MNRAS445,1430–1448(2014) doi:10.1093/mnras/stu1745 > The largest X-ray-selected sample of z 3 AGNs: C-COSMOS and ChaMP E. Kalfountzou,1,2‹ F. Civano,1,3 M. Elvis,1 M. Trichas4 and P. Green1 1Harvard–SmithsonianCenterforAstrophysics,60GardenSt,Cambridge,MA02138,USA 2CentreforAstrophysics,Science&TechnologyResearchInstitute,UniversityofHertfordshire,Hatfield,HertsAL109AB,UK 3YaleCenterforAstronomyandAstrophysics,260WhitneyAve,NewHaven,CT06520-8121,USA 4AirbusDefence&Space,GunnelsWoodRoad,Stevenage,HertfordshireSG12AS,UK Accepted2014August25.Received2014August22;inoriginalform2014May29 D o w n lo a d ABSTRACT ed Wepresentresultsfromananalysisofthelargesthigh-redshift(z>3)X-ray-selectedactive fro m galactic nucleus (AGN) sample to date, combining the Chandra Cosmological Evolution h Survey and Chandra Multi-wavelength Project surveys and doubling the previous samples. ttp s Thesamplecomprises209X-ray-detectedAGNs,overawiderangeofrest-frame2–10keV ://a luminositieslogLX =43.3–46.0ergs−1.X-rayhardnessratesshowthat∼39percentofthe cad sourcesarehighlyobscured,N >1022cm−2,inagreementwiththe∼37percentoftype-2 em AGNsfoundinoursamplebasHedontheiropticalclassification.For∼26percentofobjects ic.o u havemismatchedopticalandX-rayclassifications.Utilizingthe1/Vmax method,weconfirm p.c o that the comoving space density of all luminosity ranges of AGNs decreases with redshift m above z > 3 and up to z ∼ 7. With a significant sample of AGNs (N = 27) at z > 4, it is /m n foundthatbothsourcenumbercountsinthe0.5–2keVbandandcomovingspacedensityare ras /a consistent with the expectation of a luminosity-dependent density evolution (LDDE) model rtic atallredshifts,whiletheyexcludetheluminosityanddensityevolution(LADE)model.The le /4 measuredcomovingspacedensityoftype-1andtype-2AGNsshowsaconstantratiobetween 4 5 thetwotypesatz>3.OurresultsforbothAGNtypesattheseredshiftsareconsistentwith /2/1 4 theexpectationsofLDDEmodel. 3 0 /1 Keywords: surveys–galaxies:active–X-rays:galaxies. 39 2 7 9 3 b y bothathighredshiftsandforlowluminosities.Thisrequireslarge g 1 INTRODUCTION u samplesofAGNsspanningwiderangesofproperties.Whilemany e s Activegalacticnucleus(AGN)evolutionathighredshifts,before opticalsurveyshaveinvestigatedthespacedensityofhigh-redshift t o n theirdensitypeak,illuminatestheroleofAGNintheformationand AGNs (e.g. Richards et al. 2006; Jiang et al. 2009; Willott et al. 1 0 co-evolutionofgalaxiesandtheircentralsupermassiveblackholes 2010;Glikmanetal.2011;Ikedaetal.2011;Rossetal.2013),the M (SMBHs) during the time of rapid SMBH growth. The so-called resultsarestillcontroversialduetotheirinevitableincompleteness, arc downsizingevolutionhasbeenrevealedforbothAGN(e.g.Ueda especiallyatthefaintluminosityendduetothehostcontamination, h 2 et al. 2003; Hasinger, Miyaji & Schmidt 2005; Aird et al. 2010) and the bias against obscured sources. As compared with optical 02 3 andgalaxies(e.g.Cowieetal.1996;Kodamaetal.2004;Draper surveys,X-rayobservationsarelesscontaminatedbythehostgalaxy etal.2009).Supportingthisidea,X-raysurveyshaveshownthat emissionandincludeAGNpopulationswithawiderangeofneutral thenumberdensityofluminousAGNpeaksathigherredshiftsthan hydrogencolumndensity. lessluminousones(e.g.Uedaetal.2003;Airdetal.2010).This For the investigation of absorption evolution (e.g. Ueda et al. sortofcosmologicalco-evolutionscenarioisinferredfromthetight 2003;Hasinger2008;Draper&Ballantyne2010),X-rayselected correlationexistslocallybetweenSMBHmassandgalacticbulge samplesincludealltypesofAGN(e.g.type-1/unobscuredandtype- properties (e.g. Magorrian et al. 1998; Ferrarese & Merritt 2000; 2/obscured) and provide reduced obscuration bias in comparison Gebhardtetal.2000;McConnell&Ma2013). with optically selected AGN. Although X-ray surveys have in- Toelucidatetheco-evolutionofSMBHandgalaxies(e.g.Granato ferred the existence of an anticorrelation between the obscured etal.2001,2004;Crotonetal.2006;Hopkinsetal.2006;Menci AGN fraction and the luminosity, several of these studies have et al. 2008; Trichas et al. 2009, 2010; Kalfountzou et al. 2011, suggested that this fraction increases toward higher redshift from 2012,2014),theaccretionactivityintheUniversehastobestudied z=0toz∼2withlimitedsamplesatz>3(e.g.LaFrancaetal. 2005;Ballantyne,Everett&Murray2006;Treister&Urry2006; (cid:2)E-mail:[email protected] Ballantyne2008;Hiroietal.2012). (cid:3)C 2014TheAuthors PublishedbyOxfordUniversityPressonbehalfoftheRoyalAstronomicalSociety ThelargestX-ray-selectedsampleofz >3AGNs 1431 However, the evolution of AGN is still rife with uncertainty. OnthebasisofhardX-raysurveys,manystudiesagreedthatthe X-rayLuminosityFunction(XLF)ofAGNisbestdescribedbya luminosity-dependentdensityevolution(LDDE)model(e.g.Ueda etal.2003;Gilli,Comastri&Hasinger2007;Silvermanetal.2008; Uedaetal.2014).Airdetal.(2010)preferredinsteadaluminosity and density evolution model (LADE). In LADE, the shift in the redshiftpeakoftheAGNspacedensityversusX-rayluminosityis muchweakerthaninLDDEmodels,yetgivesasimilarlygoodfit to their data. While the z < 2 downsizing behaviour is common tobothmodels,quitedifferentnumbersofAGNsarepredictedat higherredshifts(z≥3). X-ray surveys (2–10 keV) are now sensitive enough to sample the bulk of the z > 3 AGN population. Two studies have been D performedonhigh-redshiftAGNexploitingthedeepX-raysurveys ow intheCosmologicalEvolutionSurvey(COSMOS)fieldcarriedout nlo (wNithXM=M81–;NCewivtaonno(eNtAaGl.N2=0114)0,;lBimruitseadettoa2l–.12000k9e)ValnudmCinhoasnitdireas Figure1. SkyareaversusX-rayfluxsensitivitycurvesfortheC-COSMOS aded L2A−G10NkeV >1044.2 and 1043.5ergs−1, respectively. A more recent (abreluae(bsolalicdkldinaes)headndlinCeh)a.MThPe/SvDeSrtSica(rlebdluseoldidaslhineed)lsinamepinledsicaantedstthheetflotuaxl from studybasedonthe4MsChandraDeepFieldSouth(CDF-S;Xue correspondingto10percentofthetotalC-COSMOSarea(seeSection4). h etal.2011)wasabletoinvestigatetheevolutionofz>3AGNdown TheverticalreddashedlineindicatestheChaMPX-rayfluxlimitwith>75 ttps