THE FIFTH DATA RELEASE SLOAN DIGITAL SKY SURVEY XMM-NEWTON

THE FIFTH DATA RELEASE SLOAN DIGITAL SKY SURVEYXMM-NEWTON QUASAR SURVEY

M Young12

M Elvis1 and G Risaliti

131 Harvard-Smithsonian Center for Astrophysics 60 Garden Street Cambridge MA 02138 USA myoungcfaharvardedu

2 Boston University Astronomy Department 725 Commonwealth Avenue Boston MA 02215 USA3 INAF-Osservatorio di Arcetri Lgo E Fermi 5 Firenze Italy

Received 2008 October 17 accepted 2009 May 7 published 2009 June 16

ABSTRACT

We present a catalog of 792 Fifth Data Release Sloan Digital Sky Survey quasars with optical spectra that have beenobserved serendipitously in the X-rays with the XMM-Newton These quasars cover a redshift range of z = 011ndash541 and a magnitude range of i = 153ndash207 Substantial numbers of radio-loud (70) and broad absorption line (51)quasars exist within this sample Significant X-ray detections at 2σ account for 87 of the sample (685 quasars)and 473 quasars are detected at 6σ sufficient to allow X-ray spectral fits For detected sources sim60 haveX-ray fluxes between F2minus10 keV = (1ndash10) times10minus14 erg cmminus2 sminus1 We fit a single power law a fixed power law withintrinsic absorption left free to vary and an absorbed power-law model to all quasars with X-ray signal-to-noiseratio 6 resulting in a weighted mean photon index Γ = 191 plusmn 008 with an intrinsic dispersion σΓ = 038 Forthe 55 sources (116) that prefer intrinsic absorption we find a weighted mean NH = 15 plusmn 03 times 1021 cmminus2We find that Γ correlates significantly with optical color Δ(g minus i) the optical-to-X-ray spectral index (αox) and theX-ray luminosity While the first two correlations can be explained as artifacts of undetected intrinsic absorptionthe correlation between Γ and X-ray luminosity appears to be a real physical correlation indicating a pivot in theX-ray slope

Key words accretion accretion disks ndash quasars general

Online-only material color figure machine-readable tables

1 INTRODUCTION

Catalogs are indispensable in performing statistical studies ofquasar properties The known correlations between optical andX-ray properties discussed further below imply a connectionbetween the accretion disk posited to feed the central black hole(Shakura amp Sunyaev 1973) and the hot Compton-scatteringcorona posited to lie in some unknown geometry around thedisk (Haardt amp Maraschi 1991) While quasars were firstdiscovered in early radio surveys (eg 3C 3CR PKS 4CAO) most quasar surveys since then have been conductedin the opticalUV Early quasar selection techniques includedUV-excess selection and emission line searches using slitlessspectroscopy

UV-excess selection utilizes the Big Blue Bump that dom-inates the opticalUV spectrum to distinguish quasars fromstars The Big Blue Bump is normally attributed to a multi-temperature accretion disk (Shields 1978 Malkan amp Sargent1982) UV-excess-based surveys include the Braccesi et al(1970) catalog which contains 175 objects with U minus B ltminus042 and the Palomar-Green (PG) Bright Quasar Survey(BQS Schmidt amp Green 1983) a sample of 92 objects se-lected with U minus B lt minus044 Slitless spectroscopy with prismsor grisms obtains a large number of low-resolution spectra fora single field For example the objective prism on the Cur-tis Schmidt telescope at CTIO found 174 confirmed quasarsranging in redshift from z = 01 to z = 33 (Osmer amp Smith1980 Osmer 1981) Slitless spectroscopy often incorporatedUV-excess selection as well the primary example of this tech-nique is the Markarian survey (Markarian 1967 Markarian et al1981) which searched for galaxies with unusually blue continuausing a grism The Large Bright Quasar Survey (LBQS Hewettet al 1995) used a combination of color selection and the pres-ence of emission features on objective prism plates to obtain a

homogenous sample of 1055 quasars spanning a wide redshiftrange (02 z 34)

However both of these techniques suffer from serious biasesWhile slitless spectroscopy has a high selection efficiencyeven at higher redshifts it is biased against quasars with weakemission lines and cannot reach as faint a flux limit The UV-excess selection method is biased against ldquoredrdquo sources wherered colors may be due to high redshift (the Lyα line entersthe spectrum at z sim 2) dust reddening significant host galaxycontribution or intrinsically red emission mechanisms Withthe advent of the ldquoUK Schmidt Surveyrdquo (Warren et al 1991)the Two-Degree Field (2dF Croom et al 2001) and the SloanDigital Sky Survey (SDSS York et al 2000) quasar catalogsmulticolor selection techniques that used up to five photometricbands were introduced that could select red quasars in additionto blue ones provided the sources were within the survey fluxlimit The Fifth Data Release (DR5) SDSS quasar catalog hassurpassed all previous optical surveys by providing high-qualityphotometry and spectroscopy for 77429 quasars (Schneideret al 2007) spanning redshifts from z = 008 to z = 541

As quasars emit strongly over eight decades of the spectrum(eg Elvis et al 1994) multiwavelength surveys are necessaryto relate the opticalUV accretion emission to other componentsnotably the non-thermal emission seen in the X-rays HoweverX-ray spectra are time consuming and expensive to obtain forlarge samples For this reason previous studies of optical and X-ray correlations consist largely of two types (1) small samples(N sim 20ndash50) with detailed X-ray spectral analysis have beencompiled by observing subsamples of optical surveys with X-raytelescopes (eg Laor et al 1997 Elvis et al 1994 Piconcelliet al 2005 Shemmer et al 2006 2008) The Akylas et al(2004) XMM-Newton2dF survey is larger with 96 2QZ quasarsobserved in wide field (25 deg2) shallow (2ndash10 ks per pointingf(05ndash8 keV) sim 10minus14 erg cmminus2 sminus1) XMM-Newton observations

17

18 YOUNG ELVIS amp RISALITI Vol 183

(2) Still larger samples (N sim 200ndash300) with only X-ray fluxeshave been compiled for statistical investigations These largerstudies have to assume an X-ray spectral slope (eg Vignaliet al 2003 Strateva et al 2005 Steffen et al 2006)

The ever-expanding archive of X-ray observations now pro-vides a less costly and time-consuming method of obtainingX-ray spectra for large numbers of optically selected quasarsKelly et al (2007) cross correlated the SDSS DR3 quasar cat-alog with archival Chandra observations (Weisskopf 1999) toobtain 174 quasars 44 of which have sufficient counts to fit anabsorbed power law with Γ and NH as free parameters This sam-ple was extended to a total of 318 quasars (Kelly et al 2008) byadding 149 RQ objects from an SDSSndashROSAT cross-correlation(Strateva et al 2005) As only single power-law models were fitto the ROSAT spectra with sufficient counts 153 of 318 sourceshave X-ray spectra slopes but no fits for intrinsic absorptionwere made Archival Chandra observations have also been usedto identify an X-ray-selected active galactic nucleus (AGN) pop-ulation in the Chandra Multiwavelength Project (Green et al2003) In addition a large sample (N = 1135) of optically se-lected SDSS quasars with photometric redshifts are detectedin the X-rays in Chandra fields (Green et al 2009) of which156 have sufficient counts (gt200) to fit an absorbed power lawwith Γ and NH as free parameters

XMM-Newton is a good choice for cross-correlating withoptical catalogs due to its large field of view and large effectivearea Images are made with the three European Photon ImagingCameras (EPIC) MOS-1 and MOS-2 (Turner et al 2001) eachof which has a 33prime times 33prime field of view and the pn camera(Strueder et al 2001) which has a 27prime5 times 27prime5 field of view4

XMM-Newtonrsquos large effective area (922 cm2 for MOS and1227 cm2 for PN at 1 keV)4 results in higher signal-to-noiseX-ray spectra than with Chandra for bright non-background-limited sources5 The EPIC CCDs have good spectral resolution(EΔE sim 20ndash50 for both MOS and PN) over the 05ndash10 keVband

Archival XMM-Newton observations overlapped with sim1of the SDSS DR5 coverage as of 2007 February The SDSSand XMM-Newton archives are well matched in sensitivityas discussed in Young et al (2008) The DR5 (Adelman-McCarthy et al 2007) of the SDSS covers a spectroscopicarea of 5740 deg2 and contains 90611 quasars The SDSSphotometry in the u g r i and z bands covers 3250ndash10000 Aringwhile the spectroscopy covers 3800ndash9200 Aring with a spectralresolving power of sim2000 The SDSS is sim95 complete forpoint sources to a limiting magnitude i = 191 corrected forGalactic reddening (Richards et al 2002 Vanden Berk et al2005)

A preliminary cross correlation of DR1 SDSS quasars withthe XMM-Newton public archive yielded 55 objects with expo-sure times greater than 20 ks (Risaliti amp Elvis 2005) Of these35 yielded good X-ray spectra Risaliti amp Elvis (2005) estimatedthat a cross-correlation of a final SDSS data release with the ever-growing XMM-Newton archive would yield sim1000 quasars ofwhich sim80 would have good X-ray spectra

In this paper we cross-correlate the SDSS DR5 quasarcatalog with archival XMM-Newton observations to obtain alarge (N sim 800) sample of quasars with X-ray detectionsalmost 500 of which have good optical and X-ray spectral dataBelow we outline two immediate goals for the SDSSXMM-

