction for intrinsic absorption due to the uncertainty in the absorbing columns for most of these AGNs as a result of their low X-ray counts.
8 The CHAMP XPIPE ( Kim et al. 2004b) provides energy conversion factors for two models with þ ¼ 1:7and þ ¼ 1:4. We chose the former since the photon index more closely resembles the majority of the X-ray source detections with spectroscopic identification.


No. 2, 2005

SPACE DENSITY OF X-RAY­ SELECTED AGNs

633

Fig. 3.--X-ray luminosity vs. redshift of 394 extragalactic objects found in 23 CHAMP fields (with the exception of six clusters). The dashed line marks the minimum luminosity required for the AGN sample. Optical spectroscopic classifications are shown as either BLAGNs, NELGs, or ALGs.

To construct a pure AGN sample, we require the derived (rest frame 0.3 ­ 8.0 keV ) luminosity to exceed 1042 ergs sþ1 , thereby excluding any sources that contain a significant stellar or hot interstellar medium component. As shown in Figure 3, we see that 92% of the extragalactic sources in the CHAMP not associated with an extended X-ray ­ emitting cluster satisfy this criterion. We currently have not identified any NELGs or ALGs above a redshift of 1 in the CHAMP. Our optical spectroscopic limit (r 0 $ 22) precludes classification of objects with heavy optical extinction beyond this redshift, since a $5Lö elliptical host galaxy has r 0 $ 23 at z $ 1. We detect many optical type 2 AGNs with LX < 1044 ergs sþ1 but no luminous (LX > 1044 ergs sþ1 ) optical type 2 QSOs. We do find 6% of the BLAGNs to be X-ray ­ absorbed with a substantial fraction having reddened optical colors that may prevent them from being selected in optical surveys (Silverman et al. 2004). These red BLAGNs may be similar to those found by the Two Micron All Sky Survey ( Francis et al. 2004) that tend to have significant X-ray ­ absorbing columns of gas ( Wilkes et al. 2002). Deep optical follow-up of a broadband X-ray ­ selected sample will help determine the fraction of such red quasars to the space density of X-ray ­ luminous AGNs. 5. SURVEY COMPLETENESS We have supplemented our CHAMP sample with X-ray sources from the CDF-N ( Brandt et al. 2001; Barger et al. 2002) and CDF-S (Giacconi et al. 2002; Szokoly et al. 2004) with X-ray fluxes above our chosen limit (see below; f0:3 ­ 8:0 keV > 4 ; 10þ15 ergs cmþ2 sþ1 ). These samples allow us to boost the number of spectroscopically identified sources at faint optical magnitudes (r 0 > 21) and include any additional z > 3 AGNs. These are currently the only surveys with published optical spectroscopic identifications at moderate completeness levels for sources with flux levels comparable to the CHAMP. In Figure 4 (top), we plot the flux distribution of X-ray sources found in these 25 Chandra fields. The CDF-N and CDF-S clearly contribute most of the faint sources ( log fX < þ15). In Figure 4 (bottom), we show the optical magnitude distribution of counterparts to sources with X-ray fluxes above our chosen limit and those with optical spectroscopic identifications. Optical magnitudes for the CDF-N and CDF-S sources were converted from the Johnson

Fig. 4.--Top: X-ray (0.3 ­ 8.0 keV ) flux distribution of 1989 sources in the CHAMP and CDF-N and CDF-S surveys (solid histogram). Those sources from only the CDF-N and CDF-S are shown as the long dashed line. Those with optical spectroscopic identifications are shown as the short dashed line. The vertical line marks the X-ray flux limit of f0:3 ­ 8:0keV > 4 ; 10þ15 ergs cmþ2 sþ1 . Bottom: Optical magnitude (r 0 ) distribution of counterparts to X-ray sources with fluxes above our limit. Counterparts with spectroscopic identification are shown with a short (CHAMP) or long (CDF-N and CDF-S) dashed line. The vertical line marks our chosen optical magnitude limit.