toLX∼1043ergs−1(NAGN=34;Vitoetal.2013).Theseresultsare percentcompletenessfromSDSS/UKIDSS/WISE(seeSection2.2).The ://a consistentwithadeclineoftheAGNspacedensityatz>3,butthe totalarea,aftertheappliedcuts,usedforthisworkisrepresentedbythe ca d shapeofthisdeclineremainshighlyuncertainatz>4.Toovercome shadowedgreyarea. e m theselimitations,inthisworkwecombinedthetwolargestsamples ic .o ofz >3X-ray-selected AGNs withspectroscopicredshifts,both u 2.1 TheC-COSMOSsample p derivedfromChandraX-rayObservatory(Weisskopfetal.2002) .c o surveys: the wide but shallow ChaMP survey (Kim et al. 2007; The Chandra-COSMOS survey (C-COSMOS; Elvis et al. 2009; m Greenetal.2009),andthedeeperbutnarrowerC-COSMOSsurvey Civanoetal.2012)coversthecentral0.9deg2oftheCOSMOSfield /m n (Elvis et al. 2009). This combination results in the largest X-ray uptoadepthof200ksintheinner0.5deg2,withtheACIS-ICCD ra s AGNsamplewithN =211atz>3andN =27atz>4. imager(Garmireetal.2003)onboardChandra.TheC-COSMOS /a At the same time, bAyGcNombining two surveysAwGiNth different flux X-ray source catalogue comprises 1761 point-like X-ray sources rtic le limits,weareabletodeterminethedensityevolutionofbothlow- detecteddowntoamaximumlikelihoodthresholddetml=10.8in /4 4 luminosity (L < 1044ergs−1) and high-luminosity AGNs. Our atleastoneband.Thislikelihoodthresholdcorrespondstoaproba- 5 sample includXes both obscured and unobscured AGNs, and their bilityof∼5×10−5thatacataloguesourceisinsteadabackground /2/1 4 separateevolutionhasbeendetermined. fluctuation (Puccetti et al. 2009). Given this likelihood threshold, 3 0 Thepaperisstructuredasfollows.InSection2,wediscussthe thefluxlimitreachedinthesurveyis5.7×10−16ergcm−2s−1 in /1 3 data sets used in this work and the selection of the high-z sam- the full band (0.5–10 keV), 1.9 × 10−16ergcm−2s−1 in the soft 92 ple. In Section 3, we present the optical and X-ray properties of band (0.5–2 keV) and 7.3 × 10−16ergcm−2s−1 in the hard band 79 3 the selected high-z AGN sample, and we explain the AGN type (2–10keV). b y classificationusingX-rayoropticaldata.InSections4and5,the The z > 3 C-COSMOS sample, as presented by Civano et al. g u numbercountsandspacedensityofthesamplearecomparedwith (2011),comprises107X-ray-detectedsourceswithavailablespec- es modelpredictions.Section6summarizestheconclusions.Acos- troscopic(32)andphotometric(45)redshiftsplus30sourceswith t o n mologicalmodelwith(cid:3) =0.3,λ =0.7,andaHubbleconstant aformalz <3butwithabroadphotometricredshiftprobability 1 o o phot 0 of70kms−1Mpc−1isusedthroughout(Spergeletal.2003).Errors distribution,suchthatz +1σ >3.Allofthespectroscopic M phot phot a arequotedatthe1σ level. C-COSMOS sources have a quality flag 3 (two sources) or four rc h corresponding,respectively,toasecureredshiftwithtwoormore 2 0 emissionorabsorptionlinesandasecureredshiftwithtwoormore 2 2 SAMPLE SELECTION 3 emissionorabsorptionlineswithagood-quality,highS/Nspectrum The high-redshift AGN sample used in this work has been se- (seeLillyetal.2007,2009forthoroughexplanationofqualityflags). lectedfromtheC-COSMOSX-raycatalogue,combiningthespec- TunedphotometricredshiftsfortheC-COSMOSsourceshavebeen troscopicandphotometricinformationavailablefromtheidentifi- computedandpresentedinSalvatoetal.(2011).Duetothelarge cationcatalogueofX-rayC-COSMOSsources(Civanoetal.2011, numberofphotometricbandsandthesizeablespectroscopictrain- 2012)andtheChaMP(ChandraMulti-wavelengthProject)X-ray ing sample spanning a large range in redshift and luminosity the catalogueusingonlythe323ChaMPobsidsoverlappingwithSloan estimatedphotometricredshiftsareexpectedtobequiterobustat DigitalSkySurvey(SDSS;Richardsetal.2006)DR5imaging.In z>2.5evenatthefaintermagnitudes(i >22.5).TheCOSMOS AB Fig.1,weshowtheskycoverage(theareaofasurveythatissen- photometricredshiftsforX-ray-selectedsourceshaveanaccuracy sitivetosourcesaboveagivenX-rayflux)usingtheobservedsoft of σ(cid:6)z/(1+zspec)=0.015 with a small fraction of outliers (<6 per band(0.5–2keV)sourcedetectionsforthetwosurveys,andtheir cent),consideringthesampleasawholeati<22.5.Atfaintermag- sum.Thiscorrespondsto2–8keVrestframeforz>3. nitudes,thedispersionincreasestoσ(cid:6)z/(1+zspec)=0.035with∼15 A schematic diagram of the sample selection with the detailed percentoutliers,stillremarkablygoodforanAGNsample.Forthe numberofsourcesforeachstepispresentedinFig.2. z>3C-COSMOSsample,anaccuracyofσ(cid:6)z/(1+zspec)=0.014is MNRAS445,1430–1448(2014) 1432 E.Kalfountzouetal. D o w n lo a d e d fro m h ttp s ://a c a d e m ic .o u p .c o m /m n ra s /a rtic le /4 4 5 /2 /1 4 3 0 /1 3 9 2 7 9 3 b y g u e s t o n 1 0 M a rc h 2 0 2 3 Figure2. Schematicflowdiagramofthehigh-zsampleselection. achieved with only three catastrophic outliers (<9 per cent). The the small number of bands in which these objects are detected, spectralenergydistributions(SEDs)ofthesourceswithphotomet- no photometric redshift is available for them. In X-ray-selected ricredshiftlargerthan3havebeenvisuallyinspectedtogetherwith samples,non-detectionintheopticalbandhasbeenoftenassumed the photometric fitting and the probability distribution of all the to be a proxy for high redshift (e.g. Koekemoer et al. 2004), or possiblesolutions. forhighobscuration,oracombinationofboth.4ofthe15sources Thereare91sourcesselectedinthe0.5–2keVband,14inthe have no detection in the soft band suggesting high obscuration, 2–10keV,and4inthe0.5–10keVbands.Thereare15C-COSMOS possiblycombinedwithhighredshift.Moredetailsaboutthesample sourceswithoutacounterpartintheopticalbands,butwithaK-band selectioncanbefoundinCivanoetal.