The XXL Survey - VI. The 1000 brightest X-ray point sources

30 Pages·2016·19.9 MB·English

A&A592,A5(2016) Astronomy
DOI:10.1051/0004-6361/201527402
&
(cid:13)c ESO2016 Astrophysics
Special feature
The XXL Survey: First results
The XXL Survey
VI. The 1000 brightest X-ray point sources(cid:63),(cid:63)(cid:63)
S.Fotopoulou1,F.Pacaud2,S.Paltani1,P.Ranalli3,4,M.E.Ramos-Ceja2,L.Faccioli5,M.Plionis6,7,3,C.Adami8,
A.Bongiorno9,M.Brusa10,11,L.Chiappetti12,S.Desai13,14,A.Elyiv11,16,C.Lidman17,O.Melnyk19,18,M.Pierre5,
E.Piconcelli9,C.Vignali11,10,S.Alis20,F.Ardila21,S.Arnouts22,I.Baldry23,M.Bremer24,D.Eckert1,L.Guennou25,
C.Horellou26,A.Iovino27,E.Koulouridis5,3,J.Liske28,S.Maurogordato29,F.Menanteau30,31,J.J.Mohr13,14,15,
M.Owers32,B.Poggianti33,E.Pompei34,T.Sadibekova5,A.Stanford35,R.Tuffs36,andJ.Willis37
(Affiliationscanbefoundafterthereferences)
Received18September2015/Accepted12February2016
ABSTRACT
Context.X-rayextragalacticsurveysareideallaboratoriesforthestudyoftheevolutionandclusteringofactivegalacticnuclei(AGN).Usually,a
combinationofdeepandwidesurveysisnecessarytocreateacompletepictureofthepopulation.DeepX-raysurveysprovidethefaintpopulation
at high redshift, while wide surveys provide the rare bright sources. Nevertheless, very wide area surveys often lack the ancillary information
availableformoderndeepsurveys.TheXXLsurveyspanstwofieldsofacombined50deg2 observedformorethan6MswithXMM-Newton,
occupyingtheparameterspacethatliesbetweendeepsurveysandverywideareasurveys;atthesametimeitbenefitsfromawealthofancillary
data.
Aims. This paper marks the first release of the XXL point source catalogue including four optical photometry bands and redshift estimates.
Our sample is selected in the 2−10keV energy band with the goal of providing a sizable sample useful for AGN studies. The limiting flux is
F =4.8×10−14ergs−1cm−2.
2−10keV
Methods.WeusebothpublicandproprietarydatasetstoidentifythecounterpartsoftheX-raypoint-likesourcesbymeansofalikelihoodratio
test.WeimproveuponthephotometricredshiftdeterminationforAGNbyapplyingaRandomForestclassificationtrainedtoidentifyforeach
objecttheoptimalphotometricredshiftcategory(passive,starforming,starburst,AGN,quasi-stellarobjects(QSO)).Additionally,weassigna
probabilitytoeachsourcethatindicateswhetheritmightbeastaroranoutlier.WeapplyBayesiananalysistomodeltheX-rayspectraassuming
apower-lawmodelwiththepresenceofanabsorbingmedium.
Results. We find that the average unabsorbed photon index is (cid:104)Γ(cid:105)=1.85±0.40 while the average hydrogen column density is
log(cid:104)N (cid:105)=21.07±1.2cm−2. We find no trend of Γ or N with redshift and a fraction of 26% absorbed sources (logN >22) consistent with
H H H
theliteratureonbrightsources(logL >44).Thecounterpartidentificationratereaches96.7%forsourcesinthenorthernfield,97.7%forthe
x
southernfield,and97.2%intotal.Thephotometricredshiftaccuracyis0.095forthefullXMM-XXLwith28%catastrophicoutliersestimatedon
asampleof339sources.
Conclusions.WeshowthattheXXL-1000-AGNsamplenumbercountsextendedthenumbercountsoftheCOSMOSsurveytohigherfluxesand
arefullyconsistentwiththeEuclideanexpectation.WeconstraintheintrinsicluminosityfunctionofAGNinthe2−10keVenergybandwherethe
unabsorbedX-rayfluxisestimatedfromtheX-rayspectralfituptoz = 3.Finally,wedemonstratethepresenceofasuperclustersizestructure
atredshift0.14,identifiedbymeansofpercolationanalysisoftheXXL-1000-AGNsample.TheXXLsurvey,reachingamediumfluxlimitand
coveringawidearea,isasteppingstonebetweencurrentdeepfieldsandplannedwideareasurveys.
Keywords. catalogs–surveys–galaxies:active–X-rays:general
1. Introduction phenomena that form the galactic stellar mass and make the
black holes grow are connected, although the precise physical
Supermassive black holes in the centres of galaxies par-
mechanismthatshapesthemisstillunclear.
ticipate in the evolution of their hosts. This is demon-
strated by the relationships between their masses and the host Black hole growth is observable throughout the universe
galaxy properties, such as bulge luminosity or bulge stel- in the objects collectively called active galactic nuclei (AGN).
lar velocity dispersion (Magorrianetal. 1998; Gebhardtetal. Theseobjectsaccretegalacticmatterinanaccretiondiskwhich
2000; Ferrarese&Merritt 2000). These relations show that the produces intense radiation before penetrating inside the cen-
tral black hole. The connection between stellar mass and black
(cid:63) Based on observations obtained with XMM-Newton, an ESA sci- hole mass ultimately results from the shared history of star
ence mission with instruments and contributions directly funded by
formation and accretion, which follow the same pattern of
ESAMemberStatesandNASA.BasedonobservationsmadewithESO
steady increase until a peak redshift at z ∼ 1−2, followed by
TelescopesattheLaSillaandParanalObservatoriesunderprogramme
a steep decline until present times (e.g. Hopkinsetal. 2006a;
ID089.A-0666andLP191.A-0268.
(cid:63)(cid:63) AcopyoftheXXL-1000-AGNCatalogueisavailableattheCDS Silvermanetal.2008;Watsonetal.2009).Inaddition,starfor-
viaanonymousftptocdsarc.u-strasbg.fr(130.79.128.5)orvia mationandaccretionseeminglysharethesameanti-hierarchical
http://cdsarc.u-strasbg.fr/viz-bin/qcat?J/A+A/592/A5 scenario,movingfrommassivegalaxiestolessmassivegalaxies
ArticlepublishedbyEDPSciences A5,page1of30
A&A592,A5(2016)
(Hirschmannetal. 2012). Different versions of negative feed- (Hopkinsetal. 2005; Bonolietal. 2009), or secular disk insta-
back affecting both star formation and accretion and thereby bilities for the lowest luminosity AGN (Hopkins&Hernquist
regulating the growth of stellar and black hole masses have 2006; Bournaudetal. 2011), for which luminosity dependence
beenproposed.Observationsofactivityingalaxiesinthegreen on clustering should be strong. Observational indications of
valley, transiting from the blue cloud to the red sequence, varying strength for such dependencies have been reported
put strong constraints on these mechanisms (Schawinskietal. in the literature (Plionisetal. 2008; Krumpeetal. 2010;
2009). Cappellutietal. 2010; Koutoulidisetal. 2013; Fanidakisetal.
