Mon.Not.R.Astron.Soc.000,000–000(0000) Printed2February2008 (MNLATEXstylefilev2.2) Chandra observations of five ultraluminous X-ray sources in nearby galaxies T.P. Roberts1,∗, R.S. Warwick1, M.J. Ward1 & M.R. Goad2 1X-rayandObservationalAstronomyGroup,Dept.ofPhysics&Astronomy,UniversityofLeicester,UniversityRoad,Leicester,LE17RH 4 2Dept.ofPhysics&Astronomy,UniversityofSouthampton,Highfield,Southampton,Hants.,SO171BJ 0 ∗E-mail:[email protected] 0 2 n a J ABSTRACT 5 Wereporttheresultsofaprogrammeofdual-epochChandraACIS-Sobservationsoffiveul- 1 traluminousX-raysources(ULXs)innearbyspiralgalaxies.AllfiveULXsaredetectedasun- resolved,point-likeX-raysourcesbyChandra,thoughtwohavefadedbelowthe1039ergs−1 1 luminositythresholdusedtofirstdesignatethesesourcesasULXs.Usingthissamecriterion, v wedetectthreefurtherULXswithintheimagedregionsofthegalaxies.TheULXsappearto 6 0 berelatedtothestarformingregionsofthegalaxies,indicatingthatevenin“normal”spiral 3 galaxiesthe ULX populationis predominantlyassociated with youngstellar populations.A 1 detailedstudyof theChandraACIS-Sspectraof six ofthe ULXsshowsthatfive arebetter 0 describedbyapowerlawcontinuumthanamulti-colourdiscblackbodymodel,thoughthere 4 is evidenceforadditionalverysoftcomponentsto two of the powerlawcontinua.Themea- 0 suredphotonindicesinfouroutoffivecasesareconsistentwiththelow/hardstate inblack / h holebinaries,contrarytothesuggestionthatpowerlaw-dominatedspectraofULXsoriginate p intheveryhighstate.Asimpleinterpretationofthisisthatweareobservingaccretiononto - intermediate-massblackholes,thoughwemightalsobeobservingaspectralstateuniqueto o veryhighmassaccretionratesinstellar-massblackholesystems.Short-termfluxvariability r t isonlydetectedinoneoftwoepochsfortwooftheULXs,withthelackofthischaracteristic s arguingthattheX-rayemissionofthissampleofULXsisnotdominatedbyrelativistically- a : beamedjets. TheobservationalcharacteristicsofthissmallsamplesuggestthatULXsarea v distinctlyheterogeneoussourceclass. i X Keywords: X-rays:galaxies-X-rays:binaries-Blackholephysics r a 1 INTRODUCTION lationisongoing,sinceiftheseareaccretion-poweredsourcestheir X-ray luminosities match, and in many cases greatly exceed, the Ultraluminous X-ray Sources (ULXs) can be broadly defined as Eddingtonlimitforatypical stellar-mass(∼ 10M⊙)blackhole. the most luminous point-like extra-nuclear X-ray sources located Suggestions for their physical composition currently focus upon withinnearbygalaxies,displayingX-rayluminositiesinexcessof four possibilities. The first is that many ULXs are accreting ex- 1039ergs−1. These sources were first observed in Einstein ob- amplesofanewclassof102 −105 M⊙ intermediate-massblack servations of nearby galaxies (e.g. Fabbiano & Trinchieri 1987), holes(IMBH;e.g.Colbert&Mushotzky1999).Theformationof and more than one hundred were subsequently catalogued in suchobjectsremainsamatterofsomedebate;someintermediate- ROSAThigh-resolution imager observations (Roberts& Warwick massblackholesmaybetheremnantsofprimordialPopulationIII 2000;Colbert&Ptak2002).WhilstafractionofthisobservedULX stars(e.g. Madau & Rees 2001), whilst others may be formed in population is associated with recent supernovae (e.g. SN 1986J, dense globular clusters prior to being deposited in a galaxy disc Bregman & Pildis 1992; SN 1979C, Immler, Pietsch & Aschen- (Miller & Hamilton 2002). An alternative scenario is that IMBH bach1998),ASCAstudieshaveshownthatmanyULXsappearto areformed bytherunaway mergerof stellarobjectsat thecentre displaythecharacteristicsofaccretingblackholes(e.g.Makishima of young, dense stellar clusters (Ebisuzaki et al. 2001; Portegies etal.2000; Mizuno, Kubota&Makishima2001). Crucially,high Zwart & McMillan 2002). The existence of IMBH may be sup- spatial resolution Chandra observations (e.g. Kaaret et al. 2001; portedbytherecentinferrenceof3×103 and2×104 M⊙ mas- Strickland et al. 2001), have failed to resolve most ULX targets. sivedarkobjectsinthecoresoftheglobularclustersM15andG1 Whencombinedwiththeobservedsignificantfluxvariability,this respectively(vanderMareletal.2002;Gerssenetal.2002;Geb- suggeststhepresenceofasingle,luminoussourceofX-raysrather hardt et al. 