4 httpimaginegsfcnasagovdocssats_n_datamissionsxmmhtml5 httpheasarc nasagovdocsxmmuhbnode86html

Newton Quasar Survey (1) to conduct large statistical studies tounderstand the physical basis behind opticalX-ray correlationsand (2) to investigate interesting subpopulations of quasars

11 OpticalX-ray Correlations

The relations between optical and X-ray continuum and spec-tral properties promise to reveal clues about the disk-coronastructure of quasars The large sample provided by the SDSSXMM-Newton Quasar Survey allow two correlations to be inves-tigated The first of these is the controversial αoxndashl2500 Aring relationmany studies have found that the spectral index from 2500 Aringto 2 keV defined as αox = log(L2 keVL2500) log(ν2 keVν2500)anticorrelates with the log of the monochromatic luminosityat 2500 Aring l2500 Aring (Tananbaum et al 1979 1986 Zamoraniet al 1981 Avni amp Tananbaum 1982 Kriss amp Canizares 1985Anderson amp Margon 1987 Wilkes et al 1994 Pickering et al1994 Avni et al 1995 Vignali et al 2003 Strateva et al 2005Shen et al 2006 Steffen et al 2006 Just et al 2007) How-ever some studies (Bechtold et al 2003 Kelly et al 2007)find the primary relation to be between αox and redshift whileother studies (Yuan et al 1998 Tang et al 2007) find that thecorrelation may be induced by selection effects

No physical basis for the αoxndashl2500 Aring relation has yet beenproposed and the relation itself provides little guidance In partthis is because previous studies have largely used the traditionalobservationally convenient but physically arbitrary endpointsof 2500 Aring and 2 keV They also assume an X-ray photon index(Γ sim 2 where Γ = minusα + 1 for Fν prop να) to obtain the X-ray fluxat 2 keV A systematic study with measured optical and X-rayspectra would enable an investigation of the relation at differentfrequencies than those traditionally used hopefully revealingclues about the relationrsquos physical underpinnings (M Younget al 2009 in preparation)

The second correlation is the positive relation between Γ andthe normalized accretion rate LLEdd (Shemmer et al 20062008) While early studies focused on the relation between Γ andfull width at half-maximum of the Hβ line FWHM(Hβ) (Bolleret al 1996 Laor et al 1997 Brandt et al 1997) Shemmer et al(2006 2008) have broken the degeneracy between FWHM(Hβ)and LLEdd by including highly luminous sources in order toshow that Γ depends primarily on LLEdd There have beensuggestions that this dependence may be due to the disk emittingmore and softer photons as accretion rates increase leadingto more efficient Compton cooling in the corona (Laor 2000Kawaguchi et al 2001)

Until now studies of the ΓndashLLEdd relation have consistedof small samples (N sim 40) The SDSSXMM-Newton quasarsurvey can increase this sample size by an order of magnitudeleading to a better defined relation (Risaliti et al 2009)

12 Quasar Subpopulations

Interesting quasar subpopulations can be readily investigateddue to the large number of sources in the SDSSXMM-NewtonQuasar Survey For example red quasars make up 6 ofthe SDSS sample (Richards et al 2003) Their steep opticalslopes have been attributed to dust reddening (Richards et al2003 Hopkins et al 2004) though observations of individualobjects suggest that some slopes may be intrinsically steep(Risaliti et al 2003 Hall et al 2006) In Young et al (2008)we studied 17 quasars with extreme red colors (g minus r gt05) and moderate redshifts (1 lt z lt 2) By using X-rayobservations in conjunction with optical spectra we constrained

No 1 2009 THE SDSSXMM-NEWTON QUASAR SURVEY 19

the amount of intrinsic absorption in each source therebyallowing the separation of intrinsically red from dust-reddenedoptical continua We find that almost half (7 of 17) of thequasars can be classified as probable ldquointrinsically redrdquo objectsThese quasars have unusually broad Mg ii emission lines(〈FWHM〉 = 10500 km sminus1) flat but unabsorbed X-ray spectra(〈Γ〉 = 166 plusmn 008) and low accretion rates (M ˙MEdd sim 001)

Other interesting subpopulations for future investigationsinclude broad absorption line (BAL) Type 2 and radio-loud(RL) quasars

In this paper we describe the SDSS quasar selection andthe method with which we match sources to XMM-Newtonobservations in Section 2 X-ray data reduction is described inSection 3 and the resulting sample and correlations are discussedin Section 4 We assume a standard cosmology throughout thepaper where H0 = 70 km sminus1 Mpcminus1 ΩM = 03 and ΩΛ = 07(Spergel et al 2003)

2 DATA

21 SDSS Quasar Selection

As the DR5 quasar catalog (Schneider et al 2007) wasnot yet available at the time of selection we selected quasarswith optical spectra directly from the DR5 SDSS database6

by choosing SpecClass = 3 (QSO) or 4 (QSO with z gt 23whose redshift has been confirmed using a Lyα estimator)These quasars were selected for spectroscopic follow-up bythe SDSS primarily due to their photometric colors althoughsome quasars were selected because they have a match inthe FIRST survey (White et al 1997) No X-ray selectionis involved Target selection efficiency ie the percentageof sources spectroscopically observed that are confirmed asquasars is 66 (Richards et al 2002 Vanden Berk et al 2005)The selection method is described in detail by Richards et al(2002) and is summarized briefly below

The vast majority (sim95) of SDSS quasar candidates aredetected using a multicolor selection technique Quasar candi-dates are defined to be any object at least 4σ away from thestellar locus which is defined in the (u minus g g minus r r minus i i minus z)color space In addition special colorndashcolor regions are definedto specifically include or exclude quasar candidates Inclusionregions include quasar candidates from 25 lt z lt 3 even iftheir colors cross the stellar locus Exclusion regions preventcontamination due to white dwarfs A stars and M starndashwhitedwarf pairs

For both radio and color-selected quasar candidates magni-tude limits are applied Quasar candidates brighter than i = 15are rejected for spectroscopic follow-up because bright sourcescan contaminate the spectra of objects in adjacent fibers in theSDSS spectrograph Radio and low-redshift color-selected can-didates fainter than i = 191 which have a high number densityon the sky are also rejected due to the limited number of opticalfibers for follow-up spectroscopy Since high-redshift (z 3)quasar candidates have a lower surface density on the sky afainter cutoff magnitude i = 202 is applied to these objects

SDSS selected a small number of quasar candidates (sim5)by matching point sources with the position of a radio detectionfrom the FIRST survey to within 2primeprime While the DR5 quasarcatalog (Schneider et al 2007) includes sources selected bymatching to ROSAT detections the initial SDSS quasar selectionoutlined in Richards et al (2002) does not use X-ray detectionas a criterion6 httpwwwsdssorgdr5accessindexhtml

While SDSS radio selection requires that a source be point-like their color selection includes extended sources as well inorder to include low-redshift AGN such as Seyfert galaxiesHowever in addition to having colors distinct from the stellarlocus extended sources must also have colors distinct from themain galaxy distribution (The main galaxy distribution overlapsthe stellar locus however a galaxy can be a clear outlier fromthe stellar locus both due to the shape of the stellar locus andbecause the stellar locus is determined primarily from F and Mstars that dominate the Galactic stellar density at high latitudes)Simple color cuts are applied to distinguish extended sourcesfrom galaxies rather than an additional multicolor selection

The DR5 quasar catalog (Schneider et al 2007) additionallyrequires luminosities brighter than Mi = minus220 at leastone emission line with an FWHM greater than 1000 km sminus1

or interestingcomplex absorption features magnitudes fainterthan i = 150 and highly reliable redshifts This results ina catalog of 77429 quasars We used only those quasars thatare also in the DR5 Quasar Catalog for further analysis Of the92 sources we reject from our sample sim40 are Type 2 quasarsThese will be the subject of a later study

We matched the selected SDSS quasars with the FIRSTsurvey (White et al 1997)7 using a search radius of 3primeprime inorder to calculate their radio-loudness (RL = F5 GHzF4400Kellerman et al 1989) A source is taken to be RL if RL gt 10 Apower law is interpolated between the optical magnitudes to getFλ(4400 Aring) and the 14 GHz radio flux is obtained from FIRSTsurvey detections which are extrapolated to 5 GHz using aradio power law αR = minus08 All the quasars lie in the areacovered by the FIRST survey so if there is no detection we usethe 5σ upper limit on the 14 GHz radio flux to extrapolate to5 GHz For 40 quasars the upper limit is too high to determinewhether the source is RL Of the remaining quasars 70 (93)are RL and 682 (907) are radio-quiet (RQ) a typical ratio(eg Kembhavi amp Narlikar 1999 pp 256ndash263)

The SDSS colorndashcolor selection is effective at finding avariety of BAL quasars (Schneider et al 2007) The official DR5BAL catalog (Gibson et al 2009) had not yet been publishedduring the writing of this paper so we used an incompletelist of sim4200 BALs in the DR5 Quasar Catalog (Shen et al2008) selected using traditional criteria (Weymann et al 1991)Using Shen et al (2008) we find 52 BALs in the SDSSXMM-Newton Quasar Survey Of these 15 BALs have high enoughX-ray signal to noise to obtain X-ray spectra (see Section 3)Since BALs are normally identified by C iv absorption featureswhich are visible only above a redshift z = 16 in SDSS spectraBALs are unlikely to be identified in sim56 of the SDSSXMM-Newton quasars BALs make up 145 of the SDSSXMM-Newton quasars with z gt 16 which is in line with the parentpopulation in Shen et al (2008)