(2011)andarealsopresented andIRAC (7),only IRAC(6)orno infrareddetection (2). Given inFig.2. MNRAS445,1430–1448(2014) ThelargestX-ray-selectedsampleofz >3AGNs 1433 2.2 TheChaMPsample TheChaMPisawide-areanon-continuousX-raysurveybasedon archivalX-rayimagesofthehighGalacticlatitude(|b|>20deg)sky observedwithACISonChandra.Thefluxlevels(inergcm−2s−1) reached in the survey are 9.4 × 10−16–5.9 × 10−11 in the full (0.5–8keV),3.7×10−16–2.5×10−11(0.5–2keV)inthesoftand 1.7×10−15–6.7×10−11(2–8keV)inthehardband,respectively. TheChaMPsurveyincludesatotalof392fields,omittingpointings fromdedicatedserendipitoussurveyslikeC-COSMOS,theChan- draDeepFields,aswellasfieldswithextended(>3arcmin)bright opticalorX-raysources.ThelistofChandrapointingsavoidsany overlapping observations by eliminating the observation with the shorterexposuretime.Thesurveyhasdetectedatotalof>19000 X-ray sources (Kim et al. 2007; Green et al. 2009) over 33deg2 D o with∼15350X-raysourcespositionallymatchedtoSDSSoptical wn counterparts(Greenetal.2009). Figure3. Opticaland/orinfraredChaMPsurveycompletenessasafunction loa of0.5–2keVX-rayflux.ThereddottedlineindicatestheChaMPX-ray d ThestudyoftheX-ray-detectedAGNpropertiesrequiresaccurate e estimationofredshifts,luminositiesandsourceclassificationthus, fluxcutwith>75percentcompletenessfromSDSS/UKIDSS/WISE. d fro goodqualityspectraor,whennotavailable,multibandphotometry. m h Hence for our X-ray analysis we chose only the 323 fields over- spatialresolutions,6.0arcsec(WISE;W1)and1–2arcsec(SDSS), ttp lappingwithSDSSDR5imagingforwhichthesensitivitycurveis s givenbyGreenetal.(2009),todetermineaccuratenumbercounts. weuse6arcsecasthematchingradiusforWISEcounterparts(Wu ://a etal.2012). ca Optical spectroscopy of ChaMP X-ray sources was described by d Similarly, we searched the UKIDSS Large Area Survey (LAS; e Trichas et al. (2012), where redshifts and classifications for a to- m Lawrence et al. 2007) Data Release 10 for NIR counterparts to ic tal of 1569 Chandra sources are presented. Since the ChaMP is .o ChaMP X-ray sources. The photometric system is described in u a Chandra archival survey, most ChaMP fields contain targeted p Hewett et al. (2006), and the calibration is described in Hodgkin .c sources selected by the target’s PI, and those targets are likely to o etal.(2009).WeusedtheLASYJHKsourcetable,whichcontains m bebiasedtowardsspecialX-raypopulationssuchasbrightAGN. /m onlyfieldswithcoverageineveryfilterandmergesthedatafrom Ofthetargetedsources∼90percenthaveasecurespectroscopic multipledetectionsofthesameobject.TheX-raysourcecatalogues nra redshiftwith33ofthemhavingatz>3and29atz>4(seeTrichas werethenmatchedwithin3arcsecoftheX-raypositionseparately s/a etal.2012).Thehighrateofhigh-redshift-detectedsourcesclearly to each UKIDSS band: Y (0.97–1.07µm), J (1.17–1.33µm), H rtic showsthestrongselectionbiasesthatcouldaffectouranalysisif le (1.49–1.78µm), and K (2.03–2.37µm) recovering also the areas /4 weincludedthetargetedsources.Therefore,weexcludealltargeted 4 withcoverageinasingleUKIDSSband.Theindividualbandlists 5 sources(153)toreducebiasinsamplepropertiesandsourcenumber /2 werethencombined.ForobjectsnotdetectedinaUKIDSSband, /1 counts. 4 we use the 5σ detection limits provided in Dye et al. (2006) of 3 ForSDSSpointsourceswithi<21andwithoutavailablespec- Y=20.23,J=19.52,H=18.73,andK=18.06.Matchingthe 0/1 troscopy,efficientphotometricselectionofquasarsispossibleusing ChaMPcataloguetoWISEandUKIDSS,wefind1103additional 392 anonparametricBayesianclassificationbasedonkerneldensityes- 7 WISE and/or UKIDSS counterparts which do not have a SDSS 9 timation as described in Richards et al. (2009). To select high-z 3 counterpart(thedetailednumbersarereportedinFig.2). b candidateswithoutavailablespectroscopicorphotometricredshift, In summary, ∼70 per cent of the total ChaMP X-ray sample y g SDSSdetectionisrequiredinatleasttheiandzbands,todetect u have SDSS and UKIDSS/WISE photometry (9727 SDSS and/or es Lymandropouts(e.g.Steideletal.1996). WISEand/orUKIDSSand1103WISEand/orUKIDSS).Thelimited t o SearchingtheChaMPcatalogueforX-raysourceswithin4arcsec fractionofopticalmatchesshowshowopticalcounterpartsoffaint n 1 oftheopticalSDSSquasarcoordinate(95percentofthematched 0 X-raysourcesarefainterthantheSDSSmagnitudelimit(i=21.0). M samplehasanX-ray/opticalpositiondifferenceoflessthan3arcsec; a seeGreenetal.2009),yields9727uniquematches(∼63percentof SUDVS-eSxcqeusassacroslowuerrse,widiethntaifineedxtteonis<ion19fo.1rzfo>rs3peqcutarosasrcsoptoyib=yt2h0e.i2r rch 2 thetotalChaMPX-ray-selectedsample).Weadditionallysearched 0 usingugricolourcriteria(Richardsetal.2002). 2 forcross-matchesintheWide-fieldInfraredSurveyExplorer(WISE; 3 Based on the X-ray limits, the identification completeness of Wrightetal.2010)andUKIRT(UKInfraredTelescope)Infrared ChaMPX-raysourcesfallsrapidlyforobjectswithfainteroptical DeepSkySurvey(UKIDSS;Warren,Hewett&Foltz2000;Hewett counterparts.Fig.3showstheoptical(SDSSi-bandcounterparts) etal.2006;Maddoxetal.2008).1 andinfrared(WISEandUKIDSScounterparts)completenessofthe ForasourcetobeincludedintheWISEAllSkySourceCatalog X-ray-selectedsampleasafunctionofthesoft(0.5–2keV)X-ray (Wrightetal.2010),anSNR>5detectionwasrequiredforone flux. This incompleteness can severely bias determination of the of the four photometric bands, W1, W2, W3, or W4, with central numbercountsandspacedensity,particularlyathighredshifts(e.g. wavelengthsofroughly3.4,4.6,12,and22µm,andangularres- Barger&Cowie2005). olutionsof6.1,6.4,6.5,and12.0arcsec.Becauseofthedifferent To address this issue, we set a relatively high X-ray flux limit inChaMP,wherespectroscopiccompletenessishigher,andphoto- 1TheUKIDSSprojectisdefinedinLawrenceetal.(2007).UKIDSSusesthe metriccoverageallowsgoodphotometricredshifts.Weuseasoft UKIRTWideFieldCamera(WFCAM;Casalietal.2007)andaphotometric flux limit for ChaMP at S0.5–2keV > 3 × 10−15ergs−1cm−2 as at systemdescribedinHewettetal.(2006).Thepipelineprocessingandscience thesebrighterfluxesthecompletenessishigherthan≥75percent archivearedescribedinHamblyetal.