(B, R) to the SDSS photometric system using the transformation in Fukugita et al. (1996). By including these two deep Chandra fields, our area coverage is larger than that shown in Figure 1. We add a fixed area of 0.22 deg 2, since the area coverage in these two fields does not change above this bright flux limit by more than 3%. To evaluate the status of our optical follow-up and determine effective survey limits, we measure the fraction of X-ray sources identified with optical spectroscopic observations as a function of X-ray flux and optical magnitude. Since the X-ray sources do not fully sample the fX -r 0 plane ( Fig. 2), we have implemented an adaptive binning scheme identical to that of Sanders & Fabian (2001) to generate a completeness map as a function of X-ray and optical flux for the combined sample (CHAMP + CDF-N + CDF-S). We create an array with 64 ; 64 elements covering the fX -r 0 plane ( Fig. 2). The identified fraction fID is measured at


634

SILVERMAN ET AL.

Vol. 624

Fig. 5.--Adaptively binned spectroscopic completeness map. The shading scale ranges from zero (lightest)to100%(darkest) spectroscopically classified. Only spectroscopically identified AGNs are shown with a black dot, and those at z > 3 with circled dots. Dashed lines show our chosen flux limits. A solid contour line has been overplotted to mark the 30% completeness level.

Fig. 6.--X-ray luminosity and redshift distribution of 368 AGNs (311 CHAMP + 57 CDF-N and CDF-S) with fluxes above our chosen limits.

L-z bin, equation (1) is summed over all AGNs ( N ) from our sample ( Fig. 6) that fall within this bin: n¼
N X i ¼1

each element as the number of identified objects divided by the number of X-ray sources. If 10 or more X-ray sources are in a single element, then we assign fID as the completeness level. We then increase the bin size by a factor of 2. If the binned element has 10 or more objects, the individual elements not previously assigned an identified fraction are set to this value. This procedure is iterated until the final binned element equals the size of the full array. In Figure 5, we plot the final array (i.e., completeness map) as a gray-scale image. On the basis of Figure 5, we set the flux limits of our AGN sample. We choose a single X-ray flux limit at f0:3 ­ 8:0 keV > 4 ; 10þ15 ergs cmþ2 sþ1 that is brighter than the faintest X-ray detections in the CHAMP. This limit ensures that we minimize the bias toward optically bright AGNs at the faintest X-ray flux limits. In order to limit our completeness correction to <3.3, we set an optical magnitude limit at a level at which at least 30% of the sources are identified; this leads to a magnitude limit of 22.5. With the remaining sample, the incompleteness level is used to correct the measurement of the space density (x 6). In Figure 5, we plot the location of AGNs, including those from the CDFs in the fX -r 0 plane. Since a significant number of AGNs fall in regions of low completeness at faint fluxes, we must exclude eight of the 22 AGNs with z > 3 and log fX > þ14:6 (circled points). In particular, the highest redshift AGN (CXOMP J213945.0þ234655; z ¼ 4:93; Silverman et al. 2002) found by the CHAMP has been excluded not because of its X-ray flux ( f0:3 ­ 8:0 keV ¼ 4:32 ; 10þ15 ergs cmþ2 sþ1 ) but rather because of its optical magnitude (r 0 ¼ 22:7), which falls below our limit. We do have a significant sample of 368 AGNs above our flux limits with 14 (13 from CHAMP) at z > 3, almost a factor of 3 larger than the high-redshift sample available from ROSAT. In Figure 6, we plot the LX -z distribution of the total sample of 368 AGNs selected in the 0.3 ­ 8.0 keV band. 6. COMOVING SPACE DENSITY We implement the 1/Va method (Schmidt 1968) to generate an estimate to the comoving space density (n in units of Mpcþ3 ; eq. [1]) in fixed redshift and luminosity intervals. For each

1 : fID Va (i )