Theroleofblackholesingalaxyevolutioncallsforasystem- 2013a).
aticcensusofAGNinordertostudythedistributionoftheirpa- The Ultimate XMM-Newton Extragalactic X-ray survey, or
rameters(e.g.luminosity,accretionrateorblackholemass)and XXL(Pierreetal.2016,PaperI),isanextensionoftheXMM-
their evolution, in order to relate them to the stellar properties. LSS survey at the same depth, but covering 50deg2 in two
BecauseofthediversityoftheAGNphenomenon,largelydueto 25deg2 fieldsXXL-N(RA=02h20,Dec=−5d00)andXXL-S
thepresenceofobscuringtorusonparsec-scalesintheimmedi- (RA=23h30,Dec=−55d00).WhiletheXXLSurveyhasbeen
atevicinityoftheAGNthatstronglyaffectstheirobservational primarily designed to build a consistent sample of galaxy clus-
properties, there is no unique way to build AGN samples, and tersforcosmology(Pacaudetal.2016,PaperII),animmediate
eachmethodissubjecttosomelevelofincompleteness.Numer- by-productofthesurveyistheidentificationofnumerouspoint
ousstudiesofthesedistributionshavethereforebeenperformed sources, most of which will turn out to be AGN. In this paper,
using AGN samples selected with different means, such as numberVIofthefirstXXLrelease,weintroducetheXXLpoint
optical spectroscopy (e.g. Bongiornoetal. 2007; Mastersetal. source catalogue and present the basic properties of the 1000
2012;Rossetal.2013),X-rays(e.g.Airdetal.2010;Uedaetal. brightestobjects.
2014;Miyajietal.2015),infrared(e.g.Hanetal.2012). The paper outline is as follows. In Sect. 2 we describe the
Among the different tools to select AGN, X-ray surveys X-ray observations for XXL, namely the source detection, the
are among the most efficient. With the exception of ex- extraction, and the analysis of the X-ray spectra. While X-rays
tended sources, AGN are often X-ray bright, at least above areapowerfultoolusedtouncovernuclearactivityingalaxies,
2keV, and the vast majority of high-latitude X-ray sources additionalinformationabouttheAGNandtheirgalaxyhostsare
turn out to be AGN down to extremely faint fluxes around needed.InSect.3wegiveanoverviewofthephotometricdata
10−17 erg s−1 cm−2 in the 0.5–2keV band where a signifi- sets available in the XXL field assembled from both publicly
cantpopulationofnormalgalaxiesappears(Lehmeretal.2012). available and proprietary surveys, and we describe the creation
Several important X-ray surveys have been conducted, from ofthephotometriccatalogueandthecounterpartassignmentfor
the widest all-sky surveys performed by ROSAT (Vogesetal. theX-raydetections.Wealsogiveasummaryofthephotometric
1999), ASCA (Uedaetal. 2001, 2005; Nandraetal. 2003), redshift estimation (photo-z hereafter), and the use of the Ran-
BeppoSAX (Fioreetal. 2001; Verrecchiaetal. 2007), and domForestmethodtoclassifyoursources.
MAXI(Uedaetal.2011)tothedeepestsmall-fieldsurveyslike In Sect. 5 we present the results of the X-ray spectral anal-
the COSMOS field with XMM-Newton (Hasingeretal. 2007) ysisalsoasafunctionofredshiftandthemedianSEDsforour
and Chandra (Elvisetal. 2009) or the Chandra Deep Fields sample.InSect.6,weputoursamplewithinthecontextofthe
(Giacconietal. 2001; Alexanderetal. 2003; Xueetal. 2011). cosmicweb,andwepresenttheobservednumbercounts,thede-
Thanks to its large collecting area, XMM-Newton is able to terminationoftheX-rayAGNluminosityfunctionuptoredshift
efficiently cover large sky areas reaching at the same time z = 3, and the percolation analysis in 10 Mpc and 25 Mpc. In
medium flux depth. For example, the Hard Bright Serendip- Sect.7wedescribethereleasedcataloguethataccompaniesthis
itous Survey (HBSS) covers 25deg2 reaching a flux limit of paper.Finally,inSect.8weclosewithourresultsandthefuture
F =7×10−14ergs−1cm2 (DellaCecaetal.2004).The prospectsoftheXMM-XXLsurvey.
4.5−10keV
XMM-LSS survey (Pierreetal. 2006), a survey with XMM-
Newton covering contiguously 11deg2, is a good compromise
2. X-rayobservations
between sky area and depth which has provided several re-
sultsontheenvironmentalpropertiesofAGN(Elyivetal.2012; In this section we describe the X-ray source detection, the def-
Melnyketal. 2013; Koulouridisetal. 2014, 2016, Paper XII), inition of the XXL-1000-AGN sample, and the X-ray spectral
showing in particular different behaviors between objects hav- analysis.WemodelthesourceX-rayspectraasapower-lawdis-
ingsoftandhardX-rayspectra. tributionwiththepresenceofanabsorbingmedium,determining
AGNsurveysalsoefficientlyprobethelarge-scalestructure boththephotonindex(Γ)andthehydrogencolumndensity(NH).
oftheUniverse.TheclusteringpatternofAGNanditsevolution
canprovideimportantconstraintsonthemassoftheirdarkmat-
2.1. Sourcedetectionandsampleselection
terhalohosts,leadingtosomewhatconflictingresultsdepending
ontheAGNselection(Coiletal.2007,2009;Koutoulidisetal. TheprimarygoalofXXListheaccuratedetectionofclustersof
2013), and shed light on the influence of the environment on galaxies,whichappearasextendedemissioninthe0.5−2.0keV
thenuclearactivity.Furthermore,thedependenceofAGNclus- energyband.Tothisend,weusethededicatedpipelinedescribed
amin
tering on luminosity can also provide important constraints on in Pacaudetal. (2006) (X ). As a by-product, the accurate
amin
the AGN triggering mechanism and on studies of AGN-galaxy detectionofpoint-likesourcesisalsopossible.X wasused
coevolution, since different AGN fueling modes make distinct previouslyintheXMM-LSSsurvey,thepilotofXXL,andwas
predictions for the environment of galaxies that host AGN described in detail in Chiappettietal. (2013). In this work we
amin
(Shankaretal.2009;Fanidakisetal.2013b).TheAGNtrigger- are using X 3.3, which uses the latest calibration files for
ing mechanism adopted by models of galaxy formation are ei- theXMM-Newtonobservations.
amin
thermajormergersforthemostluminousAGN(DiMatteoetal. X proceeds in three stages: (1) images from the three
2005; Hopkinsetal. 2006b; Marullietal. 2009), for which EPIC detectors are combined and a smoothed image is ob-
only weak luminosity dependence on clustering is expected tained using a multiresolution wavelet algorithm tuned to the
A5,page2of30
S.Fotopoulouetal.:TheXXLSurvey.VI.
amin
Table1.Summaryof1000brightestXXLAGNdetections,fluxesrefertoX 3.3determinedvalues.
Sample Area F F N N N N N z
0.5−2keV,min 2−10keV,min det Xspec ctp spec-z photo-z median
(deg2) (ergs−1cm−2) (ergs−1cm−2)
XXL-N 26.9 1.35×10−15 4.86×10−14 558 481 540 339 201 0.621
XXL-S 23.6 1.83×10−15 4.86×10−14 442 424 432 250 182 0.637
XXL-1000-AGN 50.5 1.35×10−15 4.86×10−14 1000 902 972 589 383 0.632
Notes.Column1:samplename,Col.2:areaofeachXXLfield,Cols.3−4:minimumfluxdetectedinthe0.5−2keVand2−10keVenergyband,
respectively,Col.5:numberofdetectedsources,Col.6:numberofsourceswithX-rayspectra,Col.7:numberofsourceswithidentifiedoptical
counterparts, Col. 8: number of sources with spectroscopic redshift, Col. 9: number of sources with photometric redshifts when spectroscopic
redshiftisnotavailable,Col.10:medianredshiftofsampleusingspec-zwhenavailableandphoto-zotherwise.
low-count Poisson regime (Starck&Pierre 1998); (2) source than 32 and C2 for the remaining candidates. The C1 class is
detection is then performed on this smoothed image via Sex- mostlyfreefromcontamination,whiletheC2classis50%con-
tractor (Bertin&Arnouts 1996) and a list of candidate sources taminated by spurious detections or blended point-like sources
is produced; and (3) a maximum likelihood (ML) fit based on (seePaperIIformoredetails).
