Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://hea-www.harvard.edu/~kim/pap/2009_ChaMPx_quasar.pdf Äàòà èçìåíåíèÿ: Mon Dec 6 20:20:13 2010 Äàòà èíäåêñèðîâàíèÿ: Tue Oct 2 01:51:40 2012 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: http astrokuban.info astrokuban

The Astrophysical Journal, 690:644­669, 2009 January 1 c 2009. The American Astronomical Society. All rights reserved. Printed in the

doi:10.1088/0004-637X/690/1/644
U.S.A.

A FULL YEAR'S CHANDRA EXPOSURE ON SLOAN DIGITAL SKY SURVEY QUASARS FROM THE CHANDRA MULTIWAVELENGTH PROJECT
Paul J. Green1 ,8 , T. L. Aldcroft1 , G. T. Richards2 , W. A. Barkhouse3,8 , A. Constantin1 , D. Haggard4 , M. Karovska1 , D.-W. Kim1 , M. Kim5 , A. Vikhlinin1 , S. F. Anderson4 , A. Mossman1 , V. Kashyap1 , A. C. Myers6 , J. D. Silverman7 , B. J. Wilkes1 , and H. Tananbaum1
1

Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138, USA; pgreen@cfa.harvard.edu 2 Department of Physics, Drexel University, 3141 Chestnut Street, Philadelphia, PA 15260, USA 3 Department of Physics, University of North Dakota, Grand Forks, ND 58202, USA 4 Department of Astronomy, University of Washington, Seattle, WA, USA 5 International Center for Astrophysics, Korea Astronomy and Space Science Institute, Daejeon, 305-348, Korea 6 Department of Astronomy, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA 7 Institute for Astronomy, ETH Zurich, 8093 Zurich, Switzerland ¨ ¨ Received 2008 May 15; accepted 2008 August 25; published 2008 December 1

ABSTRACT We study the spectral energy distributions and evolution of a large sample of optically selected quasars from the Sloan Digital Sky Survey that were observed in 323 Chandra images analyzed by the Chandra Multiwavelength Project. Our highest-confidence matched sample includes 1135 X-ray detected quasars in the redshift range 0.2 < z < 5.4, representing some 36 Msec of effective exposure. We provide catalogs of QSO properties, and describe our novel method of calculating X-ray flux upper limits and effective sky coverage. Spectroscopic redshifts are available for about 1/3 of the detected sample; elsewhere, redshifts are estimated photometrically. We detect 56 QSOs with redshift z > 3, substantially expanding the known sample. We find no evidence for evolution out to z 5 for either the X-ray photon index or for the ratio of optical/UV to X-ray flux ox . About 10% of detected QSOs show best-fit intrinsic absorbing columns greater than 1022 cm-2 , but the fraction might reach 1/3 if most nondetections are absorbed. We confirm a significant correlation between ox and optical luminosity, but it flattens or disappears for fainter (MB -23) active galactic nucleus (AGN) alone. We report significant hardening of both toward higher X-ray luminosity, and for relatively X-ray loud quasars. These trends may represent a relative increase in nonthermal X-ray emission, and our findings thereby strengthen analogies between Galactic black hole binaries and AGN. For uniformly selected subsamples of narrow-line Seyfert 1s and narrow absorption line QSOs, we find no evidence for unusual distributions of either ox or . Key words: galaxies: active ­ quasars: absorption lines ­ quasars: general ­ surveys ­ X-rays: general Online-only material: color figures, machine-readable tables

1. INTRODUCTION Interest in the properties of active galaxies and their evolution has recently intensified because of deep connections being revealed between supermassive black holes (SMBHs) and galaxy evolution, such as the relationship between the mass of galaxy spheroids and the SMBHs they host (the MBH ­ connection; Ferrarese & Merritt 2000; Gebhardt et al. 2000). A feedback paradigm could account for this correlation, whereby winds from active galactic nuclei (AGNs) moderate the SMBH growth by truncating that of their host galaxies (e.g., Granato et al. 2004). Feedback models may explain the correspondence between the local mass density of SMBHs and the luminosity density produced by high-redshift quasars (Yu & Tremaine 2002; Hopkins et al. 2006) as well as the "cosmic downsizing" (decrease in the space density of luminous AGNs) seen in AGN luminosity functions (Barger et al. 2005; Hasinger 2005; Scannapieco et al. 2005). If quasar activity is induced by massive mergers (e.g., Wyithe & Loeb 2002, 2005), then the jigsaw puzzle now assembling may merge smoothly with cosmological models of hierarchical structure formation.
8 Visiting Astronomer, Kitt Peak National Observatory, Cerro Tololo Inter-American Observatory, and National Optical Astronomy Observatory, which is operated by the Association of Universities for Research in Astronomy, Inc. (AURA), under cooperative agreement with the National Science Foundation.

Many, if not most, of the accreting SMBHs in the universe may be obscured by gas and dust in the circumnuclear region, or in the extended host galaxy. The obscured fraction may depend on both luminosity and redshift (Ueda et al. 2003; Brandt & Hasinger 2005; La Franca et al. 2005), and is indeed likely to evolve on grounds both theoretical (e.g., Hopkins et al. 2006a) and observational (Treister & Urry 2006; Ballantyne et al. 2006). Such evolution seems to be required for AGN populations to compose the observed spectrum of the cosmic X-ray background (CXRB; Gilli et al. 2007). However, a full census of all SMBHs remains observationally challenging, since some are heavily obscured, or accreting at very low rates, below the sensitivity limits of current telescopes even at low redshifts. AGN unification models explain many of the observed differences in the spectral energy distributions (SEDs) of AGNs as being due to the line-of-sight effects of anisotropic distributions of obscuring material near the SMBH (Antonucci 1993). The intrinsic number ratio of obscured-to-unobscured AGN may evolve, and is almost certainly a function of luminosity. Indeed, the ratio in the Seyfert (low-) luminosity regime is currently estimated to be 4, whereas for the QSO (high-) luminosity regime, it may be closer to unity (Gilli et al. 2007). Astronomers, like most people, usually look where they can see. Type 1 quasars are the easiest AGNs to find in large numbers via either spectroscopic or color selection because of their broad emission lines (FWHM 1000 km s-1 ) and 644


No. 1, 2009

PROPERTIES OF SDSS QUASARS IN THE ChaMP

645

generally blue continuum slopes and brighter magnitudes. Most of these show few other signs of obscuration such as infrared excess or weakened X-ray emission. Large samples of Type 1 QSOs--as the brightest high-redshift objects--have served to probe intervening galaxies, clusters, and the intergalactic material (IGM) along the line of sight, right to the epoch of re-ionization. Because observed SEDs are thought to be less affected by obscuration and therefore more representative of intrinsic accretion physics, the evolution of this Type 1 sample is of interest as well. The SEDs and clustering properties of Type 1 QSOs have been studied in increasing detail, probing farther into the universe and wider across the sky and the electromagnetic spectrum. SEDs including mid-infrared photometry from Spitzer were compiled and characterized recently for 259 quasars from the Sloan Digital Sky Survey (SDSS) by Richards et al. (2006b). These studies are most useful for calculating bolometric luminosities and K-corrections toward understanding the energetics of the accretion process, and its evolution across cosmic time. Studies that target SDSS QSO pairs with small separations find a significant clustering excess on small scales ( 40 kpc h) of varying strengths (e.g., Hennawi et al. 2006;Myersetal. 2007, 2008), which could be due to mutual triggering, or might simply result from the locally overdense environments in which quasars form (Hopkins et al. 2008). Optical luminosity function studies of optically selected quasars (e.g., Richards et al. 2006a, 2005; Croom et al. 2004) date back decades (e.g., Boyle et al. 1988). Accurate luminosity functions are needed to trace the accretion history of SMBHs and to contrast the buildup of SMBHs with the growth of galaxy spheroids. An increasing number of optical surveys not only select AGNs photometrically, but also determine fairly reliable photometric redshifts for them. These samples stand to vastly improve the available statistical reliability and the resolution available in the luminosity/redshift plane. X-ray observations have been found to efficiently select AGNs of many varieties, and at higher surface densities than ever (Hasinger 2005). Independent of the classical AGN optical emission line criteria, X-rays are a primary signature of accretion onto a massive compact object, and the observed X-ray bandpass corresponds at higher redshifts to rest-frame energies capable of penetrating larger intrinsic columns of gas and dust.9 Even for Type 1 QSOs, X-ray observations have revealed new connections (e.g., between black holes from 10 to 109 M ; Maccarone et al. 2003; Falcke et al. 2004) and new physical insights such as the possibility of ubiquitous powerful relativistic outflows (Middleton et al. 2007) or of relativistically broadened fluorescent Fe K emission (see, e.g., the review articles by Fabian et al. 2000; Reynolds & Nowak 2003). Quasars with broad absorption lines (BAL QSOs) blueward of their UV emission lines turn out to be highly absorbed in X-rays (Green et al. 1995, 2001; Green 1996; Gallagher et al. 2002). Quasars that are radio loud are 2­3 times brighter in X-rays for the same optical magnitude (Zamorani et al. 1981; Worrall et al. 1987; Shen et al. 2006), and also may have harder X-ray spectra (e.g., Shastri et al. 1993, although see Galbiati et al. 2005). The high sensitivity and spatial resolution of Chandra and XMM-Newton open other avenues for exploration of quasars and their environments. Clusters of galaxies have been discovered in the vicinity of, or along the sightlines to quasars (Green
ef The observed-frame, effective absorbing column is NH f NH /(1 + z) (Wilman & Fabian 1999). 9 2.6

et al. 2005; Siemiginowska et al. 2005). Lensed quasars have now been spatially resolved in X-rays, unexpectedly showing significantly different flux ratios than at other wavebands (Green et al. 2002; Blackburne et al. 2006; Lamer et al. 2006). The expected evolution in the environment, accretion rates, and masses of SMBHs in AGNs should correspond to observable evolution in their SEDs. The two most common X-ray measurements used are the X-ray power-law photon index 10 and the X-ray-to-optical spectral slope, ox .11 Many of the apparent correlations have been challenged as being artifacts of selection or the by-products of small, heterogeneous samples, which impede progress in our understanding of quasar physics and evolution. The archives of current X-ray imaging observatories such as Chandra and XMM-Newton are growing rapidly, and several large efforts for pipeline processing, source characterization (Ptak & Griffiths 2003; Aldcroft 2006), and catalog generation are underway. The 2XMM catalog (Page 2006) is available, and the Chandra source catalog12 is due out in 2008 (Fabbiano et al. 2007). Serendipitous wide-area surveys with Chandra were pioneered by the Chandra Multiwavelength Project (ChaMP; Green et al. 2004; Kim et al. 2004) for high Galactic latitude, while ChaMPlane (Grindlay et al. 2005) has studied the stellar content of Galactic plane fields. The ChaMP, described in more detail below, also performs multiwavelength source matching and spectroscopic characterization. These efforts are greatly augmented by large surveys such as the SDSS (York et al. 2000), which will obtain spectroscopy of 100,000 QSOs (e.g., Schneider et al. 2007), and can additionally facilitate the efficient extrapolation of photometric quasar selection with photometric redshift estimation almost a magnitude fainter, toward a million QSOs (e.g., Richards et al. 2008). The current paper studies the X-ray and optical properties of the subset of these QSOs imaged in X-rays by Chandra as part of the ChaMP (Green et al. 2004). 2. THE QUASAR SAMPLE 2.1. The SDSS Quasar Sample Most large samples of Type 1 QSOs are based on optically selected quasars confirmed via optical spectroscopy (Boyle et al. 1988; Schneider et al. 1994; Hewett et al. 1995). The largest, most uniform sample of optically selected quasars by far has been compiled from the SDSS. With the completion of the SDSS, we can expect some 100,000 spectroscopically confirmed quasars. The SDSS quasars were originally identified to i < 19.1 for spectroscopy by their UV-excess colors, with later expansion for z > 3 quasars to i = 20.2 using ugri color criteria (Richards et al. 2002). The large catalog, the broadwavelength optical photometry, and subsequent follow-up in other wavebands have meant that research results from the SDSS spectroscopic quasar sample better characterize the breadth of the quasar phenomenon than ever before. However, the limited number of fibers available, fiber placement conflicts, and above all the bright magnitude limits of SDSS fiber spectroscopy mean
is the photon number index of an assumed power-law continuum such that NE (E ) = NE0 E . In terms of a spectral index from f = f0 , we define = (1 - ). 11 is the slope of a hypothetical power law from 2500 å to 2 keV; ox ox = 0.3838 log(l2500å /l2keV ). 12 The Chandra source catalog Web page is http://cxc.harvard.edu/csc.
10


646

GREEN ET AL.

Vol. 690

that 10 times as many quasars have been imaged, and could be efficiently identified from existing SDSS photometry. Using large spectroscopic AGN samples as "training sets" can produce photometric classification and redshifts of far greater completeness and depth than spectroscopy. Without spectroscopic confirmation, photometric selection criteria strike a quantifiable balance between completeness and efficiency, i.e., a probability can be assigned both to the classification and the redshift. Deep photometric redshift surveys like COMBO-17 (Wolf et al. 2003) have found AGNs to R = 24 and z = 5 with high completeness. Efficient photometric selection of quasars in the SDSS using a nonparametric Bayesian classification based on kernel density estimation is described in Richards et al. (2004) for SDSS point sources with i < 21. An empirical algorithm to determine photometric redshifts for such quasars is described in Weinstein et al. (2004). The spectroscopic training samples for these methods now include far more high-redshift quasars, and so the algorithms have been retrained to include objects redder than (u - g ) = 1.0 to classify high-z quasars, and applied to the much larger SDSS Data Release 6 (DR6). This large catalog of 1 million photometrically identified QSOs and their photometric redshifts is described in Richards et al. (2008). They estimate the overall efficiency of the catalog to be better than 72%, with subsamples (e.g., X-ray detected objects) being as efficient as 97%. These estimates are based on an analysis of the autoclustering of the objects in the catalog (Myers et al. 2006), which is very sensitive to stellar interlopers. However, at the faint limit of the catalog some additional galaxy contamination is expected. For luminosity and distance calculations, we adopt an H0 = 70 km s-1 Mpc-1 , = 0.7, and M = 0.3 cosmology throughout. We assume = -0.5 for the optical continuum power-law slope (v , where is the emission frequency), and derive the rest-frame, monochromatic optical luminosity at 2500 å (l2500å ; units erg s-1 Hz-1 ) using the SDSS dereddened magnitude with central wavelength closest to (1 + z) â 2500 å. 2.2. The Extended Chandra Multiwavelength Project The ChaMP is a wide-area serendipitous X-ray survey based on archival X-ray images of the (|b| > 20 deg) sky observed with the AXAF CCD Imaging Spectrometer (ACIS) onboard Chandra. The full 130-field Cycle 1­2 X-ray catalog is public (Kim et al. 2007b), and the most comprehensive X-ray number counts (log N­log S) to date have been produced thanks to 6600 sources and massive source-retrieval simulations (Kim et al. 2007a). We have also published early results of our deep (r 25) NOAO/MOSAIC optical imaging campaign (Green et al. 2004), now extended to 67 fields (W. A. Barkhouse et al. 2009, in preparation). ChaMP results and data can be found online.13 To improve statistics and encompass a wider range of source types, we have recently expanded our X-ray analysis to cover a total of 392 fields through Chandra Cycle 6. We chose only fields overlapping with SDSS DR5 imaging. To ease analysis and minimize bookkeeping problems, the new list of Chandra pointings (observation IDs; "obsids" hereafter) avoids any overlapping observations by eliminating the observation with the shorter exposure time. As described in Green et al. (2004), we also avoid fields with large ( 3 ) extended sources in either optical or X-rays (e.g., nearby galaxies M101, NGC 4725, NGC 4457, or clusters of galaxies MKW8, or Abell 1240). Spurious X-ray sources (due to, e.g., hot pixel, bad bias, bright
13

Figure 1. Sky area vs. B-band (0.5­8 keV) flux limit for the 323 obsids included in our ChaMP/SDSS field sample. Flux limit is defined here as the number of counts detectable in 90% of simulation trials, converted to flux assuming a power-law = 1.7 at z = 0 and the Galactic NH appropriate to each obsid. Chip S4 (CCD ID 8) is excluded throughout. The area covered at the brightest fluxes is 32 deg2 .

source readout streaks) have been flagged and removed as described in Kim et al. (2007b). Of the 392 ChaMP obsids, 323 overlap the SDSS DR5 footprint. The ChaMP has also developed and implemented an xskycover pipeline which creates sensitivity maps for all ChaMP sky regions imaged by ACIS. This allows (1) identification of imaged-but-undetected objects, (2) counts limits for 50% and 90% detection completeness, and (3) corresponding flux upper limits at any sky position, as well as (4) flux sensitivity versus sky coverage for any subset of obsids, as needed for log N­log S and luminosity function calculations. Our method is described in the Appendix, and has been verified recently by Aldcroft et al. (2008)using the Chandra Deep Field South (CDF-S). The final sky ChaMP/SDSS coverage area (in deg2 ) for the 323 overlapping fields as a function of broadband (B band 0.5­8 keV) flux limit is shown in Figure 1 (see caption for specific definition of this limit). The area covered at the brightest fluxes is 32 deg.2 On average five CCDs are activated per obsid. We have downloaded into the ChaMP database all the SDSS photometry, and the list of photo-z quasar candidates within 20 of the Chandra aim point for each such obsid.14 Because the Chandra point-spread function (PSF) increases with off-axis angle (OAA), comparatively few sources are detected beyond this radius, and source centroids also tend to be highly uncertain. Of X-ray detected candidates, we will show in Section 2.4 that 98% of these candidates with spectra are indeed QSOs. Next we describe the identification of high-confidence ChaMP X-ray counterparts to SDSS QSOs in Section 2.3. We then discuss in Section 2.4 spectroscopic identifications for these objects. Section 3 then describes results for several interesting QSO subsamples, including our treatment of SDSS quasars that were not X-ray detected.
14

http://hea-www.harvard.edu/CHAMP

For 14 obsids, we extended to 28 radius, to achieve full coverage of the Chandra footprint. For other obsids, the SDSS imaging strips do not completely cover the Chandra field of view.