(2008). (see Fig. 3). The completeness fraction as a function of flux has MNRAS445,1430–1448(2014) 1434 E.Kalfountzouetal. been taken into account for the estimation of the number counts Table1. PhotometricredshiftreliabilitydefinedbyWuetal. andcomovingspacedensity(seeSections4and5).Forsourcesnot (2012)andnumberofsourcesforChaMPsourceswithoutspec- detected in the soft band, the 0.5–2 keV flux has been computed troscopicorSDSSphotometricredshifts. byconvertingthe2–10keVfluxusing(cid:7) =1.8(seeSection3.2). Oneofthemainadvantagesofourcompilationisthatwedonot Surveysa Reliability Nobjb Nobj−limc (percent) missthe fainthigh-redshift population, since this is recovered by theC-COSMOSsurvey.Inthisway,theChaMPsampleisusedfor W 955 359 the determination of the bright end of the luminosity function at U 27 19 highredshifts. U+W 67.4 19 16 S 70.4 637 367 S+W 77.2 733 526 2.2.1 Spectroscopicredshifts S+U 84.8 71 44 S+U+W 87.0 238 177 We compiled secure spectroscopic redshifts for a total of 1547 Total 2680 1508 sources. We have used 1056 sources (excluding target sources) D aS=SDSS;W=WISE;U=UKIDSS. ow from existing ChaMP spectroscopy (Trichas et al. 2012) for the bNumberofpoint-likeobjectsineachcombinationofsurveys. n selectedChaMPfields.Additionalspectroscopicredshiftsaregiven cNumberofpoint-likeobjectsineachcombinationofsurveys loa intheSDSS-III(N=91;Noterdaemeetal.2012)andSDSS-DR10 withS0.5–2keV>3×10−15ergs−1cm−2. ded quasarcatalogues(N=145;Paˆrisetal.2014).Wealsosearched fro theliteraturebycross-correlatingopticalpositionswiththeNASA betweensomeminimumandmaximumvalue,whichiscrucialfor m h ExtragalacticDatabase(NED),usinga2arcsecmatchradiuswhere dealingwithcatastrophicfailures.Theredshiftprobabilitydistribu- ttp weTfhoeunhdig2h5-5remdsohrieftsospuerccetrsowsciothpiscpescatmropslceocpoicnsriesdtsshoifft.44 sources tniuomnbfoerrecaocuhntssoaunrdcecoismtaokveinnginsptoacaeccdoeunnstitfyo(rsteheeSeescttiimonatsio4naonfdt5h)e. s://ac a withz>3.AllofthesesourceshaveasoftbandX-raydetection,and AsforC-COSMOSselectionofhigh-zsources,wealsoincluded d e onlythreesourceslackahard-banddetection.Amongthem,there 13sourceshavingz +1σ >3andz <3.Thisadds m aresevensourceswithz>4andonesourcewithz=6.016±0.005 another10objectstophtohtoemainzspahmotople,allofthpehmotodetectedinboth ic.o u (Jiangetal.2007).AllbutsixofthemhaveSDSSopticalspectra softandhardbands. p.c withmeanS/N>4.5(noneofthemhasS/N<2.0)withatleasttwo o m broademissionlines(LyαandCIV)significantlydetected.Forfive /m oftheremainingsources,redshiftshavebeenobtainedbyTrichas 2.2.3 High-zcandidateselectionandphotometric nra et al. (2012) while for the source with the highest spectroscopic redshiftestimation s/a r(e2d0s0h7i)f.tF(zor=306.s0o1u6rc)ewseofhtahveeCuhseadMtPhespeesctitmroastceopfriocmsamJiapnleg,ethtearle. For the remaining 7759 without a spectroscopic or photometric rticle SDSSredshift,weselectedthehigh-redshiftAGNcandidatesusing /4 areavailablephotometricredshiftsderivedbyRichardsetal.(2009, 4 5 seeSection2.2.2)withanaccuracyofσ(cid:6)z/(1+zspec)=0.013andonly t(h∼e7ir0oppetriccaelnatn),d/doerspthiteeirbieninfrgariendclcuodleodurisn. SMDoSstSoDfRth6esceatsaolougrcuees, /2/1 onecatastrophicoutlier. 4 were rejected from Richards et al. (2009) selection criteria. The 30 remainingsourcescomefromlaterSDSSdatareleases. /13 9 2.2.2 SDSSphotometricredshifts Following the same morphological criteria as Richards et al. 27 (2009), a candidate is required to be unresolved in images taken 93 For the sources without spectroscopic redshifts, we derived pho- throughthetworedderfilters(e.g.gandrforz∼3selection).This by tometric redshifts. The criteria used in SDSS DR6 have now minimizes contamination from low-z galaxies since even type-2 gu been refined to include objects redder than (u − g) = 1.0 which AGNatz>3appearpointlike.However,weavoidusinganyfaint es may well be high-z quasars. The resulting catalogue of ∼1 mil- fluxcutinordertoensurethatwedonotmissfainthigh-zcandidates t on lionphotometricallyidentifiedquasarsandtheirphotometricred- sincenon-detectioncanimplyhigh-zdropouts.Werejectsources 10 shiftsfromSDSSDataRelease6(DR6)isdescribedinRichards withflagsindicatingthattheirphotometrymaybeproblematic(e.g. M et al. (2009). Only point sources (type = 6) with i-band mag- blending of close pairs of objects, objects too close to the edge arc h nitudes between 14.5 and (de-reddened) 21.3 (psfmag i >14.5 of the frame, objects affected by a cosmic ray hit). Overall, we 2 andpsfmag i−extinction i <21.3; where psfmagarethepoint- reject5079non-pointlikesourcesorwithproblematicphotometry. 023 spread-functionmagnitudes).Theyestimatetheoverallefficiency Thisnumber(∼65percent)isingoodagreementwiththerejected ofthecataloguetobebetterthan72percent,withsubsamples(e.g. numberofsourcesbyRichardsetal.(2009)usingthesamecriteria X-ray-detected objects) being as efficient as 97 per cent. At the whichexplainthelackofaphotometricredshiftforthesesources. faintlimitofthecataloguesomeadditionalgalaxycontamination Photometric redshift criteria must strike a quantifiable balance isexpected. betweencompletenessandefficiency,i.e.aprobabilitycanbeas- Thereare1611sourceswithSDSShigh-qualityphotometricred- signedbothtotheclassificationandtheredshift.UsingtheSDSS, shifts and no spectroscopic redshifts in ChaMP (i.e. those with UKIDSS,andWISE2photometricdatacanhelpustoselectquasar good ≥ 0.0, where good is the quality flag; 6 = most robust; candidatesmoreefficientlythanusingeachsurveyindividually(see −6 = least robust; Richards et al. 2009). Among them there are Table1).Thephotometricredshiftreliability,definedbyWuetal. 14sourceswithz >3andonewithz >4,abovetheadopted phot phot ChaMPfluxlimit.Allofthesesourcesaredetectedinbothsoftand hardband.TheSDSSphoto-zcodealsogivesaprobabilityofan 2WeusethecoloursrelatedtoWISEW3andW4magnitudesonlyforsources objectbeinginagivenredshiftrange.Inthisway,wehavenotonly lackingSDSSand/orUKIDSSdetectionsbecauseWISEuncertaintiesare themostlikelyredshiftbutalsotheprobabilitythattheredshiftis substantiallylarger(Wu,Zhang&Zhou2004). MNRAS445,1430–1448(2014) ThelargestX-ray-selectedsampleofz >3AGNs 1435 (2012)asthefractionofthesourceswiththedifferencebetweenthe suggesting that even in the case of type-2 AGNs the photometric photometricandspectroscopicredshiftssmallerthan0.2isgivenin redshiftsarereliablyestimated.Overall,wefoundeightsourceswith Table1.ThehighestreliabilitycanbereachedonlyintheUKIDSS z>3atgreaterthan1σ significance,foursourceswithz>3but surveyedarea,whichismuchsmaller(4000deg2)thantheskycov- lowerthan1σsignificance,andtwosourceswithz +1σ >3. phot phot erageofbothSDSSandWISEsurveys. Richardsetal.