C−statistic(Cash1979)isperformedforeachcandidatesource; In the first release of our catalogue we present the bright-
only sources with a detection likelihood from the ML fit >15 est1000X-raypointsources,consistingofdetectionsbelonging
are considered significant, which corresponds to 0.5 spurious to neither C1 nor C2. By merging the source lists of the two
sources within 10(cid:48) of the field of view (FOV) (Pacaudetal. fields, we selected the 1000 brightest X-ray point-like sources
amin
2006). This process is performed independently for the soft basedonthe2−10keVfluxestimatedbytheX 3.3andas-
(0.5−2keV)andhard(2−10keV)bands. sumingafixedphotonindexofΓ=1.7.Wechosetoselectour
The final stage of the catalogue creation is the ingestion of sourcesinthe2−10keVbandsincewearemainlyinterestedin
the source list in the database. At this stage the association of AGN. We refer to this sample throughout the paper as XXL-
the soft and hard bands is performed. The complete procedure 1000-AGN.Eventhoughthesamplecontainsasmallnumberof
isdescribedinChiappettietal.(2013);herewebrieflydescribe unidentified objects (2.8%), stars (2.3%), and normal galaxies
themainpointsoftheprocessforcompleteness.Weuseasearch (3%), the vast majority of the sources have an estimated lumi-
radiusof10(cid:48)(cid:48)tomatchsourcesbetweenthetwodetectionbands. nosity of L >1042ergs−1 (>90%), the typical threshold
2−10keV
WealsoallowforasourcetobebelowtheMLfit<15,atmost usedtoseparateAGNfromstarburstgalaxies.
in one band. If a source is present in more than one pointing, In Table 1 we give an overview of the XXL-1000-AGN
we favour sources present in good pointings (>7ks exposure sample. The minimum flux, as estimated by the pipeline, is
time and background level <1.5×10−5ctss−1pixel−1, quality F =1.35×10−15ergs−1cm−2 and F = 4.85 ×
0.5−2keV 2−10keV
flag 0), over sources originating from pointings with low ex- 10−14ergs−1cm−2 forthesoftandhardbands,respectively.Col-
posure time (>3ks exposure time, quality flag 1) and/or1 high umnN givesthenumberofsourceswithgood-qualityX-ray
Xspec
backgroundlevel(backgroundlevel<4.5×10−5 ctss−1pixel−1, spectra (signal-to-noise ratio S/N > 3; see Sect. 2.3). Column
qualityflag2).Ifasourceisfoundinmorethanonegoodpoint- N givesthenumberofidentifiedcounterpartsusingeitherop-
ctp
ing,wefavourthedetectionthathasthesmalleroff-axisangle. ticalornearinfraredimages(seeSect.4.2).ColumnsN and
spec-z
Finally, in order to define the count-to-flux conver- N give the number of sources with spectroscopic redshift
photo-z
sion factors we assume a single power-law spectrum with (spec-z) and photometric redshift (photo-z) determination, re-
fixed photon index Γ=1.7 and hydrogen column den- spectively.Finally,fromN weseethatthemedianredshift
median
sity N =2.6×1020cm−2. The following conversion fac- ofthisfluxlimitedsampleisapproximately0.63.
H
tors, corresponding to 1 cts−1, are used: for the PN cam-
era 1.5×10−12ergs−1cm−2 and 7.9×10−12ergs−1cm−2 for
2.2. Spectralextraction
the 0.5−2keV and 2−10keV energy bands, respectively,
while for the MOS cameras the corresponding factors are For each source in the XXL-1000-AGN sample, a circular ex-
5×10−12ergs−1cm−2 and 23×10−12ergs−1cm−2 (see also traction region and an annulus-shaped background region were
Table 3 in Chiappettietal. 2013). Each source is assigned a computed. The radii were chosen in order to maximise the ex-
unique flux in the soft (0.5−2keV) and hard (2−10keV) bands pectedS/Nofthespectrum,asdescribedbelow.
by taking the average flux from all respective PN and MOS Since XXL is a mosaic of many overlapping XMM-Newton
observations. pointings,anysourcemaybepresentonmorethanoneofthem
SincetheXXLsurveywasdesignedforcosmologicalstudies and at different off-axis angles. For the spectral analysis, we
usingX-rayselectedgalaxyclustersasprobes,theidentification make use of all pointings. However, our pipeline analyses each
of extended X-ray sources is of central importance in our col- pointingindividually,whichdoesnotguaranteethatadetection
laboration. The definition of extended sources is based on the is available on all pointings for each source. Therefore, the ex-
0.5−2keV energy band and it is detailed in Paper II. Briefly, a pectedtotalcountsC availableforspectralanalysisfromallex-
detectionenterstheextendedcandidatelistwhenithasanextent posureswereestimatedas
greater than 5(cid:48)(cid:48) and extension likelihood greater than 15. This
list is then split into two categories: C1 for objects with exten- C = A×(C +C ), (1)
B,ref CD,ref
sion likelihood greater than 33 and detection likelihood greater
whereC andC arethecountsinthereferenceexposure
B,ref CD,ref
1 When both conditions of low exposure time and high background forthe0.5–2and2–10keVbands,respectively,andAisafactor
levelaremet,qualitylevel3isassigned. whichcorrectsforthepresenceofotherexposuresinadditionto
A5,page3of30
A&A592,A5(2016)
thereferenceone.Theexpectedtotalbackgroundsurfacebright- the relatively shallow depth of the XXL survey, spectra gener-
nesswasalsodefinedinthesameway. ally have few counts, even for the brightest sources. This is es-
The A factor depends on the number of overlapping ex- peciallytrueforbackgroundspectra,meaningthattheyaresub-
posures, on their relative length, and on their relative overlap ject to large relative uncertainties. We therefore do not subtract
(through the effective area of the EPIC CCDs, which depends thebackgroundspectrafromthesource+backgroundspectra,but
on the off-axis angle2). For simplicity, we used A = 1.5 for all we use them as source-free fields to constrain the background
sources, which is justified in the following cases which we as- parameters.
sumedasrepresentativeofsourcesintheXXLmosaic: Following the approach of Leccardi&Molendi (2008), we
model the background in all observations (source and source-
– asourceisdetectedinthereferencepointingwithin3(cid:48)ofthe free) using two components: X-ray background (XB) and non-
boresight position, and in a second pointing, with the same X-ray background (NXB). The XB itself consists of three
length,at9(cid:48)–10(cid:48)off-axis; components: the cosmic X-ray background (CXB), the Lo-
– an exposure has been repeated, e.g. because of high back- cal Bubble emission, and the Galactic Halo (see Fig. B.1 in
ground. Therefore, a source is detected in the reference Leccardi&Molendi 2008). We model the first with a power
pointing, and then in another one at the same off-axis an- law and the last two with a thermal plasma emission (apec
gle,butthecleantimeonthenon-referenceexposurewillbe model in xspec). The slope of the CXB component is fixed to
shorter. the value of 1.46 (DeLuca&Molendi 2004), but its normal-
isation is allowed to vary for any given sky position (source
Had we computed A for each source individually, the expected
and corresponding source-free regions are fixed to the same
variationonAwouldonlyhavehadaminoreffectontheradiiof
normalisation). The shape of the Local-Bubble and Galactic
the extraction regions. A non-optimal radius would either have
Halo emissions are fixed, but their normalisations are deter-
included more background (if the radius had been larger than
mined using ROSAT data. We extracted from HEASARC’s
optimal)orcausesomefluxloss(iftheradiushadbeensmaller X-RayBackgroundTool3 theROSATspectrafromregionscov-
thanoptimal).However,forthebrightsourcesdiscussedinthis
ering the XXL-N and XXL-S areas (Snowdenetal. 1997). The
paper,suchaneffectislikelynegligible.