No. 1, 2009

PROPERTIES OF SDSS QUASARS IN THE ChaMP

647

Figure 2. Positional offsets of matched sample. Left: the histogram of X-ray/optical centroid separations in the full matched sample of 1376. The mode is 0. 3, the median is 0. 76, and the mean is 1. 1. Right: the ratio of the X-ray/optical centroid separation to the 95% X-ray positional uncertainty vs. separation. The fraction of objects with separations larger than the 95% uncertainty (i.e., with ratio greater than 1) remains relatively constant, and virtually no separations wider than 3 are larger than the positional uncertainty XP95 .

2.3. X-Ray/Optical Matching The positional uncertainty of ChaMP X-ray source centroids has been carefully analyzed via X-ray simulations by Kim et al. (2007a) and depends strongly on both the number of source counts and the OAA. Cross-correlating ChaMP X-ray source centroids with QSOs, we find that 95% of the X-ray/ optical separations dXO of matched QSOs are smaller than the 95% X-ray positional uncertainty XP95 . We first perform an automated matching procedure between each optical QSO position and the ChaMP's X-ray source catalog. We adopt 4 as our matching radius criterion. Figure 2 shows that 95% of the matched sample has an X-ray/optical position difference of less than 3 , and expansion to include match radii above 4 would not substantially increase the sample size. Searching the ChaMP catalog for X-ray sources within 4 of the optical SDSS quasar coordinate, yields 1376 unique matches in the "Matched" sample. Although the overall efficiency of the SDSS photometric QSO catalog is only expected to be 72% (Richards et al. 2008), the rarity of luminous Type 1 quasars and X-ray sources means that matched objects should be quite clean. By repeatedly offsetting the SDSS coordinates of each QSO by 36 and rematching to the ChaMP X-ray catalog, we derive a spurious match rate of just 0.7%.15 This excellent result is due to Chandra's arcsec spatial resolution, which at SDSS depths allows for unambiguous counterpart identification, given that both Type 1 QSOs and X-ray sources are relatively sparse on the sky. We identify and remove a variety of objects with potentially poor data, including overlapping multiple sources (some of which are targeted lenses), bright X-ray sources suffering from pile-up, optical sources with photometry contaminated by close brighter sources or within large extended galaxies, or stellar diffraction spikes. In addition to the automated matching procedure, we also perform visual inspection (VI) of both X-ray and optical
15

images, overplotting the centroids and their associated position errors. We retain only the highest-confidence matches (matchconf=3). Most of the 105 objects we thereby eliminate have large ratios of dXO /XP95 , or multiple candidate optical counterparts. We note that some of the most interesting celestial systems may be found among sources with matchconf<3. For example, these might include QSOs that are lensed, have bright jets, or are associated with host or foreground optical clusters or galaxies. Systems that are poorly matched, multiply matched, or photometrically contaminated may account for up to 10% of the full X-ray-selected sample. We therefore caution against blind cross-correlation of large source catalogs (e.g., the Chandra Source Catalog)16 without such detailed quality control and visual examination of images. However, since we seek here to analyze the multiwavelength properties of a large clean sample of QSOs, and since most of these more complicated systems require significant further analysis or observation, we defer their consideration to future studies. 2.4. Spectroscopic Redshift Information After cross-correlation with the X-ray catalog, we sought spectroscopic redshifts for any objects in the photometric QSO catalog. For this purpose, we obtained redshifts from existing ChaMP spectroscopy, from the SDSS (DR6) database itself, and then finally we searched the literature by cross-correlating optical positions with the NASA Extragalactic Database (NED), using a 2 match radius. Of 1376 matched objects, we found high confidence spectroscopic redshifts for 407, of which 43 spectra were observed by the ChaMP. In striking testimony to the quality of the quasar selection algorithm (especially once candidates have been matched to X-ray sources) only eight of these (2%) are not broad-line AGNs. Three are narrow emission line galaxies, two are absorption line galaxies (no emission lines of equivalent width W > 5 å), and three are known BL Lac objects. We exclude these objects from all samples described below. Because
16 The Chandra Source Catalog, available at http://cxc.cfa.harvard.edu/csc, contains all X-ray sources detected by the Chandra X-ray Observatory.

The ratio of the number of X-ray matches to the optical control sample to the number of matches in the actual sample is 0.0067.


648

GREEN ET AL.

Vol. 690

Figure 3. SDSS i magnitude vs. broadband (0.5­8 keV) X-ray flux for the SDSS/ChaMP QSO the Main sample. The flux shown is based on the bestfit PL-free yaxx model, with the effect of Galactic absorption removed from both X-ray and optical. X-ray detections are marked by black dots and flux upper limits by green arrows. The open red triangles show QSOs with existing spectroscopic redshifts, clearly biased toward brighter optical mags. Radioloud quasars (large open blue circles), BAL QSOs (open black squares), NAL QSOs (open black diamonds) and NLS1s (open black triangles), and Chandra PI targets (large asterisks) are also indicated as shown in the legend. (A color version of this figure is available in the online journal.)

the magnitude distribution (and therefore the photometric color errors) for objects lacking spectroscopic classifications is fainter (see Figure 3), we expect that the overall fraction of misclassified photo-z QSOs is larger than 2%, but it would be difficult to estimate it without deeper spectroscopic samples. A plot of the photometric versus spectroscopic redshifts of these quasars is shown in Figure 4. 3. SAMPLES, SUBSAMPLES, AND DETECTION FRACTIONS The definition of our Main sample and a variety of subsamples are described in subsequent sections and summarized in Table 1. The size of our Main sample allows us to investigate the effects of luminosity or redshift limits, X-ray nondetections, PI target bias, strong radio-related emission (RL QSOs), broad- and narrow-line absorption (BAL and NAL, respectively, QSOs), and Narrow-Line Seyfert 1s (NLS1s). In Table 1, samples to which we refer frequently are arrayed under "Primary Samples" in decreasing order of the number of detections. Samples mentioned only once or twice in this paper are listed in similar order under "Other Samples." Tables listing bivariate statistical results (Tables 5­7) later in the paper list samples in this same order for reference. We define the Main sample to be the 2308 SDSS QSOs that fall on Chandra ACIS chips in a region of effective exposure greater than 1200 s (excluding CCD 8; see below), regardless of Chandra X-ray detection status. Our cleaned matched sample (the MainDet sample) of X-ray/optical matched QSOs contains 1135 distinct X-ray sources with high optical counterpart match confidence, where we have removed all sources (1) with significant contamination by nearby bright optical sources, (2) with significant overlap with other X-ray sources, (3) with detected

on ACIS-S chip S4 (CCD 8), because of its high background and streaking, (4) with dithering across chips (which renders unreliable the yaxx X-ray spectral fitting described below), or (5) with spectroscopy indicating that the object is not a Type 1 QSO. The mean exposure time for the MainDet sample is 25.9 ks per QSO, with an average of 3.6 QSOs detected on each Chandra field.17 For the 1173 nondetections in the Main sample, the mean exposure time is 17.6 ks. A histogram of exposure times for all QSOs in the Main sample is shown in Figure 5. We publish key data for 1135 QSOs in the MainDet sample in Table 2, marking 82 sources that are the intended Chandra principal investigator (PI) targets. X-ray sources in plots include only the MainDet sample or subsets of it. Figure 6 shows luminosity versus redshift for the MainDet sample. A large fraction of the z > 4 objects are Chandra targets (large black stars). Strong redshift­luminosity trends are seen both in optical and X-ray, as is expected from any flux-limited survey. However, the factor 30­50 range in luminosity is unusual for a single sample; such breadth is usually only achieved using sample compilations encompassing diverse selection techniques. In Figure 6, the large number of objects in our sample makes it difficult to distinguish the point-types presenting object class information, so Figure 7 shows a zoom-in on the most densely populated regions of the L­z plane. Since we start with Type 1 SDSS QSOs, we are studying an optically selected sample, and the selection function is complex (Richards et al. 2006a). If we limit the analysis to detections only, then the sample is both optically and X-ray selected, and the selection function becomes increasingly complex. If instead we include all X-ray upper limits in the analysis, the sample remains fundamentally optically selected, but then statistical analyses must incorporate the nondetections (see Section 6.1). The ChaMP's xskycover pipeline allows us to investigate the detection fraction for the full SDSS QSO sample, shown in Figure 8. Of 2308 SDSS QSOs that fall on an ACIS chip (the Main sample) in our 323 ChaMP fields, 1135 (49%) are detected in the MainDet sample. Detection fractions as a function of SDSS QSO mag and redshift for this sample are shown in Figure 9. To minimize sample biases, we can also examine detection fractions as a function of X-ray observing parameters like exposure time and OAA. To simultaneously optimize detected sample size and detection fraction, we simply 2 maximize Ndet Nlim , where Ndet and Nlim are the number of X-ray detections and nondetections (flux upper limits), respectively. We find that an X-ray-unbiased subsample with a significantly higher detection rate is achieved by limiting consideration to the 1269 QSOs with OAA < 12 and exposure time T > 4 ks, of which 922 (72%) are detections. This high detection fraction sample is called the D2L sample (Table 1). Data for X-ray nondetections is available in Table 3. We include in Table 3 only the 347 limits in the D2L sample, where the flux limits are sensitive enough to be interesting. By "interesting," we mean that the limits are close to or brighter than the faint envelope of detections. A flux limit several times brighter than that envelope provides no statistical constraints whatsoever on the derived distributions or regressions. For X-ray nondetections, data are more sparse all around for several reasons. QSOs with limits are optically fainter (mean and median i = 20.4 mag for the D2L sample limits compared to i = 19.9
17

For14of323 Chandra fields, no QSOs are detected.


No. 1, 2009

PROPERTIES OF SDSS QUASARS IN THE ChaMP

649

Figure 4. Left: photometric redshift vs. spectroscopic redshift for QSOs detected in ChaMP fields. Filled (red) circles show QSOs for which the formal photometric redshift probability is greater than 95%. A fraction of objects have large errors in their photometric redshifts. About 18% of QSOs with zspec < 1 have zphot > 2. This drops to 13% using only Prob(zphot ) > 0.5. Right: difference between the photometric and true (spectroscopic) redshift for QSOs in our sample, plotted against photometric redshift probability, again illustrates the reliability of these probabilities. (A color version of this figure is available in the online journal.)

Figure 5. Left: histogram of effective Chandra exposure times for QSOs in the Main sample, in units of ks, with bin size 6 ks for detections (solid black line) and limits (dashed blue line). For detections, the mean and median exposure times are 25.9 and 17.6 ks, respectively. The inset shows detail at low exposure times using bin size 1 ks. Right: histogram of Chandra OAAs for the Main sample, in units of arcmin. The black solid histogram shows how detections trend toward small OAA. Mean and median for detections are 6.4 and 5.8, respectively. Blue histogram shows that limits trend toward large OAA. The dashed histograms show the corresponding cumulative fractions. About 90% of the detections (compared to 50% of the limits) are at OAA < 12 off-axis. (A color version of this figure is available in the online journal.)

for detections). Being fainter, fewer have SDSS spectroscopy. Also, as nondetections, none have been targeted for spectra by the ChaMP, so globally only about 10% of nondetected QSOs have optical spectra. The fraction of radio detections is also smaller (1.2% versus 4.8%). None are Chandra PI targets. Finally, X-ray nondetections lacking optical spectroscopy are somewhat less likely to be QSOs. The selection efficiency (fraction of QSO candidates that are actual QSOs) between about 0.8 < z < 2.4 is 95% (Richards et al. 2004; Myers et al. 2006), but Richards et al. (2008) estimate that near the faint

limit of i 20.4 mag, the overall QSO selection efficiency is 80%. Particular attention must be paid to possible galaxy contamination at the faint end as the autoclustering estimates of the efficiency do not include galaxy interlopers at faint limits where SDSS star­galaxy separation begins to break down. However, many of these "spurious" cross-matches may turn out to be (e.g., low-luminosity) AGNs. In any case, the increased level of contamination by non-QSOs is another rationale for limiting the number of nondetections to those with sensitive X-ray limits. The nature of the statistical analysis (as discussed


650

GREEN ET AL.
Table 1 Quasar Sample Definitions Sample Main MainDet noTDet D2L D2LNoRB hiLo HiCtNoTRB NoRB NoRBDet D2LNoTRB hiLoLx zLxBox LoBox zBox zBoxDet D2LSy1 HiCt HiCtNoTRB Limsa y ... ... y y y ... y ... y ... ... ... y ... y ... ... Targets ... . .. n ... ... ... n ... . .. n . .. . .. . .. ... . .. ... . .. n RL ... ... ... ... n n n n n n n ... ... ... ... n ... n Absb Tmin OAA ... ... ... 12 12 ... ... ... ... 12 ... ... ... ... ... ... ... ... N
det

Vol. 690

N

lim

N

total

% Det 49 100 100 72 71 46 100 47 100 71 100 100 100 53 100 68 100 100

Primary Samples ... ... ... ... ... ... ... 4 n 4 n ... ... ... Other Samples n ... n ... n 4 n ... ... ... ... ... ... ... ... ... n ... ... ... n ...

1135 1135 1053 922 866 847 129 1054 1054 828 801 817 530 420 420 176 157 129

1173 0 0 347 338 961 0 1144 0 338 0 0 0 360 0 84 0 0

2308 1135 1053 1269 1204 1808 129 2198 1054 1166 801 817 530 780 420 260 157 129

Notes. a If "y," sample includes X-ray nondetections. b If "n," sample excludes QSOs with evident BALs or NALs and also NLS1s.

in Section 6.1) is such that nondetections are included, but are assumed to follow the distribution of detections, and so effectively have a lower weight in the results. Amongst the undetected QSOs in the Main sample, 165 have SDSS spectroscopy, of which 144 are high confidence spectroscopic QSOs. The 21 non-QSOs comprise 16 stars and five galaxies. The higher (13%) rate of nonstar spectroscopic classifications amongst undetected QSOs is not surprising, since X-ray detection greatly increases the probability that an optical AGN candidate is indeed an AGN. From the upper limit QSO sample, we remove the 22 non-QSOs, and use the SDSS spectroscopic redshifts instead of the photometric redshifts wherever applicable. 3.1. Targets The "Nontarget" detected sample (the noTDet sample) of 1053 QSOs further eliminates 82 objects (7.2% of the MainDet sample) that are the intended targets of the Chandra observation wherein they are found. Targets are on average brighter than most of the QSO sample (see Figure 3), but more importantly were chosen for observation for a variety of reasons unrelated to this study. In particular, targets tend to be more luminous than serendipitous QSOs (Figures 6 and 7), and several are known lenses (e.g., HS 0818+1227, PG 1115+080, UM 425 = QSO 1120+019). The bias in sample characteristics is largely mitigated in the subsamples excluding targets (see Table 1). However, the exclusion of targets also produces a (much smaller) bias because some objects with similar characteristics would have been included (at a lower rate) were the Chandra pointings all truly random. Because many of the targets are indeed of interest (e.g., high-z QSOs), we include them in most discussions, but always check that results are consistent without them. We also note that some target bias probably affects the X-ray sample even after the exclusion of PI target QSOs, because PI targets may cluster with other categories of X-ray sources such as other AGN, galaxies, or clusters. Overall,

a comparison of regression results18 for several of our subsamples that differ only in target exclusion does not indicate a significant target bias, due at least in part to our large sample sizes. 3.2. Radio Loudness Quasars with strong radio emission are observed to be more X-ray luminous (e.g., Green et al. 1995; Shen et al. 2006). At least some of the additional X-ray luminosity is likely to originate in physical processes related to the radio jet rather than to the accretion disk, so it may be important to recognize those objects that are particularly radio loud. The Faint Images of the Radio Sky at Twenty Centimeters (FIRST) survey (Becker et al. 1995) from the NRAO Very Large Array (VLA) has a typical (5 ) sensitivity of 1 mJy, and covers ´ most of the SDSS footprint on the sky. Following Ivezic et al. (2002), we adopt a positional matching radius of 1. 5, which should result in about 85% completeness for core-dominated sources, with a contamination of 3%. We thereby match 69 sources to the Matched sample. Jiang et al. (2007) matched the FIRST to SDSS spectroscopic quasars and found that about 6% matched within 5 . We might expect a lower matched fraction because our optical photometric sample extends 1­2 mag fainter. On the other hand, we are looking at X-ray-detected quasars, so the actual matched fraction of 5% is similar. We also matched all the quasar optical positions to the FIRST within 30 , and visually examined all the FIRST images to look for multiple matches and/or lobe-dominated quasars. There are 26 sources that we judged to have reliable morphological complexity that are resolved by FIRST into multiple sources. The NRAO VLA Sky Survey (NVSS; Condon et al. 1998), lists detections for 19 of these. Comparing NVSS with summed FIRST fluxes, we found NVSS fluxes slightly larger: less 2% difference in the mean (20% max). Since the NVSS beam is larger (45 ) than
18 Table 6 and 7 shows similar results comparing, e.g., the MainDet sample and the noTDet sample, or the D2LNoRB sample and the D2LNoTRB sample.