2002), though thepresence of such an object in M15 thanagroupingoflessluminoussources. isfar fromproven (Baumgardt et al. 2002). Further evidence has However,debateoverthenatureofthebulkoftheULXpopu- recentlyemergedwiththediscoveryoftheX-rayspectralsignature 2 T.P.Robertset al. of“cool”accretiondiscs,consistentwiththepresenceofIMBH,in Table2.Thehostgalaxies. severalULXs(e.g.Milleretal.2003,Roberts&Colbert2003). The remaining models focus upon interpreting ULXs as ex- Galaxy Hubbletypea db ib NHc treme examples of “ordinary” stellar-mass (i.e. ∼ 10 M⊙) black (Mpc) (◦) (×1020cm−2) holeX-raybinaries.ThesecondphysicalmodelisthatmanyULXs IC342 SAB(rs)cd 3.9 20 30.3 areordinaryX-raybinariesinanunusuallyhighaccretionmode,in NGC3628 SAbpecsp 7.7 87 2.0 whichtheiraccretiondiscbecomesradiationpressure-dominated, NGC4136 SAB(r)c 9.7 0d 1.6 producingphotonbubbleinstabilitiesthatallowthedisctoradiate NGC4559 SAB(rs)cd 9.7 69 1.5 at a truly super-Eddington X-ray flux (Begelman 2002). Thirdly, NGC5204 SA(s)m 4.8 53 1.5 many ULXs may only appear to exceed the Eddington limit, but Notes: couldinfactbeX-raybinariesemittinganisotropically(Kingetal. aDatafromtheNASA/IPACextragalacticdatabase(NED). 2001),withonlyamildbeamingfactorb∼0.1(whereb=Ω/4π, bDistance(d)andinclination(i)datafromTully(1988). and Ω is the solid angle of the X-ray emission) required in most c Foregroundabsorption,interpolated atthepositionofeachgalaxyfrom casestoreducetheenergyrequirementsbelowtheEddingtonlimit theHImapsofStarketal.(1992). for a conventional stellar-mass black hole. The fourth model is d There is no recorded inclination, so a face-on aspect is adopted after a variation on the third scenario, suggested by Reynolds et al. inspectionofDigitisedSkySurveyimages. (1997) andmorerecentlyKo¨rding,Falcke&Markoff (2001) and Georganopoulos,Aharonian&Kirk(2002),inwhichULXsaremi- mined from WilliamHerschel Telescope/INTEGRAL IFU obser- croquasarsinnearbygalaxiesthatweareobservingdirectlydown vations,willbedetailedinafuturepaper(Robertsetal.,inprep.). thebeamoftheirrelativisticjet(“microblazars”,c.f.Mirabel&Ro- Thelayoutofthispaperisasfollows.Inthenextsectionwe driguez1999). Observational support for thislastscenario comes brieflyoutlinethedetailsoftheChandraobservationsandthedata fromthedetectionofradioemission,potentiallythesignatureofa reduction.Thisisfollowedbyadiscussionofthedetectionofthese relativistically-beamedjet,emanatingfromanULXinNGC5408 (andthreeother)ULXs,andtheirlocationswithintheirhostgalax- (Kaaretetal.2003). ies.Insection4wedetailthespatial,temporalandspectralprop- erties of each source, before discussing the implications of these Current observations do not completely rule out any of the resultsforpossiblephysicalmodelsofULXsinsection5.Ourfind- above scenarios. However, they do suggest that there may be at ingsandconclusionsaresummarisedinsection6. least two separate underlying populations of ULXs, since alarge numberareseencoincidentwithactivestarformationregions(e.g. Zezas & Fabbiano 2002; Roberts et al. 2002) and hence are pre- sumablyassociatedwithnascentstellarpopulations,whereassome 2 CHANDRAOBSERVATIONSANDDATAREDUCTION ULXsare found in elliptical galaxies (e.g. Irwin, Athey & Breg- The details of the ten Chandra observations of ULXs that form man 2003) and so must be associated with an older stellar pop- thebasisofthispaperarelistedinTable1.Theobservationswere ulation. King (2002) suggests that the population associated with pointedattheROSATHRIpositionsoftheULXs,afteracorrection star formation are high-mass X-ray binaries (HMXBs) undergo- totheastrometrycalculatedfromX-ray/opticalmatchesintheHRI ing an episode of thermal-timescale mass transfer (see also King field-of-view. All coordinates listed in the Table, and throughout etal.2001),whilsttheolderpopulationmaybelong-lastingtran- thispaper,areepochJ2000.Wetabulatesomebasicparametersfor sientoutburstsinlow-massX-raybinaries.Localexamplesofeach thehostgalaxiesinTable2.Theobservationswereperformedbe- suggestedclassareSS433andGRS1915+105respectively.This tween2001January9and2002August26,andrangeinexposure heterogeneityissupportedbythefirstreportedopticalstellarcoun- timefrom9.7to22.5 ks. Eachtargetwas observed ontwoocca- terpartstoULXs.Robertsetal.