normalisations of the two components were fixed by fitting
The spectral extraction regions were computed using the
the emission models (including CXB) over these two areas.
autoregions software (Ranallietal. 2015) which for every
Depending on the instrument, the NXB consists of a combi-
sourcei
nation of several emission lines typically resulting from inter-
nal fluorescence and continuum components phenomenologi-
– definesthesourceregionsascirclesofradiusr andtheback-
i cally represented with a combination of power laws and/or a
ground regions as annuli of inner and outer radii 1.5r and
i blackbody. The CXB and NXB models and parameters were
2r,respectively;
i derived from detailed studies of the XMM-Newton/EPIC back-
– given the expected source counts and background surface
ground (DeLuca&Molendi 2004; Leccardi&Molendi 2008;
brightness, finds the r which maximises the expected S/N
i Kuntz&Snowden2008;Eckertetal.2014).
ofthespectrum;
– checks for neighbour sources. If there are any within 12(cid:48)(cid:48), Consideringtheverycomplexdatasets,backgroundmodels,
andthelargenumberofsourcesthatpreventmanualanalysisof
source i is considered confused and dropped from the sam-
ple. Else, for any neighbour j within 30(cid:48)(cid:48) of source i, it ex- eachsourceseparately,weadopthereaBayesianapproach.We
avoidfittingthemodelsinthetraditionalsensewiththegoalof
cises a circular area of radius r from both the source and
j providingpointestimatesforallmodelparameters,whichwould
backgroundregionsofi.
includetheparametersoftheCXBandNXB.Instead,webuild
Thespectrawerefinallyextractedusingthecdfs-extractsoft- credibleintervalsfortheparametersofinterestandmarginalise
ware (Ranallietal. 2015), which for any pair of source i and overtheparametersoftheCXBandNXB,whichareconsidered
pointingk(treatingPN,MOS1,MOS2exposuresindividually) asnuisanceparametersforthiswork.Non-informativepriorsare
selectedforallparameters(usingtheJeffreyspriorinthecaseof
– checksifsourceiispresentintheFOVofk; norm parameters). We determine the Bayesian evidence using
– checksthatnomorethan40%ofthefluxofiislostinchip thenestedsamplingalgorithm(Skilling2004).Forthispurpose,
gaps,missingMOS1CCDs,andbordersoftheFOV,other- we use the BXA xspec interface (Buchneretal. 2014) of the
wisethe(i,k)pairisdropped; MultiNest implementation (Feroz&Hobson 2008; Ferozetal.
– extracts source and background spectra, and computes re- 2009,2013)ofthisalgorithm.MultiNestdirectlybuildsaposte-
sponses using the standard XMM SAS tools (evselect, riorsamplefromthefulldistribution,whichwemarginaliseover
rmfgen,arfgen). thenuisanceparameterstorecoverthejointposteriordistribution
oftheparametersofinterest(thesource’sparameters).
We perform a fit appropriate for AGN, assuming a power
2.3. Spectralanalysissetup
law absorbed by our Galaxy, allowing the presence of ab-
For each detected source we produced their spectral products sorption in the host galaxy at the redshift of the source
(PHA and response files) per camera and per exposure as de- (phabs*zphabs*powerlawinxspec);therefore,theredshiftof
scribedintheprevioussection.Inaddition,abackgroundregion the host galaxy absorber needs to be determined (see Sect. 4).
wasassociatedtoeachsourceanditsspectralproductswereex- It is either fixed to the spectroscopic redshift when available,
tracted.Thus,weobtainavaryingnumberofsourcespectraand or let free in the 68% confidence interval of the photometric
corresponding background spectra for each source. Because of redshift. The full 0.5–12keV energy range is used, and the al-
lowed range for the intrinsic photon index is Γ=[1,3], while
2 See the XMM-Newton Users’ Handbook, Sect. 3.2.2.2 and
Fig. 13; http://xmm.esa.int/external/xmm_user_support/ 3 http://heasarc.gsfc.nasa.gov/cgi-bin/Tools/xraybg/
documentation/uhb/\effareaoffaxis.html xraybg.pl
A5,page4of30
S.Fotopoulouetal.:TheXXLSurvey.VI.
Table2.X-rayspectralqualityclassesdefinedbasedontheXMMpointingqualityandthesignal-to-noiseratio(S/N)ofthespectrum.
Qualityclass N FXamin/Ffit Flag PNcounts MOS1counts MOS2counts
Pointing S/N med mean σ min max med min max med min max med
0 >6 547 1.00 1.0 0.3 1 25 11194 268 13 2626 105 6 2651 107
0/1/2 >4 281 1.01 1.1 0.5 2 8 1879 103 3 244 43 6 415 40
0/1/2 3−4 75 1.07 2 3 3 6 246 43 2 49 23 4 81 20
0/1/2 <3 96 0.78 52 191 4 3 177 16 1 34 6 1 55 5
NoData – 1 – – – 5 – – – 2 2 – 1 1 –
Notes. The XMM pointing quality classes are: 0 = good quality (>7ks exposure time and <1.5×10−5ctss−1cm−2 background level, 1 = low
exposuretime(>3ks),2=highbackground<4.5×10−5ctss−1cm−2).Pointingswithqualityclass3correspondingtobothlowexposuretimeand
highbackgroundwerenotusedforspectrumextraction.
amin
by X 3.3 and the spectral fit. The black line shows the
10 one-to-one relation. The scatter of the points around the one-
to-one relation is due to (1) the combined quality of the XMM
pointing and the extracted spectrum and (2) the real spectrum
t 11 of the source. Both are expected to cause a scatter of the esti-
i
f mated flux around the one-to-one relation since lower quality
l
a pointings introduce noise in the spectra, while the real spec-
r
t 12 trum of the observed sources will deviate from the power law
c
e of constant photon index assumed in the pipeline. In Table 2,
p
s wesummarisethemedian,mean,andstandarddeviationofthe
) 13 ratioF /F .ThescatterbetweenF
V 2−10keV,pipe 2−10keV,fit 2−10keV,pipe
e andF actuallyseemstoincreaseaswemovetofainter
k 2−10keV,fit
0 sources, but no systematic offset is easily observed. The open
1
−14 squaresshowsourceswithS/N < 3(flag=4).Thechallenging
2
F natureofflag=4sourcesistheresultofacombinationoffactors
g( includingthelowqualityoftheX-rayspectrum,thedifficultyin
lo 15 constrainingthebackground,andtheuncertaintyontheredshift
ofthesource.Wesuggestavoidingtheuseofthesesourcesina
source-by-sourcescientificanalysisbasedontheX-rayspectral
16 fits.ThesesourcesarenotusedinSect.5.1wheretheresultsof
theX-rayspectralanalysisarediscussedindetail.