(2002)useda3Dmulticolourspacetoselecthigh- 2.2.4 TheChaMPhigh-zsample redshift QSO candidates in SDSS: griz (g − r, r − i, i − z) for candidates with z > 3.0. Following the SDSS group, we search The total z > 3 ChaMP sample includes 87 sources with z > 3. forhigh-zcandidatesinthreeredshiftintervals(z(cid:5)3.0–3.5,z(cid:5) Among them there are 44 sources with secure spectroscopic red- 3.5–4.5, z (cid:5) 4.5). The details of the selection criteria are given shift,15sourceswithSDSSz >3and13sourceswithSDSS phot intheappendix.Ourselectioncriteriarequirethatoursourceslie z + 1σ > 3 available from Richards et al. (2009), and 15 phot phot outsideofa2σ regionsurroundingthestellarlocus.Westillexpect sources with estimated photometric redshifts based on optical/ thesampletobecontaminatedbystarsandlow-zgalaxies.Forthis infrared colour–redshift relations (13 with z > 3 and 2 with phot reason,weusesomeadditionalcriteriadescribedbyRichardsetal. zphot+1σphot>3). D (2002) to exclude objects in colour regions containing predomi- ow nantly white dwarfs, A stars and unresolved red–blue star pairs. 3 THE C-COSMOS AND CHAMP nloa Duringthecolourselectionprocess,nospecificlineisdrawnbe- z > 3 AGN SAMPLE de tween optically selected quasars (type-1 AGN) and type-2 AGN. d Taking into account that both type-1 and type-2 AGNs are unre- Insummary,wehaveassembledasampleofX-ray-selectedAGNs fro m solved in optical images at z > 3 and type-2 AGNs should lie at z > 3 in the C-COSMOS and ChaMP on the basis of both h outsidethestellarlocusduetotheirredopticalcolours,weexpect spectroscopicandphotometricredshifts.Thetotalsampleincludes ttp s that the above criteria efficiently select both high-redshift AGN 209sourceswithz>3.Ofthese,45areselectedtobeatz>3from ://a populations.Wefound53SDSS-detectedhigh-zcandidates. theirbroadP(z).Therearealso15C-COSMOSsourcesconsidered ca d To increase the reliability of the photometric estimation, we to be at z > 3 on the basis of their optical non-detection these e m also combine the SDSS selection with the redder baselines from are included only in the derivation of the upper boundary of the ic .o UKIDSS and WISE, where the contamination of the stellar locus logN–logScurve.Thepropertiesofthesamplemembersaregiven u p and low-redshift galaxies is lower. We used the combination of in Table A1 (Appendix A)and the detailed numbers are given in .c o UKIDSS and SDSS colours in the Y − K versus g − z colour– Fig.2.Fig.4showstheopticalandnear-infrared(i,K,and3.6µm) m /m colordiagramsuggestedbyWu&Jia(2010)toefficientlyseparate observedmagnitudedistributionsforthetotalhigh-zpopulationand n quasarswithredshiftz<4fromstars.Similarly,Wuetal.(2012) forsourceswithspectroscopicandphotometricredshifts,separately. ra s suggestedthatz−W1andg−zcolourscouldbeusedtoseparate Sourcesselectedasi-dropoutsarealsopresented. /a stars from quasars. Based on these criteria, we have rejected 10 Thehard(2–10keVrest-frame)X-rayluminosityversusplaneis rtic le sources associated with stars based on both SDSS–UKIDSS and showninFig.5togetherwiththefluxlimitoftheC-COSMOSand /4 4 SDSS–WISEcolour–colordiagrams.Forsourcesdetectedonlyby ChaMPsurveys(dashedline)andtheappliedfluxcutforChaMP 5 /2 UKIDSS, we used the i = 21.3 upper limit and a Y − K versus (dottedline).Luminositieswerecomputedfromineverycaseas- /1 4 i − Y colour–color diagram to separate stars and low-z galaxies suminganintrinsic(cid:7)=1.8. 3 0 from high-z candidates. We found four high-z candidates. In the The C-COSMOS and ChaMP high-z sample is a factor of 4– /1 3 case of sources detected only by WISE, there is no efficient way 5 larger than all the previous individual X-ray-selected samples 92 7 detailedintheliteraturetoseparatehigh-zquasarsfromstars. at z > 3 (e.g. Brusa et al. 2009; Hiroi et al. 2012; Vito et al. 9 3 Photometricredshiftshavebeenestimatedforthehigh-zcandi- 2013).Mostimportantly,thisisthefirsttimethatasignificantsam- b y datesbycomparingtheobservedcolourswiththeoreticalcolour– ple of 29 X-ray-selected AGNs at z > 4 is assembled. At these g u redshift relations derived from samples with known redshifts redshifts previous studies had a maximum of nine sources. The es (Richards et al. 2002; Wu & Jia 2010; Wu et al. 2012). A stan- z > 3 X-ray-selected AGN sample also covers more than a fac- t o n dardχ2minimizationmethodisusedtoestimatethemostprobable tor of 2 of soft (2–10 keV rest-frame) X-ray luminosity, and in- 1 0 photometric redshifts. Here, the χ2 is defined as (see Wu et al. cludesasignificantnumberofbothbroad-lineandnon-broad-line M a 2004) AGNs. rc h χ2=(cid:2) [(mi,cz−mj,cz)−(mi,ob−mj,ob)]2, (1) ourTosadmispcluestsothbee colbasscsuifireeddAasGoNbsfcruarcetdionorrueqnuoibrsecsuereadch. Tohbejreectairne 2023 σ2 +σ2 twocommonlyadoptedmethodsforclassification:oneisbasedon ij mi,ob mj,ob the optical emission line widths (‘optical type’) or, if a spectrum wherethesumisobtainedforallfourSDSScoloursand/orWISE isnotavailable,bythetypeoftemplatethatbestfitstheoptical– and/orUKIDSScolours,m −m isthecolourinthecolour– infrared SEDs of the sources. The other is based on the column i,cz j,cz redshiftrelations,m −m istheobservedcolourofaquasar, densities, N , in the X-ray spectra (‘X-ray type’) or, if an X-ray i,ob j,ob H andσ andσ aretheuncertaintiesofobservedmagnitudes spectrumisunavailable,bythehardnessratio(HR;e.g.Hasinger mi,ob mj,ob intwobands.Theuncertaintyinthemeasurementwasobtainedby etal.2001).X-rayabsorptionshouldtypicallycorrelatewithoptical mappingthe(cid:6)χ2 error.Sincetheabovestudiesaredominatedby AGN type. In the unified scheme (e.g. Lawrence & Elvis 1982; optically selected quasars, we would expect that the photometric Antonucci1993;Urry&Padovani1995)asthenarrowemissionline redshiftsuncertaintiesintype-1AGNsaresmaller.However,since