(2001)reportthedetectionofablue sions,separatedby3–5months.Inordertomitigatetheanticipated continuumsourcecoincidentwithNGC5204X-1,whichHSTre- effectsofdetectorpile-upintheACIS-SS3chip,severaloftheob- solvedintothreeseparatesourceswithcoloursthatareconsistent servationswereperformedinasub-arraymode;thesearelistedin with young, compact stellar clusters in NGC 5204 (Goad et al. Table1.Thechoiceofsub-arraywasgovernedbytheanticipated 2002). HST observations also show an ULX in M81 (NGC 3031 countrates,modelledfromthepreviousROSATHRIobservations X-11) to have a possible O-star counterpart (Liu et al. 2002). In oftheseobjects(RW2000;Lira,Lawrence&Johnson2000),with contrast, ULXs have been found with potential globular cluster theaimoflimitingthepile-upfractionto10%orlessineachobser- counterparts inboth NGC 4565 (Wuet al. 2002) and NGC 1399 vation.Onthisbasis,sub-arraysweredeemednecessaryforthree (Angelini,Loewenstein&Mushotzky2001),suggestingthatthese ofthefivetargetsintheprogramme. ULXsareassociated withtheolder stellarpopulation, orperhaps DatareductionwasperformedusingtheCIAOsoftwaresuite, massivecentralblackholes,oftheseglobularclusters. versions 2.1 and 2.2. The reduction started in each case with the Inthispaper,wepresenttheresultsofdual-epochChandraob- leveltwoeventfile,fromwhicheventswithenergiesoutsidethe0.3 servationsoffivedifferentULXslocatedinnearby(d < 10Mpc) –10keVrangewererejected.Alldatasetsweresearchedforperi- galaxies, awarded Chandra time in AO-2 & AO-3. These ULXs odsofhighbackgroundflaring,butnonewereobserved,allowing are listed in Table 1. They were selected from the catalogues of thefullscienceexposuretobeutilisedineachcase.Furthersteps ROSATHRIpoint-likeX-raysourcedetectionsinnearbygalaxies intheanalysisofthedataareoutlinedinthefollowingsections. presented by Roberts & Warwick (2000; hereafter RW2000) and Lira, Lawrence & Johnson (2000) on the basis of their high X- rayluminosities(LX > 1039ergs−1 intheROSATHRI)andthe 3 ULXDETECTIONSANDLOCATIONS lackofpreviousChandraobservations.Adiscussionoftheobser- vationalhistoryoftheseULXsispresentedinAppendixA.Further A0.3–10keVimageofthecentral8.1′×8.1′regionofeachfield discussion of the environment of each of these ULXs, as deter- wasconstructedfromthecorrespondingcleanedeventfile.Thisim- Chandraobservationsoffive ultraluminousX-raysources innearbygalaxies 3 Table1.DetailsofthetenChandraobservations. Target Observationaimpoint Chandrasequencenumber Observationdate ACIS-S3subarray Exposure Rightascension Declination (yyyy-mm-dd) (ks) IC342X-1 03h45m55.2s +68◦04′55′′ 600253 2002-04-29 1 9.9 8 600254 2002-08-26 9.9 NGC3628X-2 11h20m37.5s +13◦34′28′′ 600255 2002-04-06 full 22.3 600256 2002-07-04 22.5 NGC4136X-1 12h09m22.6s +29◦55′49′′ 600257 2002-03-07 full 18.8 600258 2002-06-08 19.7 NGC4559X-1 12h35m52.0s +27◦56′01′′ 600160 2001-01-14 1 9.7 4 600161 2001-06-04 11.1 NGC5204X-1 13h29m39.2s +58◦25′01′′ 600162 2001-01-09 1 10.1 8 600163 2001-05-02 9.5 agewassearchedforpointsourcesusingWAVDETECT,awavelet- populations,wemightexpecttofindthemco-locatedwiththere- based source detection algorithm available in the CIAO package, gionsofthegalaxymostpronetohostingstarformation,suchasthe andtheX-raysourcedetectionscoincidentwiththeopticalextent spiralarms,whereasiftheyareassociatedwiththeolderstarsthen ofeachgalaxy(orthatpartofitwithintheACIS-S3field-of-view) their distribution is more likely to be centrally peaked, or spread were catalogued. The target ULXs were detected in every obser- evenlythroughoutthedisc.Table2showsthatthemajorityofthe vation, and their Chandra nomenclature1, including their refined ULXs(fiveoutofeight)arelocatedmorethanhalfwayoutfromthe position,andobservedcountratesarelistedinTable3.Countrates nucleustowardstheedgeofthegalaxy.Fouroutofthesefiveappear wereconvertedtoapproximatefluxesbasedonasimplepowerlaw coincidentwithspiralarms;thefifth(CXOUJ112037.3+133429)is continuummodelwithastandardphotonindex(Γ)of2,subjectto locatedinanedge-onsystemwherewecannotdistinguishwhether a foreground absorption column appropriate for each galaxy (see itisinanarm,oraninter-armregion. Table2).Inadditiontothetargets,threeothersourceswithlumi- TheidentificationoffouroftheULXswithspiralarmsisrel- nositiespotentiallyintheULXregimewerefound;oneisaprevi- atively robust, as inboth IC 342 and NGC 4136 the galaxies are ouslycataloguedsource(NGC4559X-4inRW2000),buttheother face-onandsoline-of-sightconfusion through thegalaxyismin- twoarenew ULXidentifications.Thedetailsofthesenew ULXs imised. This is also true for CXOU J123551.7+275604 in NGC arealsogiveninTable3. 