In this paper we measure the optical color of a quasarby its relative color (Richards et al 2003) Relative colorscompare a quasarrsquos measured colors with the median colorsin its redshift bin where redshift bin sizes are 01 in redshiftso that Δ(g minus i) = (g minus i) minus 〈(g minus i)〉z The use of relativecolors corrects for the effect of typical emission lines on thephotometry in a particular band The relative (g minus i) colors ofthe SDSS quasars match a Gaussian distribution on the blue sidebut require the addition of a tail on the red side (see Figure 3 inRichards et al 2003)

7 httpsundogstsciedu

22 Matching with the XMM-Newton Archive

We matched the SDSS quasars with the XMM-Newton archivefrom 2007 February choosing only those quasars that fellwithin 14prime of XMM-Newton observation field centers (typically2ndash3 quasars per field) As part of the extraction processdescribed below a source region is defined around each set ofSDSS quasar coordinates Depending on the SN of the sourcethe extraction radius can range from 10primeprime to 85primeprime with a typicalradius of 19primeprime Low SN objects are extracted with smaller radiito minimize the effect of high background levels while largerradii for high SN objects allow for an increased encircledenergy fraction in the presence of relatively low backgroundlevels The large extraction radii take into account the XMM-Newton point-spread function (PSF) which is characterized bythe radius at which 90 of the total energy is encircled Thisradius increases from 48primeprime (MOS) and 51primeprime5 (PN) at 0prime off-axisangle to 52primeprime5 (MOS) and 66primeprime (PN) at 12prime off-axis angle8 Theextraction radii also take positional accuracy into account SDSSpositional errors are negligible (0primeprime1 at the survey limit of r = 22for typical seeing Pier et al 2003) while the XMM-Newtonpositional accuracy is 3primeprime at 3σ for offset angles 0 lt θ lt 10arcmin and 6primeprime for 5 lt θ lt 10 arcmin (Pierre et al 2007)

Multiple X-ray observations exist for 265 sources In thesecases all observations were retrieved and reduced but onlythe observation with the longest exposure time was used forfurther analysis To avoid biases we did not select the highestSN observations although in sim80 cases the two selectionswould be effectively the same

The chance of including an unrelated random source withinthe extraction region is small but non-zero Within a sim14prime radiusfield of view and using the average extraction radius 19primeprime thereare 1954 ldquobeamsrdquo in an observation Since there are sim70 sourcesin a typical XMM-Newton observation (Watson 2008 whichhas an exposure time distribution similar to that in this paper)this results in a 36 chance of extracting a random sourcerather than the SDSS-selected quasar Since a random sourceis likely to be faint in the X-rays any contamination is onlysignificant for sources under 100 net counts We have 792 uniquesources of which 390 have less than 100 net counts so sim14(2 of the SDSSXMM-Newton quasar sample) have significantcontamination from an unrelated ldquointerloperrdquo source

23 X-ray Data Reduction

The 582 X-ray observations were processed using the XMM-Newton Science Analysis System SAS v7029 We reprocessedthe events to ensure that each observation has the same up-to-date calibration and then filtered the observations to removetime intervals of flaring high-energy background events usingthe standard cutoff of 035 counts sminus1 for the MOS cameras and10 counts sminus1 for the PN camera10 Source and backgroundregions were defined in a semiautomatic process The SAStask eregionanalyse was used to optimize the source extractionradius for signal to noise Most radii include at least 80 ofthe source counts Background regions were defined by eyeavoiding obvious X-ray sources and chip edges These regionswere typically a circle of radius 2000ndash2500 pixels (100primeprimendash125primeprime) selected to lie at the same off-axis angle as the sourceand as close to the source as possible without overlapping

8 httpxmmvilspaesaesexternalxmm_user_supportdocumentationuhbnode17html9 httpxmmesacesaintsas10 httpxmmvilspaesaessas710documentationthreads

the source extraction region Once source and backgroundregions were defined for every SDSS quasar in an observationspectra were extracted for a total of 1380 non-unique quasars in582 observations

To check for biases in the data reduction Figures 1(a) and(b) show the net counts and the X-ray photon index (Γ) plottedagainst the extraction radius Figures 1(c)ndash(d) plot the off-axisangle and the X-ray SN against Γ In Figure 1(a) the netcounts are expected to correlate with the extraction radius sincethe extraction radius will increase to larger encircled energyfraction for sources that stand higher above the backgroundThe lack of correlations in Figures 1(b) and (c) show that theencircled energy correction takes the extraction radius and off-axis angle into account correctly Figure 1(d) shows that objectswith flat Γ are primarily found among low SN objects whereabsorption may be undetected in a spectral fit

Where possible observations were processed for all threeXMM-Newton EPIC CCDs In sim40 of the observations asource lies in a bad region in one or two of the three cameraseither in a strip between two chips or because the MOS andPN cameras have different shapes outside the field of view inone of the cameras In these cases we use the remaining imagesfrom the other cameras for analysis

Table 1 contains observational data for each quasar in theDR5 SDSSXMM-Newton Survey the SDSS name XMM-Newton observation ID redshift Galactic column density inthe direction of the source X-ray signal to noise observationexposure time off-axis angle net source counts backgroundcounts (from a background region that is scaled to the area ofthe source region) and two flags indicating if a source is RL orBAL

Figure 2 summarizes the survey characteristics The X-rayexposure times (Figure 2(a)) range from 16 to 294 ks though themajority of observations lie between 20 and 100 ks This rangein exposure times results in a wide range in sensitivity Whilemost sources have low signal to noise a significant fraction haveSN 10 where more complex models can be fit (Figure 2(b))The detection fraction is 80 until z gt 35 (Figure 2(c)) butspectral coverage drops off fairly quickly for sources with z gt 2(Figure 2(d))

3 X-RAY ANALYSIS SPECTRAL FITS

We made fits to the extracted spectra using the Sherpa pack-age11 within CIAO12 For each source the available MOS+PNspectra were fit simultaneously over the 05ndash10 keV band Theobservations were fit according to their SN with more com-plicated models being applied as SN increased All the modelsincluded local absorption fixed to the Galactic hydrogen col-umn density (NHgal) at each source location Values for NHgal

were taken from the NH tool available at WebPIMMS13 whichis based on the 21 cm H i compilation of Dickey amp Lockman(1990) and Kalberla et al (2005)

For the 319 low SN sources (SN 6) we fix a powerlaw to the weighted mean obtained for the high SN (SN 6) quasars in the sample Γ sim 19 (Section 41) and allowonly the normalization to vary in order to obtain the fluxFor the 101 sources with SN lt 2 we obtain a 90 upperlimit to the flux We use the Cash (1979) statistic which givesmore reliable results for low-count sources to fit sources with

11 httpcxcharvardedu sherpathreadsindexhtml12 httpcxcharvardeduciao13 httpheasarcgsfcnasagovToolsw3pimmshtml

Figure 1 (a) Net counts vs source extraction radius (arcsec) (b) Γ vs source extraction radius (c) X-ray photon index (Γ) vs off-axis angle (arcmin) (d) Γ vs SN

Table 1SDSS Quasars with Archival XMM-Newton Observations

SDSS Name XMM-Newton ID z NHgala (SN)X

b Exp Θd Net Background RL BALTimec Countse Countsf flagg flagh

SDSS J12593097+2827055 0204040101 1094 092 180 2210 134 469 1030 0 0SDSS J13012013+2821372 0204040101 1369 094 458 2210 132 2780 4490 1 0SDSS J11485656+5254252 0204260101 1633 137 230 94 58 620 521 1 0SDSS J21541970minus0917536 0204310101 1212 371 125 813 109 218 435 0 0SDSS J16422122+3903334 0204340101 1713 122 113 457 119 182 391 0 0SDSS J02100022minus1003542 0204340201 1960 220 132 316 109 243 470 1 0SDSS J02110083minus0951384 0204340201 0767 217 144 316 118 268 406 0 0SDSS J12350819+3930200 0204400101 0968 149 99 655 48 153 428 0 0SDSS J12352736+3928240 0204400101 2158 149 201 898 60 553 1040 0 0SDSS J13381297+3915271 0204651101 0439 086 187 231 80 416 384 0 0

Notes The upper limits are at the 16σ levela Galactic hydrogen column density (1020 cmminus2) in the direction of the sourceb X-ray signal-to-noise ratioc Exposure time (ks)d Off-axis angle (arcmin)e Net source countsf Background counts (counts in background region scaled to source extraction area)g Radio-loud flag is 0 for RQ quasars 1 for RL quasars and 2 if the radio upper limit is too high to determine whether source is RLh BAL flag is 0 for non-BAL quasars 1 for BAL quasars

(This table is available in its entirety in a machine-readable form in the online journal A portion is shown here for guidance regardingits form and content)

SN 6 As the background is not subtracted and is insteadfit simultaneously with the source we apply a backgroundmodel with three components as described in Lumb et al(2002) and in the XMM-Newton Users Handbook14 a powerlaw (for the extragalactic X-ray spectrum) a broken power law