13 12 In what follows, unless otherwise stated, we also consider
log(F ) pipeline
2 10keV sources with flag 3 since estimated flux shows a slightly larger
− scatter(σ=3)thanthepipeline,butnotasignificantoffset(flux
amin
Fig.1.Comparisonbetweenthe2−10keVfluxestimatedbyX 3.3 ratio median = 1.07). In Table 2 we also present the observed
assuming a power-law spectrum with Γ=1.7, N =2.6×1020cm−2,
H counts for PN, MOS1, and MOS2 split according to the qual-
andthespectralfitvalue(greypoints).Theblacklineshowstheone-to-
ity of the X-ray spectra. For the highest quality spectra (flag =
onerelation.Theopensquaresdenotethesourcesthathavelow-quality
X-rayspectra,(flag=4,seeTable2). 1) the PN counts range from 25 to 104 counts, while the me-
dianobservedcountsare268,105,and107forPN,MOS1,and
MOS2, respectively. Only one source belongs to the category
for the intrinsic absorption logN =[19,24.5]. We include the
H with flag = 5, for which it was not possible to extract spectral
10–12keVrange,asitefficientlyconstrainsthecontinuumpart
data on any ofthe pointings. The sample considered in the rest
oftheNXB.Werebinallspectraindividuallysothattheyhaveat
ofthepaper(flag<4)hasmediancountsC =175,C =70,
leastthreecountsineachbintoavoidbinswithzerocounts.The PN MOS1
and C =65. Further results of the X-ray spectral analysis
Galactichydrogencolumndensitywasdeterminedfromthehy- MOS2
arediscussedinSect.5.1.
drogenmapsofDickey&Lockman(1990).Theresultingfixed
Galactic absorption, modelled withthe phabs xspec model, is
appliedtoallcomponentsexcepttheLocalBubbleemissionand
3. Multiwavelengthobservations
the NXB. All components except the source are scaled to the
areaoftheextractionregion.AstheNXBdoesnothaveanas- TheXXLfieldsbenefitfromancillaryphotometricobservations
trophysicalorigin,wedidnotapplytheeffectiveareacorrection ranging from the ultraviolet to infrared wavelengths consisting
(ARF)ontheNXBmodels;however,weappliedtheredistribu- of both private and public surveys. A summary of all observa-
tionmatrix(RMF)totakeintoaccounttheinstrumentresolution tionsavailableacrosstheelectromagneticspectrumtargetingthe
andredistribution. XXLfieldisgiveninPaperI,Table2.Here,wedescribebriefly
In the released catalogue we provide a flag (see Table 2) the data sets used for this work. In Tables 3 and 4 we gather
showing the combined quality of the X-ray pointing and the thesurveyinformationweusedtoconstructthespectralenergy
S/N of the X-ray spectrum. From Table 2, we see that sources distributions (SEDs) of sources in the XXL-N and XXL-S, re-
with S/N above four have the highest quality spectra. Figure 1 spectively.Thelimitingmagnitudesrefertothethirdquantileof
shows the comparison between the 2−10keV flux determined theaperturemagnitudedistribution.
A5,page5of30
A&A592,A5(2016)
Table3.XXL-N:ancillaryphotometricdatasetusedinthisworkandmeasuredlimitingmagnitudes.
Telescope Survey Version Limitingmagnitude
FUV NUV
AIS 22.18 22.47
GALEX DIS GR6/7 24.57 24.18
GI 23.51 23.69
u g r i yb z
W1 24.96 25.30 24.74 24.39 24.50 23.40
T0007
D1 26.26 26.62 26.34 26.08 25.61 25.21
CFHT WA – 24.82 24.49 – – 22.86
WB PI – 24.98 24.52 – – 23.55
WC – 24.88 – – – –
SDSS DR10 20.73 19.77 19.92 19.96 – 20.13
z Y J H K
VHS DR2 – 20.85 20.98 20.70 20.27
VISTA VIKING DR1 22.81 22.10 21.50 21.36 21.25
VIDEO DR3 25.45 24.65 24.65 24.25 23.90
WIRcam PI – – – – 22.20
DXS – – 22.37 – 22.17
UKIDSS DR10
UDS – – 24.95 24.37 24.64
3.6µm 4.5µm
IRAC PI 21.50 21.34
W1 W2 W3 W4
WISE ATLAS ALLWISE 20.38 20.35 18.85 17.50
Notes.MagnitudesgiveninAB.Aslimitingmagnitudewequotethethirdquantileoftherespectivemagnitudedistribution. CFHTreplacement
i-filter.
Table4.XXL-S:ancillaryphotometricdatasetusedinthisworkandmeasuredlimitingmagnitudes.
Telescope Survey Version Limitingmagnitude
FUV NUV
AIS 22.42 22.82
GALEX MIS GR6/7 – 23.61
GII 23.92 24.03
g r i z
BCS 24.14 24.06 23.23 21.68
DECam PI 25.73 25.78 25.6 24.87
J H K
VISTA VHS DR2 21.1 20.77 20.34
3.6µm 4.5µm
IRAC SSDF 21.5 21.45
W1 W2 W3 W4
WISE ATLAS ALLWISE 20.32 20.35 18.68 17.26
Notes.MagnitudesgiveninAB.Aslimitingmagnitudewequotethethirdquantileoftherespectivemagnitudedistribution.
3.1. XXL-NandXXL-S DeepImagingSurvey(DIS),andguestinvestigatorprograms
(GI). The DIS and GI surveys mainly focus on the XMM-
ObservationsthatareincommonforthetwoXXLfields:
LSS area. The XXL-S field is fully covered by GALEX
– GALEXGalaxyEvolutionExplorer4 (GALEX)releasedthe GR6/7withAIS,MediumImagingSurvey(MIS),and/orGI
final mission catalogue (GR6/7) with full sky coverage in programs.
the far ultraviolet (λ = 1516 Å) and near ultraviolet – VISTAThreepublicEuropeanSouthernObservatory(ESO)
FUV
(λ = 2267Å).Weretrievedtherelevantimagesthrough large programme surveys have observed the XXL field in
NUV
the Mikulski Archive for Space Telescopes5 (MAST). The the near infrared (z, Y, J, H, K filters: 0.8−2.1µm) with
the Visible and Infrared Survey Telescope for Astronomy6
full XXL-N field is covered by one or more of the fol-
(VISTA) in various depths. XXL-N is covered by all three
lowing surveys: the All-Sky Imaging Survey (AIS), the
4 http://www.galex.caltech.edu/ 6 http://www.eso.org/public/teles-instr/
5 http://archive.stsci.edu/ surveytelescopes/vista/
A5,page6of30
S.Fotopoulouetal.:TheXXLSurvey.VI.
surveys: VISTA Hemisphere Survey (VHS, PI: R., McMa- the XXL-S field using the MOSAIC II imager at the Cerro
hon),VISTAKilo-degreeInfraredGalaxySurvey (VIKING, TololoInter-AmericanObservatory17 (CTIO).Theobserva-
PI:W.Sutherland),andVISTADeepExtragalacticObserva- tions cover the g, r, i, z bands (4850−9000Å). The im-
tions Survey (VIDEO, PI: M. Jarvis), while the XXL-S is ages used in this work were analysed as described in
coveredonlybytheVHSsurvey. Menanteauetal.(2009).
– IRACBothXXLfieldshavebeentargetedwiththeInfrared – DECam The Dark Energy Camera18 (DECam) is the suc-
ArrayCamera(IRAC)onboardtheSpitzerSpaceTelescope7 cessor of the MOSAIC II camera at CTIO. We observed
providingimagingat3.6µmand4.5µmoverthewholearea the XXL-S field (PI: C. Lidman) in the g, r, i, z bands
(XXL-N; PI: M. Bremer, XXL-S; SPT-Spitzer Deep Field, (4850−9000Å), but at deeper depth compared to BCS (see
SSDF,Ashbyetal.2014). Table4).Thedetailsoftheobservationsanddatareduction
– WISE The Wide-field Infrared Survey Explorer8 (WISE) is areinGardineretal.(inprep.).Thestackedimagesusedin
anall-skymissionobservingtheskyfrom3.4µmto22µm. thisworkwereperformedasdescribedinDesaietal.(2012).