4559,asthisislocatedinafaintouterspiralarmclearlyseparated One excellent capability of the Chandra observatory is that in projection from the main body of the galaxy. The co-location it provides an absolute astrometry solution toan accuracy of one of the ULXswith spiral arms itself argues that they could be as- arcsecond or better. This positional accuracy facilitates detailed sociated with a young stellar population. If they are ordinary X- follow-upoftheChandraX-raysourcedetectionsthroughthechar- raybinaries,withtypicalkicksimpartedfromtheformationofthe acterisation of their multi-wavelength counterparts. This avenue compactprimaryinasupernovaexplosionof∼ 100kms−1 (c.f. has already borne fruit inthestudy of ULXs,asdiscussed inthe recentresultsforGRO1655-40, Mirabeletal.2002),thenin107 introduction.However,herewelimitourselvestoabriefdiscussion yearsthiswouldmovethepositionoftheULXrelativetoitsbirth oftheenvironmentoftheChandraULXdetectionsonthebasisof placebyamaximumof1kpc,consistentwithanobservedposition PalomarDigitisedSkySurvey(DSS)data. inornearaspiralarmifbornthere.However,iftheULXareas- Thepositionsof theULXsrelativetotheirhost galaxiesare sociatedwithanolderpopulationthenintheir>108yearlifetime illustratedinFigures1&2,whereweoverlaytheChandraX-ray theycouldbedisplacedby10kpcormore,clearlyleavingnore- emissioncontoursontoDSS-2(blue)imagesoftheequivalentfield- quirement for anassociationwiththespiralarms. Thearguments of-view.IneachcasethetargetULXisatthecentreofthefield-of- infavouroftheseULXsbeingrelatedtoyoungstellarpopulations view.ThepositionofeachULXwithrespecttothenucleusofthe aresupportedbyfollow-upobservationsoftheimmediateenviron- parent galaxy is quantified in Table 3. The observed offset from ments of three of these “outer” ULXs, with HII regions detected thepositionofthegalaxynucleus,takenineachcasefromFalcoet closeto(within200pcof)CXOUJ123551.7+275604 andcoinci- al.(1999),issimplytheprojecteddistancemeasuredinarcseconds; dentwithCXOUJ120922.6+295551(Robertsetal.,inprep.),and wecorrectthistoadeprojectedradius,assumingthateachsourceis thedetectionofapossiblesupernovaremnantencirclingtheposi- intheplaneofitshostgalaxy,usingthehost’sinclinationasgiven tionofCXOUJ034555.7+680455(Robertsetal.2003).However, inTable2andapositionangletakenfromtheRC3catalogue(de the kick argument above requires that either the ULXs are very Vaucouleursetal.1991).Thisdeprojectedradiusisalsoshownas young, or possess a small kick velocity and/or a velocity almost f(R25), the fraction of the distance between the nucleus and the entirelyprojectedalongourline-of-sightforthesetobetruephysi- edgeofthegalaxy,definedbythesemi-majoraxisofthe25mag calassociationsratherthanline-of-sightcoincidenceswiththestar arcsec−2isophotalellipse,atwhichtheULXisfound.Finally,we formation regions we might expect to observe within a galaxy’s also give a qualitative description of the region of the galaxy in spiralarms. whichtheULXisfound. Line-of-sight confusion is a much bigger issue for the three IftheULXsinthesegalaxiesareassociatedwithyoungstellar ULXs that are located closer to the centre of their respec- tive host galaxies. Despite this, detailed follow-up of CXOU J132938.6+582506 reveals it to be associated with young stars 1 WerefertotheULXsbytheirChandranamesthroughoutthispaper.The (Goad et al. 2002). The remaining two sources are located close cross-identificationwithpreviousnamesisshownwherenecessary. to the centre of NGC 4559. However, the nucleus of this galaxy 4 T.P.Robertset al. Table3.TheULXdetections. CXOUJ Previous Countrate(ctks−1) Offsetfromgalaxynucleus Location designation Epoch1 Epoch2 observed(′′) deprojected(kpc) f(R25) TargettedULX 