14 httpxmmvilspaesaesexternalxmm_user_ supportdocumentation

(for the quiescent soft proton spectrum) and two spectral lines(for cosmic-ray interactions with the detector) All parametersexcept for the normalization of the spectral lines were fixed For13 undetected sources and 5 detected sources this fit resultsin bad or null flux values In these cases we flag the source asundetected (flag = minus1 in Column 9 of Table 2) and we list theαox values as 999 (Column 8 of the same table)

Figure 2 Survey characteristics (a) exposure time histogram where exposure times are summed over up to three EPIC cameras Exposure times do not includehigh-background intervals filtered out during data reduction (b) SN histogram where 16 sources have X-ray SN gt 100 (c) Fraction of detected sources vs redshiftand (d) redshift histogram For (c) and (d) the open dotted-line histogram is for all SDSS-selected quasars the open solid-line histogram is for all detected sourcesthe hatched histogram is for sources with X-ray SN gt 6 and the double-hatched histogram is for sources with X-ray SN gt 10

For the 473 sources with enough SN to fit spectral parameters(SN 6) we use the χ2 statistic to fit three models a singlepower law (SPL) with no intrinsic absorption a fixed powerlaw (FPL) where intrinsic absorption is left free to vary andan intrinsically absorbed power law (APL) The F-test15 whichmeasures the significance of the change in χ2 as componentsare added to a model is used to determine whether the dataprefer the APL model To compare the SPL and FPL modelswe simply compare the respective χ2

ν values since both modelshave the same number of parameters Therefore the best-fitmodel is the one preferred by the F-test that also has the lowestχ2

ν valueFor sources with an unacceptable χ2

ν values for all threemodels we by default make the SPL model the best fit butwe also list the APL 90 upper limit on intrinsic absorptionWe plot the reduced χ2 distribution in Figure 3 A later paperwill look in more detail at sources with bad fits (χ2

ν gt 12)Results for the best-fit models are listed in Table 2 including

observed-frame and rest-frame fluxes (or 90 upper limits) αoxa flag indicating the best-fit X-ray spectral model the photonindex intrinsic absorption (or 90 upper limit) and the χ2

values and degrees of freedom for the best-fit model The best-fit flag indicates which values are listed for each best-fit modelFor sources that prefer the APL model (flag = 3) both Γ andNH are from the APL fit If the FPL model is preferred (flag =2) the SPL Γ and FPL NH are listed For sources that prefer the

15 httpcxcharvardeduciaoahelpftesthtml

Figure 3 χ2ν histogram The open histogram represents sources that prefer a

power-law model The solid histogram represents sources that prefer an absorbedpower-law model Five sources have χ2

ν gt 2 and are not included in the plot

SPL model (flag = 1) Γ is from the SPL fit and NH is the 90upper limit from the APL fit

4 RESULTS AND DISCUSSION

The SDSSXMM-Newton Quasar Survey contains 792 sources685 of which are detected in the X-rays and 473 of which haveX-ray spectra (All have optical spectra) The catalog coversredshifts z = 011ndash541 and optical magnitudes range from i= 153 to i = 207 Figures 4 shows the survey sensitivity in

Figure 4 X-ray and optical sensitivity (a) optical 2500 Aring luminosity vs redshift and (b) X-ray 2ndash10 keV luminosity vs redshift Flux limits are plotted in both plotsFor (a) i band magnitude limits (dotted lines) were set during selection (Richards et al 2002) For (b) the flux limit (log F2minus10 keV = minus140 erg cmminus2 sminus1 dottedline) is for an observation of 20 ks (Watson et al 2001) (c) The observed-frame F2minus10 keV distribution

Table 2X-ray Spectral Data of SDSS Quasars

SDSS Name f05minus2a f2minus10

a f05minus2b f2minus10

b L05minus2c L2minus10

c αox Fit Γe NHf χ2νg

(obs) (obs) (rest) (rest) (rest) (rest) flagd

SDSS J12593097+2827055 094 520 041 246 027 161 minus177 1 099+013minus014 lt310 38928

SDSS J21541970minus0917536 151 238 093 210 079 177 minus212 1 187+029minus027 lt068 5012

Notes Luminosities are computed using H0 = 70 km sminus1 Mpcminus1 ΩM = 03 and ΩΛ = 07 The upper limits are at the 16σ levela Observed-frame X-ray flux in the observed band is given in units of 10minus14 erg cmminus2 sminus1b Rest-frame X-ray flux in the soft and hard bands is given in units of 10minus14 erg cmminus2 sminus1c Rest-frame X-ray luminosity in the soft and hard bands is given in units of 1044 erg sminus1d A flag indicating the X-ray fit An undetected source is flagged as minus1 and upper-limit flux values are listed A detected source withSN lt 6 is flagged as 0 and detected flux values are listed For sources with SN gt 6 three models can be applied If a single power-lawmodel (SPL) is preferred the flag = 1 and the SPL parameters are listed as well as the 90 upper limit on intrinsic absorption from theAPL model If an FPL model with variable NH is preferred the flag = 2 The best-fit slope from the SPL is listed as well as the best-fitNH from the FPL model If the APL model is preferred the flag = 3 and the APL power law and absorption parameters are listede Photon index for the best-fit model (If the APL model is preferred then the photon index from that model is quoted otherwise thephoton index is from the SPL model)f Intrinsic absorption or 90 upper limit in units of 1022 cmminus2g The χ2 value and degrees of freedom for the best-fit model

Figure 5 (a) l2 keV vs l2500 The dotted line shows l2 keV prop l2500 constrained to pass through the average values of l2 keV and l2500 (b) The αoxndashl2500 relation for theSDSSXMM-Newton quasar sample The linear fit to the SDSSXMM-Newton sample is plotted as a solid line and the Steffen et al line is overplotted as a dotted linefor comparison (c) αox vs redshift Red circles represent detected sources with X-ray SN lt 6 and blue circles represent sources with X-ray SN gt 6 BALs and RLquasars were removed from all three plots

the optical (a) and X-ray (b) bands and the observed-frame2ndash10 keV flux distribution (c) The FWHM of the distributionspans F2minus10 keV = (1ndash10) times 10minus14 erg cmminus2 sminus1 and containssim60 of the detected sources

Results of standard αox analysis (ie using the conventional2500 Aring and 2 keV fiducial points) are shown in Figure 5We test for correlations with αox using the Kendall correlationtest available in ASURV a survival statistics package (Lavalleyet al 1992) Figure 5(a) shows log L2 keV versus log L2500 witha dotted line of slope unity for reference The best-fit lineobtained from the EM (estimate and maximize) algorithm withinASURV is flatter than unity with a slope of 064 plusmn 003This deviation from unity is clear in Figure 5(b) which showsthe αoxndashl2500 Aring correlation significant at the 114σ level in theSDSSXMM-Newton Quasar Survey The solid line is the best-fit EM regression line for our data

αox = (3080 plusmn 0376) + (minus0153 plusmn 0012) log(L2500 Aring)

The slope is consistent within errors with the best-fit regressionfrom Steffen et al (2006) (dotted line in Figure 5(b))

Figure 5(c) plots αox against redshift We find no correlationbetween αox and redshift as in R05 The lack of an αoxndashzcorrelation agrees with some previous studies (Vignali et al2003 Strateva et al 2005 Risaliti amp Elvis 2005 Steffen et al2006) but not all (Bechtold et al 2003 Shen et al 2006 Kellyet al 2007) In particular Kelly et al (2007) apply a moresophisticated statistical analysis by allowing for nonlinear fitswith multiple variables As a result they find that αox depends onboth l2500 and redshift so that quasars become more X-ray loudat low luminosities and higher redshifts We have applied only

linear regressions in this paper and will apply more complexstatistics in a later publication

The weighted mean of the X-ray spectral slope for the473 sources with an X-ray SN 6 is 〈Γ〉 = 191 plusmn 008with a standard deviation of 040 eg Shen et al (2006) Thetypical 1σ error on Γ is 015 resulting in an intrinsic dispersionσΓ = 037 Figure 6 shows the αox and Γ distributions

We calculate the percentage of sources with significantintrinsic absorption by testing whether sources with SN 6 prefer the APL model over the SPL model with F-testprobability PF gt 095 and acceptable χ2

ν resulting in 34 (72)absorbed sources However this method is biased against lowSN sources since the APL model has more parameters thanthe SPL model Since the FPL and SPL models have the samenumber of parameters we count sources as absorbed if theyprefer either the FPL or APL models with PF gt 095 andacceptable χ2

ν This method gives 55 sources (116) that areintrinsically absorbed which is comparable to the percentagefound among Type 1 quasars in previous surveys (Mateos et al2005 Green et al 2009) The XMM-Newton COSMOS surveyobtained a higher percentage (20 Mainieri et al 2007) usinga lower confidence threshold (PF gt 09) when applying theF-test When we redo our method using the same confidencethreshold we find 82 absorbed sources (174)

Nevertheless the amount of absorption in this sample is onlya lower limit for two reasons First undetected absorption maystill exist particularly in low SN sources For example in thecorrelation plots (eg Figures 7 9 and 11) three sources canbe seen with Γ sim 05 These sources do not formally prefereither absorption model but their X-ray spectra are cut off at

Figure 6 (a) αox distribution for all sources (b) Γ distribution for sources with SN gt 6 Open histograms represent RQ non-BAL quasars and have been scaleddown by factors of 3 and 4 respectively Histograms with vertical hatching represent RL quasars and histograms with diagonal hatching represent BAL quasars Thespace above the upper limit arrows indicates the number of RQ+non-BAL upper limits in a given bar