AGNsareviewedthroughthedustytorus,andhencehavehigher the Lyα break enters the g band at z ∼ 3.5, the g − r colours absorption column densities than broad emission line AGNs. In quicklyreddenwithredshiftforbothpopulations.Alexandroffetal. fact,evidencehasbeenmountingovertheyearsthattheoptical-and (2013)foundthatg−rcoloursareindistinguishableatan84per X-ray-basedclassificationsoftengivecontrastingresults(Lawrence centconfidencelevelbetweentype-1andtype-2quasarsatz>2 &Elvis2010;Lanzuisietal.2013;Merlonietal.2014). MNRAS445,1430–1448(2014) 1436 E.Kalfountzouetal. D o w n lo a d e Fpliagnuerefo5r.thTehoebhjeacrtdsXin-roauyrlsuammipnloes.itBylu(ecosmqupaurteesd=wiCth-C(cid:7)O=SM1O.8S)sreadmsphlieft. d fro m Redcircles=ChaMPsample.Filled=spectroscopicredshift.Open=pho- h tometric redshift. The dashed lines represent the 2–10 keV luminosity ttp s limit of the surveys computed from the 0.5–2 keV limiting flux. The ://a dotted red line represents the completeness flux cut we have adopted at c 3×10−15ergcm−2s−1.Thedottedblacklinescorrespondtothefluxlimits ad e weimposedforthecomputationofthespacedensityandtheirassociated m ic areas,purple(43.4<logLX<44.0),green(44.0<logLX<44.7),orange .o (logLX>44.7). up .c o m numbers of broad-line and non-broad-line AGNs are found. The /m classification for the 75 AGNs in C-COSMOS with photometric n redshifts is obtained by the Salvato et al. (2011) photometric fit- ras t(iAnrgnmouettsho&dIfilbtteinrtg2t0h1e1S)E.3DMvoiareχd2emtaiinlsimonizathtieonfitwtiintghccoandebLeEfPoHuAnRdE /article in Salvato et al. (2011). Briefly, two libraries of templates were /44 5 used,dependingonmorphology,opticalvariability,andX-rayflux /2 ofthesource.Thefirstlibrary(definedinSalvatoetal.2009,table /1 4 2)consistsofAGNtemplates,hybrid(host+AGN)templates,and 30 /1 a few normal galaxies and was used for all the point-like optical 3 9 sources and for the extended sources with an X-ray flux brighter 2 7 than8×10−15ergcm−2s−1.Thesecondlibrary(asdefinedinIl- 93 bertetal.2009)includesonlynormalgalaxytemplatesanditwas by usedfortheremainingsources(i.e.extendedandwithX-rayflux gu <8×10−15ergcm−2s−1).Theflowchartinfig.6ofSalvatoetal. es t o (2011)summarizestheprocedure.Civanoetal.(2012),according n Fanigdu3r.e64µ.mObbasnerdve(fdroAmBtmopagtnoitbuodtetodmis)trhibiguhti-oznoobfjeaclltst.heBlia-bckansdo,lKid-,bbanlude, tuonothbisscfiurtteidngA,GdiNvi.deAtbhoeusto4u0rcpeesrincteontob(2sc8urseodurAceGsN),ogfatlhaexipeshoatnod- 10 M dot–dashed,reddashed,andgreensolidlinesrepresentthetotal,spectro- a metric sample is best fitted with an unobscured quasar template, rc scopic,photometricredshift,andi-dropoutsamples,respectively.Thei-band h dropoutsarenotincludedinthei-bandhistogram. and 47 sources with an obscured quasar template. For 29 AGNs 20 withspectroscopicidentification,thephotometricandspectroscopic 23 typesmatch.Giventhemismatchrateof∼9percent,weestimate 3.1 Opticaltypes that∼7outofthe75AGNscouldhavebeenassignedthewrong SEDclassification. The optical type of the sources has been determined by the IntheChaMPz>3spectroscopicsample,asexpectedatthese measured full width at half-maximum (FWHM) of the permitted fluxes(e.g.Brusaetal.2009),andduetothepredominantlySDSS emissionlines.ThoseobjectswithemissionlineshavingFWHM spectroscopictargetselection,only2/44sourcesarenon-BLAGN. >1000kms−1 (e.g.Stern&Laor2012)areclassifiedas‘optical The characterization of these sources based on their SED fittings broad-line’ (BLAGN), and all others as ‘optical non-broad-line’ hasbeenobtainedbyTrichasetal.(2012).Inordertobeinagree- (non-BLAGN),i.e.theyshownarrowemissionlinesorabsorption mentwiththespectroscopicChaMPsample,wefollowedthesame linesonly,followingCivanoetal.(2011,2012). SEDfittingmethodforthecharacterizationofthe43sourceswith- IntheC-COSMOSspectroscopicz>3sample,21of32sources out a spectroscopic classification. According to this fitting, 11 of are classified as BLAGN. These are mainly associated with the brighter optical sources (i ∼ 22–23) of the spectroscopic sam- AB ple (see Fig. 7). At fainter optical magnitudes (iAB > 23), equal 3http://www.cfht.hawaii.edu/arnouts/LEPHARE/lephare.html MNRAS445,1430–1448(2014) ThelargestX-ray-selectedsampleofz >3AGNs 1437 43sourcesarebestfittedwithanobscuredquasartemplate(non- BLAGN).MoredetailsonthefittingcanbefoundinTrichasetal. (2012)andRuizetal.(2010).Briefly,atotalof16templateshas been used including QSO, Seyfert-2 galaxies, starburst galaxies, absorptionlinegalaxiesandcompositetemplatesthatareknownto harbourbothanAGNandastarburst.TheRuizetal.(2010)model has been adopted, which fits all SEDs using a χ2 minimization techniquewithinthefittingtoolSHERPA(Freeman,Doe&Siemigi- nowska2001).Thefittingallowsfortwoadditivecomponents,one associated with the AGN emission and the other associated with thestarburstemission.Thefitwiththelowestreducedχ2hasbeen chosenasthebest-fittingmodel. Ageneralproblemofrelyingontheopticaltypeisthattheclassi- ficationmaydependonthequalityoftheavailableopticalspectra, D sincegoodsignal-to-noiseratioisrequiredtodetectless-luminous ow broad-emissionlinesabovethestellarcontinuumemissionofhost nlo a galaxies.Also,atz>3,theHα emissionlinemovesintothein- d e fAraGrNedtaynpdesso(1,.u8n,t1il.9r;ecOesnttelryb,rwocaks&difKficouslktit1o9o7b6s)erravpe.idIlnytelromseedthiaetier Fcliegsu=reC6h.aMHRPvsaemrsupsler.eFdislhleidft.