034555.7+680455 IC342X-1 212±5 226±5 302 6.1 0.68 Spiralarm 112037.3+133429 NGC3628X-2 13±1 13±1 292 10.8 0.66 Outerdisc 120922.6+295551 NGC4136X-1 3±1 4±1 65 3.0 0.54 Spiralarm 123551.7+275604 NGC4559X-1 153±4 192±4 123 15.2 0.96 Faintouterspiralarm 132938.6+582506 NGC5204X-1 411±6 159±4 17 0.7 0.21 Innerdisc FieldULX 120922.2+295600 - 26±1 18±1 62 2.9 0.52 Spiralarm 123557.8+275807 - 11±1 21±1 31 2.3 0.15 Innerdisc 123558.6+275742 NGC4559X-4 62±3 119±3 12 1.7 0.11 Innerdisc/bulge isknowntohostactivestarformation(Ho,Filippenko&Sargent 4.1 Spatialextent 1997),implyingthatitispossible(thoughbynomeanscertain)that A primary goal of this programme was to use the unique 0.5- theseULXsmayalsoberelatedtothepresenceofyoungstars.It arcsecond spatial resolution of Chandra to investigate whether isentirely plausible, then, that all the ULXsinthis small sample these ULXsareresolved into complexes of many X-ray emitting maybe associatedwithyoung stellar populations. Thisisconsis- sources, or whether they remain a single, point-like object at the tentwithpreviousstudiesfindingcomparativelylargepopulations highest available X-ray resolution. Hence, the target ULXs were of ULX in the most active star forming galaxies (e.g. Fabbiano, placedattheon-axisaimpointoftheACIS-S3chipineachobser- Zezas & Murray 2001; Lira et al. 2002; Zezas, Ward & Murray vationtoprovidethebestpossiblespatialdata. 2003),indicatingarelationshipbetweenmanyULXandactivestar Toinvestigatethequestionofspatialextension,thetwoobser- formation,thoughinterestinglyinthiscaseweco-locatethemwith vationsofeachfieldwerecombinedtoenhancethesignal-to-noise thestarformingregionsofotherwiserelativelynormalgalaxies. ratioof the images. Theexcellent spatial precision of thedata (1 Toputthisincontext,itisworthnotingthatthehostgalaxies pixel ≡ 0.492 arcsecond) was maintained by lining-up the peak inoursamplearealltypeSborlater,indicatingthattheyaredisc- inintensityofeachtargetsource.Thismethodwasverifiedbyex- dominated, and so do not possess large bulges dominated by old amining the peaks in fainter sources within the field, which also stars.WecaninvestigatetheULX–youngstellarpopulationlink lined-upaccurately.WethenderivedtheradialprofileofeachULX furtherbyreferencetotheULXsurveyofColbert&Ptak(2002). andfititscorewithaGaussianfunction,followingtheprocedure Theydetectmorethan30candidateULXscoincidentwith,orinthe describedinsection4.2.1ofRobertsetal.(2002)2.Theresulting haloesof,ellipticalgalaxies,whichmustbedominatedbyanold fitsgaveFWHMofbetween1.8–2.2pixelsforallon-axissources, stellarpopulation.Averagingoverthefourteenellipticalgalaxiesin and∼2.4pixelsforthe2-arcminoff-axisULXsinNGC4559,all whichULXsarelocatedwithrespecttotheblueluminosityofeach consistentwiththenominalChandrapoint-spreadfunctionatthose galaxy,LB,gives0.8ULXper1010 LB.Thisisverymuchanup- positions.Noevidencewasfoundforafaint,extendedcomponent perlimitonthenumberofULXfromanoldstellarpopulation,as surrounding the position of any of the ULXs. This demonstrates Colbert&Ptaksearchouttoaradiusof2R25,potentiallyallowing thattheULXsarepoint-likeatthe0.5-arcsecondspatialresolution amuchgreatercontaminationfrombackground objectsthansim- ofChandra,corresponding tomaximumphysicalsizesfortheX- plyfocussingwithinthestandarddefinitionofthegalaxysize,the 25magarcsec−2 isophotalellipse.Also,sincemanyoftheircan- rayemittingregionsof∼ 9–23parsecsinthehostgalaxies.The spatialdataisalso,ofcourse,consistentwithalltheobjectsbeing didateULXsarepossiblehaloobjects,thesemaythereforeconsti- single,point-likeX-raysources. tuteaphysicallyseparatepopulationtothatassociatedwiththeold stellarpopulation.Nevertheless,bycomparingtheiraverageULX toL ratiowiththe blueluminositiesof our targetgalaxies (and B correctingforthecoverageofeachgalaxyintheChandraobserva- 4.2 Spectralproperties tions),wecanestablishaconservativeupperlimitonthenumberof Spectrawereextractedinaneightarcseconddiameteraperturecen- ourULXsassociatedwiththeolderstellarpopulations.Thisturns tred on the position of each ULX using the PSEXTRACT script, outtobeanupperlimitoftwooutoftheeightULXsweobserve. which also retrieves the appropriate response matrices and