Figure 7 Top X-ray photon index (Γ) vs monochromatic X-ray luminosity at rest frame (a) 07 keV (b) 2 keV (c) 10 keV and (d) 20 keV In (b) the Green et al(2009) correlation is plotted for comparison as a dashed line and the quasars from Dai et al (2004) are overplotted as open black squares Only RQ non-BAL quasarsare plotted (blue circles) The weighted least-squares regression line is calculated using RQ non-BAL quasars and is plotted as a solid line with errors shown asdotted lines Bottom the weighted mean Γ values are given for bins of width Δ log L2minus10 keV = 05 Error lines mark the 1σ dispersions in both axes Blue linesrepresent RQ+non-BAL quasars while green lines represent RL+non-BAL quasars

(A color version of this figure is available in the online journal)

soft energies suggesting absorption as the likely cause of flatspectra Second we cannot test for absorption in sources withSN lt 6 and it is possible that these low SN sources have ahigher percentage of absorption

For the 55 sources that prefer absorption we calculatethe weighted mean of the intrinsic column density NH =15 plusmn 03 times 1021 cmminus2 The absorbed population consists of6 BAL objects 7 RL and 42 RQ non-BAL objects

Figure 8 The slope of the ΓndashLX correlation vs energy E with 1σ errors on theslope for (a) all sources and (b) sources where the rest-frame energy lies in theobserved range The slopes of the ΓndashLX correlations are obtained for E = 071 15 2 4 7 10 and 20 keV and are marked as solid circles The Green et al(2009) correlation slope is marked as an open circle at 2 keV A dashed linemarks zero

Table 3 summarizes the weighted means of αox Γ and NHfor the three subsets of quasar populations RQ+non-BALRL+non-BAL and BAL (There are four quasars that are bothRL and BALs) Because of the small numbers of absorbedspectra in the subpopulations comparing the respective NHdistributions is not meaningful but αox and Γ KolmogorovndashSmirnov (KS) tests show that the RQ distribution is significantlydifferent from the RL distribution with probabilities PKS lt05 that the two samples are drawn from the same parentdistribution (Figure 6) RL quasars are known to be brighter inthe X-rays for a given 2500 Aring luminosity (eg Zamorani et al1981) However while RL quasars are also known to have flatterX-ray slopes than RQ quasars (Wilkes amp Elvis 1987 Williamset al 1992 Reeves et al 1997 Reeves amp Turner 2000 Pageet al 2005 Piconcelli et al 2005) the average Γ value for RLquasars in the SDSSXMM-Newton sample (〈Γ〉 = 185plusmn004)is steeper than that found in previous samples (〈Γ〉 sim 15ndash175)Figure 6(b) shows that RL quasars do follow the same Gaussiandistribution as RQ quasars for Γ lt 19 but for Γ gt 19 the RLdistribution falls off rapidly Only 7 of 49 RL quasars (14)have Γ gt 2 compared to 48 of RQ quasars

The Γ distribution is not significantly different for BAL versusnon-BAL quasars (PKS = 75) once the X-ray spectra arecorrected for intrinsic absorption However the αox distributionof BAL quasars which includes 37 sources with SN lt 6that cannot be corrected for absorption remains significantlydifferent (PKS = 02) from non-BAL quasars Previous studieshave noted the difference in X-ray brightness for BAL quasars(eg Green amp Mathur 1996)

41 Correlations with the X-ray Spectral Slope

The large number of X-ray spectra in the SDSSXMM-Newton Quasar Survey allows us to test for correlations withthe X-ray photon index (Γ) Since we use only sources detectedwith a high enough SN to fit a power law we do not usesurvival statistics and instead use the Kendallrsquos rank correlationcoefficient to test for correlations We use only those sourcesthat do not prefer either absorption model and we also requirethat the SPL model result in a reasonably good fit (χ2

ν lt 12)In addition we restrict the selection to those sources with SNgt 10 in order to reduce the effect of undetected absorption inlower SN spectra

When correlations are significant we plot a Weighted Least-Squares (WLS) regression line to take into account the mea-surement errors in Γ which are much larger than the errors inthe independent variable RQ quasars show two correlations ofΓ with luminosity and optical color with probabilities less than05 that they are due to chance (1) Γ versus log L2 keV (PK =24eminus5) and (2) Γ versus Δ(g minus i) (PK = 75eminus4) where PKis the Kendall rank two-sided significance level for each cor-relation RQ quasars also show a marginal correlation betweenΓ and αox (PK = 16) The ΓndashL2 keV ΓndashΔ(g minus i) and Γndashαoxrelations are not significant (PK gt 10) for RL quasars (greenstars in Figures 7 9 and 11) We now discuss each of thesecorrelations in turn

411 X-ray Slope versus X-ray Luminosity

An anticorrelation exists between the X-ray slope (Γ) and the2 keV luminosity (log L2 keV) hereafter l2 keV such that X-rayslopes harden as X-ray luminosity increases (Figure 7(a)) TheWLS regression line (for RQ non-BAL quasars) is

Γ = (639 plusmn 13) + (minus016 plusmn 005)l2 keV (1)

To investigate the possibility of a pivot point in the X-ray spec-trum we next examined seven additional correlations betweenthe X-ray slope (Γ) and the monochromatic X-ray luminositiesat 07 1 15 4 7 10 and 20 keV The monochromatic fluxes areobtained for sources preferring the SPL model by normalizingthe fit at each energy in turn The results of these correlationsare summarized in Figures 7(a)ndash(d) and in Table 4 The strengthof the correlation increases with energy as the slope steepensso the strongest steepest correlation is between Γ and l20 keV(Figure 7(d)) Correlation strength decreases at lower energiesbottoming out at 1 keV where the slope is consistent with zeroAt 07 keV the slope flips to a positive correlation and thestrength of the correlation increases again The flip in the corre-lation slope indicates a pivot in the X-ray spectrum near 1 keV(Figure 8(a))

For the correlations above we have used Γ to extrapolatethe rest-frame monochromatic luminosities for sources withredshifts out of range of the observed spectrum To check thatthis extrapolation does not affect the correlations we performthe fits again this time excluding any sources where the rest-frame luminosity is not in the observed range The results areshown in Table 5 and plotted in Figure 8(b) The errors arelarger due to the smaller sample sizes and the narrower rangeof luminosities observed but the spectrum still pivots between1 and 15 keV At the lowest energies the slopes become muchsteeper possibly due to the influence of a soft excess componenton the power-law fit

The Γ determined from the SPL model is used to determinethe monochromatic flux so it is important to check that the

Figure 9 Top (a) Γ vs αox with the Γndashαox correlation found by Green et al (2009) plotted for comparison as a dashed line (b) Γ vs Δ(g minus i) the relative (g minus i)color for sources with z lt 23 Symbols for (a) and (b) are as in Figure 7 though the correlations are determined only for RQ non-BAL and non-absorbed quasarsWeighted least-square regressions are plotted as solid lines with 3σ errors shown as dotted lines (c) αox vs relative (g minus i) color for sources with z lt 23 Symbolsfor (c) are as in Figure 5 We plot the OLS regression as a solid line with 1σ errors shown as dotted lines Bottom the weighted mean Γ values are given for bin widthsas follows (a) Δαox = 02 (b) Δ[Δ(g minus i)] = 03 and (c) Δ[Δ(g minus i)] = 04 Error lines mark the 1σ dispersions in both axes Blue lines represent RQ+non-BALquasars while green lines represent RL+non-BAL quasars Figure (c) does not include BAL or RL quasars

Table 3Weighted Averages of X-ray Spectral Quantities

Ndeta αox Nspec

b Γ σΓc σΓintr

d NHe

All 685 minus160 473 190 plusmn 002 040 037 015 plusmn 003RQ+non-BAL 589 minus161 411 191 plusmn 008 041 038 014 plusmn 004RL+non-BAL 62 minus146 47 185 plusmn 003 030 026 015 plusmn 003BAL 34 minus178 15 196 plusmn 005 037 034 23 plusmn 06

Notesa Number of detected quasars in each sample (SN 2)b Number of quasars in each sample with X-ray spectra (SN 6)c Observed dispersion of X-ray sloped Intrinsic dispersion of X-ray slopee Intrinsic absorption in units of 1022 cmminus2

model assumptions do not induce the observed anticorrelationTo do this we re-fit the sources this time with a power lawfixed to the sample mean (Γ = 191) and intrinsic absorptionfixed to zero We then obtain the monochromatic flux at 2 keVand compare the F2 keV obtained via a fixed power law tothe F2 keV obtained via the best-fit power law We find thatchanging the model assumptions changes the log flux values by14 which is not enough to explain the observed correlationat 2 keV

Even with the selection restricted to sources with SN gt10 there is still the possibility of undetected absorption Forexample Figures 7 show two sources with Γ sim 05 and anotherfour RQ sources with Γ sim 1ndash15 The X-ray spectra of thesesources show curvature in the soft X-rays that while notsignificant enough for the sources to prefer an absorption modelnevertheless suggests that intrinsic absorption is the likelycause of flat X-ray slopes Therefore we test for correlationsagain this time using sources with SN gt 20 in order to

Figure 10 (a) Intrinsic absorption (NH) vs Γ (b) NH vs αox (c) NH vs Δ(g minus i) The solid circles in (a) and (b) represent sources preferring the APL model while in(c) the solid circles represent sources preferring either absorption model (FPL or APL) The arrows represent sources with only a 90 upper limit on NH NH is givenin log units of 1022 cmminus2