InthisworkweusetheALLWISEdatarelease9,whichpro-
vides imaging for the whole XXL field in all four WISE We retrieved all the images available for these surveys. All the
filters. optical and near-infrared images were rescaled to zero-point
30 for consistency and ease during the photometry extraction.
GALEX, IRAC, and WISE data already have homogeneous
3.2. XXL-N
zero-points(persurvey)sotherewasnoneedforrescaling.
ObservationscoveringexclusivelytheXXL-Narea:
– CFHTTheCanada-France-HawaiiTelescope10 (CFHT)has 4. Multiwavelengthcatalogue
observedthepartsoftheXXL-NareaaspartoftheCHFT-
Legacy Survey11 (CFHTLS) in two configurations a) wide The photometry of publicly available catalogues is usually a
area (W1 field) and b) deep field (D1 field) observed with generic extraction which suffers from false detections, non-
MEGACam, providing imaging in the filters: u, g, r, i/y, z detected real objects, blended photometry, etc. To make the
most out of the existing observations, we performed our own
(3800−8800Å).InthisworkweusetheT0007datarelease
downloaded from Terapix12. We complemented the cover- photometric extraction. Since AGN are rare objects (we have
ageofCFHTwiththreeadditional1deg2 fieldstothenorth about 104 X-ray detections in the 50deg2 of XXL compared
to ∼106 optically detected galaxies in the same area), it is cru-
of W1. The WA and WB fields were observed in the g, r,
cial that the photometry is 1) accurate, in terms of photomet-
zfilterswhiletheWCfieldwasobservedinthegfilter(PI:
riccalibration;and2)complete,intermsofdetectedsourcesin
M.Pierre).Additionally,XXL-Nhasbeenobservedwiththe
each filter. At the same time, the computation of photometric
WIRcamcameraonCFHTintheK band(2.2µm,VIPERS-
s redshiftforAGNrequiresspecialtreatment,particularlytruein
Multi-LambdaSurvey).Detailsontheobservationsanddata
the case of bright X-ray sources (Salvatoetal. 2011). The de-
reductionareprovidedinMoutardetal.(2016a,b).
– SDSSTheSloanDigitalSkySurvey13 (SDSS)providesob- tailedprocedureofthephotometrypipelineandthephotometric
redshift estimation is outside the scope of this work and it will
servationsinthewholeXXL-Nareainthefilters:u,g,r,i,z
be presented in detail in a companion paper (Fotopoulou et al.,
(3800−8800Å).InthisworkweusetheDR10datarelease
inprep.).
(Ahnetal.2014).
– UKIDSSTheUKIRTInfraredDeepSkySurvey14(UKIDSS,
Dyeetal.2006)hastwofieldsoverlappingwithXXL-N:the 4.1. Photometry
Ultra Deep Survey (UDS), and the Deep Extragalactic Sur-
Similarly to the XMM-Newton pointings, the ancillary multi-
vey(DXS)targetingtheXMM-LSSareain J, H, K and J,
s
wavelength observations are organised in tiles corresponding
K bands, respectively. Here we use the DR10 data release
s
downloadedfromtheWFCAMScienceArchive15. to the FOV of each telescope. Since the photometry is not ob-
tainedsimultaneously,differentobservingconditionswillcause
variations to the PSF and the zero-point of each image. We
3.3. XXL-S extracted the photometry per filter and per tile for each sur-
vey.EachcatalogueisthencorrectedforGalacticextinctionac-
ObservationscoveringexclusivelytheXXL-Sarea:
cordingtotheSchlegeletal.(1998)maps.Theaperturemagni-
– BCS The Blanco Cosmology Survey16(BCS; Desaietal. tudes are corrected to total from the difference between the 3(cid:48)(cid:48)
2012) has targeted an area of 80deg2 which overlaps with aperture magnitude and the mag estimated with Sextractor
auto
(Bertin&Arnouts1996).Foreachsurveyandfilter(seeTables3
7 http://ssc.spitzer.caltech.edu/
and 4) we built a source catalogue keeping track of the PSF
8 http://wise.ssl.berkeley.edu/
size,aperturecorrection,andzero-pointcalibrationforeachtile
9 http://wise2.ipac.caltech.edu/docs/release/allwise/
separately.
10 http://www.cfht.hawaii.edu/
Inordertomergethedetectionswithagivensurvey,itisnec-
11 http://www.cfht.hawaii.edu/Science/CFHTLS/
12 http://terapix.iap.fr/rubrique.php?id_rubrique=268 essary to remove duplicate detections due to overlapping tiles.
13 http://www.sdss.org/ For each tile, we identify the distance of the nearest neigh-
14 TheUKIDSSprojectisdefinedinLawrenceetal.(2007).UKIDSS bour for each source. We then concatenate the individual tile
usestheUKIRTWideFieldCamera(WFCAM,Casalietal.2007).The catalogues and search for sources that lie within that radius. If
photometricsystemisdescribedinHewettetal.(2006),andthecalibra- there are positive matches, they are considered as one source
tionisdescribedinHodgkinetal.(2009).Thepipelineprocessingand
sciencearchivearedescribedinHamblyetal.(2008). 17 http://www.ctio.noao.edu/noao/
15 http://surveys.roe.ac.uk/wsa/ 18 http://www.ctio.noao.edu/noao/content/
16 http://www.usm.uni-muenchen.de/BCS/ dark-energy-camera-decam
A5,page7of30
A&A592,A5(2016)
and thebest iskept in thecatalogue. We consideras “best” the The XXL-1000-AGN sample has 540 high-quality spectro-
sourcewiththesmallestphotometricuncertainty.Withthispro- scopic redshifts. For the remaining 44.4% of the sample, we
cedure we create the catalogue of primary detections for each computed photometric redshifts adopting the following proce-
filter. dure. Building on the legacy of the COSMOS survey on pho-
In order to create the multiwavelength catalogue, we im- tometric for AGN described in detail in Salvatoetal. (2009,
posetheconditionthatasourcemustbedetectedinatleasttwo 2011);insteadofthreeclasses–namelynormalgalaxies,AGN,
filters.Sinceouropticalandnear-IRimageshaveatypicalsee- quasi-stellar objects (QSO), – we increased the number to five
ing of 0.8(cid:48)(cid:48) it is sufficient to use positional matching to asso- classes: passive, star forming, starburst, AGN, QSO. We used
ciate the sources detected on our images. We adopted a match- theCOSMOStemplatestoestimateaphotometricredshiftsolu-
ingradius0.7(cid:48)(cid:48).Fromthisstepweidentifymatchedsourcesand tionforagivenclass.Morespecifically,weusedtemplates1−7
single-band detections. As a second step, we match GALEX, fromIlbertetal.(2009)todescribethepassiveclass,templates
IRAC, and WISE catalogues within 1(cid:48)(cid:48) of the multiwavelength 8−18forthestarformingclass,andtemplates19−31forthestar-
catalogue from the previous step and to the individual single- burstclass.FortheactivegalaxiesfromSalvatoetal.(2009)we
detectioncatalogues. usedthepureQSOtemplateswiththeirextensiontotheUVto
describetheQSOclassandthehybridtemplatestodescribethe
AGNclass.Thesameset-upofextinctionlawsandredshiftstep
4.2. X-raycounterpartassociation wasusedasstatedinthesepapers.Namely,wesearchedforthe
best fit model within each class in the redshift range z = (0,6)
SourceassociationistrivialwhentheimagePSFissmall(<1(cid:48)(cid:48))
with z = 0.01. The extinction laws of Calzettietal. (2000)
for which a positional match usually suffices. This is the case step
was applied to the star forming and starburst classes, while the
for our optical and near-IR. However, in order to ensure a cor-
extinctionlawofPrevotetal.(1984)wasappliedtotheQSOand
rect association between X-ray detections and sources detected
AGNclasses.Therefore,foreachobjectwehavefivephoto-zes-
intheopticalwemustemployastatisticalapproach.Wechoseto
timateswiththecorrespondingprobabilitydistributionfunction
usethemethodofSutherland&Saunders(1992)whichhasbeen
(PDF). According to the COSMOS schema, X-ray flux, mor-
usedfrequentlyintheliterature.Thebasisofthemethodliesin
phology(point-likeness),andvariabilityareusedtodistinguish
thefactthatX-raysourcesarerareevents;brightopticalsources
betweenthethreecategories:normal,AGN,QSO.Hereinstead
arealsorareevents,sotheobservationofanX-raysourceanda
weuseanextendedattributesetconsistingof84attributes:
brightopticalsourceinthesameregionoftheskyisconsidered
anon-randomevent. – Allcolourcombinationsusingthebands,g,r,i,z, J, H, K,
Since the available optical observations are highly inhomo- 3.6µm;
geneous we have to search for counterparts in multiple cata- – FWHMscaledtothecorrespondingPSF;
logues.Westartfromthebrightestcatalogues(i.e.SDSS),which – Half-lightradiusscaledtothecorrespondingPSF;
usually cover a larger area on the sky, and we use succes- – X-rayflux;
sively surveys with deeper photometric observations available – Hardnessratio;
(i.e. CFHTLS-D1). Following the procedure used in previous – χ2valuesofthefivemodelcatagoriesfittedtothedata.