No. 1, 2009

Table 2 Properties of SDSS Quasars Detected by Chandra SDSS Obj ID (1) 587731187277889693 587731187277955083 588015510343385196 587731186204606566 587731186204606704 588015508733231171 588015508733231262 588015508733231265 587731186204737774 587730773889974538 587731186742198291 588015509270822924 588015509270823186 588290881639350481 588290881639350569 587730775501504810 588290881639350397 588015507661324390 588015509809266720 588015509809659937 587727227305066749 587727227305197873 587731185135648990 587731185135648996 588015508201144501 588015508201144513 587731186209783863 588015509275803698 588015509275869378 R.A. (J2000) Decl. (2) (3) 0.50832 0.62797 0.64800 1.59420 1.64276 1.72545 1.72868 1.74704 1.82682 2.81349 3.19633 3.27563 3.30851 5.07661 5.08519 5.08732 5.10509 5.82850 6.96833 7.88101 10.07393 10.30287 12.47624 12.48751 12.62757 12.65482 13.47981 14.77296 14.84438 0.761275 0.833065 0.889224 -0.073452 -0.085489 -0.259281 -0.230998 -0.294690 -0.088692 14.767168 0.210979 0.075532 0.053632 15.715105 15.735262 15.914392 15.681860 -1.050280 0.437687 0.572282 -9.190477 -9.238581 -0.939257 -0.968442 -0.780014 -0.808028 -0.052600 0.114358 0.050395 i (4) 19.043 17.955 20.249 19.575 20.770 17.878 20.345 19.428 20.734 18.276 18.970 18.463 20.625 20.375 20.942 21.076 17.173 19.666 17.733 18.474 20.752 20.770 20.389 20.758 20.845 20.411 17.984 17.489 19.194 zphot (5) 1.395 1.275 1.895 1.285 1.605 1.675 2.485 1.975 1.205 4.665 2.145 0.815 2.055 0.885 2.175 0.175 1.985 1.215 0.145 1.875 0.145 1.295 2.145 0.385 1.265 1.855 0.485 0.745 4.385 Pz (6) 0.962 0.973 0.841 0.993 0.450 0.919 0.552 0.857 0.811 0.976 0.703 0.510 0.539 0.538 0.608 0.958 0.921 0.990 0.933 0.868 0.649 0.635 0.702 0.551 0.961 0.883 0.778 0.842 0.998 zlo (7) 0.980 1.030 1.410 1.010 1.440 1.440 2.110 1.600 0.900 4.490 1.880 0.670 1.430 0.640 1.430 0.060 1.440 0.980 0.140 1.620 0.100 1.000 1.440 0.250 0.900 1.620 0.390 0.660 4.200 zhi (8) 1.550 1.480 2.140 1.440 2.000 2.080 2.670 2.180 1.520 5.070 2.240 0.970 2.300 1.200 2.360 0.240 2.160 1.450 0.240 2.040 0.250 1.570 2.260 0.500 1.470 2.110 0.700 0.950 4.560 zbest (9) 1.3950 1.3527 1.895 1.0370 1.6050 1.7195 2.4850 1.9750 1.2050 4.9672 2.1528 2.1453 2.0550 0.8850 2.1750 0.1750 2.0091 1.2150 0.2053 1.7354 0.1450 1.2950 2.1450 0.3850 1.2650 1.8550 1.7189 0.7189 4.1544 Spec Ref (10) S R R CXOMP (11) J000202.0+004541 J000230.7+004959 J000235.5+005321 J000622.6-000424 J000634.3-000510 J000654.1-001533 J000654.9-001351 J000659.2-001740 J000718.5-000522 J001115.2+144601 J001247.0+001241 J001306.1+000431 J001314.0+000313 J002018.3+154254 J002020.4+154406 J002020.7+155451 J002025.2+154054 J002318.8-010301 J002752.4+002615 J003131.4+003420 J004017.7-091125 J004112.6-091417 J004954.3-005620 J004957.0-005806 J005030.6-004649 J005037.2-004829 J005355.1-000309 J005905.4+000651 J005922.6+000301 srcid (12) XS04861B2_001 XS04861B7_001 XS04861B7_008 XS04096B5_002 XS04096B5_001 XS04096B7_001 XS04096B7_003 XS04096B7_002 XS04096B2_002 XS03957B7_001 XS04829B6_007 XS04829B7_005 XS04829B7_001 XS01595B7_004 XS01595B6_001 XS01595B5_004 XS01595B7_001 XS04079B7_001 XS04080B7_003 XS02101B7_002 XS04888B3_010 XS04888B1_010 XS04825B7_018 XS04825B7_012 XS04825B2_003 XS04825B2_001 XS04830B7_001 XS02179B6_001 XS02179B7_003 OAA (13) 8.2 0.6 4.1 13.1 10.9 0.6 1.2 3.0 11.5 0.6 8.9 0.6 2.8 1.6 2.1 12.7 1.3 1.7 4.5 1.2 7.6 11.0 4.7 5.8 8.5 8.8 0.6 5.5 0.6 cts (14) 4 1. 2 132.7 7. 5 217.3 4 2. 8 4 1. 7 8. 6 2 2. 8 1 3. 6 129.8 1 0. 9 1 5. 6 4 0. 6 4. 5 1 5. 6 4 4. 9 494.3 2 3. 9 3 9. 2 2 8. 8 1 9. 8 1 6. 1 1 0. 8 2 6. 9 2 1. 9 2 3. 8 2 7. 7 324.7 1 0. 8 cts_err (15) 7. 6 12.7 3. 8 16.6 8. 9 7. 5 4. 1 5. 9 5. 3 12.5 4. 7 5. 1 7. 5 3. 4 5. 1 10.9 23.4 6. 0 7. 6 6. 5 5. 8 6. 5 4. 4 6. 6 6. 2 6. 2 6. 4 19.6 4. 4 Exposure (16) 5. 6 5. 0 5. 0 4. 2 4. 2 4. 2 4. 2 4. 2 3. 6 3. 4 6. 6 6. 6 6. 6 1 3. 3 1 6. 6 1 2. 7 1 3. 3 1. 6 1. 5 3. 8 8. 8 8. 6 1 2. 9 1 2. 9 1 2. 3 1 2. 3 7. 0 2. 2 2. 6
Gal NH (17)

S S R

S S S

S S S

2. 5 2. 5 2. 5 3. 0 3. 0 3. 0 3. 0 3. 0 3. 0 9. 6 5. 6 5. 6 5. 6 4. 2 4. 2 4. 2 4. 2 2 1. 2 4 4. 9 2. 4 3. 4 3. 4 2. 1 2. 1 2. 1 2. 1 1. 9 3. 0 3. 0

PROPERTIES OF SDSS QUASARS IN THE ChaMP 651


652

Table 2 (Continued.) SDSS Obj ID (1) 587731187277889693 587731187277955083 588015510343385196 587731186204606566 587731186204606704 588015508733231171 588015508733231262 588015508733231265 587731186204737774 587730773889974538 587731186742198291 588015509270822924 588015509270823186 588290881639350481 588290881639350569 587730775501504810 588290881639350397 588015507661324390 588015509809266720 588015509809659937 587727227305066749 587727227305197873 587731185135648990 587731185135648996 588015508201144501 588015508201144513 587731186209783863 588015509275803698 588015509275869378 Nintr H (18)
hi NH (19) lo NH (20)

(21) 1.53 2.40 2.14 1.74 1.31 2.05 3.50 2.27 1.82 1.88 2.13 2.10 1.47 2.05 1.72 1.69 1.77 1.94 1.28 2.23 1.91 1.16 0.95 2.30 2.79 1.77 0.90 1.61 1.32

hi (22) 0.45 0.42 1.25 0.30 0.49 0.46 1.87 0.68 0.91 0.37 1.02 0.87 0.43 1.48 0.73 0.70 0.16 0.60 0.50 0.58 0.73 0.79 0.89 0.65 0.90 0.71 0.48 0.26 0.84

lo (23) -0.44 -0.38 -1.03 -0.24 -0.46 -0.44 -1.26 -0.62 -0.79 -0.35 -0.89 -0.74 -0.41 -1.24 -0.67 -0.63 -0.12 -0.57 -0.46 -0.54 -0.67 -0.73 -0.86 -0.59 -0.80 -0.62 -0.48 -0.16 -0.77

log fx (24) -12.960 -12.850 -13.990 -12.321 -12.870 -13.180 -14.000 -13.470 -13.230 -12.580 -13.730 -13.820 -13.220 -14.640 -14.010 -13.450 -12.609 -13.000 -12.540 -13.360 -13.540 -13.410 -13.950 -13.810 -13.690 -13.580 -13.300 -11.751 -13.410

log l2ke (25)

V

log l2500 (26) 30.590 30.978 30.391 30.020 29.899 31.173 30.650 30.758 29.765 32.063 31.020 31.220 30.316 29.639 30.241 27.578 31.676 30.231 29.033 30.939 27.718 29.783 30.449 28.568 29.830 30.307 31.118 30.690 31.542

å

ox (27) 1.383 1.502 1.576 1.041 1.025 1.605 1.569 1.500 1.230 1.290 1.666 1.778 1.218 1.850 1.470 1.238 1.515 1.317 1.388 1.581 1.395 1.276 1.532 1.456 1.412 1.395 1.630 1.230 1.480

f20 cm (28)

Ext (29) 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0

R (20) 1.06 0.62 1.54 4.86 1.75 0.59 1.58 1.21 1.73 0.75 1.03 0.82 1.69 1.59 1.82 1.87 0.31 1.31 0.53 0.83 1.74 1.75 1.60 1.74 1.78 1.60 0.63 3.84 1.12

Class (31) 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 3 0 0 0 0 0 0 0 2 0 0

Targ (32) 0 1 0 0 0 1 0 0 0 1 0 1 0 0 0 0 1 0 1 0 0 0 0 0 0 0 1 0 1

Comments (33)

0.00

0.27

0.00

0.00

0.47

0.00

0.00

0.47

0.00

26.988 27.065 26.286 27.308 27.228 26.992 26.563 26.850 26.561 28.701 26.681 26.587 27.142 24.819 26.412 24.354 27.729 26.799 25.418 26.822 24.084 26.458 26.457 24.775 26.153 26.673 26.871 27.486 27.688

3897.60

NED: FBQS J0006-0004 NED: LBQS 0004-0032

HiBAL NED: LBQS 0010-0012

GREEN ET AL.

NLSy1

Many strong NALs

2508.80

Notes. (1) SDSS Object ID, (2) SDSS R.A. (J2000), (3) SDSS decl. (J2000), (4) SDSS asinh mag_psf i, dereddened, (5) photometric redshift (see, Weinstein et al. 2004), (6) photometric redshift range probability, (7) lower limit of photometric redshift range, (8) upper limit of photometric redshift range, (9) best redshift: spectroscopic if different than zphot , (10) reference for spectroscopic redshift--S: SDSS, O: ChaMP, R: published reference from NED, (11) ChaMP IAU source name, (12) ChaMP internal source ID, format XSoooooBc_nnn where ooooo is Chandra obsid, c is ACIS CCD ID, and nnn is source ID on that CCD, (13) Chandra OAA in arcmin, (14) net 0.3­8 keV source counts, (15) rms uncertainty on net counts, (16) vignetting-corrected exposure time in ks, (17) galactic column in units 1020 cm-2 , (18) best-fit X-ray intrinsic column in 1022 cm-2 , only included four counts > 200, (19) 90% upper limit on intrinsic column in 1022 cm-2 , (20) 90% lower limit on intrinsic column in 1022 cm-2 , (21) best-fit X-ray power-law index , (22) 90% upper limit on , (23) 90% lower limit on , (24) log X-ray flux (0.5­8 keV) in erg cm-2 s-1 , (25) log X-ray luminosity at 2 keV in erg s-1 Hz-1 , (26) log optical/UV luminosity at 2500 å in erg s-1 Hz-1 , (27) ox , the optical/UV to X-ray spectral index, (28) 20 cm radio flux in mJy from FIRST or NVSS, (29) radio extent flag, (30) radio loudness R, (31) spectral class: 1, BAL; 2, NAL; 3, NLS1, (32) 1, intended Chandra PI target, (33) comments. (This table is available in its entirety in a machine-readable form in the online journal. A portion is shown here for guidance regarding its form and content.)

Vol. 690


No. 1, 2009

PROPERTIES OF SDSS QUASARS IN THE ChaMP

653

Figure 6. Luminosity at 2500 å (left) and 2 keV (right) vs. redshift for the SDSS/ChaMP in the Main sample. Above z 2.5, the number of QSOs declines steeply, due to the SDSS magnitude limit and due to the decreased efficiency of the photometric selection algorithm as it crosses the stellar color locus (see Figure 8 and Richards et al. 2002). These plots show 56 QSOs with z > 3, of which 34 are new serendipitous detections. X-ray upper limits are shown as small green triangles here. See Figure 3 for symbol types. The dashed black rectangle surrounds the zBoxDet sample, a portion of the l2500 å -z plane chosen to test for redshift dependence. The green rectangle "LoptBox" surrounds the LoBox sample, used to test for dependence on l2500 å . The blue rectangle surrounds the zLxBox sample, chosen to avoid X-ray flux limit bias. (A color version of this figure is available in the online journal.)

Figure 7. Left: zoom-in of 2500 å luminosity vs. redshift. Objects with spectroscopic redshifts (open red triangles) tend to be at high optical luminosities by selection. Detectably radio-loud QSOs are shown with open blue circles. See Figure 3 for symbol types. BAL QSOs (open black squares) are mostly detected at z > 1.6 where the CIV region enters the optical bandpass. There appears to be no preference of BAL QSOs for high optical luminosity, apart from the bias caused by Chandra target selection. Right: zoom-in of 2 keV luminosity vs. redshift. Here, the RL QSOs clearly populate the upper luminosity envelope. BAL QSOs are preferentially X-ray quiet, unlike the QSOs with NALs only (open diamonds). (A color version of this figure is available in the online journal.)

FIRST (5 ), FIRST detection algorithms may exclude some of the extended source flux as background, so we include the NVSS fluxes for these 19 objects, and summed FIRST fluxes for those remaining. ´ Following Ivezic et al. (2002), we adopt a radio-loudness parameter R as the logarithm of the ratio of the radio to optical monochromatic flux: R = log(F20 cm /Fi ) = 0.4(i - m20 cm ), where m20 cm is the radio AB magnitude (Oke & Gunn 1983), m20 cm = -2.5log(F20 cm /3631 Jy) calculated from the integrated radio flux density, and i is the SDSS i-band magnitude, corrected for Galactic extinction. We adopt a radio-loudness threshold R = 1.6. Thus there are 72 QSOs in the Main sample with radio detections, of which 57 (79%) are radio loud. For the MainDet sample (detections only), there are 55 radio-detected QSOs, of which 43 (78%) are radio-loud. Figure 10 shows radio loudness versus redshift for the MainDet sample. Many of the radio upper limits are near our adopted radio loudness threshold.

Given the ( 1 mJy) source detection limit of the FIRST Survey, all RL QSOs will be detected to about i 20.4 mag. For the magnitude range 17 < i < 20 where the statistics are good and the FIRST is sensitive to all RL QSOs, we find 41 of 529 (8 ± 1%) such QSOs from the MainDet sample are detected by the FIRST, with 29 (5.4%) that are radio loud. Since the 34% of our full the MainDet sample that is fainter than i = 20.4suffers from incomplete radio-loudness measurements, some 2% may be unidentified RL QSOs. A similar fraction pertains if we count X-ray nondetections as well (the Main sample). 3.3. Broad and Narrow Absorption Line Quasars We identified QSOs with BALs and NALs directly by visual inspection of QSOs with spectroscopy, finding 16 BAL and 11 NAL QSOs in the MainDet sample. Ten (two) of the BAL (NAL) QSOs were the Chandra PI targets.


654

GREEN ET AL.