Ï 1÷

As mentioned in x 3, our survey has noncontiguous sky coverage with varying flux limits. To account for this, we determine the accessible volume Va (eq. [2]) over which an individual AGN i will be included in our sample, given the X-ray and optical flux limits Z z2 dVc dz: Ï 2÷ Va ¼ dz z1 We calculate the X-ray and optical limiting redshifts (z X , z O ) lim lim for each object. The integrand in equation (2) is summed from z1 to z2 , where z2 is the smaller of z X and z O if the AGN cannot be lim lim detected over the full redshift interval. The solid angle () is a function of X-ray flux as shown in Figure 1, plus an additional 0.22 deg2 contributed by the CDF observations. A correction factor ( fID ), determined from the location of an individual AGN in the fX -r 0 plane ( Fig. 5), compensates for the incompleteness in our spectroscopic identifications as detailed in the previous section. Each AGN contributes ( fID Va )þ1 to the space density in a specific luminosity and redshift interval. We estimate 1 errors on the basis of a Poisson distribution because of the small number of objects per bin. While there are known biases inherent in the 1/Va method (e.g., Miyaji et al. 2001; Page & Carrera 2000), the overall evolutionary trends can be discerned (e.g., Barger et al. 2003; Cowie et al. 2003; Fiore et al. 2003). A detailed X-ray luminosity function will be presented in a future publication (J. Silverman et al. 2005, in preparation) while implementing more sophisticated analysis techniques (e.g., maximum likelihood method). 6.1. Broad Band (0.3 ­ 8.0 keV ) We measure the comoving space density (eq. [1]) of AGNs detected in the broad band from 25 Chandra fields. In Figure 7, the number density is plotted in three luminosity bins (42:0 <


No. 2, 2005

SPACE DENSITY OF X-RAY­ SELECTED AGNs

635

for a drop in the density that is inconsistent (at 3.4 ) with the peak value found at 2 < z < 3. This result is not strongly affected by our bright optical magnitude limit since most AGNs at z > 3 do not all fall near the flux limit, as shown in Figure 5. However, given the large numbers of unidentified sources, the upper limits to the space density at z > 2 ( Fig. 7) show considerable uncertainty. A higher fraction of identified X-ray sources is required to determine if a decline is present with high significance. The peak in the space density appears to shift to lower redshift with decreasing luminosity as modeled by luminositydependent density evolution ( Ueda et al. 2003; Miyaji et al. 2000), in contrast to pure luminosity evolution reported by the optical surveys of bright quasars (e.g., Croom et al. 2004). 6.2. Soft Band (0.5 ­ 2.0 keV ): Chandra+ROSAT With Chandra's broad energy range, we can select an AGN sample in the soft band (0.5 ­ 2.0 keV ) to directly compare with the ROSAT results. The CHAMP AGNs, supplemented by those in the CDFs, can be combined with the ROSAT sample ( Miyaji et al. 2000) to measure the number density of AGNs in the soft band with significant numbers of type 1 AGNs up to z $ 4. The ROSAT sample is a compilation of AGNs found in surveys spanning the pencil beam (0.3 deg 2), deep Lockman Hole ( Lehmann et al. 2001) to the wide-area (2:0 ; 104 deg 2), shallow ROSAT Bright survey (Schwope et al. 2000). To directly compare our sample with the published results from the ROSAT and optical surveys such as the Two-Degree Field (2dF; Croom et al. 2004), SDSS ( Fan et al. 2001), and COMBO-17 ( Wolf et al. 2003), we use cosmological parameters M ¼ 1, ö ¼ 0, and H0 ¼ 50 km sþ1 Mpcþ1 . The luminosity is calculated in the observed frame (no k-correction) to compare with the ROSAT results ( Miyaji et al. 2000). We have assembled a total of 1004 AGNs with L0:5 ­ 2:0 keV > 1042 ergs sþ1 , fX > 2 ; 10þ15 ergs cmþ2 sþ1 , and r 0 < 22:5 ( Fig. 8). The CHAMP AGNs boost the numbers at high redshift and lower luminosity because of Chandra's faint limiting flux. Following the analysis of Miyaji et al. 2000), we measure the comoving space density for luminous ( log L0:5 ­ 2:0keV > 44:5) AGNs to ensure that we are sensitive to these objects out to z $ 4. This minimum luminosity ensures that our measurements are not highly biased by our X-ray flux limit.