Figure 11 X-ray photon index (Γ) vs (a) optical 2500 Aring monochromatic luminosity and (b) redshift Blue circles represent RQ non-BAL quasars green starsrepresent RL quasars and red triangles represent BAL quasars Filled symbols are given for those sources that prefer an absorbed power law with an F-test probabilityP gt 095 The bottom plots show the weighted mean Γ values for bins of width (a) Δ log L2500 = 05 and (b) Δz = 1

minimize the chance of undetected absorption The significanceof the correlation increases with the probability of a chancecorrelation falling below PK = 1eminus7 However with manylow-luminosity sources eliminated the correlation is biased toa steeper slope Having shown that the correlations do not relyon undetected absorption we continue to use the correlationsfor sources with SN gt 10 for further analysis

The slope of the Γndashl2 keV anticorrelation is equivalent withinerrors to the slope found in Green et al (2009) They also finda correlation between Γ and the 2 keV luminosity based on 156

RQ and RL quasars fit with a power law plus intrinsic absorptionplus 979 quasars with lower SN that were fit with an SPL andzero intrinsic absorption A similar anticorrelation between Γand 2ndash10 keV luminosity was found in Page et al (2005) for16 RL quasars While the Page et al (2005) correlation was notfound to be significant for RQ quasars there were only seven RQquasars in their sample However a previous study by Dai et al(2004) found the opposite correlation between Γ and l2minus10 keVat 986 significance for a sample of 10 quasars observed withChandra and XMM-Newton The Γ and extrapolated l2 keV values

Table 4ΓndashLX Correlations

E (keV)a PKb Z-levelc Slope Intercept Dispersion

07 00062 273 0095 plusmn 0058 minus0500 plusmn 1564 03610 01683 016 minus0015 plusmn 0054 2462 plusmn 1449 03715 00033 177 minus0106 plusmn 0051 4903 plusmn 1363 03720 00000 293 minus0162 plusmn 0048 6386 plusmn 1286 03740 00000 561 minus0265 plusmn 0041 9042 plusmn 1082 03570 00000 750 minus0316 plusmn 0035 1028 plusmn 0917 033100 00000 860 minus0335 plusmn 0032 1073 plusmn 0818 031200 00000 961 minus0365 plusmn 0027 1138 plusmn 0690 029

Notesa Energy at which monochromatic luminosity is takenb Kendallrsquos probability that correlation is due to chancec Significance level of Kendall rank correlation coefficient in units of 1σ

Table 5ΓndashLX Correlations with Redshift Cutoffs

E (keV)a PKb Sigc Slope Intercept Dispersion

07 00112 254σ 0583 plusmn 0285 minus1329 plusmn 7587 04610 00386 143σ 0196 plusmn 0129 3077 plusmn 3447 04415 00024 207σ minus0120 plusmn 0058 5270 plusmn 1363 03820 00000 303σ minus0162 plusmn 0053 6800 plusmn 1417 03840 00000 564σ minus0178 plusmn 0044 9596 plusmn 1176 03570 00000 749σ minus0337 plusmn 0038 1084 plusmn 0986 033100 00000 860σ minus0355 plusmn 0034 1125 plusmn 0874 031200 00000 703σ minus0293 plusmn 0129 9560 plusmn 3342 022

Notesa Energy at which monochromatic luminosity is takenb Kendallrsquos probability that correlation is due to chancec Significance level of Kendall rank correlation coefficient

for the Dai et al (2004) quasars are plotted in Figure 7(a) asopen black squares where they are consistent with the trendobserved in SDSSXMM-Newton quasars16 The SDSSXMM-Newton Quasar Survey increases the sample size of previousstudies by factors of 3ndash30 and covers sim3 decades of X-rayluminosity

To explain the observed correlations between Γ and X-rayluminosity we must answer the following two questions

1 Why does the X-ray slope change with luminosity2 Why does the slope change such that the pivot point is near

1 keVWe address two possible answers to the first question First

an increased hard component at higher X-ray luminosities mayexplain the trends observed here The hard component maybe due to non-thermal emission associated with a jet (egZamorani et al 1981 Wilkes amp Elvis 1987) or due to a reflectioncomponent (Krolik 1999) RL quasars do not show a significantcorrelation between Γ and LX except for marginal correlationsat high energies (7 and 10 keV) but this may be because ajet already dominates the X-ray emission The spectrum of thehard component may flatten the X-ray spectrum by covering upthe steeper power law due to inverse Compton scattering whilesimultaneously increasing the 2ndash10 keV luminosity

Alternatively quasars with low X-ray luminosities may havesteeper slopes due to a component linked to high accretion rates

16 More recently Saez et al (2008) find the same correlation to be significantto gt995 for 173 bright RQ quasars in the Chandra Deep Fields Howeverthe trends found in Saez et al (2008) are dominated by Type 2 quasarsparticularly at low X-ray luminosities As we have excluded Type 2 quasarsthe two samples are not in conflict

As discussed in Section 11 previous studies have found a strongcorrelation between Γ and the Eddington ratio (LbolLEdd)where steeper sources are associated with higher accretion ratesTherefore sources with high accretion rates (and steep X-rayspectra) must be associated with low X-ray luminosity in orderto produce the ΓndashLX correlations However it is not clear if highaccretion rates and steep X-ray spectra tend to coincide withlow X-ray luminosity Studies of X-ray binary (XRB) accretionstates have shown that high accretion states tend to be associatedwith steep X-ray spectra and high X-ray flux compared to the lowaccretion state (Remillard amp McClintock 2006) However thedistinction is not clear in every sourcemdasheven ldquohighrdquo accretionstates can be associated with both low and high X-ray fluxNarrow-line Seyfert 1 (NLS1) galaxies are an extreme exampleof high accretion rate objects (Boroson 2002) but while theytypically have steep X-ray slopes their X-ray flux can rangefrom X-ray bright to weak (eg αox = 07ndash22 Leighly 1999)The ΓndashLbolLEdd relation for the SDSSXMM-Newton QuasarSurvey will be discussed further in Risaliti et al (2009)

Inverse Compton scattering of UV photons from an accretiondisk in a hot corona could explain why the X-ray spectrum pivotsat low X-ray energies In the corona model the X-ray spectrumwill change shape if the temperature (Te) andor optical depth(τ ) of the corona vary (Rybicki amp Lightman 1979) since theoutput spectrum flattens as the y parameter increases The yparameter is the average fractional energy gained by a photonfor a thermal non-relativistic electron distribution

y = 4kTe

mec2max(τ τ 2) (2)

For example if the disk emission brightens increasing thesoft photon supply the corona will increase radiative coolingto maintain temperature balance producing a steeper X-rayspectrum For a T sim 108ndash109 K corona opacity is notnecessarily dependent on luminosity so opacity variationscan result in a pivot in the 2ndash10 keV band without a largeaccompanying change in luminosity (Haardt et al 1997)

Because of the relation between the physical parameters andthe output spectrum it is possible to calculate Teτ though thedegeneracy is not breakable without knowing the cutoff of thehigh-energy spectrum From Krolik (1999 p 227ndash8)

Teτ sim a(lh ls)14 sim a

minus16

Γ minus 1

where a is a coefficient dependent on geometry (a = 006for slabs) and lh ls is the compactness ratio equivalent to theheating rate over the soft photon seed supply Therefore as Γchanges from 23 at low luminosities to 18 at high luminositiesTeτ approximately doubles from 416 times 108 to 713 times 108 K

Several studies have found evidence of pivot points inindividual objects The previous paragraph describes the modelof Mrk766 a NLS1 which was found to pivot near sim10 keV(Haardt et al 1997) In Cygnus X-1 a black hole X-ray binaryZdziarski et al (2002) find a negative correlation between X-ray flux and Γ for the 015ndash12 keV band no correlation for the20ndash100 keV band and a positive correlation for flux greaterthan 100 keV implying a pivot point near sim50 keV Zdziarskiet al (2002) model this spectrum as a variable supply of softseed photons irradiating a thermal plasma where optical depthis constant An increase in seed flux results in a decrease inthe corona temperature to satisfy energy balance resulting ina steeper spectrum with a pivot somewhere below 100 keVTwo model-independent analyses of the spectral variability inSeyfert 1 galaxies also show evidence of pivot points NGC 4051was found to have a pivot near 50 keV using correlationsbetween the 2ndash5 keV and 7ndash15 keV fluxes (Taylor et al 2003)and NGC 5548 studied with an 8-day BeppoSAX observation(Nicastro et al 2000) can be entirely explained by a pivot inthe medium energy band (sim6 keV) While these examples showthat pivot points may exist in X-ray spectra it is clear thatfurther exploration must be done to provide a clear picture ofthe underlying physics

412 X-ray Slope versus Optical Color and αox (for RQ QuasarsOnly)

X-ray slope (Γ) and optical color [Δ(g minus i)] are correlatedat 34σ such that quasars with redder colors are more likelyto have flat X-ray slopes (Figure 9(a)) This is likely due toundetected absorption which is discussed further below Weexclude sources with redshifts greater than 23 from the ΓndashΔ(g minus i) correlation as Lyα absorption artificially reddens high-redshift quasars The linear bisector is

Γ = (204 plusmn 002) + (minus045 plusmn 010)Δ(g minus i) (3)