Vol. 690

Figure 8. Left: histogram of redshifts for detected (solid blue), nondetected (red dashed), and all (black solid) QSOs in the redshift for each object (i.e., spectroscopic redshifts are always used when available). Above z 2.5, the number of QSOs efficiency of the photometric selection algorithm as it crosses the stellar color locus (Richards et al. 2002). The inset shows redshift. Right: histogram of SDSS i mag for detected (solid blue), nondetected (red dashed), and all (black solid) QSOs. The function of magnitude. (A color version of this figure is available in the online journal.)

MainDet sample. We tally the "best" declines steeply, due to the decreased the detected fraction as a function of inset shows the detected fraction as a

Figure 9. Left: histogram of redshifts for detected (solid blue), nondetected (red dashed), and all (black solid) QSOs, after restriction to obsids with T > 4 ks and QSOs with OAA less than 12 (the D2L sample). The inset shows the detected fraction as a function of redshift. Right: histogram of SDSS i mag for detected (solid blue), nondetected (red dashed), and all (black solid) QSOs. The inset shows the detected fraction as a function of magnitude. (A color version of this figure is available in the online journal.)

The best estimates to date of the raw BAL QSO fraction among optically selected quasars range from about 13­20% (Reichard et al. 2003; Hewett & Foltz 2003). From the SDSS DR3 sample of Trump et al. (2006), Knigge et al. (2008) carefully define what is a BAL QSO and correct for a variety of selection effects to derive an estimate of the intrinsic BAL QSO fraction of 17% ± 3. The vast majority of BAL QSOs in the SDSS are above redshift 1.6 because only then does the CIV absorption enter the spectroscopic bandpass.19 If we determine our BAL QSO fraction in the MainDet sample by only counting the serendipitous (nontarget) QSOs with
19

z > 1.6 and spectroscopic redshifts, we find just 4 out of 119 QSOs with BALs. Even with sensitive X-ray observations such as these, X-ray selection is strongly biased against the highly ionized absorbing columns along the line of sight toward the X-ray emitting regions of BAL QSOs. Of the 24 absorbed (BAL or NAL) QSOs, two are detectably radio loud; SDSS J171419.24+611944.5--a BAL QSO--and SDSS J171535.96+632336.0--a NAL QSO--are targets selected (by Chandra PI Richards) as reddened QSOs. 3.4. Narrow-Line Seyfert 1s X-rays from NLS1s are of particular interest because they were thought to show marked variability and strong soft

A much smaller number of the rare low-ionization BAL QSOs (with BALs just blueward of Mg ii) are found at lower redshifts.


No. 1, 2009

PROPERTIES OF SDSS QUASARS IN THE ChaMP 4. OPTICAL COLORS AND REDDENING

655

Figure 10. Radio-loudness vs. redshift for the MainDet sample (detections). Radio-loud objects R > 1.6 are shown with open blue circles. Radio-quiet but FIRST radio-detected objects are shown as filled green circles. All other symbols (described in Figure 3) have radio flux upper limits only. Note that most Chandra targets are distinctly either loud or quiet, highlighting a bias in the target subsamples. (A color version of this figure is available in the online journal.)

In Figure 11, we plot the (g - i) colors of the matched SDSS/ChaMP QSO sample as a function of redshift, and compare to the optical-only sample.20 The Chandra-detected sample does not show significantly different colors from the full optical sample. This likely attests to (1) the sensitivity of the Chandra imaging relative to the magnitude limit of the optical sample and (2) the fact that Type 1 QSOs are largely unabsorbed in both the optical and X-ray regimes. The right panel of Figure 11 shows that 10 of 14 BAL QSOs are above the (g - i ) = 0 line. This reflects that SDSS BAL QSOs tend to be redder than average (Reichard et al. 2003;Dai et al. 2007). Most of the RL QSOs are also redder than average. Richards et al. (2001) found a higher fraction of intrinsically ´ reddened quasars among those with FIRST detections. Ivezic et al. (2002) found that RL QSOs are redder than the mean (at any given redshift) in (g - i) by 0.09 ± 0.02 mag. Figure 11 confirms a similar trend in the X-ray detected SDSS/ ChaMP sample. At the same time, a small number of RL QSOs are found on the blue extreme of the color-excess distribution. These trends are fully consistent with the detailed results found by stacking FIRST images of SDSS quasars (White et al. 2007), independent of X-ray properties. 5. X-RAY SPECTRAL FITTING WITH yaxx Besides comparing the broadband multiwavelength propertiesofQSOs, Chandra imaging provides X-ray spectral resolution capable of yielding significant constraints on the properties of emission arising nearest the SMBH. While the ChaMP calculates hardness ratios (HRs) and appropriate errors for every source, these can be difficult to interpret, since HR convolves the intrinsic quasar SED with telescope and instrument response Gal and does not take redshift or Galactic column NH into account. A direct spectral fit of the counts distribution using the full inGal strument calibration, known redshift, and NH provides a much more direct measurement of quasar properties. Note that even in the low-count regime, one can obtain robust estimates of fit parameter uncertainties using the Cash (1979) fit statistic. We use an automated procedure to extract the spectrum and fit up to three models to the data. For all objects in the Matched sample, we first define a circular source region centered on the X-ray source which contains 95% of 1.5 keV photons at the given OAA. An annular background region is also centered on the source with a width of 20 . We exclude any nearby sources from both the source and background regions. We then use CIAO21 tool psextract to create a PHA (pulse height amplitude) spectrum covering the energy range 0.4­8 keV. Spectral fitting is done using the CIAO Sherpa22 tool in an automated script known as yaxx23 (Aldcroft 2006). All of the spectral models contain an appropriate Galactic neutral absorber. For all sources we first fit two power-law models which include a Galactic absorption component frozen at the 21 cm value24 : (1) fitting photon index , with no intrinsic absorption
The optical-only sample refers to all SDSS QSOs within 20 of all Chandra pointings, regardless of whether its position falls on an ACIS CCD. We use a 9th-order polynomial fit to the optical-only sample with the following coefficients: 0.698892, 3.011733, -20.358267, 37.850353, -33.617121, 16.652032, -4.861889, 0.833027, -0.077526, and 0.003026. 21 http://cxc.harvard.edu/ciao 22 http://cxc.harvard.edu/sherpa 23 http://cxc.harvard.edu/contrib/yaxx 24 Neutral Galactic column density N Gal taken from Dickey & Lockman H (1990)for the Chandra aim-point position on the sky.
20

X-ray excesses (e.g., Green et al. 1993; Boller et al. 1996). NLS1s are proposed to be at one extreme of the so-called (Boroson & Green 1992) "eigenvector 1," which has been suggested to correspond to low SMBH masses (Grupe et al. 2004) and/or high (near-Eddington) accretion rates (Boroson 2002). For objects with SDSS spectra encompassing H (z < 0.9), we identify as NLS1s (Osterbrock & Pogge 1985) those objects with FWHM(H ) < 2000 km s-1 and line flux ratio [O iii]/H< 3. The FWHM measurements are obtained via FWHM = 2.35 c /0 /(1 + z), where 2 is the variance of the Gaussian curve that fits the H emission line, z is the redshift, and 0 is the rest-frame wavelength of the H line (4863 å). We extract measurements of (in å) and line fluxes from the SDSS DR6 SpecLine table. Our MainDet sample contains at least 19 NLS1s (those with spectroscopy in the redshift range to include H ). Among the nondetections, we identify three more. Six of these 22 objects are targets described in Williams et al. (2004), and SDSS J125140.33+000210.8 was a target selected (by Chandra PI Richards) as a dust-reddened QSO. The Williams et al. (2004) sample consisted of 17 SDSS NLS1s selected from the ROSAT All Sky Survey to be X-ray weak. Their study confirmed earlier suggestions that strong, ultrasoft X-ray emission is not a universal characteristic of NLS1s. We therefore present here new results for a sample of 15 SDSS NLS1s observed by Chandra. This admittedly small sample is nevertheless the largest published sample of optically selected NLS1s with unbiased X-ray observations. The sample of Williams et al. (2004) was selected to be X-ray weak, while the Grupe et al. (2004) ROSAT sample was selected to have strong soft X-ray emission. We find no evidence for unusual SEDs from the distributions of either ox or .


656

Table 3 Properties of SDSS Quasars with Chandra Limits
SDSS Obj ID (1) 588015510343385295 588015510343385302 588015509270823376 588015509270888717 588015509270888725 587730775501504550 587727180601295054 587727180601557048 587727227305394556 587727180601622916 587731185135649019 588015508201144424 588015509275279576 587731186209849918 587731511532454189 587731511532454208 588015507667419343 587727884161581293 587727883893211318 587727178999398623 587731512611897674 587731513691013244 587731512083349864 587728906098377011 587728906098376844 588007005767532860 588007005767532880 587731680110052241 587725470127095818 R.A. (J2000) Decl. (2) (3) 0.56337 0.909429 0.63485 0.870704 3.32406 0.057045 3.40998 0.141416 3.45158 0.098009 5.00208 15.852781 10.04659 -9.111831 10.60443 -8.983756 10.73418 -9.141124 10.82587 -9.032790 12.55619 -0.978559 12.68671 -0.838214 13.48778 0.041851 13.50874 -0.132789 19.71203 -0.963115 19.75317 -0.964191 19.83900 -1.181540 29.96522 -8.803312 30.05424 -8.839227 30.28336 -9.408542 32.80721 -0.133716 45.00239 0.807781 51.75811 -0.573625 115.26597 31.160109 115.31801 31.184168 116.33380 39.501996 116.44541 39.466243 116.73920 27.665360 118.81215 41.058484 i (4) 20.956 20.819 20.984 20.457 20.804 19.190 20.220 19.739 20.920 20.550 20.734 20.077 21.014 20.138 20.433 20.711 19.716 20.815 20.380 20.874 20.592 16.531 20.969 18.733 20.169 20.363 20.179 20.778 19.824 zphot (5) 4.6050 2.2250 0.0650 1.4450 1.4950 1.6850 2.0950 1.8750 0.1450 2.8550 0.3950 0.5850 2.3550 4.6750 2.3950 0.1250 1.4350 0.1250 1.8250 2.4850 1.4450 4.3850 1.1150 3.8450 4.8350 0.8250 2.0250 4.6050 1.6150 Pz (6) 0.949 0.415 0.616 0.615 0.774 0.915 0.653 0.916 0.370 0.866 0.575 0.525 0.485 0.958 0.849 0.503 0.927 0.597 0.479 0.686 0.561 0.990 0.702 0.971 0.983 0.582 0.738 0.984 0.725 zlo (7) 4.440 1.840 0.060 0.960 0.950 1.440 1.290 1.590 0.140 2.520 0.240 0.420 2.080 4.480 2.190 0.120 1.140 0.120 1.440 2.060 0.960 4.190 0.620 3.370 4.510 0.390 1.820 4.440 1.440 zhi (8) 5.22 2.70 0.12 1.61 2.22 2.11 2.26 2.13 0.25 3.26 0.98 0.68 2.73 4.87 2.92 0.25 1.53 0.26 2.01 2.76 1.66 4.51 1.66 4.35 5.52 1.01 2.19 5.19 1.95 zbest (9) 4.6050 2.2250 0.0650 1.4450 1.4950 1.7510 2.0950 1.8750 0.1450 2.8550 0.3950 0.5850 2.3550 4.6750 2.3950 0.1250 1.4350 0.1250 1.8250 2.4850 1.4450 4.3850 1.1150 3.8450 4.8350 0.8250 2.0250 4.6050 1.6150 Spec Ref (10) obsid (11) 4861 4861 4829 4829 4829 1595 4888 4886 4886 4886 4825 4825 4830 4830 4963 4963 4963 6106 6106 3772 2081 4145 5810 0377 0377 6111 6111 3561 3032 ccdid (12) 7 7 7 3 3 5 3 2 1 0 7 3 6 7 7 7 5 3 2 7 7 7 2 7 7 1 1 7 2 OAA (13) 6.4 2.8 3.5 9.0 10.8 10.7 7.5 9.9 2.3 8.1 6.2 10.0 5.1 5.7 4 .1 4.2 11.1 3.2 5.0 3.5 3.2 0.6 8.4 3.4 1.8 5.6 10.7 3.2 11.5 counts < (14) 16.0 10.0 11.1 20.2 22.7 25.3 17.2 22.6 9.4 17.7 15.1 22.7 12.8 14.0 15.0 14.8 43.3 12.0 13.9 11.4 10.2 7.7 18.3 11.7 9.8 14.5 34.7 10.4 25.7 Exposure (15) 4.5 4.6 6.0 4.4 4.6 5.3 7.8 6.9 7.9 7.0 11.4 5.9 5.0 6.2 36.9 36.5 20.3 32.7 29.1 13.2 4.0 4.0 7.8 26.1 25.5 28.7 35.6 4.4 4.1
Gal NH (16) 2.5 2.5 5.6 5.6 5.6 4.2 3.4 3.6 3.6 3.6 2.1 2.1 1.9 1.9 14.1 14.1 14.1 15.9 15.9 24.2 2.7 7.0 6.7 4.2 4.2 8.3 8.3 4.7 8.5

S

log fx < (17) -13.650 -13.865 -13.926 -13.239 -13.211 -13.405 -13.668 -13.498 -13.930 -13.605 -14.075 -13.317 -13.635 -13.843 -14.575 -14.577 -13.737 -14.453 -14.342 -14.278 -13.872 -13.875 -13.624 -14.530 -14.595 -14.298 -14.014 -13.833 -13.144

log l2keV < (18) 27.554 26.581 22.955 26.747 26.811 26.786 26.714 26.766 23.694 27.104 24.536 25.702 26.871 27.376 25.949 22.908 26.241 23.032 25.893 26.284 26.113 27.278 26.083 26.488 26.658 25.086 26.333 27.371 26.960

log l2500 (19) 30.856 30.311 26.290 30.023 29.915 30.742 30.495 30.586 27.029 30.574 28.869 29.374 30.327 31.176 30.534 27.428 30.318 27.465 30.304 30.470 29.957 32.749 29.515 31.669 31.268 29.585 30.480 31.001 30.331

å

ox > (20) 1.268 1.432 1.280 1.258 1.191 1.519 1.452 1.466 1.280 1.332 1.663 1.410 1.327 1.459 1.760 1.735 1.565 1.702 1.693 1.607 1.475 2.100 1.317 1.989 1.770 1.727 1.592 1.394 1.294

f20cm (21)

R (22) 1.82 1.77 1.83 1.62 1.76 1.12 1.53 1.34 1.81 1.66 1.73 1.47 1.85 1.49 1.61 1.72 1.33 1.77 1.59 1.79 1.68 0.05 1.83 0.93 1.51 1.59 1.51 1.75 1.37

Class (23) 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Notes (24)

GREEN ET AL.

Notes. (1) SDSS object ID, (2) SDSS R.A. (J2000), (3) SDSS decl. (J2000), (4) SDSS asinh mag_psf i, dereddened, (5) photometric redshift (see Weinstein et al. 2004), (6) photometric redshift range probability, (7) lower limit of photometric redshift range, (8) upper limit of photometric redshift range, (9) best redshift: spectroscopic if different than zphot , (10) reference for spectroscopic redshift--S: SDSS, O: ChaMP, R: published reference from NED, (11) Chandra observation ID (obsid), (12) ACIS CCD id, (13) Chandra OAA in arcmin, (14) 99% counts upper limit the 0.3­8 keV range, (15) vignetting-corrected exposure time in ks, (16) galactic column in units 1020 cm-2 , (17) log upper limit to the X-ray flux (0.5­8 keV) in erg cm-2 s-1 , (18) log upper limit to the X-ray luminosity at 2 keV in erg s-1 Hz-1 , (19) log optical/UV luminosity at 2500 å in erg s-1 Hz-1 , (20) ox , the optical/UV to X-ray spectral index, (21) 20 cm radio flux in mJy from FIRST, (22) radio loudness, (23) class: 1, BAL; 2, NAL; 3, NLS1, (24) comments. (This table is available in its entirety in a machine-readable form in the online journal. A portion is shown here for guidance regarding its form and content.)