Fig. 7.-- Comoving space density of AGNs selected in the broad (0.3 ­ 8.0 keV ) band. Vertical bars represent estimates of the 1 error. Horizontal bars, centered on each data point, mark the redshift bin size. Horizontal bars with a downward arrow show the highly conservative upper limits that take into account the optically faint (r 0 > 22:5), unidentified X-ray sources. The upper limits correspond to the data that have the same line type connecting the points.

log LX < 43:0, 43:0 < log LX < 44:5, and 44:5 < log LX < 46:0; L is in units of ergs sþ1 ) as a function of redshift. We must keep in mind that we are limited to the detection of obscured AGNs (type 2) at luminosities below 1044 ergs sþ1 and redshifts below 1 owing to our optical spectroscopic limit, as shown in Figure 3. As a result, the highest luminosity range (44:5 < log LX < 46:0) is restricted to type 1 AGNs only. To account for the optically faint (r 0 > 22:5), spectroscopically unidentified X-ray sources, we remove the optical flux limit and implement the method of Cowie et al. (2003) and Barger et al. (2003) to measure upper limits to the space density. We have assigned all unidentified sources to each feasible L-z bin with the redshift and luminosity set to the mean value. With this method, a specific source usually falls within multiple L-z bins. At z < 1, the low-luminosity ( log LX < 44:5) AGNs are more than an order of magnitude more numerous than those with high luminosity, thus confirming recent results (Cowie et al. 2003; Fiore et al. 2003; Ueda et al. 2003). This is evident even when considering the unidentified population, since the upper limit to the space density of luminous AGNs ( log LX > 44:5) at z < 1is fairly constrained. From Figure 3, we see that most of the sources contributing to the space density at z < 1 are non-BLAGNs. These results agree with those reported by Steffan et al. (2003) that show type 2 AGNs as the dominant population at 0:5 < z < 1. The peak values of the two lower luminosity curves [n $ Ï3 7÷ ; 10þ5 Mpcþ3 ] agree with those from the soft (0.5 ­ 2.0 keV ) X-ray luminosity function ( Hasinger et al. 2005).9 We find the space density to be a factor of $3 less, for AGNs with 43:0 < log LX < 44:5 at z $ 0:8, than the hard (2 ­ 10 keV ) X-ray survey results of Ueda et al. (2003). These comparisons seem reasonable since our broadband (0.3 ­ 8.0 keV ) selected sample will be affected by absorption, as is known to occur in samples selected in the soft band (0.5 ­ 2.0 keV ). For the luminous AGNs ( log LX > 44:5), we measure a rise and fall of the number density with a peak at z $ 2:5. With 14 AGNs in the high-redshift bin (3 < z < 4:2), we have evidence
Also reproduced in Fig. 8a of a review article on deep extragalactic X-ray surveys by Brandt & Hasinger (2004).
9

Fig. 8.--X-ray luminosity vs. redshift of 1004 AGNs selected in the soft (0.5 ­ 2.0 keV ) band with ROSAT and Chandra. The horizontal line shows our threshold for measuring the space density of highly luminous AGNs.