XMMsurveys(Brusaetal.2005,2007;Rovilosetal.2011)we
For the classification we used a Random Forest classifier
usetheopticalibandasastartingpointandsuccessivelyweuse
(Breiman 2001) through the sci-kit.learn package in
theKband,and3.6µmincaseswhereacounterpartisnotfound.
Python. We used first the spectroscopic sample from the whole
Withthismethod,weareabletoidentify540counterpartsoutof
XXLfieldkeepingonlysourceswithhigh-qualityspec-zdeter-
558 X-ray detected sources for the XXL-N and 432 out of the
mination. We split the sample into three equal parts: training,
442 X-ray detected sources for the XXL-S. In total we reach a
test,validation.Eachsampleconsistsofabout8000galaxies(in-
97.2%identificationratefortheXXL-1000-AGN.
cludingAGN/QSO).Thetrainingsampleisusedbytheclassifier
tocreatethedecisiontreesandthetestsampleisusedagainby
4.3. Redshiftsandclassification theclassifiertoestimatethequalityoftheclassification.Inorder
to label our sources as belonging in one of the five categories,
A significant effort has been made to gather a large number of we use the proximity of the photo-z solution to the true spec-z
spectroscopic redshifts for the XXL sources. We use a com- value.Thevalidationsampleisneverseenbytheclassifier,and
bination of publicly available data (e.g. SDSS19) and data ob- it is used to assess the performance once the Random Forest is
tainedthroughcollaborationswithotherconsortia,namelywith created.Theforestcanbeeasilysavedandappliedtoallsources
the VIMOS Public Extragalactic Redshift Survey20 (VIPERS; inourfield.
Guzzoetal. 2014) and Galaxy and Mass Assembly21 (GAMA; We used two more Random Forest classifiers to assign
Driveretal.2009,2011;Hopkinsetal.2013;Liskeetal.2015). probability for a source to be (1) a star or (2) an out-
Additionally, through the efforts of our collaboration, the XXL lier. Going through the same procedure as described above
benefits from more than 20 000 high-quality spectra. As is the of class assignment-training-testing, we used the spectroscop-
caseforthephotometricobservations,theXXL-Nfieldalsohas ically identified stars for the star classifier (1600 objects) and
a better spectroscopic coverage (Adami et al., in prep). For the the sources that had photo-z solution in all five categories
XXL-Nthereisalargecompilationofspectrafornormalgalax- |z −z |>0.15·(1+z )asoutliers(3000objects).These
spec phot spec
ies, while in the XXL-S through targeted observations pursued classifiersareabletopredictthecorrectclassforabout98%and
by our collaboration, more than 3000 spectra were obtained to 96%oftherespectivetestsampleofthestarandoutlierclass.
studyX-rayselectedAGNLidmanetal.(2016,PaperXIV). Figure2ashowsthephoto-zperformanceoftheXXL-1000-
AGN sample, using the validation sample, i.e. the sample that
19 https://www.sdss3.org/dr10/ wasneverseenbyourclassifierduringtraining.Inordertovisu-
20 http://vipers.inaf.it/ alisealltheinformationincludedintheredshiftPDF,wecreate
21 http://www.gama-survey.org/ spec-zslices,∆z =0.1.WethentakethePDFsofallsources
spec
A5,page8of30
S.Fotopoulouetal.:TheXXLSurvey.VI.
(a) (b)
Fig.2.a)Spectroscopicversusphotometricredshiftforthe339XXL-1000-AGNsourcesnotincludedintheRandomForesttraining.Theplotis
a2Dhistogramofthestackedphoto-zprobabilitydistributionfunctionsandthecolourbarshowstheprobabilityenclosedineachcellforagiven
spectroscopicredshiftslice(∆z=0.1).Thesolidredlineisthediagonal,thedashedanddottedlinesare0.10×(1+z )and0.15×(1+z )
spec spec
respectively.Sourcesoutsidethedottedlineareconsideredoutliers.b)RedshiftdistributionofXXL-N(blue)andXXL-S(orange)fields.The
solidhistogramsshowthedistributionofspectroscopicredshift,thestackedhatchedbarsshowthedistributionofadditionalphotometricredshift
whenspectroscopicredshiftsarenotavailable.Thecombinedheightshowsthetotalredshiftdistributionineachfield.
thatfallinaspecificbin.Usingthesamebinning∆z =0.1on 4.4. Classificationresults
phot
thephoto-zaxis,weintegratethePDFinagivenbin.Theintegral
givestheprobabilityofeachsourcefallinginthatparticularbin Our classifier returns the optimal photometric class per source
cell. Since our sample consists of independent sources, we add for 70% of the sources in the test sample. We note that the
the calculated probabilities of each source within a given bin. scope of the current set-up is not to identify correctly the ex-
Wethendividethecellvaluebythenumberofsourcespresent act nature of the source, but rather to identify the class that
inthespec-zslice.Thegreyscalecolourbarshowsthepercent- will give us the optimal photo-z solution. Therefore, misclas-
age of this probability. In the ideal case of perfect photometric sification between normal galaxy classes would not necessar-
redshiftestimates,ourcellswouldhave100%probabilityalong ily cause a dramatic effect on the photo-z estimates, if (a)
thediagonaland0%elsewhere.Thisisnotachievableusingonly enough photometric bands are available and (b) characteris-
broadbandphotometry.Instead,asseeninFig.2a,therearear- tic features such as the Balmer break are present in the SED.
easaboveandbelowthediagonalwithlowprobabilityvaluesup Figure 3a shows the fx/fopt diagram (Maccacaroetal. 1988;
toz=2.5. Lehmannetal. 2001; Hornschemeieretal. 2003; Brusaetal.