The ΓndashΔ(g minus i) correlation is consistent with previous findingsof a correlation between Γ and the opticalUV spectral slope(αuv) (Kelly et al 2007) because color correlates tightly withthe optical slope (Richards et al 2003 Δ(g minus i) prop 05αuv) Theslope of the correlation in this paper is in the same direction asbut steeper than in Kelly et al (2007) who find a slope = minus025plusmn007

X-ray slope (Γ) anticorrelates with αox at marginal (24σ )significance such that X-ray faint quasars are more likely to

have flat X-ray slopes (Figure 9(b)) The WLS regression line is

Γ = (28 plusmn 02) + (06 plusmn 01)αox (4)

This anticorrelation between Γ and αox contrasts with theslightly positive correlation found by Green et al (2009)(overplotted as the almost horizontal dashed line in Figure 8(b))The Green et al (2009) correlation may be affected by theinclusion of RL quasars which lie at the X-ray bright end andtypically have flatter X-ray slopes than RQ quasars

Completing the triangle of relations a correlation is foundbetween αox and Δ(g minus i) at 47σ significance (Figure 9(c))Again sources with redshifts greater than 23 are excluded inorder to avoid contamination by the Lyα forest Since the αoxndashΔ(g minus i) relation includes censored data we use the EM methodto give the best-fit line

αox = (minus1613 plusmn 0009) + (minus016 plusmn 004)Δ(g minus i) (5)

The EM algorithm assumes that all of the error lies in thedependent variable so the algorithm reduces to an OLS Y versusX fit for uncensored data

In an attempt to disentangle the relationships between ΓΔ(g minus i) and αox we perform the Kendall partial correlationtest for sources with SN gt 6 and z lt 23 but each correlationis significant at the sim3σ level even when the third variable istaken into account The dispersions in all three relations arelarge so the sources are binned along the x-axis to show therelationships more clearly The binning shows that the relationsare dominated by outliers optically red and X-ray weakflatsources

Detected absorption cannot explain the observed correlationswith Γ since sources are only included if they do not prefera model with intrinsic X-ray absorption Γ is not correlatedwith NH (Figure 10(a)) for sources that prefer the APL modelindicating that once the X-ray slope is corrected for absorptionno intrinsic correlation between X-ray slope and absorptionremains For the same sources αox and NH (Figure 10(b))correlate at the 33σ level which is surprising since the 2 keVflux should also be corrected for absorption Since the detectionstrace the upper limits this correlation is likely not intrinsic butis due to the survey flux limit The optical color also depends onabsorption (Figure 10(c)) at the 26σ level as is expected sinceX-ray absorbed sources are more likely to have redder colorsdue to dust reddening

Undetected absorption is a possible explanation for all threerelations causing red optical colors and X-ray weakness whileflattening the X-ray spectral slope in low SN spectra A simplecalculation shows that the amounts of absorption obtained fromEquations (3) and (4) are consistent with observed gas-to-dustratios observed for quasars First we assume that both relationsare driven entirely by the effects of absorption (ie ignoring anypossible effects due to intrinsic properties such as black holemass and accretion rate) We also assume that an unabsorbedquasar has a typical X-ray slope (Γ = 19) zero absorption(NH = 0 and EBminusV = 0) and typical blue optical colors(Δ(g minus i) = 0) If the X-ray slope changes by a given amountwe can calculate the change in color by Equation (3) whichcorresponds to a dust reddening (Richards et al 2003) whichin turn leads to a change in optical luminosity Since the X-rayslope also gives αox by Equation (4) we can calculate the changein X-ray luminosity at 2 keV which in turn gives the intrinsicX-ray absorption So for an absorbed quasar with Γ = 14 thedust reddening is E(B minusV ) = 0096 from Equation (3) the gas

column is NH sim 6times1022 cmminus2 from Equation (4) giving a gas-to-dust ratio sim100 times the Galactic value Quasars typicallyhave gas-to-dust ratios in the range of 10ndash100 times the Galacticvalue (Maccacaro et al 1981 Maiolino et al 2001 Wilkeset al 2002) so relations (3) and (4) are consistent with intrinsicabsorption that remains undetected in the X-ray spectra

Intrinsic absorption is less likely to remain undetected inspectra with high SN As a second test we restrict thecorrelation tests to those sources with X-ray SN gt 20 Asa result all three correlations disappear which again supportsundetected absorption as the cause of correlations that includelow SN sources

413 X-ray Slope versus Redshift and Optical Luminosity

Neither redshift nor 2500 Aring luminosity correlate significantlywith Γ (PS gt 10 Figures 11) confirming previous studieswith smaller samples (Page et al 2004 Risaliti amp Elvis 2005Shemmer et al 2005 Vignali et al 2005 Kelly et al 2007)Note however Bechtold et al (2003) who find that X-ray slopesare flatter at lower redshifts for a sample of 17 RQ high-redshift (37 lt z lt 63) quasars observed with Chandra TheSDSSXMM-Newton sample (473 sources with X-ray spectra)is at least an order of magnitude larger than these previoussamples though it does not have homogenous spectral coveragefor redshifts above 25

5 CONCLUSIONS AND FUTURE WORK

We have cross correlated the DR5 SDSS Quasar Catalog withthe XMM-Newton archive creating a sample of 792 quasars witha detection rate of 87 Almost 500 quasars have X-ray spectrathe largest sample available for analysis of opticalX-ray spec-tral correlations We find that the X-ray photon index Γ corre-lates significantly with lX where X = 2 4 7 10 and 20 keV Op-tical color and αox also correlate with Γ but these correlations arelikely due to the effect undetected intrinsic absorption rather thanintrinsic physical changes The X-ray slope does not correlatesignificantly with redshift or optical luminosity With a samplesize at least an order of magnitude larger than previous studieswe confirm a highly significant correlation between αox and themonochromatic luminosity at 2500 Aring and we also confirm thatαox does not correlate significantly with redshift or optical color

Future studies of the sample will include αox Γ versus op-tical properties and studies of subpopulations such as BALsand Type 2 quasars Variability studies will also be pursuedfor 265 objects with multiple XMM-Newton observations Morecomplex spectral analysis on high X-ray SN sources will in-clude a thermal component for the soft excess warm absorbersand emission line detection

The authors thank the anonymous referee for insightful com-ments that improved this paper The authors also thank Gor-don Richards for his excellent assistance in navigating theSDSS database This paper is based on observations obtainedwith XMM-Newton an ESA science mission with instrumentsand contributions directly funded by ESA Member States andNASA and the Sloan Digital Sky Survey (SDSS) Funding forthe SDSS and SDSS-II has been provided by the Alfred P SloanFoundation the Participating Institutions the National Sci-ence Foundation the US Department of Energy the NationalAeronautics and Space Administration the Japanese Monbuk-agakusho the Max Planck Society and the Higher EducationFunding Council for England This research also made use of

the NASA IPAC Infrared Science Archive which is operatedby the Jet Propulsion Laboratory California Institute of Tech-nology under contract with the National Aeronautics and SpaceAdministration This work has been partially funded by NASAgrants NASA NNX07AI22G and NASA GO6-7102X

REFERENCES

Adelman-McCarthy et al 2007 ApJS 172 634Akylas A Georgakakis A amp Georgantopoulos I 2004 MNRAS 353 1015Anderson S F amp Margon B 1987 ApJ 314 111Avni Y amp Tananbaum H 1982 ApJ 262 L17Avni Y Worrall D M amp Margon W A 1995 ApJ 454 673Bechtold J et al 2003 ApJ 588 119Boller T Brandt W N amp Fink H 1996 AampA 305 53Boroson T A 2002 ApJ 565 78Braccesi A Formiggini L amp Gandolfi E 1970 AampA 5 264Brandt W N Mathur S amp Elvis M 1997 MNRAS 285 L25Cash W 1979 ApJ 228 939Croom S M Smith R J Boyle B J Shanks T Loaring N S Miller L

amp Lewis I J 2001 MNRAS 322 L29Dai X Chartas G Eracleous M amp Garmire G P 2004 ApJ 605 45Dickey J M amp Lockman F J 1990 ARAampA 28 215Elvis M et al 1994 ApJS 95 1Gibson R R et al 2009 ApJ 692 758Green P amp Mathur S 1996 ApJ 462 637Green et al 2003 Astron Nachr 324 93Green et al 2009 ApJ 690 644Haardt F amp Maraschi L 1991 ApJ 380 L51Haardt F Maraschi L amp Ghisellini G 1997 ApJ 476 620Hall P Gallagher S Richards G T Alexander D M Anderson S F Bauer

F Brandt W N amp Schneider D P 2006 AJ 132 1977Hewett P C Foltz C B amp Chaffee F H 1995 AJ 109 1498Hopkins P et al 2004 AJ 128 1112Just D W Brandt W N Shemmer O Steffen A T Schneider D P Chartas

G amp Garmire G P 2007 ApJ 665 1004Kalberla et al 2005 AampA 440 775Kawaguchi T Shimura T amp Mineshige S 2001 ApJ 546 966Kellerman K Sramek R Schmidt M Shaffer D amp Green R 1989 AJ 98

1195Kelly B C Bechtold J Siemiginowska A Aldcroft T amp Sobolewska M

2007 ApJ 657 116Kelly B C Bechtold J Trump J R Vestergaard M amp Siemiginowska A

2008 ApJS 176 355Kriss G A amp Canizares C R 1985 ApJ 297 177Krolik J H 1999 Active Galactic Nuclei From the Central Black Hole to the