=Bslupeecstqruoascreosp=icCre-dCsOhiSftM.OOpSensa=mpphleo.tRomedectriirc- d from broadHβ emission,andwithoutHαthesemaybemisclassifiedas redshift.Sourceswithnohardbandorsoftbanddetectionareshownwith h type-2.Nevertheless,sucheffectsarenotexpectedtobesignificant arrows.FourcurvesofconstantNH(1020,1022,5×1022,and1023cm−2) ttps inoursample,astconsistspredominantlyofluminousAGNsfor arereportedfor(cid:7)=1.8(dashedlines)and(cid:7)=1.4(solidlines). ://a whichcontaminationfromthehostgalaxiesisnegligible. ca d tobemade.UsingtheCIAO4spectralanalysispackage,SHERPA,5we em have simulated X-ray spectra for AGN populations at 3 < z < 7 ic .o in order to quantify the evolution of X-ray spectral slopes due u 3.2 X-raytypes p to the k-correction of the observed AGN spectra towards high-z. .c o Mostsourcesinoursamplehavealownumberofdetectedcounts Based on these simulations, we find that the HR distribution for m (median∼25inthe0.5–8.0keVfullband).Inthiscountregime, theChaMPsamplepeaksat(cid:7)∼1.8–2.0whiletheHRdistribution /mn spectralfitresultsarenotreliable,especiallyifmorethanonefree forC-COSMOSsamplepeaksat(cid:7)∼1.4–1.9.Hereafter,tobetter ras pthaerapmaeratemreitserfisttaerde;leavrgeen.iFfotrhethfiestecroenavseorngse,swtheeuusenctehretaBinatyieessioann cwointhstrparienvitohuescsotuludmiens,dwenesfiitxyedantdhefoprhtohteonpuinrdpeoxsetoof(cid:7)co=m1p.a8riasnodn /article EstimationofHardnessRatios(BEHR)method(Parketal.2006) convertedallsourcefluxes. /4 4 toderiveX-rayspectraltype.Hardnesscountratios(HR),defined Inthiswork,weadoptNH=1022cm−2asthedividingcriterion; 5/2 asHR=(CHB −CSB)/(CHB +CSB),whereCSB andCHB arethe AGNswithNH<1022cm−2and>1022cm−2areclassifiedasX-ray /14 countsinthesoftbandandhardband,respectively. unobscuredandobscured,respectively.Thiscriterionisadoptedby 30 BEHRisparticularlypowerfulinthelow-countPoissonregime, manyauthors,andisknowntobegenerallyingoodagreementwith /13 becauseitcomputesarealisticuncertaintyfortheHR,regardlessof theopticaltype(seee.g.Uedaetal.2003;Hiroietal.2012). 92 7 whethertheX-raysourceisdetectedinbothenergybands.Sources 9 3 with unconstrained upper or lower limits due to non-detections b 3.3 X-ray/opticalfluxratio y (14 hard-only and 49 soft-only detections) have been computed g u byconvertingthe3σ fluxupperlimitintheundetectedbandinto TheX-ray/optical(X/O)fluxratioisaredshiftdependentquantity es counts. for obscured AGN, given that the k-correction is negative in the t o n Toestimatethecolumndensity,curvesofconstantN asafunc- opticalbandandpositivefortheX-rays(Comastrietal.2003;Fiore 1 H 0 tion of redshift have been derived for two spectral slope values, etal.2003;Brusaetal.2010).Asaresult,obscuredsourceshave M (cid:7) = 1.4 and (cid:7) = 1.8. The flatter spectral slope has been chosen higherX/Oathighredshift.Ontheotherhand,unobscuredsources arc to be consistent with the assumptions adopted in producing the havesimilark-correctionsinthetwobands,andthedistributionin h 2 original X-ray catalogues (Kim et al. 2007; Puccetti et al. 2009). X/Oisnotcorrelatedwiththeredshift(Civanoetal.2012).Usually, 02 3 Thesteepervalueismorerepresentativeoftheintrinsicvalueifthe ther-ori-bandfluxisused(e.g.Brandt&Hasinger2005)while spectrumisnotaffectedbyobscuration(Nandra&Pounds1994). a soft X-ray flux was used originally used for this relation with TherelationshipbetweenHRandredshiftofourC-COSMOSand themajorityofluminousspectroscopicallyidentifiedAGNsinthe ChaMPAGNsamplesisshowninFig.6.CurvesofN =1020,1022, EinsteinandASCAsurveys characterized byX/O=0±1(e.g. H 5×1022,and1023cm−2arereportedfor(cid:7)=1.8(dashedlines)and Stockeetal.1991;Schmidtetal.1998;Lehmannetal.2001).The (cid:7)=1.4(solidlines).WeobservethatC-COSMOSsampletendto samerelationhasbeenusedalsointhehardband,withoutreally bemoreobscuredasexpectedduetothefainterX-raysensitivity accountingfortheX-raybandusedorthechangeinspectralslope limit,thantheChaMPsample(Lawrence&Elvis1982;Uedaetal. (e.g. Alexander et al. 2001; Brusa et al. 2003; Fiore et al. 2003; 2003;Hasinger2008;Brusaetal.2010;Burlonetal.2011). Civano,Comastri&Brusa2005;Lairdetal.2009;Xueetal.2011). Though the two samples (C-COSMOS and ChaMP) of z > 3 Fig.7showsthedistributionofX-raysoft(left)andhard(right) AGNsshowdifferenttrendsregardingtheirobscuration,thelarge flux versus optical magnitude to illustrate the parameter space HR errors and the similarity in this redshift range of the curves withwidelydifferentNHvaluesforthesamespectralslope,donot 4http://cxc.harvard.edu/ciao/ allowanaccurateestimateofthecolumndensityforeachsource 5http://cxc.harvard.edu/sherpa/ MNRAS445,1430–1448(2014) 1438 E.Kalfountzouetal. D o w n lo a Figure7. X-rayflux(soft-left,hard-right)versusthei-bandmagnitudeforalltheX-raysourceswithani-bandcounterpart.Thegreyshadedregionrepresents d e tphheotloomcuestroicfrAedGsNhisftabloynogpethnesycmorbreollast.ioOnraXng/eOc=ircl0es±an1d.bSloaucrkcseqsuwariethsrseepcruerseenstpnecotnr-oBscLoApGicNreadnsdhiBftLsAaGreNr,erpersepseecnttievdelbyy.Gfirleleednusypmpebrollismaitnsdresporuersceenstwi-bitahnda d fro m dropoutsandblackleftpointingarrowsrepresentsoftandhardX-rayfluxupperlimitsforundetectedsourcesineachband.TheC-COSMOSsampleis h representedbytheopenbigbluecircles. ttp s ://a spannedbythebroad-lineandnon-broad-linepopulations.TheX/O scheme, we have separated our total sample into broad-line and c a d ratio(Maccacaroetal.1988)isdefinedas non-broad-lineAGNs(asdescribedinSection3.1). e m X/O=log10(fX/fopt)=log10(fX)+C+mopt/2.5, (2) (NIn=th1e02c2acsme−o2f)tghieveBsL28A/1G2N4sX,-trhaeyXob-rsacyurceldassosiufirccaetsiofonrc(cid:7)ri=ter1io.8n. ic.o H u p wherefXistheX-rayfluxinagivenenergyrange,moptisthemag- Halfofthesesourceshaveaspectroscopicredshiftandallbutthree .co nitudeatthechosenopticalwavelength,andCisaconstantwhich come from C-COSMOS sample. If we also take into account the m dependsonthespecificfilterusedintheopticalobservations.For HRerrors,thenforthelowerHRlimits,11BLAGNsareclassified /m n both X-ray bands, the X/O = ±1 locus (grey area) has been de- as X-ray obscured sources and 41 are classified X-ray obscured ra s fined using as C(i) = 5.91 (Civano et al. 2012), which was com- sources for the upper HR limits. In the non-BLAGN subset, the /a putedtakingintoaccountthewidthofthei-bandfiltersinSubaru, above criterion gives 49/71 X-ray obscured sources (detected in rtic CFHT (Canada–France–Hawaii Telescope), or for bright sources bothsoftandhardbands)for(cid:7)=1.8.The27softbandsourcesin le/4 4 SDSS.Inthehardband,thelocusisplottedtakingintoaccountthe non-BLAGNsamplewithnodetectioninthehardband(reportedas 5 /2 bandwidthandthespectralslopeusedtocomputetheX-rayfluxes downwardarrowsinFig.6)haveveryhighupperlimitsontheHR, /1 ((cid:7) =1.8).ThemajorityofBLAGNswithasecurespectroscopic duetotheconservativefluxupperlimitcomputedbyPuccettietal. 