Vol. 690


No. 1, 2009

PROPERTIES OF SDSS QUASARS IN THE ChaMP

657

Figure 11. Left: SDSS (g - i) color vs. best redshift. Symbols show individual QSOs in the SDSS/ChaMP the MainDet sample. The black line shows the mean color at each redshift (in redshift bins of 0.12 for z< 2.5 and 0.25 for higher redshifts) for the full optical QSO sample (regardless of Chandra imaging). We derive a smooth (9th-order) polynomial fit to those means, whose value is plotted as the "expected" (g - i) with a green asterisk (at the redshift of each actual QSO). Right: the color excess (g - i) (the difference between the actual and "expected" (g - i) color) is plotted against redshift for a limited redshift range, to show highlights. The residuals of the polynomial fit to the mean binned (g - i) of the full optical sample are shown connected by a solid green line. See Figure 3 for symbol types. Most targets, BAL QSOs, and RL QSOs are redder than average. (A color version of this figure is available in the online journal.)

component (model "PL") and (2) fitting an intrinsic absorber intr with neutral column NH at the source redshift, with photon index frozen at = 1.9 (model "PLfix"). Allowed fit ranges intr are -1.5 < < 3.5 for PL and 1018 < NH < 1025 for PLfix. These fits use the Powell optimization method, and provide a robust and reliable one-parameter characterization of the spectral shape for any number of counts. Spectra with less than 100 net counts25 were fit using the ungrouped data with Cash statistics (Cash 1979). Spectra with more than 100 counts were grouped to a minimum of 16 counts per bin and fit using the 2 statistic with variance computed from the data. Finally, X-ray spectra with over 200 counts were also fit with Gal a two-parameter absorbed power law where both and the NH were free to vary within the above ranges (model "PL_abs"). 5.1. X-Ray Spectral Continuum Measurements We compile "best-PL" measurements, where for fewer than intr 200 counts, we use from the PL (NH fixed at zero) fits and intr for higher count sources we use PL_abs (both and NH free). In the MainDet sample there are 156 sources with 200 counts or more. High-count objects are found scattered at all luminosities below z 2.5. QSOs with more than 200 counts (0.5­8 keV) intr with both and NH fits in yaxx, are well-distributed in l2keV amongst the detections due to the wide range of exposure times. The mean for all the 1135 QSOs in the MainDet sample is 1.94 ± 0.02 with median 1.93. Means and medians for the MainDet sample and the subsamples discussed in this section are listed in Table 4. The typical (median) error in , 0.5, is similar to the dispersion 0.54 in best-fit values of . If we limit the sample to the 314 sources with more than 100 counts,
25

Source counts derived from yaxx may differ at the 1% level from those derived by ChaMP XPIPE photometry, due to slightly different background region conventions.

the typical error is 0.32, with no change in mean or median . The distribution that we find is similar to that found recently for smaller samples of broad-line AGNs (BLAGNs). Just et al. (2007) studied a sample of luminous optically selected quasars observed by Chandra, ROSAT, and XMM-Newton, and found = 1.92 ± 0.09 for 42 QSOs. Mainieri et al. (2007) studied a sample of 58 X-ray-selected BLAGNs in the XMM-COSMOS fields, and found = 2.09 with a dispersion of 0.26. Page (2006) found = 2.0 ± 0.1 with a dispersion of 0.36 for 50 X-ray-selected BLAGN in the 13H XMM-Newton/Chandra deep field. Figure 12 shows a histogram of best-fit power-law slopes for several interesting subsamples of QSOs from both the MainDet sample and from the noTDet sample which omits PI targets. The mean and median values for these subsamples are listed in Table 4. We do not separately plot the radio-quiet (RQ) QSO sample, since it follows quite closely the shape of the full sample histogram. For 43 detected RL QSOs in the MainDet sample, the nominal mean slope is = 1.73 ± 0.05 with median 1.65, with a distribution significantly flatter than for the 704 definitively RQ QSOs ( = 1.91±0.02) in the MainDet sample, using the twosample tests described in Section 6.1.1. RL QSOs are known to have flatter high energy continua from previous work (e.g., Reeves & Turner 2000). For the 15 known BAL QSOs in the MainDet sample, the nominal mean slope is = 1.35 ± 0.15 with median 1.3, and the distribution is significantly different than for the full MainDet sample (minus BALs, NALs, and NLS1), using the two-sample tests described in Section 6.1.1. The difference is even more significant (Pmax < 0.01%) when comparing only to the 667 definitively RQ QSOs (RL < 1.6). Since the mean (median) number of (0.5­8 keV) counts for BAL QSOs is just 27 (15), the nominally lower likely reflects undetected absorption (Green et al. 2001; Gallagher et al. 2002).


658

GREEN ET AL.

Vol. 690

Figure 12. Left: histogram of best-fit X-ray spectral power-law slope (best-PL) for the MainDet sample. The full sample (black solid) histogram has been divided by 15, and the RL QSO (long-dashed black) histogram by 2, for ease of comparison with smaller subsamples. BAL QSOs (blue down-slash shading) show very flat slopes, due to strong intrinsic absorption. The NAL distribution (green up-slash shading) suggests a possible bimodality. Neither the NAL nor the NLS1 (magenta dense shading) nor the high redshift (short-dashed red) histograms are significantly different, by the Kolmogorov­Smirnov (K-S) test from the full sample distribution. Right: the same plot, but for the noTDet sample which omits PI targets. The RL QSO distribution is less distinct here, and the softest (largest ) NLS1s disappear. (A color version of this figure is available in the online journal.) Table 4 Quasar Sample Univariate Results Samplea MainDet RQ RL BAL NAL NLS1 z>3 D2L RQ RL BAL NAL NLS1 z>3 N 1135 704 43 15 9 19 56 1269 680 31 23 8 19 47 Mean tionsd Error
b

Median 1.93 1.86 1.65 1.30 1.75 1.95 1.76 1.365 1.452 1.392 1.664 1.500 1.433 1.786

P

max

c

(%)

Distribu 1.94 0.02 1.91 0.02 1.73 0.05 1.35 0.15 1.68 0.13 2.01 0.15 1.80 0.07 ox Distributions 1.421 0.005 1.527 0.008 1.382 0.030 1.717 0.028 1.463 0.056 1.540 0.080 1.817 0.077

... 0.2 0.2 0.05 19 35 21 ... 0.0 0.0 0.0 49 24 55

intrinsic NAL quasars are indistinguishable from those of the larger quasar population. For the 19 known NLS1s in the MainDet sample, the nominal mean slope is = 2.01 ± 0.15 with median 1.95, indistinguishable from the comparison sample (MainDet sample minus BALs, NALs, and NLS1).
5.1.1. X-Ray Spectral Evolution

Notes. a For each parameter tested, the numeric sample at the top is the parent for comparison subsamples below. The RL subsamples are tested against non-RL QSOs from the parent sample. The remaining three QSO subsamples for each parameter are tested against the parent sample excluding all four QSO subtypes. b Error in the mean from the Kaplan­Meier estimator as implemented in ASURV. An estimate of the dispersion can be obtained by multiplying this by N - 1. c The maximum probability for the null hypothesis (of indistinguishable samples) from three tests described in Section 6.1.1. Only for Pmax < 5 do we consider the distributions significantly different. RL and RQ samples are contrasted to each other. Other samples are compared to their parent sample (MainDet or D2L) -X , where X = BALs + NALs + NLS1s, except for BALs, whose parent sample is RQ QSOs only. d These are distributions of "best-PL" measurements, best-fit , which always Gal includes NH , and also includes Nintr for 0.5­8 keV counts > 200. H

The mean slope = 1.68 ± 0.13 with median 1.75 for the nine known NALs in the MainDet sample is indistinguishable from the full MainDet sample (minus BALs, NALs, and NLS1), but the NAL statistics are poor. On the other hand, a smaller, nonoverlapping sample of NALs observed by Chandra published by Misawa et al. (2008) agrees that the X-ray properties of

No signs of evolution have been detected for the intrinsic power-law slope of QSOs: z > 4 samples (Vignali et al. 2005; Shemmer et al. 2006) show 2, just like those at lower redshifts (Reeves & Turner 2000). In a recent small sample of high (optical) luminosity QSOs (Just et al. 2007) also found no trend of with redshift. A larger compilation also shows at best marginal signs of evolution, and only for well-chosen redshift ranges (Saez 2008). Figure 13 shows versus redshift both for the individual QSOs (top), and for subsamples (bottom) in fixed bins of z. We also try another binning, where we grow redshift bins from z = 0 until each bin has between 100 and 150 objects (except for the high-redshift bin z > 3 which has just 58). The largest difference between any of the bins is 0.17, whereas the typical rms dispersion in all bins is 0.5. The noTDet sample and the NoRBDet sample remove contributions from targets, and from RL or absorbed QSOs, respectively. The HiCt sample is the subset of the MainDet sample intr with counts greater than 200, where fits for NH are also performed. This decreases the effect of QSOs that may appear to have hard but are actually absorbed. The HiCtNoTRB sample also has only counts greater than 200, but further removes targets, RL, BAL, and NAL QSOs. We find no evidence for evolution; the null hypothesis (no correlation) cannot be rejected with P < 8% for any of samples 0, 1, 2, 20, or 21, which test possible contamination or biases from targets, RL, and/or absorbed QSOs. Given that the typical uncertainty in for a single QSO ( 0.5) is comparable to the sample dispersion, simultaneous spectral fitting of subsamples of detected QSOs in bins of redshift (in the manner of Green et al. 2001) might improve


No. 1, 2009

PROPERTIES OF SDSS QUASARS IN THE ChaMP

659

Figure 13. Top: best-fit X-ray spectral power-law slope vs. redshift for the MainDet sample. The large open green circles show QSOs with more than intr 200 (0.5­8 keV) counts and simultaneous ­NH fits. No evidence is seen for evolution in . See Figure 3 for symbol types. Dots to distinguish detections from limits in other plots are omitted here, since by definition all objects are detected. Radio-loud QSOs (open blue circles) tend to have flatter slopes. Some of the flattest slopes seen are for BAL QSOs (open black squares), due to their strong intrinsic absorption. Typical mean errors on are 0.5 below z = 2, rising to 0.8 at the highest redshifts. Bottom: mean values are shown at the mean redshift for QSOs in five redshift bins of width z = 1 for the MainDet sample. Error bars show the error in the means for both axes. (A color version of this figure is available in the online journal.) Table 5 Quasar Sample Bivariate Regression Results: (l2keV ) OLS Sample MainDet noTDet HiCtNoTRBt NoRBDet hiLoLx HiCt Slope - - - - - - 0.1465 0.1424 0.1213 0.1528 0.2336 0.1454 Error 0.0199 0.0188 0.0497 0.0262 0.0388 0.0472 Intercept 5.8575 5.7579 5.2473 6.0343 8.1937 5.8540 Error 0.5276 0.4959 1.3333 0.6980 1.0337 1.2679

Figure 14. Top: best-fit X-ray spectral power-law slope vs. X-ray luminosity for the MainDet sample. See Figure 3 for symbol types. The large open green circles show QSOs with more than 200 (0.5­8 keV) counts and simultaneous intr ­NH fits. A significant but shallow trend toward harder spectra (smaller ) is evident from the best-fit regression lines shown: the MainDet sample (red solid), the HiCtNoTRB sample with counts greater than 200 (blue short-dashed) or HiLoLx sample with log l2500å > 29.8and l2keV > 26 (blue long-dashed). Fit parameters are shown in Table 5. Bottom: mean values are shown at the mean luminosity for QSOs in six bins of width lX = 0.75 for the MainDet sample. Error bars show the error in the means for both axes. (A color version of this figure is available in the online journal.)

Notes. Samples tested are arranged in the same order as in Table 1. OLS refers to the ordinary least-squares regression.

constraints, but such a project is both beyond the scope of this paper and probably unwarranted given the scarce evidence for evolution to date.
5.1.2. X-Ray Spectral Trend with X-Ray Luminosity

There is conflicting evidence as to whether correlates with X-ray luminosity LX (Dai et al. 2004; Page et al. 2004), or perhaps with Eddington ratio (Porquet et al. 2004; Kelly 2007b; Shemmer et al. 2008; Kelly et al. 2008). Saez (2008) find significant softening (increase) of with luminosity in selected redshift ranges, for a combination of Type 1 and Type 2 X-ray-selected AGNs from the CDFs. Type 2 objects dominate numerically, especially at low LX , and show larger scatter in at all luminosities, so Type 2 objects dominate the trends that they observe. Our sample is complementary in that it treats only Type 1 QSOs, and provides the largest, most uniform such spectral study to date. We detect a significant anticorrelation between and l2keV across a several subsamples (see Table 5 and Figure 14), but none

between and either redshift or l2500 å . The best-fit continuum slope hardens (decreases) with increasing luminosity for the MainDet sample. Since is itself used to calculate l2keV via the standard X-ray power-law k-correction, there might be some danger that the apparent anticorrelation is induced. We investigate this in two ways. First we examine the range in the ratio of l2keV calculated assuming our best-fit k -correction to that calculated with a fixed = 1.9. Across the full X-ray luminosity range, the different assumptions induce a change of 0.1 dex (25%), insufficient to account for the observed trend. Second, we examine whether correlates significantly with l2500 å , which is clearly unaffected by the X-ray k-correction. There is indeed a significant trend in the MainDet sample (see Table 5). One possibility is that the observed trend of with luminosity intr might be due to an undetected increase in NH . To test this, we examined the relationship in the HiCtNoTRB sample, the subset of the MainDet sample with counts greater than 200, intr where fits for NH are also performed. We do not claim that all absorption is detected in the HiCtNoTRB sample. However, if indeed undetected absorption caused the trend in the MainDet sample, we would expect the anticorrelation to weaken or flatten in the HiCtNoTRB sample. Instead, the anticorrelation is still significant, and the slope is actually steeper (see Table 5). It is also conceivable that soft X-ray emission associated with star-formation activity might contaminate the sample at low luminosities. To test this, we create the HiLoLx sample with definitively QSO-like luminosities in both wavebands, i.e., log l2500 å > 29.8 and l2keV > 26. For this sample, the anticorrelation remains strong and steep (see Table 5). We speculate that the balance between thermal (accretion disk) and nonthermal X-ray emission may shift toward higher luminosities. If at high l2keV , nonthermal emission represents


660

GREEN ET AL.

Vol. 690

a larger fraction of the emitted X-rays, we would expect a hardening (decrease) of with luminosity, as is seen. We would also expect to find more RL QSOs at high l2keV , and we do. Since quasars such as RL QSOs with a larger fraction of nonthermal emission also show stronger X-ray emission relative to optical (smaller ox ), we might expect a correlation between and ox such that as the spectrum hardens, ox decreases. This is shown to be the case in Section 7.2. The idea of a significant nonthermal, probably jet-related, emission component even in RQ QSOs is not new. Blundell & Beasley (1998) found strong evidence from very long baseline array (VLBA) observations for jet-producing central engines in eight of the 12 RQ QSOs in their sample. Barvainis et al. (2005) find similarities between RL and RQ quasars spanning a host of radio indicators (variability, radio spectral index, and VLBI-scale components), suggesting that the physics of radio emission in the inner regions of all quasars is essentially the same, involving a compact, partially opaque core together with a beamed jet. Czerny et al. (2008) similarly find evidence for a blazar component in RQ QSOs by modeling their variability. More sensitive empirical tests of whether the observed trend is due to substantial (but not directly detectable) absorption depressing the observed soft X-ray continuum at high luminosities, or to an increasing thermal fraction at lower luminosities could be performed by stacking counts in narrow energy bands from Chandra images of all QSOs (detected or not). Stacking has been used effectively this way in the CDFs (e.g., Steffen et al. 2007; Lehmer et al. 2007), but the task is more daunting for the ChaMP, where care must be taken to properly account for the Gal effects of a much wider range of NH , CCD quantum efficiency values, and exposure times. A study of 35 Type 1 QSOs by Shemmer et al. (2008) finds that increases (softens) with L/LEdd , the latter derived from FWHM(H ) and L (5100 å). Kelly et al. (2008) examine a larger sample, and find complicated relationships between MBH ,LUV /LEdd , and LX /LEdd that change direction depending on the emission line used to estimate MBH . In a subsequent paper, we are extending our current analysis to include estimates for MBH and L/LEdd for the spectroscopic subsample of the current paper. 5.2. X-Ray Intrinsic Absorption Measurements The quality of X-ray measurements of intrinsic absorbing columns depends strongly on the number of counts in the quasar, but also on the redshift of the object, as illustrated in a simple test in Figure 15. Figure 16 shows our best-fit intrinsic absorption column measurements, which are overwhelmingly upper limits, either when assuming (as for counts less than 200) that = 1.9 or when fitting both and Nintr (for QSOs with more than 200 H counts). Just one of the RL QSOs has detectable Nintr , which H is nevertheless not large (1021 cm-2 ). Jet-related X-ray emission may reduce any absorption signatures. Strong intrinsic absorption is relatively rare in RQ Type 1 QSOs. In the strictest interpretation of AGN unification models, none of these broad-line AGNs should be significantly Xray-absorbed. Small obscured fractions might be expected by selection, which requires that the broad-line region (BLR) not be heavily dust-reddened, i.e. our view of the BLR is unobscured. X-ray absorbed BLAGN have therefore sometimes been called "anomalous." The obscured fraction of BLAGN from the literature spans a wide variety of samples and analysis methods, but most define "obscured" as Nintr > 1022 cm-2 and H

Figure 15. Best-fit log intrinsic absorption vs. (0.5­8 keV) counts for 10 random exposure-time subsamples scaled to achieve 2%, 5%, 10%, 20%, and 50% of the full original exposure time (100%, at the far right) for one bright X-ray source observed by Chandra. Best-fit values of Nintr are shown as blue dots, H with the corresponding 90% upper confidence limit shown as black arrows. This plot shows a clear trend, reflecting the skewed (one-sided positive definite and intr logarithmic) nature of the NH parameter. (A color version of this figure is available in the online journal.)

find fractions of about 10% or less (Perola et al. 2004; Page 2006; Mainieri et al. 2007). By contrast, selecting optically unobscured AGNs only from optical/IR photometry, and using X-ray HRs from XMM , Tajer et al. (2007) find 30%. To estimate the obscured fraction of SDSS QSOs from the ChaMP, we first limit to the D2L sample and z < 2 to maximize the fraction that are X-ray detected (see Figure 9). We further limit to QSO-like optical luminosities log l2500 å > 29.8, yielding 630 QSOs, for which 498 (79%) are X-ray detections. If we rather stringently require that the 90% lower bound to the bestintr fit log NH exceeds 22, then 50 of the 498 (10%) of X-ray detected QSOs are obscured. If instead we simply use the best-fit intr NH , 98 of 498 (20%) of detected QSOs are obscured. However, if we make the unlikely assumption that all X-ray-undetected QSOs are missed due to heavy intrinsic absorption, the obscured fraction could be as high as 29% and 36% for these two different methods, respectively. We note that the correlation of dust to total column of X-ray absorbing metals is not strict. For instance, we know that a significant fraction of optically selected QSOs (13­20%; Reichard et al. 2003;Hewett&Foltz 2003; Knigge et al. 2008) appear as BAL QSOs, which are thought to be seen along a sightline piercing warm (ionized) absorbers. BAL QSOs are highly absorbed in X-rays (Green et al. 1995, 1996, 2001; Gallagher et al. 2002). All the BAL QSOs here have Nintr H detections or upper limits that are greater than 3 â 1022 cm-2 . An even larger fraction may harbor undetected warm absorbing material that is too smooth in its velocity and/or geometric ´ distribution to show distinct absorbing troughs (Gierlinski & Done 2004; Green et al. 2006). The frequency of warm Xray absorbers has been shown to be about 50% (Porquet et al. 2004) in low-redshift (PG) QSOs. Again, the bias of (colorplus broad-line-based) optical selection decreases their number intr here. The fraction of detectably large NH in the fully X-rayselected ChaMP is larger, but nearly all such examples are found in objects which appear optically as narrow-line AGN or absorption line galaxies (XBONGs), as shown by Silverman et al. (2005a).