ancil- Wethereforeconfirmthatoursampleisverylikelytobedominated laryresponsefilesforeachobservation.Asource-freelocalback- byULXsassociatedwithayoungstellarpopulation. groundregionofequivalentsizewasusedineachcase,thoughits impact was of littlesignificance since it typically contained only 0.1−1%ofthecountsaccumulatedfromthesource.Theancillary responsefileswerecorrectedforthegradualin-orbitdegradationin thequantumefficiencyoftheACISdetectorsusingtheCORRARF 4 X-RAYCHARACTERISTICS Inthissection weinvestigatethedetailed X-rayproperties of the 2 Someadjustmentstothesizeandpositionofthebackgroundannuliwere sources.Unlessspecificallystated,weperformtheanalysesonall requiredtoeliminatenearbyX-raysourcesinseveralfields.Anexampleis fivetargetplusallthreefieldULXs.Discussionofthesecharacter- theNGC4136field,inwhichthetwoULXsareseparatedbyonly∼ 22 isticsintermsofphysicalmodelsisdeferreduntillatersections. pixels. Chandraobservationsoffive ultraluminousX-raysources innearbygalaxies 5 Figure1.ThelocationsoffourofthefivetargetULXs.Eachfigureshowsan8.1′×8.1′DSS-2blueimagecentredonthepositionoftheULX,rotatedslightly tomatchtheChandraprojection.CoordinatesareshowninJ2000.TheChandraX-rayemissioncontoursareoverlayedinredtohighlightthepositionsofthe X-raysourcesineachfield-of-view.TheChandradatahasbeensmoothedbyaHWHM3pixelGaussianmasktoaidvisibility,andthecontoursareplotted at0.3and10countpixel−1.ThetargetULXsarehighlightedbytheverticalarrows,andthefieldULXsbyhorizontalarrows.Thespatialcoverageofthe Chandrasub-arraysusedforobservationsoftheIC342X-1andNGC4559X-1fieldsarehighlightedinblueforeachrespectivefield,withthefirstepoch coverageshownbythedashedlines,andthesecondepochbythedottedlines. tool3.Thespectraweregroupedtoaminimumof20countsperbin, tolimitthedistortingeffectofdetectorpile-upontheobservedX- andthenanalysedinXSPECv.11.2,excludingdatabelow 0.5keV rayspectrumofeachULX.AsstatedinSection2,theROSATHRI duetotheuncertaintyinthecalibrationattheseenergies.Inthefol- countratesofeachULXwereusedtodeterminetheoptimumsub- lowinganalysis,allquotederrorsarethe90%confidenceerrorsfor array size to lower the anticipated pile-up fraction below 10%. oneinterestingparameter. However, this did not guarantee that each source would have an Amajorconsiderationinthedesignoftheseobservationswas acceptablelevelofpile-up,duetothelong-termvariablenatureof theULXsthemselves(seebelow).Wecanderivealowerlimiton the observed pile-up fraction of each observation by reference to 3 Seehttp://asc.harvard.edu/cal/Acis/Calprods/qeDeg. 6 T.P.Robertset al. Figure2.Asperfigure1,fortheNGC5204X-1field. Figure6.24oftheChandra Proposer’sObservatoryGuide(v.4)4, improved χ2 was the 2001 January observation of CXOU which shows pile-up fraction as a function of counts per readout J132938.6+582506,with∆χ2=16.3foroneextradegreeoffree- frame (wecalculate a lower limitsince our input isthe observed dom,andastatisticallysignificantsofteningofthespectralslope. countsperframe,whichalreadyincludespiled-upevents).Thisis Wediscussthiscasefurtherbelow. very much a first order approximation, since it does not include, TheULXobservations inwhichlessthan250sourcecounts for example, the effect of variations in the source spectra on the wereaccumulatedwerenotconsideredforspectralanalysis,which degreeofpile-up.Onlyoneobservation,the2001Januaryobserva- ruled out both observations of CXOU J120922.6+295551 and tionofCXOUJ132938.6+582506, turnsouttohavealowerlimit CXOUJ123557.8+275807.Theremainingspectrawereinitiallyfit inexcessof10%,andthisisonlymarginalat11%.Allotherobser- withsimpleabsorbedsinglecomponentspectralmodels,namelya vationshavelimitsof8%pile-uporless,implyingthatthepolicy powerlawcontinuum, athermalbremsstrahlungmodel,themulti- ofusingsub-arrayswassuccessful. colourdiscblackbodymodel(hereafterMCDBB)usedtodescribe We investigated the presence of residual pile-up effects us- anaccretiondiscaroundablackholeinitshigh(soft)state(Mit- ing the XSPEC v.11.2 parameterisation of the CCD event pile- sudaetal.1984), aMEKALopticallythinthermal plasmamodel, up model of Davis (2001). A simple absorbed powerlaw con- and a conventional blackbody spectrum. The best-fit parameters tinuum