Galactic Environment (Princeton NJ Princeton Univ Press) 198Laor A 2000 New Astron Rev 44 503Laor A Fiore F Elvis M Wilkes B J amp McDowell J C 1997 ApJ 477

93Lavalley M Isobe T amp Feigelson E 1992 in ASP Conf Ser 25 Astro-

nomical Data Analysis Software and Systems I ed D M Worrall CBiemesderfer amp J Barnes 245

Leighly K M 1999 ApJS 125 317Lumb D H Warwick R S Page M amp De Luca A 2002 AampA 389 93Maccacaro T Perola G C amp Elvis M 1981 ApJ 246 L11Mainieri et al 2007 ApJS 172 368Maiolino R Marconi A Salvati M Risaliti G Severgnini E O La Franca

F amp Vanzi L 2001 AampA 365 28Malkan M amp Sargent W 1982 ApJ 254 22Markarian B E 1967 Astrofizika 3 55Markarian B E Lipovetskii V A amp Stepanyan D A 1981 Astrophysics

17 321Mateos et al 2005 AampA 433 855Narlikar J V 1999 Quasars and Active Galactic Nuclei An Introduction

(Cambridge Cambridge Univ Press)Nicastro F et al 2000 ApJ 536 718Osmer P S 1981 ApJ 247 762Osmer P S amp Smith M G 1980 ApJS 42 333Page K L Reeves J N OrsquoBrien P T amp Turner M J L 2005 MNRAS

364 195Page K L Reeves J N OrsquoBrien P T Turner M J L amp Worrall D M

2004 MNRAS 353 133Pickering T E Impey C D amp Foltz C B 1994 AJ 108 1542

Piconcelli E Jimenez-Bailn E Guainazzi M Schartel N Rodrguez-Pascual P M amp Santos-Lleo M 2005 AampA 432 15

Pier J R Munn J A Hindsley R B Hennessy G S Kent S M LuptonR H amp Izezic Z 2003 AJ 125 1559

Pierre M et al 2007 MNRAS 382 279Reeves J N amp Turner M J L 2000 MNRAS 316 234Reeves J N Turner M J L Ohashi T amp Kii T 1997 MNRAS 292 468Remillard R A amp McClintock J E 2006 ARAampA 44 49Richards et al 2002 AJ 123 2945Richards et al 2003 AJ 126 1131Risaliti G amp Elvis M 2005 ApJ 629 L17Risaliti G Elvis M Gilli R amp Salvati M 2003 ApJ 587 L9Risaliti G Young M amp Elvis M 2009 ApJL submittedRybicki G B amp Lightman A P 1979 Radiative Processes in Astrophysics

(New York Wiley) 208Saez C Chartas G Brandt W N Lehmer B D Bauer F E Dai X amp

Garmire G P 2008 AJ 135 1505Schmidt M amp Green R F 1983 ApJ 269 352Schneider D P et al 2007 AJ 134 102Shakura N I amp Sunyaev R A 1973 AampA 24 337Shemmer O Brandt W N Netzer H Maiolino R amp Kaspi S 2006 ApJ

646 L29Shemmer O Brandt W N Netzer H Maiolino R amp Kaspi S 2008 ApJ

682 81Shemmer et al 2005 ApJ 630 729Shen Y Strauss M A Hall P B Schneider D P York D G amp Bahcall

N A 2008 ApJ 677 858Shen S White S D M Mo H M Voges W Kauffmann G Tremonti C

amp Anderson S F 2006 MNRAS 369 1639Shields G A 1978 Nature 272 706Spergel D N et al 2003 ApJS 148 175

Steffen A Strateva I Brandt W Alexander D Koekemoer A Lehmer BSchneider D amp Vignali C 2006 AJ 131 2826

Strateva I V Brandt W N Schneider D P Vanden Berk D G amp VignaliC 2005 AJ 130 387

Strueder L et al 2001 AampA 365 L18Tananbaum H Avni Y Green R F Schmidt M amp Zamorani G 1986 ApJ

305 57Tananbaum H et al 1979 ApJ 234 L9Tang S M Zhang S N amp Hopkins P 2007 MNRAS 377 1113Taylor R D Uttley P amp McHardy I M 2003 MNRAS 342 L31Turner M et al 2001 AampA 365 L27Vanden Berk D E et al 2005 AJ 129 2047Vignali C Brandt W N amp Schneider D P 2003 AJ 125 433Vignali C Brandt W N Schneider D P amp Kaspi S 2005 AJ 129 2519Warren S J Hewett P C Irwin M J amp Osmer P S 1991 ApJS 76 1Watson M G 2008 Astron Nachr 329 131Watson M et al 2001 AampA 365 L51Weisskopf M 1999 arXivastroph9912097v1Weymann R J Morris S L Foltz C B amp Hewett P C 1991 ApJ 373 23White R Becker R Helfand D amp Gregg M 1997 ApJ 475 479Wilkes B J amp Elvis M 1987 ApJ 323 243Wilkes B J Schmidt G D Cutri R M Ghosh H Hines D C Nelson

B amp Smith P S 2002 ApJ 564 L65Wilkes B J Tananbaum H Worrall D M Avni Y Oey M S amp Flanagan

J 1994 ApJS 92 52Williams O R et al 1992 ApJ 389 157York D G et al 2000 AJ 120 1579Young M Elvis M amp Risaliti G 2008 ApJ 688 128Yuan W Siebert J amp Brinkmann W 1998 AampA 334 498Zamorani G et al 1981 ApJ 245 357Zdziarski A A Poutanen J Paciesas W S amp Wen L 2002 ApJ 578 357

ERRATUM ldquoTHE FIFTH DATA RELEASE SLOAN DIGITAL SKY SURVEYXMM-NEWTON QUASARSURVEYrdquo (2009 ApJS 183 17)

M Young12

131 Harvard-Smithsonian Center for Astrophysics 60 Garden St Cambridge MA 02138 USA myoungcfaharvardedu

2 Boston University Astronomy Department 725 Commonwealth Ave Boston MA 02215 USA3 INAF-Osservatorio di Arcetri Largo E Fermi 5 Firenze Italy

We have discovered an error in Column 9 of Table 2 in the original paper This column reports the fit flags indicating which modela source prefers a simple power-law (SPL flag = 1) a fixed power-law plus intrinsic absorption (FPL flag = 2) or an absorbedpower-law (APL flag = 3) The table of the original paper mistakenly reports all flag = 3 sources as having flag = 2 and all flag = 2sources as having flag = 1 so that only 32 sources prefer an absorption model

We have updated Table 2 to print out the correct fit flags for each source resulting in 55 sources that prefer an absorption modelSince the fit flags determine which numbers are reported for the remaining columns of Table 2 these numbers are updated as wellThe abstract and text of the original paper report the correct number of absorbed sources so the conclusions are unaffected

In addition a minor rounding error was found in the SDSS names of some objects and so we replace both Tables 1 and 2 withcorrected versions

Online-only material machine-readable tables

b Exp Θd Net Background RL BALTimec Countse Countsf Flagg Flagh

(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 formand content)

250

No 1 2009 ERRATUM 251

SDSS Name f05ndash2a f2ndash10

a f05ndash2b f2ndash10

b L05ndash2c L2ndash10

(Obs) (Obs) (Rest) (Rest) (Rest) (Rest) Flagd

Notes Luminosities are computed using H0 = 70 km sminus1 Mpcminus1 ΩM = 03 and ΩΛ = 07 The upper limits are at the 16σ levela Observed-frame X-ray flux in the observed band is given in units of 10minus14 erg cmminus2 sminus1b Rest-frame X-ray flux in the soft and hard bands is given in units of 10minus14 erg cmminus2 sminus1c Rest-frame X-ray luminosity in the soft and hard bands is given in units of 1044 erg sminus1d A flag indicating the X-ray fit An undetected source is flagged as minus1 and upper-limit flux values are listed A detected source with SN lt 6 is flagged as 0and detected flux values are listed For sources with SN gt 6 three models can be applied If a single power-law model (SPL) is preferred the flag = 1 andthe SPL parameters are listed as well as the 90 upper limit on intrinsic absorption from the intrinsically APL model If an FPL model with variable NH ispreferred the flag = 2 The best-fit slope from the SPL is listed as well as the best-fit NH from the FPL model If the APL model is preferred the flag = 3and the APL power law and absorption parameters are listede Photon index for the best-fit model (If the APL model is preferred then the photon index from that model is quoted otherwise the photon index is from theSPL model)f Intrinsic absorption or 90 upper limit in units of 1022 cmminus2g The χ2 value and degrees of freedom for the best-fit model

(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)

M Young12

250

1 INTRODUCTION
- 11 OpticalX-ray Correlations
- 12 Quasar Subpopulations
- - 2 DATA
  - - 21 SDSS Quasar Selection
    - 22 Matching with the XMM-Newton Archive
    - 23 X-ray Data Reduction
    - - 3 X-RAY ANALYSIS SPECTRAL FITS
      - 4 RESULTS AND DISCUSSION
      - 41 Correlations with the X-ray Spectral Slope
        
        
        REFERENCES

Page 2: THE FIFTH DATA RELEASE SLOAN DIGITAL SKY SURVEY XMM-NEWTON