No. 1, 2009

PROPERTIES OF SDSS QUASARS IN THE ChaMP

661

Figure 16. Left: best-fit intrinsic absorption vs. (0.5­8 keV) counts for detected QSOs. Object type symbols are as described in Figure 3. The green arrows indicate upper limits, while detections (where the 90% lower limit exceeds log Nintr = 20) are red dots with 90% error bars. The strong decrease in the upper limit envelope H traces the increased sensitivity as a function of detected counts. Only a handful of objects show significant absorption above 200 counts where simultaneous fits to intr are performed. Right: best-fit intrinsic absorption vs. redshift. The strong increase in the upper limit envelope reflects both the decrease in counts and the and NH decreased sensitivity as a function of redshift, due to the rest-frame-absorbed region below 2 keV redshifting out of the Chandra bandpass. (A color version of this figure is available in the online journal.)

6. X-RAY TO OPTICAL SEDs Beyond the desire to know of real trends in the SEDs of QSOs for the sake of understanding accretion physics, such trends influence other scientifically important research. Corrections to derive bolometric luminosities (Lbol ) from measured LX depend strongly on SED trends. Lbol in turn is key to constraining such fundamental parameters as accretion efficiency and/or SMBH densities in the universe (e.g., Marconi et al. 2004), or active accretion lifetimes or duty cycles (e.g., Hopkins et al. 2005; Adelberger & Steidel 2005). As another example, to compare the AGN space densities found via optical versus X-ray selection, Silverman et al. (2005b, 2008) and others use the measured trend in ox to convert the X-ray luminosity function (LF) to an optical LF. Numerous studies have debated the strength and origin of trends in the X-ray-to-optical luminosity ratio of optically selected quasars. This ratio is herein (and typically) characterized by the spectral slope ox . Most studies include large samples mixing both targeted and serendipitous X-ray observations (e.g., Avni & Tananbaum 1982; Wilkes et al. 1994; Vignali et al. 2003; Strateva et al. 2005; Steffen et al. 2006; Kelly et al. 2007a). All these studies, using widely varying techniques and sample compilations, have found a strong correlation between ox and optical luminosity, and several have also suggested a residual evolution of ox with redshift or lookback time, even after the luminosity correlation is accounted for. While large sample sizes and appropriate statistical techniques (see Section 6.1) have been sought in these studies, the impact of sample selection effects combined with large intrinsic sample dispersions in luminosity may still dominate the observed correlations. These effects have been modeled and simulated by Yuan et al. (1998) and Tang et al. (2007) among others, who retrieve seemingly statistically significant but artificially induced correlations. They highlight the need for large, well-defined samples to alleviate this problem, and they also note the importance of minimizing the effects of a

strong L­z correlation induced by a single sample flux limit. The ChaMP/SDSS sample is a large step toward alleviating these problems. Contamination by unrelated physical processes should be eliminated wherever feasible. As mentioned above, it is wellknown that RL QSOs tend to be X-ray bright, and there is evidence that a distinct, jet-related physical process produces that extra X-ray emission. BAL QSOs, on the other hand, tend to be X-ray weak, because of intrinsic absorption from the warm (ionized) winds (Green et al. 2001; Gallagher et al. 2002). When investigating distributions and correlations amongst ox , lo , and lx , most previous authors have chosen to eliminate both BAL and RL QSOs to minimize contamination of (presumably) pure accretion-dominated X-ray emission. It is important to note that this precaution is neither complete, nor perhaps correct. For example, the intrinsic radio-loudness distribution of quasars does appear to be well modeled by a quasi-normal distribution with a 5% tail of RL objects (Cirasuolo et al. 2003), but the relationship of that distribution to X-ray emission is not well characterized. Even after removal of quasars with a radio/optical flux ratio above some limit, radio-related X-ray emission may pervade quasar samples, and affect the distributions we study. Similarly, the best quantitative measures of absorption profiles via, e.g., "balnicity" (Weymann et al. 1991) indicate that the fraction of quasars with intrinsic outflows may be significantly underestimated (Reichard et al. 2003), with classic BAL QSOs just the tip of the iceberg. Equally important may be that BALs to date are mainly detected only beyond z 1.4 in groundbased optical spectroscopic quasar samples (z 1.5 in SDSS spectra), so the low-redshift, low-luminosity end of the quasar distribution may harbor significant undetected warm absorption. We therefore examine a variety of subsamples within and without these classes. When we compare bivariate regression results l2500 å , l2keV , and ox , directly to Steffen et al. (2006; S06 hereafter), we follow their convention of excluding known RL QSOs and BAL QSOs, and expand somewhat to also eliminate QSOs with evident NALs, and NLS1s as well. We note that their


662

GREEN ET AL.

Vol. 690

ox definition is the negative of the more conventional one we adopt here. 6.1. Statistical Tools and Methods Many previous studies have emphasized the importance of including X-ray upper limits in the samples, which require appropriate statistical treatment using, e.g., survival analysis. Results from survival analysis can depend strongly on the number of detections, and therefore on the detection threshold, since in flux-limited samples, most detections are near that threshold. The Chandra fields used here span a wide range of effective exposure times, substantially alleviating the powerful effects that a single survey flux limit can have (e.g., by creating a very tight Lx ­z correlation). Previous studies have managed to limit the fraction of upper limits by using mostly targeted X-ray observations (e.g., Wilkes et al. 1994), where exposure times are often chosen based on optical flux, or by judicious choice of subsamples (e.g., Strateva et al. 2005; Steffen et al. 2006) designed to fill the luminosity-redshift plane. For example, Strateva et al. (2005) uses SDSS spectroscopic AGNs from the SDSS Data Release 2 (DR2), observed by ROSAT PSPC for more than 11 ks. For this purpose, the DR2 AGN sample is already biased, since many spectroscopic targets are chosen as X-ray ROSAT or FIRST radio detections. Other AGNs they include to fill in the L­z plane were specifically targeted for Chandra or XMM -Newton observations for a variety of reasons. A slightly larger study by Steffen et al. (2006) added objects using mostly photometric AGN classifications and redshifts from COMBO-17, together with the Extended Chandra Deep Field (CDF) South (ECDF-S), but also several other small bright samples for the low L­z region. Even after selecting and combining various samples with high detection fractions, the use of Survival Analysis techniques to incorporate limits must sometimes be abandoned.26 In summary, if significant selection biases may affect either the constituent subsamples or their ensemble, neither the inclusion of upper limits, nor the use of complex statistical analysis methods should convey the impression that statistical results are as robust as those from a uniform, complete, and well-characterized sample. It is also worthwhile to consider the fact that some previous studies have appended low-luminosity subsamples (Seyferts 1s). Our SDSS sample excludes some of these objects because they would be spatially resolved. This may bias our sample in the sense that for low-redshift AGNs (e.g., 107 objects in the MainDet sample with z 0.55) we include only those that are compact (optical light distribution consistent with the expected SDSS PSF). Most previous studies have instead attempted to include such objects, but then they correct for the inclusion of substantial host galaxy emission via, e.g., spectroscopic template fitting (Strateva et al. 2005). This may bias those samples in a different way by excluding the host contribution only for nearby objects.
6.1.1. Univariate Analyses

tests that we use are the Gehan, or generalized Wilcoxon test, and the log-rank test in ASURV (Survival Analysis for Astronomy; LaValley et al. 1992). Again, we require Pmax < 0.05 to call the distinction significant. For Pmax < 0.10, we call the difference "marginal." The Kaplan­Meier method we employ to estimate the mean of a distribution allows for the inclusion of censored values (upper or lower limits).
6.1.2. Bivariate Regressions

Except where noted, all the correlations studied between X-ray and optical luminosity, and between ox and lo are highly significant: P < 10-4 by Cox Proportional Hazard, Kendall's , or Spearman's tests, as implemented in the ASURV package (LaValley et al. 1992). We deem a correlation significant if the maximum probability Pmax from all three tests is 0.05 or less. We perform bivariate linear regressions using the two-dimensional Kaplan­Meier (2KM) test (Schmitt 1985) as implemented in ASURV, which permits linear regression with limits in either axis.27 We use 20 bootstrap iterations for error analysis, (more than sufficient given the large sample size here) and 20 bins in each axis, with origins 27.3 (23.0, 1.0) for l2500 å (l2keV , ox ), except where samples have been explicitly restricted in luminosity, or by object type to have N < 200, whereupon we use 10 bins. For bivariate regressions between X-ray and optical luminosity, we make no assumptions about which luminosity constitutes the dependent or independent variable, and calculate the mean (bisector) of the ordinary least-squares (OLS) regression lines which minimize residuals for Y (X), X(Y ), respectively.28 While the intrinsic dependence of ox (lo ) or ox (lx ) is unknown, most studies assume ox to be the dependent variable, and quote slopes accordingly, so we will follow suit largely for purposes of comparison. We further caution that in samples with large dispersions, different regression methods can yield different results (see a recent mathematical review of related issues by Kelly 2007b). For completeness, we provide statistical results in table format for a variety of subsamples which may be of interest, even beyond those discussed in the text. For ease of reference, tables listing bivariate statistical results (Tables 5­7) list samples in the same order as the table defining samples (Table 1). 6.2. ox for QSO Subsamples Here we use the high detection fraction (D2L) sample, with high detection fraction, to examine the ox distributions of several subsamples, including targets. The mean ox for 1269 QSOs in the D2L sample is 1.429 ± 0.005 with median 1.370. Means and medians for the D2L sample and for subsamples discussed in this section are listed in Table 4, as are result for two-sample tests. For 31 RL QSOs in the D2L sample, the mean ox = 1.377 ± 0.028 with median 1.392. Using the two-sample tests described in Section 6.1.1 above, this distribution is only marginally different than for the 1238 non-RL QSOs in the D2L sample. RL QSOs are thought to be more X-ray loud than RQ QSOs. For the 23 known BAL QSOs in the D2L sample, the mean ox is 1.717 ± 0.028 with median 1.66, and the
27

To compare two-sample distributions, we test the "null" hypothesis that two independent random samples subject to censoring (e.g., for RL and RQ QSO) are randomly drawn from the same parent population. The programs for two-sample
26

For example, Steffen et al. (2006) simply drop the X-ray limits when regressing with l2keV as the independent variable. We choose instead to incorporate the limits using the two-dimensional Kaplan­Meier (2KM) test (Schmitt 1985).

Our results from the other two bivariate regression algorithms in ASURV (the Buckley­James method and the parametric EM Algorithm) are quite consistent. 28 In the absence of limits, these results reduce reliably to the bisectors found by the SLOPES program (Feigelson & Babu 1992; Babu & Feigelson 1992; Isobe et al. 1990).


No. 1, 2009

PROPERTIES OF SDSS QUASARS IN THE ChaMP
Table 7 Quasar Sample Bivariate Regression Results: ox (l2500 å ) OLS Sample Main MainDet NoTDet D2L D2LNoRB hiLo NoRB NoRBDet D2LNoTRB zLxBox LoBox zBox zBoxDet D2LSy1 S06 Slope 0.0598 0.0826 0.0732 0.0610 0.0516 0.1284 0.0513 0.0804 0.0482 0.1943 0.1895 0.1019 0.1119 0.0058 0.137 Error Primary Samples 0.0066 0.0066 0.0089 0.0085 0.0078 0.0070 Other Samples 0.0073 0.0085 0.0104 0.0067 0.0105 0.0242 0.0267 0.0175 0.008 Intercept -0.2776 -1.0331 -0.7496 -0.3189 -0.0358 -2.3754 -0.0239 -0.9667 0.0652 -4.4484 -4.2956 -1.5709 -1.9335 1.3253 -2.638

663

Table 6 Quasar Sample Bivariate Regression Results: l2keV (l2500 å ) OLS Bisector Sample Main MainDet NoTDet D2L D2LNoRB hiLo NoRB NoRBDet zLxBox LoBox zBox zBoxDet D2LSy1 D2LNoTRB S06 Slope 1.1171 0.9372 0.9719 1.1350 1.1667 1.2976 1.1502 0.9359 0.8421 0.9907 1.6658 1.4088 1.2299 1.1908 0.72 Error Primary Samples 0.0170 0.0266 0.0259 0.0209 0.0238 0.0340 Other Samples 0.0146 0.0253 0.0173 0.0310 0.1125 0.0803 0.0504 0.0196 0.01 Intercept -7.5929 -1.9178 -2.9508 -8.1162 -9.0641 -13.1316 -8.5922 -1.8745 1.0177 -3.5191 -24.3936 -16.3512 -10.6975 -9.7838 4.53 Error 0.6365 0.8797 0.8418 0.6329 0.7199 1.0356 0.6125 0.8903 0.6695 0.9357 3.4355 2.4564 1.4746 0.5921 0.69

Error 0.1988 0.1978 0.2681 0.2580 0.2377 0.2151 0.2217 0.2546 0.3146 0.2033 0.3176 0.7404 0.8172 0.5160 0.240

Notes. Samples tested are arranged in the same order as in Table 1. OLS refers to the ordinary least-squares regression. S06 are results from Steffen et al. (2006) for comparison.

Notes. Samples tested are arranged in the same order as in Table 1. OLS refers to the ordinary least-squares regression. S06 are results from Steffen et al. (2006) for comparison.

distribution is significantly different than for the non-BAL the D2L sample. The apparent X-ray weakness has been shown to be consistent with intrinsic absorption of a normal underlying X-ray continuum (Green et al. 2001; Gallagher et al. 2002). For the eight known NALs in the D2L sample, the mean ox is 1.463 ± 0.056 with median 1.5, but the poor statistics render the distribution indistinguishable from the overall D2L sample. For the 19 known NLS1s in the D2L sample, the mean ox is 1.54 ± 0.08 with median 1.43, again indistinguishable from the overall D2L sample. 6.3. X-Ray Luminosity l
2keV

Versus Optical l

2500 å

Figure 17 shows a highly significant correlation of l2keV with l2500 å , and plots our best-fit regression lines. The bisector regression relationship for the SDSS/ChaMP sample (the Main sample, limits included, 2308 QSOs), is log(l
2keV

) = (1.117 ± 0.017) log(l

2500 å

) - (7.59 ± 0.64).

This slope is close to linear, and significantly different than the bisector slope = 0.72 ± 0.01 derived by S06 from their smaller, more diverse sample. The subsample philosophically closest to that of S06 is the NoRB sample, or its high detection fraction version the D2LNoRB sample, since they exclude known RL and BAL QSOs (as well as NALs and NLS2s), but include targets. The D2LNoRB sample bisector slope is = 1.115 ± 0.015. Our bisector slope results in Table 6 are closer to linear than S06 for all samples tested across the full luminosity range, including those that omit limits altogether. Figure 18 shows the MainDet sample (detections only) across a smaller luminosity range, to highlight the different QSO types. A nearly linear result was also found by Hasinger (2005)for a sample of Type 1 QSOs that spanned a similarly large range of luminosities as our own. That work used an X-ray-selected sample with a high ( 95%) completeness for spectroscopic identifications, and concluded that the non-linear trends seen in optically selected samples probably result from selection effects.