636

SILVERMAN ET AL. 7. CONCLUSION

Vol. 624

We have measured the broadband (0.3 ­ 8.0 keV ) comoving space density using a sample of 368 X-ray ­ emitting AGNs detected by Chandra. Our spectroscopic magnitude limit to date allows the inclusion of optically obscured AGNs to z 1 and unobscured AGNs out to z $ 4. Our primary results are as follows: 1. We confirm that low-luminosity AGNs ( log LX < 44:5) are more prevalent at z < 1 than higher luminosity AGNs, as seen by Cowie et al. (2003), Fiore et al. (2003), and Ueda et al. (2003). Non-BLAGNs are the major contributor to the comoving space density at these low redshifts, as reported by Steffan et al. (2003). 2. The space density of type 1, highly luminous ( log LX > 44:5) X-ray ­ selected AGNs rises from the present epoch to a peak at z $ 2:5 and then declines at z > 3. This behavior is similar to that of optically selected surveys. This is the first X-ray ­ selected survey to detect a decline at high redshifts. The evolution is evident using a sample of 311 AGNs from the CHAMP selected in the broad (0.3 ­ 8.0 keV ) Chandra band and 217 luminous QSOs selected in the soft band (0.5 ­ 2.0 keV ) from Chandra and ROSAT. Our results support a more rapid depletion of fuel for the highluminosity AGNs from z $ 2:5 to the present epoch ( Menci et al. 2004), perhaps because of their higher accretion rate. At z < 1, the lower luminosity AGNs evolve slowly, which may be attributable to a low accretion rate, compared to the more luminous AGNs. At z > 3, the decline in the space density of highly luminous AGNs represents the growth phase of SMBHs during a period of rapid galaxy assembly. The uncertainties in the X-ray luminosity function will improve with larger samples and a higher fraction of source classification, either through spectroscopic or photometric techniques. With the inclusion of obscured, highly luminous QSOs, X-ray surveys are well on the way to presenting a more comprehensive view of AGN evolution.

Fig. 9.-- Comoving space density of 217 Chandra + ROSAT AGNs selected in the soft (0.5 ­ 2.0 keV ) band with log LX > 44:5 compared to the optical surveys. The optical space densities have been scaled to match the X-ray points at z ¼ 2:5 for ease of comparison.

In Figure 9, we plot the comoving space density of the 217 most X-ray ­ luminous AGNs. With these highly luminous AGNs, we can compare the space density to past ROSAT results and optical surveys out to high redshift (z < 5). To easily compare the space density evolution from optical surveys, in Figure 9, we have renormalized their curves to match the CHAMP at z ¼ 2:5. For all but the COMBO-17 survey, optical space densities (MB < þ26) have been scaled up by a factor of 3.2. The COMBO-17 survey, which reaches much greater depths (R < 24), was scaled down by a factor of 7.1. Some caution must be taken when comparing optical surveys with vastly different magnitude limits. For example, the AGNs in the 2dF survey evolve faster than those from X-ray ­ selected samples at z < 2 ( Fig. 9). The COMBO-17 survey, which probes similar absolute magnitudes (MB < þ22:2) as the CHAMP, may show evolution rates comparable to the X-ray surveys over the redshift interval 1:5 < z < 2:1. The number density of our combined sample is similar to the ROSAT results with z < 3. Our larger z > 3 sample (12 AGNs), while statistically consistent with the earlier results of Miyaji et al. 2000), has smaller errors and shows a decrease in the space density from 2 < z < 3 to 3 < z < 4:2, significant at the 4.6 level. Even with the addition of ROSAT AGNs, we have a slightly smaller sample at z > 3 than in the broadband sample (x 6.1) as a result of lower fluxes, luminosity, or counts in the soft band for the CHAMP AGNs. We find reasonable agreement with a slight excess between the relative drop in the space density of the X-ray and optical surveys at z > 2; our data are consistent with a decline in the X-ray luminosity function similar to that observed in the optical.

We are greatly indebted to NOAO and the SAO TACs for their support of this work. We thank the staffs at KPNO, CTIO, Las Campanas, W. M. Keck Observatory, FLWO, and MMT for assistance with optical observations. We thank John Huchra for his suggestions on dealing with complex selection effects. We also thank Takamitsu Miyaji and Guenther Hasinger for providing us with the ROSAT AGN catalog. We gratefully acknowledge support for this project under NASA CXC archival research grants AR3-4018X and AR4-5017X. T. L. A., R. A. C., P. J. G., D. W. K., A. E. M., H. T., and B. W. also acknowledge support through NASA contract NAS8-39073 (CXC ).