model, both with and without the pile-up model, was to the powerlaw continuum and MCDBB models are listed in fit to the observed spectrum of the brightest ULXs (namely Table 4, and individual cases are discussed below. The thermal CXOUJ132938.6+582506,CXOUJ123551.7+275604andCXOU bremsstrahlung model provided an adequate fit in most cases, J034555.7+680455inbothepochs,andCXOUJ123558.6+275742 though never asstatisticallyacceptable asapowerlaw continuum initssecondobservationalepoch).Inthepile-upmodelweallowed (orMCDBBinthecaseofCXOUJ123558.6+275742).Weomitit α,thegrademorphingparameter,tovaryfreely,andsettheframe from the table in favour of the two models that have more gen- timeto0.7secondsand1.1secondsforthe 1 and 1 sub-arraysre- 8 4 erally been used to describe ULX spectra in past analyses. The spectively,withtheotherparametersfixedasdefault.Insixofthe spectra were generally too hard to provide meaningful fits to the seven casesthe changes tothespectral fit whenthepile-up algo- MEKALmodel,whoseparameterstendedtowardsthoseofthether- rithmwasappliedwereminimal(∆χ2 <3foroneextradegreeof malbremsstrahlungfit.Similarly,simpleblackbodymodelsdidnot freedom).Thoughinmostcasesthepowerlawphotonindicesbe- provideagoodfittoanyofthespectra. cameslightlysofter,asexpectedsinceCCDpile-upactstoharden AsshowninTable4,mostdatasetsareadequatelydescribed theobservedsourcespectra,thischangewasoffsetbymuchworse byasingle(absorbed)spectralmodelcomponent,withareduced- constraintsontheparameters.Hence,theabsorptioncolumnsand χ2 value at, or below, unity. However, in several instances the powerlawphotonindiceswereconsistentbothwithandwithoutthe reduced-χ2 isstillconsiderablyaboveone.Wehaveattemptedto inclusionofthepile-upmodel,withinthederivederrors,inallsix fitthesedatasetswithmorecomplex two-component models(see cases.Noneofthesesixfitswereabletoplacestrongconstraints Table5).Allthreedatasetsforwhichthespectralfitwasimproved onα.Wethereforeconsiderpile-upeffectstobenegligibleinthe bytheadditionofasecondspectralcomponent(excludingthepile- X-rayspectraofallbutoneoftheULXs. upcorrectiontoCXOUJ132938.6+582506) requiredthepresence The only case in which the pile-up model gave a much ofaverysoftspectralcomponent. Thesearediscussedonacase- by-casebasisinthefollowingsub-sections, wherewediscussthe 4 Seehttp://asc.harvard.edu/proposer/POG/index.html. spectralfitstoeachindividual ULX.Wedemonstratethespectral Chandraobservationsoffive ultraluminousX-raysources innearbygalaxies 7 Table4.Bestfitsoftwosimplemodelstothetwo-epochULXX-rayspectra. ULX(CXOUJ) Epoch WA*POa WA*DISKBBa LXb NHc Γd χ2/dof NHc kTine χ2/dof 034555.7+680455 2002-04-29 0.52±0.07 1.63+0.13 78.8/81 0.30+0.05 1.81+0.22 97.4/81 4.4(5.9) −0.12 −0.04 −0.18 2002-08-26 0.61±0.08 1.70+0.12 92.0/87 0.36±0.05 1.76+0.19 105.7/87 4.4(6.4) −0.13 −0.16 112037.3+133429 2002-04-06 0.022f 1.57±0.24 8.4/10 0.022f 0.95+0.3 12.3/10 0.7(0.8) −0.2 2002-07-04 0.16+0.09 2.20+0.34 4.4/10 0.022f 0.86+0.18 6.4/11 0.6(0.7) −0.13 −0.19 −0.14 120922.2+295600 2002-03-07 0.13+0.08 1.68+0.27 30.1/19 0.016f 1.46+0.27 37.7/20 2.3(2.6) −0.07 −0.22 −0.25 2002-06-08 <0.21 1.55+0.30 13.9/13 0.016f 1.43+0.44 18.4/14 1.7(1.9) −0.33 −0.24 123551.7+275604 2001-01-14 0.015f 1.91±0.09 69.9/52 0.015f [∼0.84]g 169/52 9.8(10.0) 2001-06-04 0.04±0.03 2.16+0.15 100/72 0.015f [∼0.66]g 222/73 11.6(12.5) −0.14 123558.6+275742 2001-01-14 0.17+0.08 1.98+0.23 26.3/22 0.015f 1.14+0.15 24.6/23 4.2(4.3) −0.07 −0.24 −0.13 2001-06-04 0.16±0.06 1.82+0.17 63.6/51 0.015f 1.30+0.12 49.2/52 8.9(9.1) −0.15 −0.10 132938.6+582506 2001-01-09 0.10±0.02 2.38+0.10 141/109 0.014f [∼0.7]g 260/110 5.7(6.9) −0.11 2001-05-02 0.10+0.05 2.96+0.25 41.1/46 0.014f 0.44+0.04 86.5/47 1.8(2.4) −0.04 −0.21 −0.03 Notes:aSpectralmodelcomponentsareshownaspertheXSPECsyntax,with“WA”representingacoldabsorptionmodel,“PO”a powerlawcontinuumand“DISKBB”theMCDBBmodel.bObservedluminosityinthe0.5–8keVbandinunitsof1039ergs−1. Figuresinparenthesesgivetheintrinsic(unabsorbed)luminosity.Wecalculatethesevaluesusingthebest-fittingmodelhighlighted byshowingitsχ2/dofinbold.cAbsorptioncolumn,inunitsof1022atomscm−2.dPowerlawphotonindex.eInneraccretiondisc