Figure 17. X-ray (2 keV) vs. optical (2500 å) luminosity for the the Main sample. See Figure 3 for symbol types. The long-dashed magenta line is of slope unity, normalized to the sample means. The flattest fit is the best-fit (OLS bisector) relation from Steffen et al. (2006) (their Equation 1c), shown as a solid cyan line, spanning the luminosity range of their compilation. The best-fit OLS bisector regression line to the SDSS/ChaMP in the Main sample (including nondetections) is shown in blue, spanning the full plot. The short-dashed lines plotted using 1 statistical errors to the ChaMP fits are so close to the best fit as to be barely discernible on the plot. The red line is a Y (X ) regression on the same data, illustrating the sensitivity of fits to regression method in high dispersion data. (A color version of this figure is available in the online journal.)

6.4. ox Versus Optical Luminosity Figure 19 (left) shows the trend of ox with l2500 å for the Main sample. While the detected QSOs appear as a large "blob," this is due to a combination of the SDSS depth, the QSO optical luminosity function, and the highly efficient QSO selection in the corresponding redshift range 1 < z < 2.5 (see Figure 8). There is a highly significant correlation: P < 10-4 by Cox Proportional Hazard, Kendall's , or Spearman's for any of


664

GREEN ET AL.

Vol. 690

Figure 18. Zoom-in of the X-ray (2 keV) vs. optical (2500 å) luminosity plot for the MainDet sample. See Figure 3 for symbol types. Here we show a smaller luminosity range, and omit limits to highlight the classes indicated in the key. RL QSOs clearly populate the upper end of the distribution in X-ray luminosity, while BAL QSOs are underluminous in X-rays. The fits shown here are performed on the MainDet sample with detections only. Line types are the same as in Figure 17. (A color version of this figure is available in the online journal.)

luminosity. For example, the MainDet sample ox (lo ) regression that results from instead using fixed = 1.9 and best-fit Nintr H has slope 0.077 and intercept -0.872. Inasmuch as ox probes intrinsic accretion processes, it probably samples the balance between optical/UV blackbody thermal emission from a geometrically thin but optically thick accretion disk, against X-ray emission from a hot, optically thin corona that upscatters the seed UV photons from the disk. Various physical models can explain an ox -l2500 å correlation with plausible parameters, e.g., a disk truncation radius that increases with luminosity (Sobolewska et al. 2004). We also know that ox measurements can be affected by intrinsic absorption. This is proven by the extreme example of the BAL QSOs here and elsewhere (Green et al. 2001; Gallagher et al. 2002). Further evidence may come from a recent sample of AGN host galaxies of five clusters observed by Chandra and the Advanced Camera for Surveys (ACS) onboard HST, where Martel et al. (2007) found that the X-ray-to-optical flux ratio of the AGNs correlates with the inclination angle of the host galaxies29 . If the observed trends are not dominated by selection effects (e.g., Tang et al. 2007), it seems likely that intrinsic absorption acts to increase the dispersion of an intrinsic relationship which is dominated by accretion physics. 7. EVOLUTION OF OX In a sample with a strong ox (l2500 å ) correlation, ox will naturally correlate strongly with redshift as well, due to the powerful redshift­luminosity trend shown in Figure 6. To determine whether any redshift evolution of ox occurs independent of its luminosity dependence, we use two methods. First, we examine a subsample with a narrow range in l2500 å but a reasonably broad range in redshift. The zBox sample (Table 7) contains all the MainDet sample objects with 30.25 < log (l2500 å ) < 31 and 0.5 < z < 2.5. This sample shows no significant correlation between ox and redshift (Pmax = 19%). Accordingly, the nominal best-fit regression has a slope consistent with zero (-0.001 ± 0.014). Next, we can subtract the best-fit ox (l2500 å ) regression to the more luminous hiLo sample (log (l2500 å ) > 29.8), where a simple linear fit seems applicable, and look for any residual ox (z) dependence (evolution). The expected ox (z) trend is significant for the hiLo sample, induced by the ox (l2500 å ) and l2500 å (z)relationships. We then subtract the best-fit regression trend from Table 7. The residual ox shows no trend with redshift in Figure 20 (bottom) or in correlation tests (Pmax = 0.80). The large subsample sizes afforded by the SDSS/ChaMP QSO sample allow us to conclude, without resort to more elaborate statistical analyses, that any apparent evolutionary trend can be accounted for by the l2500 å (z) correlation in our sample, and that such evolution is at best very weak in the range 0.5 < z < 2.5. 7.1. ox Versus X-ray Luminosity A weaker trend has also been noted in the correlation between ox and l2keV (Green et al. 1995; Steffen et al. 2006). For the Main sample, we find ox = (0.003 ± 0.010) log(l
2keV

the samples considered (with or without limits). Our best-fit regression line minimizing residuals only in ox (e.g., a Y (X) regression) for the Main sample is ox = (0.061 ± 0.009) log(l
2500 å

) - (0.319 ± 0.258).

This is significantly flatter than the results of S06 (0.137 ± 0.008), consistent with the more closely linear relationship we find above between X-ray and optical luminosity. As can be seen from Figure 19 (right) and Table 7, keeping targets but removing RL and BAL QSOs (the D2LNoRB sample) yields similarly flat slopes. The classical Seyfert 1/QSO dividing line at MB -23 corresponds here to log (l2500 å ) 29.8(or log (2500 l2500 å ) 44.9). While this conventional partition is essentially arbitrary, it does represent a sharp discontinuity in the luminosity histogram of the current sample: six times as many objects have "QSOlike" optical luminosity as "Seyfert-like." The hiLo sample, restricted to luminous QSOs as above, yields a slope for the ox (l2500å ) relation of 0.128 ± 0.007, most similar to S06 and previous work. The SDSS/ChaMP sample does boast a larger number of low optical luminosity objects than most previous analyses, in large part because of the sensitivity of the Chandra observations, and these lower luminosity objects may be responsible in part for our different results across the full l2500å range. Analysis of the (much smaller, with Ndet = 176 and Ntot = 260) the D2LSy1 sample with Seyfert-like luminosities, suggests that no significant correlation exists (P > 0.6 for all tests). This highlights that the trend of ox (l2500 å ) may not be linear, as also found by S06 and Kelly et al. (2007a), or may only apply for high luminosities. Note that we have investigated whether the details of our luminosity and ox calculations affect the results. No significantly different scientific conclusions result from our use of best-fit (which affects the X-ray K-correction) to calculate the X-ray

)+(1.384 ± 0.261)

29

The X-ray/optical versus inclination correlation holds for late- but not early-type galaxies, so may not apply directly to Type 1 QSOs if they are mostly in elliptical hosts.


No. 1, 2009

PROPERTIES OF SDSS QUASARS IN THE ChaMP

665

Figure 19. ox vs. optical 2500 å log luminosity for the Main sample (left) and the D2LNoRB sample (right). See Figure 3 for symbol types. The best-fit OLS Y (X ) regression for each ChaMP sample is shown as a red line with errors. The best-fit relation from Steffen et al. (2006) is shown as a solid cyan line. (A color version of this figure is available in the online journal.)

Figure 20. Top: ox vs. redshift for the zBox sample, restricted in log l2500 å to the range 30.25­31. No significant trend exists. See Figure 3 for symbol types. Bottom: ox vs. redshift for the hiLo sample (log (l2500 å > 29.8) after subtraction of its best-fit ox (l2500 å ) relation in Table 7. Although the redshift range remains wide for this subsample, no trend is apparent. (A color version of this figure is available in the online journal.)

which (while the correlation is significant) is consistent with zero slope. With a higher detection fraction, the the D2L sample yields slope -0.0280 ± 0.0087, so that objects more luminous in X-rays are also X-ray brighter (relative to l2500 å ). Further removing RL and BAL QSOs (the D2LNoRB sample) does not change the best-fit parameters. While the effects of different samples on the measured regression is significant, the ox (l2keV ) relationship is particularly affected by limits, since they affect both axes. An apparent luminosity dependence of ox is generated artificially if the intrinsic dispersion in optical luminosity o exceeds that for X-rays x (Yuan et al. 1998; Tang et al. 2007).30
30 Any correlation between a dependent variable B which is derived via B A-1 from an independent variable A will be similarly affected in samples with large dispersion.

The significance of the induced correlation is proportional to 2 2 o lo where lo is the optical/UV luminosity range of the sample. The magnitude of the biases also depends on the luminosity function and sample flux limits. Given these effects, the apparently strong correlations so far published are all consistent with no intrinsic dependence (Tang et al. 2007). The most convincing remedy is likely to be a volume-limited sample that spans a large range of both redshift and luminosity with high detection fraction. This requires careful treatment of large combined samples, similar to Silverman et al. (2005b, 2008), including detailed pixel-by-pixel flux (and consequently volume) limits. While the groundwork has been laid by the ChaMP's xskycover analysis, such a project is well beyond the scope of this paper.


666

GREEN ET AL.

Vol. 690
Table 8 Quasar Sample Bivariate Regression Results: (ox )OLS Sample MainDet NoTDet NoRBDet HiCt HiCtNoTRB Slope 0.188 0.274 0.340 0.342 0.358 Error 0.106 0.108 0.109 0.202 0.179 Intercept 1.676 1.566 1.466 1.507 1.529 Error 0.153 0.154 0.154 0.259 0.225

Notes. Samples tested are arranged in the same order as in Table 1. OLS refers to the ordinary least-squares regression.

Figure 21. Top: best-fit X-ray spectral power-law slope vs. ox for the MainDet sample. The best-fit OLS regression relation for this sample is shown with a red line, and associated errors in dashed lines. The large open green circles show intr QSOs with more than 200 (0.5­8 keV) counts and simultaneous ­NH fits. See Figure 3 for other symbol types. Bottom: mean values are shown at the mean redshift for QSOs in four redshift bins of width ox = 0.25 for the MainDet sample. Error bars show the error in the means for both axes. (A color version of this figure is available in the online journal.)

7.2. ox Versus Figure 21 shows best-fit plotted against ox for the MainDet sample. We detect for the first time a significant but shallow correlation between and ox . Quasars that are relatively X-ray weak (larger ox ) tend to have softer continuum slopes (larger ). Figure 21 shows the best-fit regression relation for the MainDet sample, = (0.188 ± 0.106) ox +(1.676 ± 0.153). The measured slope of the correlation is likely somewhat suppressed by the warm absorbers commonly found even in Type 1 AGNs. In Figure 21, for the hard/weak region bounded by ox > 1.6 and < 1, we find a BAL (indicated by open black squares) is visible for every object for which a spectrum exists covering blueward of rest-frame CIV 1550. It is likely that most if not all objects in this region are BALs, as could be determined, e.g., from rest-frame UV spectroscopy. Again, these objects are probably not intrinsically flat, but rather have a hard best-fit due to undetected intrinsic absorption. Other samples shown in Table 8 omit targets, and RL or possibly absorbed QSOs all show steeper slopes. The steepest slope shown is for the HiCtNoTRB sample, which also includes only objects with counts greater than 200, where Nintr is fitted H independently of . This further supports that absorption if anything flattens the apparent relation compared to the intrinsic relation. 8. DISCUSSION In this study, we have presented the largest homogeneous study to date of optically selected broad-line quasars (from the SDSS) with sensitive X-ray flux limits (from Chandra; mode 2 â 10-14 erg cm-2 s-1 ; 0.5­8 keV). Our large sample highlights the large dispersion in quasar properties that is

unveiled whenever sensitive limits and wide sky areas are combined. We confirm and extend several well-known multiwavelength relations. The relationship between ox and 2500 å luminosity is confirmed for high luminosities (MB -23, or log (l2500 å ) 29.8(log(2500 l2500 å ) 44.9) with slopes similar to those found previously (e.g., S06). Including a wider luminosity range inevitably produces a flatter relation across a range of subsamples which test the effects of excluding Chandra PI targets, RL QSOs, and QSOs with BALs, or NALs as well as NLS1s. No significant ox (l2500 å ) correlation exists for objects (68% detected) at lower luminosities. Another possibility is that the relation simply flattens at low luminosities, or has a higher-order luminosity dependence (e.g., Kelly et al. 2007a). We find for the first time a significant, robust and rather steep dependence of X-ray continuum slope on X-ray luminosity l2keV in the sense that the spectrum hardens with increasing l2keV . A trend in the opposite sense has been reported recently for AGNs in the Chandra Deep Fields (Saez 2008), but may be dominated by differences between Type 1 and Type 2 AGNs. We also report a shallow but significant trend that is harder for relatively X-ray bright (low ox ) QSOs. We note that X-ray bright QSOs include the RL QSOs, and that RL QSOs also have predominantly flatter (harder) . We thus speculate that the overall trends of (ox ) and (l2keV ) both reflect an increase in the nonthermal emission fraction toward higher X-ray luminosities. Not all QSOs with a strong nonthermal Xray emission component are necessarily radio loud-detectable radio loudness may pertain only to a fraction of these objects. Radio bright phases may be short compared to the overall QSO lifetime and/or episodic. The black hole masses in AGNs are 5­8 orders of magnitude larger than those in Galactic black hole (GBH) X-ray binaries. Since for a given L/LEdd the disc temperature scales with mass as M -1/4 , AGN disks are cooler than in GBHs. The thermal accretion disk emission component that dominates the soft X-ray emission ( 2 keV) in GBHs corresponds to optical/ UV ( 2500 å) emission in AGNs. Similarly, the nonthermal emission (probably from Comptonized emission from the disk's corona) sampled as X-ray emission in AGNs comes from harder (20 keV) X-rays in GBHs. Type 1 QSOs may be analogous to GBH binaries in the high/soft state (Sobolewska et al. 2008) where a similar trend is seen in versus a disc/Comptonization index GBH (analogous to ox for QSOs) as we report here. Jester et al. (2006) also compare the disk versus nonthermal emission fraction of GBHs and AGNs, and find that AGNs segregate by radio loudness similarly to GBHs in regions where luminosity and/or nonthermal fraction are high. While the correlations we


No. 1, 2009

PROPERTIES OF SDSS QUASARS IN THE ChaMP APPENDIX

667

find lend significant support to these interpretations, the scatter is large, and could be significantly reduced if extrinsic effects can be identified and corrected for. To better understand the intrinsic physics of accretion requires identifying and quantifying extrinsic effects such as absorption and nonthermal processes. Absorption may occur close to the SMBH gravitational radius, in the BLR, a molecular torus, surrounding star-forming regions, the outer host galaxy, or at intervening redshifts. The intrinsic absorption may be orientationdependent, may evolve with redshift, and may be a function of luminosity as well. Contributions from nonthermal processes certainly play a role, whether or not it is reflected in detectable radio emission, and that role may change with SMBH mass, spin, and environment, so consequently with look-back time as well. Let us face it--quasars are complicated, but on average they do not change much. If quasar SED changes with luminosity were large compared to the dispersion in the population and relatively immune to selection effects, we would have been using them for standard candles a long time ago. Large samples, uniformly observed and analyzed, offer the greatest hope to disentangle the intertwining mysteries. Support for this work was provided by the National Aeronautics and Space Administration through Chandra Award Number AR4-5017X and AR6-7020X issued by the Chandra X-ray Observatory Center, which is operated by the Smithsonian Astrophysical Observatory for and on behalf of the National Aeronautics Space Administration under contract NAS8-03060. G.T.R. was supported in part by an Alfred P. Sloan Research Fellowship and NASA Grant 07-ADP-7-0035. D.H. is supported by a NASA Harriett G. Jenkins Predoctoral Fellowship.We acknowledge use of the NASA/IPAC Extragalactic Database (NED), operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. Funding for the SDSS and SDSS-II has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, the U.S. Department of Energy, the National Aeronautics and Space Administration, the Japanese Monbukagakusho, the Max Planck Society, and the Higher Education Funding Council for England. The SDSS Web site is http://www.sdss.org/. The SDSS is managed by the Astrophysical Research Consortium for the Participating Institutions. The Participating Institutions are the American Museum of Natural History, Astrophysical Institute Potsdam, University of Basel, Cambridge University, Case Western Reserve University, University of Chicago, Drexel University, Fermilab, the Institute for Advanced Study, the Japan Participation Group, Johns Hopkins University, the Joint Institute for Nuclear Astrophysics, the Kavli Institute for Particle Astrophysics and Cosmology, the Korean Scientist Group, the Chinese Academy of Sciences (LAMOST), Los Alamos National Laboratory, the Max Planck Institute for Astronomy (MPIA), the Max Planck Institute for Astrophysics (MPA), New Mexico State University, Ohio State University, University of Pittsburgh, University of Portsmouth, Princeton University, the United States Naval Observatory, and the University of Washington. Facilities: CXO, Mayall (MOSAIC-1 wide-field camera), Blanco (MOSAIC-2 wide-field camera), FLWO:1.5 m (FAST spectrograph), Magellan:Baade (LDSS2 imaging spectrograph), Magellan:Clay (IMACS), WIYN (Hydra).