2005; Fotopoulouetal. 2012). The AGN and QSO are ex-
Tocomparethesedatawithpreviousresultsweusethenor- pected to lie largely between log(f /f ) = ±1, calculated
x opt
malised median absolute deviation (NMAD) as an estimator of as log(f /f )=log f +I/2.5+5.5 (black lines). The colour-
theaccuracydefinedasσ =1.48×|z −z |/(1+z ). X opt x
TheaccuracyinthevalidaNtiMoAnDsampleof3p3h9otsousrpceecsisσ spec= codingdenotestheclassassignedbytheRandomForest.Wesee
0.095 with η = 28.3% catastrophic outliers (i.e. percentNaMgAeDof thattheAGN(orange)andQSO(red)liebetweenthetwolines,
while at the same time the galaxies labelled as passive objects
sourceswith|zphot−zspec|/(1+zspec)>0.15).Asexpected,very (darkblue)liebelowthelog(f /f )=−1lineandstarforming
x opt
luminous X-ray sources with photometry observed not simul-
(cyan)andstarburst(green)objectsappeartohavelowerX-ray
taneously in all bands is affected by intrinsic variability, which
fluxesthanAGNandQSOinagreementwiththeobservationof
in turn makes the photometric redshift estimation challenging.
Salvatoetal.(2011)intheCOSMOSsurvey.Thisdiagramgives
Onceweexcludesourcesthathaveaprobabilityofbeingoutliers
usadditionalconfidenceontheclassifier’sresults.
ofmorethan0.2%estimatedusingourclassifier,theaccuracyis
bσyNMhAaDlf=. 0.071withη=16.3%,butthesamplehasbeenreduced splitItninFgigth.e3bsawmeplpeloatcctohredihnigstotograthme oRfanthdeomintFrionrseisctLc2la−s1s0kaeVs-,
signment.TocalculatetheL weusedtheabsorptioncor-
2−10keV
Figure 2b shows the redshift distribution of the rectedfluxestimatedbytheX-rayspectrumwiththecorrespond-
XXL-1000-AGN in the XXL-N (top) and the XXL-S (bot- ingphotonindexusingspectroscopicredshiftwhenavailableor
tom). The solid histograms represent the spec-z sources, while photometric redshift otherwise. We observe that the two dom-
the stacked hatched bars show the additional photo-z. We inant classes are indeed QSO (red) and AGN (orange). Even
see that the photo-z distribution follows loosely the spec-z by using information such as broadband colours and morphol-
distribution. ogyestimates,ourclassifiercorrectlyidentifiedQSOs,whicha
A5,page9of30
A&A592,A5(2016)
(a) (b)
Fig.3.a)F /F diagramfortheXXL-1000-AGN.Themajorityoftheobjectsclassifiedasquasi-stellarobject(QSO)/AGN(redandorangedots)
X opt
fallbetweenthelog(F /F )=±1lines,whilepassiveobjects(darkblue)andstarsaregatheredinthebottomleftcorneroftheplot.b)X-ray
X opt
luminosityperclass.QSOclassifiedobjectshavethehighest2−10keVluminosity.Thecolour-codingindicatesthephoto-zclassassignedbythe
RandomForestinbothplots:QSO−red,AGN−orange,starburst−green,starforming−blue,passive−black.
posteriori we find to be high-luminosity objects (logL ∼ Sect. 2.3. For this work we adopt a power law with absorbing
xpeak
44.5). Similarly, AGNs show logL ∼ 43.9, while the star medium as a sufficient description of the X-ray spectrum. We
xpeak
forming and passive galaxies, show logL ∼ 43.0. Inter- findthattherearetensourcesthatwouldclearlyrequireanextra
xpeak
estingly, the starburst population shows a broad distribution in modellingcomponentforthesoftX-rayemission,butsevenout
logL extending up to high luminosities. This population in- of these ten sources are associated with stars based on visual
x
cludes objects that show strong absorption in the blue part of inspectionoftheSEDsandopticalimages.
the SED, while at the same time enhanced infrared emission.
In Fig. 4 we show the distribution of the X-ray spectral pa-
Therefore, they are indeed AGN as we concluded by the X-ray
rameters determined by the fitting (top to bottom), log Flux, Γ,
luminosity,butthephotometricredshiftisbestconstrainedbya
log N , and the distribution of the intrinsic luminosity L . The
starburstgalaxytemplate. H x
photon index Γ is in the range [1.0, 3.0] during the fitting. We
In Appendix A we present a subsample of the XXL-1000-
findanaveragevalueof(cid:104)Γ(cid:105)=1.85±0.4,inagreementwithpre-
AGNsources.ForeachoftheRandomForestcategories–QSO,
viousobservations(Mainierietal.2002,2007;Tozzietal.2006;
AGN,starburst,starforming,andpassive–weshowtheX-ray
Buchneretal. 2014; Corraletal. 2015). The relative uncertain-
spectrum,themultiwavelengthSEDalongwiththebestfitmodel
tiesassociatedwithourfittingresultsvaryasafunctionofX-ray
forthephoto-zsolution(redline)andthestarmodel(greyline).
spectral quality flag (as defined in Sect. 2.3). We find that the
Additionally, we show a single filter image and a false colour
2−10keVfluxshowsrelativeuncertaintyof10%,20%,and35%
image.ThegreendashedcircleiscentredattheX-rayposition,
for classes 1−3. Similarly, the photon index shows relative un-
whilethechosencounterpartismarkedwitharedcircle.
certaintyof5%,10%,and20%,forclasses1−3.Thehydrogen
columndensityistheleastwellconstrainedparameterwithrel-
5. XXL-1000-AGNmultiwavelengthproperties ative uncertainty of about 80% for class 1. Only ∼10% of our
sourceshaveN relativeuncertaintybelow30%.Thisisacom-
In this section we discuss further the XXL-1000-AGN X-ray H
binationoflowcountrateX-rayspectraandthefactthatweare
spectralanalysisresultsandSEDfitting.Weshowthatonaver-
incorporatingtheuncertaintyintervalofthephotometricredshift
age,theXXL-1000-AGNsamplecomprisesunabsorbedsources
estimationinthefittingprocess.However,usingthefullproba-
both in the X-rays (logN ∼ 1021 cm−2) and the optical
(E(B−V)<0.1),withphotHonindex(cid:104)Γ(cid:105)=1.85,andaveragein- bilitydistributionfunctioninourscientificanalysisweareable
topropagatecorrectlytheknowledge(orlackthereof)ofthein-
trinsic luminosity logL ∼ 44. We also provide a recipe for
x trinsicabsorption.
creating an expected N distribution, given a hardness ratio
H
value. Figure5ashowstheestimatedphotonindexΓ,asafunction
ofredshift(upperpanel)andintrinsicluminosity(lowerpanel).
Thepointsshowthevaluesobtainedforeachsourceinoursam-
5.1. X-rayspectralproperties
ple (XXL-N: blue points, XXL-S: orange points). In order to
The selection of the XXL-1000-AGN sample was based on the investigateifthereisanytrendofphotonindexwithredshiftor
2−10keV estimated by our pipeline, where the spectrum of a luminosity, we also plot boxplots. We use five bins in redshift
sourceisconsideredtobeapowerlawwithauniversalslopeof andluminosityandshowthemedianofthephotonindexperbin
Γ=1.7.Withthebrightestsampleweareinapositiontomore (redline).Theboxesenclose50%ofthedistribution,whilethe
accurately determine the shape of the spectrum as described in dashedlinesextendtotheminimumandmaximumvalues.The
A5,page10of30