REFERENCES Alexander, D. M., et al. 2003, AJ, 126, 539 Croom, S. M., Smith, R. J., Boyle, B. J., Shanks, T., Miller, L., Outram, P. J., & Barger, A. J., Cowie, L. L., Capak, P., Alexander, D. M., Bauer, F. E., Brandt, Loaring, N. S. 2004, MNRAS, 349, 1397 W. N., Garmire, G. P., & Hornschemeier, A. E. 2003, ApJ, 584, L61 Dickey, J. M., & Lockman, F. J. 1990, ARA&A, 28, 215 Barger, A. J., et al. 2002, AJ, 124, 1839 Fabian, A. C. 1999, MNRAS, 308, 39 Brandt, W. N., & Hasinger, G. 2005, preprint (astro-ph /0501058) Fan, X., et al. 2001, AJ, 121, 54 Brandt, W. N., et al. 2001, AJ, 122, 2810 ------. 2004, AJ, 128, 515 Castander, F. J., Treister, E., Maccarone, T. J., Coppi, P. S., Maza, J., Zepf, Fiore, F., et al. 2003, A&A, 409, 79 S. E., & Guzman, R. 2003, AJ, 125, 1689 Francis, P. J., Nelson, B. O., & Cutri, R. M. 2004, AJ, 127, 646 Cavaliere, A., & Vittorini, V. 2000, ApJ, 543, 599 Fukugita, M., Ichikawa, T., Gunn, J. E., Doi, M., Shimasaku, K., & Schneider, Cowie, L. L., Barger, A. J., Bautz, M. W., Brandt, W. N., & Garmire, G. P. D. P. 1996, AJ, 111, 1748 2003, ApJ, 584, L57 Giacconi, R., et al. 2002, ApJS, 139, 369


No. 2, 2005

SPACE DENSITY OF X-RAY­ SELECTED AGNs

637

Gilli, R., Salvati, & M., Hasinger, G. 2001, A&A, 366, 407 Gilli, R., et al. 2003, ApJ, 592, 721 Green, P. J., et al. 2004, ApJS, 150, 43 Hasinger, G., Miyaji, T., & Schmidt, M. 2005, A&A, submitted Hasinger, G., et al. 2001, A&A, 365, L45 Kauffmann, G., & Haehnelt, M. 2000, MNRAS, 311, 576 Kim, D.-W., et al. 2004a , ApJ, 600, 59 ------. 2004b, ApJS, 150, 19 Lehmann, I., et al. 2001, A&A, 371, 833 Menci, N., Fiore, F., Perola, G. C., & Cavaliere, A. 2004, ApJ, 606, 58 Miyaji, T., Hasinger, G., & Schmidt, M. 2000, A&A, 353, 25 ------. 2001, A&A, 369, 49 Page, M. J., Carrera, F. J. 2000, MNRAS, 311, 433 Rosati, P., et al. 2002, ApJ, 566, 667 Sanders, J. S., & Fabian, A. C. 2001, MNRAS, 325, 178 Schmidt, M. 1968, ApJ, 151, 393

Schmidt, M., Schneider, D. P., & Gunn, J. E. 1995, AJ, 110, 68 Schwope, A., et al. 2000, Astron. Nachr., 321, 1 Silverman, J. D., et al. 2002, ApJ, 569, L1 ------. 2005, ApJ, 618, 123 Steffan, A. T., Barger, A. J., Cowie, L. L., Mushotzky, R. F., & Yang, Y. 2003, ApJ, 596, L23 Szokoly, G. P., et al. 2004, ApJS, 155, 271 Treister, E., Castander, F. J., Maccarone, T. J., Herrera, D., Gawiser, E., Maza, J., & Coppi, P. S. 2004, ApJ, 603, 36 Ueda, Y., Akiyama, M., Ohta, K., & Takamitsu, M. 2003, ApJ, 598, 886 Warren, S. J., Hewett, P. C., & Osmer, P. S. 1994, ApJ, 421, 412 Wilkes, B. J., Schmidt, G. D., Cutri, R. M., Ghosh, H., Hines, D. C., Nelson, B., & Smith, P. S. 2002, ApJ, 564, L65 Wolf, C., Wisotzki, L., Borch, A., Dye, S., Kleinheinrich, M., & Meisenheimer, K. 2003, A&A, 408, 499