The ChaMP has developed and implemented an xskycover pipeline which creates sensitivity maps for all ChaMP sky regions imaged by ACIS. This allows (1) identification of imaged-but-undetected objects, (2) counts limits for 50% and 90% detection completeness, (3) flux sensitivity versus sky coverage for any subset of obsids, necessary for log N­log S and luminosity function calculations, and (4) flux upper limits at any sky position. The basic recipe is as follows. We use the wavdetect detection algorithm in CIAO (Freeman et al. 2002) to generate threshold maps at each wavdetect kernel scale actually run (1, 2, 4, 8, 16, and 32 pixels). The threshold maps, computed from the local background intensity, determine the magnitude of the source counts necessary for a detection at each pixel with a detection threshold of P = 10-6 (corresponding to 1 false source per 106 pixels). Thus, when a source is not detected at a given location, the threshold map value serves as an upper limit to the source counts. Nominally, a source with true intensity equal to this counts limit would be detected in approximately half the instances that the source is observed under the same conditions. To retain fidelity yet create a reasonably sized and easily sampled sensitivity table covering the full ChaMP, we first average the threshold map values in sky pixels, each 10 â 10 arcsec, whose boundaries are chosen to match a regular commensurate grid across the sky. The final value of this counts limit in any given sky pixel is interpolated from the two threshold maps computed at wavelet scales that bracket the size of the PSF at that location.31 Note that the identification of the threshold map value oneto-one with the counts limit is valid only for a specific shape of the PSF (see Equation (6) of Freeman et al. 2002). In particular, the strength of the putative source can vary widely based on the correspondence between the PSF size and the wavelet scale, as well as the shape of the PSF. For nonGaussian PSF shapes (such as are found with Chandra at larger OAAs), the threshold values must be corrected before a source counts limit is determined. We calibrate this correction factor by comparing the detection threshold map values with simulated source-retrieval experiments conducted on a subsample of ChaMP fields--the 130 Cycle 1­2 obsids studied by Kim et al. (2007a). While the threshold values give us a reliable map of variation on the sky, we must find a normalization from these simulations. Summing over a large number ( 50,000) of simulated sources, we compare the actual detection fraction from the simulations to the ratio of input (simulated) counts to derived (threshold) counts. From the fine binning available in these data, we interpolate to find the normalization that yields the correct counts values for 50% and 90% completeness, across a range of exposures, background levels, chip types, and OAAs. Our experiments show that the only significant dependence of the correction factor (normalization) is on OAA, and that dependence warrants only a linear correction dependence with best fit N50 = 1.32 + 0.198 â OAA (N90 = 2.08 + 0.331 â OAA) for 50% (90%) source detection probability. Our method has been verified recently by Aldcroft et al. (2008) using the CDF-S data available from the CXC. These data include the full 1.8 Msec from 2000 (Giacconi et al. 2002), and Director's Discretionary Time observations in 2007. The full
31

The 39% radius is determined using the PSF enclosed counts fraction, which corresponds to the 1 two-dimensional PSF size, from the library generated by the Chandra X-Ray Center (CXC) Calibration Group as documented at http://cxc.harvard.edu/cal/Hrma/psf.


668

GREEN ET AL.

Vol. 690

1.8 Msec exposure was used to generate a source list which extended to faint fluxes. The 21 individual obsids comprising the full exposure were then treated as realizations of an experiment to detect these sources and the ensemble of detection statistics were analyzed by the same method as used for the 50,000 simulated sources. We found excellent agreement for the 50% detection threshold and a slight disparity (10%) for the 90% detection threshold which could be explained by source variability. REFERENCES
Adelberger, K. L., & Steidel, C. C., ApJ, 629, 636 Aldcroft, T. L. 2006, Bull. Am. Astron. Soc., 38, 376 Aldcroft, T. L., Green, P. J., Connelly, J., Kashyap, V., & Kim, D.-W. 2008, Bull. Am. Astron. Soc., 40, in press Antonucci, R. 1993, ARA&A, 31, 473 Avni, Y., & Tananbaum, H. 1982, ApJ, 262, L17 Babu, G. J., & Feigelson, E. D. 1992, Simul. Comput., 21, 533 Ballantyne, D. R., Everett, J. E., & Murray, N. 2006, ApJ, 639, 740 Barger, A. J., Cowie, L. L., Mushotzky, R. F., Yang, Y., Wang, W.-H., Steffen, A. T., & Capak, P. 2005, AJ, 129, 578 ´ Barvainis, R., Lehar, J., Birkinshaw, M., Falcke, H., & Blundell, K. M. 2005, ApJ, 618, 108 Becker, R. H., White, R. L., & Helfand, D. J. 1995, ApJ, 450, 559 Blackburne, J. A., Pooley, D., & Rappaport, S. 2006, ApJ, 640, 569 Blundell, K. M., & Beasley, A. J. 1998, MNRAS, 299, 165 Boller, T., Brandt, W. N., & Fink, H. 1996, A&A, 305, 53 Boroson, T. A. 2002, ApJ, 565, 78 Boroson, T. A., & Green, R. F. 1992, ApJS, 80, 109 Boyle, B. J., Shanks, T., & Peterson, B. A. 1988, MNRAS, 235, 935 Brandt, W. N., & Hasinger, G. 2005, ARA&A, 43, 827 Cirasuolo, M., Celotti, A., Magliocchetti, M., & Danese, L. 2003, MNRAS, 346, 447 Condon, J. J., Cotton, W. D., Greisen, E. W., Yin, Q. F., Perley, R. A., Taylor, G. B., & Broderick, J. J. 1998, AJ, 115, 1693 Croom, S. M., Smith, R. J., Boyle, B. J., Shanks, T., Miller, L., Outram, P. J., & Loaring, N. S. 2004, MNRAS, 349, 1397 Cash, W. 1979, ApJ, 228, 939 Czerny, B., Siemiginowska, A., Janiuk, A., & Gupta, A. C. 2008, MNRAS, 386, 1557 Dai, X., Chartas, G., Eracleous, M., & Garmire, G. P. 2004, ApJ, 605, 45 Dai, X., Shankar, F., & Sivakoff, G. R. 2008, ApJ, 672, 108 Dickey, J. M., & Lockman, F. J. 1990, ARA&A, 28, 215 Fabian, A. C., Iwasawa, K., Reynolds, C. S., & Young, A. J. 2000, PASP, 112, 1145 Fabbiano, G., et al. 2007, Astron. Data Anal. Softw. Syst. XVI, 376, 172 ¨ Falcke, H., Kording, E., & Markoff, S. 2004, A&Ap, 414, 895 Feigelson, E. D., & Babu, G. J. 1992, ApJ, 397, 55 Ferrarese, L., & Merritt, D. 2000, ApJ, 539, L9 Freeman, P. E., Kashyap, V., Rosner, R., & Lamb, D. Q. 2002, ApJS, 138, 185 Galbiati, E., et al. 2005, A&A, 430, 927 Gallagher, S. C., et al. 2002, ApJ, 567, 37 Gebhardt, K., et al. 2000, ApJ, 539, L13 Giacconi, R., et al. 2002, ApJS, 139, 369 ´ Gierlinski, M., & Done, C. 2004, MNRAS, 349, L7 Gilli, R., Comastri, A., & Hasinger, G. 2007, A&A, 463, 79 Granato, G. L., et al. 2004, ApJ, 600, 580 Green, P. J. 1996, ApJ, 467, 61 Green, P. J. 2006, ApJ, 644, 733 Green, P. J., Infante, L., Lopez, S., Aldcroft, T. L., & Winn, J. N. 2005, ApJ, 630, 142 Green, A. R., McHardy, I. M., & Lehto, H. J. 1993, MNRAS, 265, 664 Green, P. J., et al. 1995, ApJ, 450, 51 Green, P. J., et al. 2001, ApJ, 558, 109 Green, P. J., et al. 2002, ApJ, 571, 721 Green, P. J., et al. 2004, ApJS, 150, 43 Grindlay, J. E., et al. 2005, ApJ, 635, 920 Grupe, D., Wills, B. J., Leighly, K. M., & Meusinger, H. 2004, AJ, 127, 156 Hewett, P. C., & Foltz, G. B. 2003, AJ, 125, 1784 Hewett, P. C., Foltz, C. B., & Chaffee, F. H. 1995, AJ, 109, 1498 Hasinger, G. 2005, in Growing Black Holes: Accretion in a Cosmological Context, ed. A. Merloni, S. Nayakshin, & R. Sunyaev (Berlin: Springer) Hennawi, J. F., et al. 2006, AJ, 131, 1

Hopkins, P. F., Hernquist, L., Cox, T. J., Di Matteo, T., Robertson, B., & Springel, V. 2006, ApJS, 163, 1 Hopkins, P. F., Hernquist, L., Cox, T. J., Di Matteo, T., Robertson, B., & Volker, S. 2005, ApJ, 630, 705 Hopkins, P. F., et al. 2008, ApJS, 175, 356 Isobe, T., Feigelson, E. D., Akritas, M., & Babu, G. J. 1990, ApJ, 364, 104 ´ Ivezic, Z., et al. 2002, AJ, 124, 2364 ¨ Jester, S., Kording, E., & Fender, R. 2006, VI Microquasar Workshop: Microquasars and Beyond ´ Jiang, L., Fan, X., Ivezic, Z., Richards, G. T., Schneider, D. P., Strauss, M. A., & Kelly, B. C. 2007, ApJ, 656, 680 Just, D. W., Brandt, W. N., Shemmer, O., Steffen, A. T., Schneider, D. P., Chartas, G., & Garmire, G. P. 2007, ApJ, 665, 1004 Kelly, B. C. 2007, ApJ, 665, 1489 Kelly, B. C., Bechtold, J., Siemiginowska, A., Aldcroft, T., & Sobolewska, M. 2007, ApJ, 657, 116 Kelly, B. C., Bechtold, J., Trump, J. R., Vestergaard, M., & Siemiginowska, A. 2008, ApJS, 176, 355 Kim, D.-W., et al. 2004, ApJS, 150, 19 Kim, M., Wilkes, B. J., Kim, D.-W., Green, P. J., Barkhouse, W. A., Lee, M. G., Silverman, J. D., & Tananbaum, H. D. 2007a, ApJ, 659, 29 Kim, M., et al. 2007b, ApJS, 169, 401 Knigge, C., Scaringi, S., Goad, M. R., & Cottis, C. E. 2008, MNRAS, 386, 1426 La Franca, F., et al. 2005, ApJ, 635, 864 LaValley, M., Isobe, T., & Feigelson, E. D. 1992, ADASS, 1, 245 Lamer, G., Schwope, A., Wisotzki, L., & Christensen, L. 2006, A&A, 454, 493 Lehmer, B. D., et al. 2007, ApJ, 657, 681 Maccarone, T. J., Gallo, E., & Fender, R. 2003, MNRAS, 345, L19 Mainieri, V., et al. 2007, ApJS, 172, 368 Marconi, A., Risaliti, G., Gilli, R., Hunt, L. K., Maiolino, R., & Salvati, M. 2004, MNRAS, 351, 16 Martel, A. R., Menanteau, F., Tozzi, P., Ford, H. C., & Infante, L. 2007, ApJS, 168, 19 ´ Middleton, M., Done, C., & Gierlinski, M. 2007, MNRAS, 381, 1426 Misawa, T., Eracleous, M., Chartas, G., & Charlton, J. C. 2008, ApJ, 677, 863 Myers, A. D., et al. 2006, ApJ, 638, 622 Myers, A. D., et al. 2007, ApJ, 658, 99 Myers, A. D., et al. 2008, ApJ, 678, 635 Oke, J. B., & Gunn, J. E. 1983, ApJ, 266, 713 Osterbrock, D. E., & Pogge, R. W. 1985, ApJ, 297, 166 Page, C. G. 2006, Astron. Data Anal. Softw. Syst. XV, 351, 85 Page, K. L., Reeves, J. N., O'Brien, P. T., Turner, M. J. L., & Worrall, D. M. 2004, MNRAS, 353, 133 Perola, G. C., et al. 2004, A&A, 421, 491 Porquet, D., Reeves, J. N., O'Brien, P., & Brinkmann, W. 2004, A&A, 422, 85 Ptak, A., & Griffiths, R. 2003, Astron. Data Anal. Softw. Syst. XII, 295, 465 Reeves, J. N., & Turner, M. J. L. 2000, MNRAS, 316, 234 Reichard, T. A., et al. 2003, AJ, 126, 2594 Reynolds, C. S., & Nowak, M. A. 2003, Phys. Rep., 377, 389 Richards, G. T., et al. 2001, AJ, 121, 2308 Richards, G. T., et al. 2002, AJ, 123, 2945 Richards, G. T., et al. 2004, ApJS, 155, 257 Richards, G. T., et al. 2005, MNRAS, 360, 839 Richards, G. T., et al. 2006a, AJ, 131, 2766 Richards, G. T., et al. 2006b, ApJS, 166, 470 Richards, G. T., et al. 2008, ApJS, in press (arXiv:0809.3952R) Saez, C., et al. 2008, AJ, 135, 1505 Scannapieco, E., Silk, J., & Bouwens, R. 2005, ApJ, 635, L13 Schmitt, J. H. 1985, ApJ, 293, 178 Schneider, D. P., Schmidt, M., & Gunn, J. E. 1994, AJ, 107, 1245 Schneider, D. P., et al. 2007, AJ, 134, 102 Shastri, P., Wilkes, B. J., Elvis, M., & McDowell, J. 1993, ApJ, 410, 29 Shemmer, O., Brandt, W. N., Netzer, H., Maiolino, R., & Kaspi, S. 2006, ApJ, 646, L29 Shemmer, O., Brandt, W. N., Netzer, H., Maiolino, R., & Kaspi, S. 2008, ApJ, 682, 81 Shen, S., White, S. D. M., Mo, H. J., Voges, W., Kauffmann, G., Tremonti, C., & Anderson, S. F. 2006, MNRAS, 369, 1639 Siemiginowska, A., Cheung, C. C., LaMassa, S., Burke, D. J., Aldcroft, T. L., Bechtold, J., Elvis, M., & Worrall, D. M. 2005, ApJ, 632, 110 Silverman, J. D., et al. 2005a, ApJ, 618, 123 Silverman, J. D., et al. 2005b, ApJ, 624, 630 Silverman, J. D., et al. 2008, ApJ, 679, 118 ´ Sobolewska, M. A., Gierlinski, M., & Siemiginowska, A. 2008, MNRAS, submitted Sobolewska, M. A., Siemiginowska, A., & Zdotycki, P. T. 2004, ApJ, 617, 102


No. 1, 2009

PROPERTIES OF SDSS QUASARS IN THE ChaMP

669

Steffen, A. T., Strateva, I., Brandt, W. N., Alexander, D. M., Koekemoer, A. M., Lehmer, B. D., Schneider, D. P., & Vignali, C. 2006, AJ, 131, 2826 Steffen, A. T., et al. 2007, ApJL, 667, L25 Strateva, I., Brandt, W. N., Alexander, D. M., Koekemoer, A. M., Lehmer, B. D., Schneider, D. P., & Vignali, C. 2006, AJ, 131, 2826 Strateva, I. V., Brandt, W. N., Schneider, D. P., Vanden Berk, D. G., & Vignali, C. 2005, AJ, 130, 387 Tajer, M., et al. 2007, A&A, 467, 73 Tang, S., Zhang, S. N., & Hopkins, P. F. 2007, MNRAS, 377, 1113 Treister, E., & Urry, C. M. 2006, ApJ, 652, L79 Trump, J. R., et al. 2006, ApJS, 165, 1 Ueda, Y., Akiyama, M., Ohta, K., & Takamitsu, M. 2003, ApJ, 598, 886 Vignali, C., Brandt, W. N., & Schneider, D. P. 2003, AJ, 125, 433 Vignali, C., Brandt, W. N., Schneider, D. P., & Kaspi, S. 2005, AJ, 129, 2519 Weinstein, M. A., et al. 2004, ApJS, 155, 243 Weymann, R. J., Morris, S. L., Foltz, C. B., & Hewett, P. C. 1991, ApJ, 373, 23

White, R. L., Helfand, D. J., Becker, R. H., Glikman, E., & de Vries, W. 2007, ApJ, 654, 99 Wilkes, B. J., Tananbaum, H., Worrall, D. M., Avni, Y., Oey, M. S., & Flanagan, J. 1994, ApJS, 92, 53 Williams, R. J., Mathur, S., & Pogge, R. W. 2004, ApJ, 610, 737 Wilman, R. J., & Fabian, A. C. 1999, ApJ, 522, 157 Wolf, C., Wisotzki, L., Borch, A., Dye, S., Kleinheinrich, M., & Meisenheimer, K. 2003, A&A, 408, 499 Worrall, D. M., Tananbaum, H., Giommi, P., & Zamorani, G. 1987, ApJ, 313, 596 Wyithe, J. S. B., & Loeb, A. 2002, ApJ, 581, 886 Wyithe, J. S. B., & Loeb, A. 2005, ApJ, 634, 910 York, D. G., et al. 2000, AJ, 120, 1579 Yu, Q., & Tremaine, S. 2002, MNRAS, 335, 965 Yuan, W., Brinkmann, W., Siebert, J., & Voges, W. 1998, A&A, 330, 108 Zamorani, G., et al. 1981, ApJ, 245, 357