43 0 redshift,followthetrendof−1<log10(fX/fi)<1.However,given (2009),butmostofthemdonottherebysatisfytheNH>1022cm−2 /13 thevariationinαOXwithluminosity(e.g.Vignali,Brandt&Schnei- criterion. 92 der2003;Young,Elvis&Risaliti2010;Trichasetal.2013),there Forthetotalsample,wefindagreementbetweentheopticaland 79 3 canbesomeshiftinthelocationsofQSOswithluminositywithin X-rayclassificationfor∼74percent:∼77percentfortheBLAGNs b theso-calledBLAGNregion.ThisshiftisconsistentwiththeX/O and∼69percentforthenon-BLAGNs.Theseratesareconsistent y g u relationbeingoriginallycalibratedonsoft-X-ray-selectedsources, withrecentstudies(e.g.Lanzuisietal.2013;Merlonietal.2014). es brightintheopticalandalsointheX-rays.Thismightexplainthe Possibleexplanationcanbeamisclassificationoffainttype-1with t o n mildshiftbetweentheChaMPandC-COSMOSBLAGNs. strongoptical/IRcontaminationfromhostgalaxylight. 1 0 Apart from the AGN population found in the BLAGN region, To improve the statistics and gain information on the average M there is also a significant population that lie at log10(fX/fi) > 1 propertiesofthetwosubclasses,wecomparedtheirmeanHRval- arc suggestingobscurednuclei.Themaincharacteristicsofthissample ues as a function of redshift (Fig. 8). Despite the ∼30 per cent h 2 areasfollows:(1)lackofspectroscopicredshifts(opensymbols), misclassificationfortheindividualsources(seeTable2),themean 02 3 (2)non-BLAGNopticalclassification(greensymbols),and(3)low propertiesoftheBLAGNsandnon-BLAGNsseemtoagreewith X-ray luminosities (1043ergs−1 < L2–10keV < 1044ergs−1) with theNH ∼1022cm−2 division.Theseresultsdoesnotchangeeven N > 1022cm−2 for ∼65 per cent of them, which is consistent ifweuseonlysourceswithspectroscopicredshifts.Theupperand H withpreviousstudiesfindingthatmildobscurationiscommonat lowerlimitsdetectedonlyinthesoftorthehardbandwereused theseluminosities(e.g.Silvermanetal.2010).Furthermore,nearly tocomputetheupperandlowerboundaryoftheshadedarea.We 75 per cent of all the sources with X/O > 1 are obscured, thus discusstheresultsinSection5. confirmingthatselectionsbasedonhighX/Oratioareefficientin findingsamplesofobscuredAGNs. 4 THE logN−logS OF THE z > 3 AGN We derived the soft band number counts of the z > 3 and z > 4 3.4 ComparisonofopticalandX-raytypes samplesbyfoldingtheobservedfluxdistributionthroughthesky X-rayabsorptionisanalternativegoodindicatorofAGNtype.In coverageareaversusfluxcurveoftheC-COSMOSsurvey(Puccetti ordertocompareouropticalclassificationtotheexpectedobscu- et al. 2009) and the ChaMP’s 323 fields (Green et al. 2009). rationofBLAGNsandnon-broad-lineAGNsbasedontheunified Additionally, we have corrected the number counts for ChaMP MNRAS445,1430–1448(2014) ThelargestX-ray-selectedsampleofz >3AGNs 1439 D o w n lo a d e d fro Figure8. ThemeanHRasafunctionofredshiftforBLAGN(blacksquares) m andnon-BLAGN(orangecircles)z>3AGNsubsamples.Theerrorbars http representthe68percentdispersion.Onlysourceswithbothsoftandhard s banddetectionsaretakenintoaccountfortheestimationofthemeanHR ://a c ineachbin.Undetectedsourcesinoneofthesebandsareincludedonlyfor a d estimationoftheupperandlowerlimits(dashedareas).Thezphot =6.88 em isnouorrcdeerwtiothbaeninucplpuedreldiminitthHeRfiugpu=re0..74hasbeenshifteddowntoHR=0.2 ic.ou p .c o Table2. ComparisonofopticalandX-raytypes.The m upperandlowerlimitshavebeenestimatedtakinginto /m n accountonlytheerrorrangesinHR. ra s /a Number Unobscured Obscured rtic ofsources NH<1022cm−2 NH≥1022cm−2 le/4 4 NoBnL-BALGANGN 9262+−+−1119643 2489+−+−1191463 5/2/143 0 /1 3 9 incompleteness in the spectroscopic/photometric coverage as a 2 7 9 functionofX-rayflux(i.e.Fig.3). 3 b To minimize the error associated with the most uncertain part y g ofthesensitivitycurve,wetruncatetheC-COSMOSsampleatthe u e flux corresponding to 10 per cent of the total area (blue dashed s t o line in Fig. 1). All the sources with a 0.5–2 keV flux above n 3 × 10−16ergcm−2s−1 have been considered (73 objects out of Figure9. Top:thebinnedlogN–logSrelation(withassociatederrors)of 10 thez>3(orangecircles)andz>4(bluesquares)QSOspopulation.The M the81softbanddetected).Thefluxlimitappliedtothesampleis consistentwiththesignal-to-noiseratiothresholdschosenbyPuc- grey shaded area represents the maximum and minimum number counts arc undertheassumptionsdescribedinSection4.Theblueandorangecurves h cetti et al. (2009), on the basis of extensive simulations, to avoid correspondtothepredictionbasedontheLDDE+exp(thicksolid),LDDE 20 theEddingtonbiasinthecomputationofthenumbercountsofthe (dotted),LADE(dashed),andCTAGNs(dashcross)modelsforeachred- 23 entireC-COSMOSsample.Thus,byapplyingafluxlimitcut,we shiftrange,respectively.Thesmallopencirclesrepresentsthenumbercounts alsoreduce the Eddington bias affecting our sample. For ChaMP estimatedbyBrusaetal.(2009),thesmallopensquaresarefromCivano thiswouldbeatS0.5−2keV >2×10−15ergcm−2s−1,belowtheflux etal.(2011),andsmallfilledtrianglesfromVitoetal.(2013).Middle:the limitalreadyapplied. ratiooftheobservednumbercountsforz>3relativetotheLDDE+exp ThebinnedlogN–logSrelationsfortworedshiftranges(z>3; model (thick solid line at the top panel). The thick solid line represents orange points and z > 4; blue points, with associated errors) are theLDDE+expmodelN/Nexp =1,thedottedlinerepresentstheLDDE relativetotheLDDE+expmodel,thedashedlinestheLADErelativetothe plottedinFig.9(toppanel).Inintegralform,thecumulativesource LDDE+expmodel,andthedashcrosslinestheCTAGNsmodelrelative distributionisrepresentedby totheLDDE+expmodel.Bottom:theratiooftheobservednumbercounts N(>S)=(cid:2)NS 1 , (3) mfoerdziu>mp4arneella.tiTvheetroattihoeoLfDthDeEL+DeDxEprmeloadtievle.tSoytmheboLlDsDarEe+siemxpilamrotdoetlhies i=1 (cid:3)i N/Nexp>8.0andisnotpresented. where N(>S) is the number of sources with a flux greater than S and(cid:3),isthelimitingskycoverageassociatedwiththeithsource. i MNRAS445,1430–1448(2014)