temperatureinkeV.fValuefixedattheforegroundGalacticabsorptioncolumn(seeTable2).gParametervaluenotconstrainedby modelfit. Table5.Two-componentfitstotheULXX-rayspectra. ULX(CXOUJ) Epoch Modela α NH kT/kTin Γ χ2/dof ∆χ2b 120922.2+295600 2002-03-07 WA*(DISKBB+PO) - 0.81+0.37 0.12+0.05 1.77+0.36 16.7/17 13.4 −0.34 −0.04 −0.34 123551.7+275604 2001-01-14 WA*(DISKBB+PO) - <0.25 0.20+0.15 1.71+0.23 65.0/49 4.9 −0.07 −0.16 2001-06-04 WA*(DISKBB+PO) - 0.19+0.08 0.15+0.10 2.10+0.18 92.4/70 7.6 −0.14 −0.03 −0.26 WA*(MEKAL+PO) - 0.41+0.13 0.18+0.05 2.34+0.10 76.7/70 23.3 −0.19 −0.02 −0.14 132938.6+582506 2001-01-09 PILEUP*WA*PO 0.57+0.32 0.14±0.03 - 2.79+0.16 124/108 16.3 −0.18 −0.14 Note:aThemodelcomponentsandparametersareasperTable4,exceptfortheMEKALcomponentwhichisathermalplasma modelwithitsmetallicityfixedatsolarabundance,andPILEUPwhichisasdescribedinthetext.bImprovementinthe χ2statisticoverthesinglecomponentmodelbest-fit,fortwoextradegreesoffreedom(oneextrainthepile-upmodel). qualityofeachobservationofeachsourceinFigures3and4.These at∼5×1039ergs−1(versus6×1039ergs−1inASCA)whenex- showthedataandthebest-fitsinglecomponentmodelintheupper trapolatedtothe0.5–10keVband.HenceitappearsthatthisULX windowofeachpanel,andtheresidualswhenthedataisdivided hasbeeninaconstantspectralstateinthreeseparateobservations bythebest-fittingmodelinthelowerpanel. over2.5years. 4.2.1 CXOUJ034555.7+680455(IC342X-1) This ULX showed a fairly constant X-ray spectrum between the two observations, with an absorbed powerlaw continuum provid- ing a good fit to the data from both epochs. In each case the ab- 4.2.2 CXOUJ112037.3+133429(NGC3628X-2) sorption was roughly twice the foreground value (c.f. Table 2), withtheadditionalabsorptionof2–3×1021atomscm−2inex- ThisULXhadfadedtoaluminosityofonly∼ 7×1038ergs−1 cess of the integrated column through IC 342 at the ULX posi- in our Chandra observations, hence the spectra are of low qual- tion(∼ 8×1020atomscm−2;Crosthwaite,Turner&Ho2000). ity,andthespectralmodelswerenotstronglyconstrained.Apow- This implies a source of additional absorption intrinsic to, or in erlaw continuum was again the preferred model in both epochs, the environment of, CXOU J034555.7+680455. This may origi- though the MCDBB also gave a statistically-acceptable fit in the natein thenebula surrounding thisULX described by Roberts et second epoch. The source flux appeared to change littlebetween al. (2003). The Chandra spectra are consistent with the low/hard the epochs, perhaps fading slightly in the three months between spectral stateobserved by ASCAin2000 February24 - March1, observations.However,despitethelowquality,theintrinsicX-ray describedbyanabsorbedpowerlawcontinuumwithNH=0.64± spectrumshowedchangesbetweentheepochs,withthefirstobser- 0.07 ×1022atomscm−2 and Γ = 1.73 ±0.06 (Kubota et al. vation showing an intrinsically hard (Γ ∼ 1.57) but unobscured 2001). This state has also been interpreted as an anomalous/very powerlaw continuum, whereas the second observation showed a high state(Kubota, Done &Makishima 2002), whichwe discuss softer intrinsic slope (Γ ∼ 2.2) with a low-energy turnover due furtherinSection5.1.Theluminosityalsoappearslittlechanged, toabsorption. 8 T.P.Robertset al. Figure3.TheChandraACIS-S3spectraoftheULXs.Weshowthedataandbest-fitsinglecomponentmodelinthetopwindowofeachpanel,andtheratio ofthedatatothemodelinthelowerwindow.WeplotthefirstandsecondepochdataforeachULXintheleftandrightcolumnsrespectively,usingthesame normalisationscalingforeachpairofspectratoaidtheirdirectcomparison.Allspectraareshownusingthesameenergyscale,andtheratiovaluesareall displayedoverthesamerange. 4.2.3 CXOUJ120922.2+295600 fittothedata.Thepowerlawcontinuumslopeisconsistent(within thelargeerrors)inbothepochs,thoughtheinferredabsorptioncol- ThisnewULXhasaspectrumwhichisbest-fitbyapowerlawcon- umnismuchlowerintheabsenceofaverysoftcomponentinthe tinuum model in both epochs. However, the powerlaw fit is only secondobservation. statistically-acceptableinthesecondepoch.Inordertoimprovethe fittothefirstepochdatawetriedavarietyoftwo-componentspec- 4.2.4 CXOUJ123551.7+275604(NGC4559X-1) tralfitstothedata.Thebestfitscamefrommodelswithahighly absorbed,verysoftcomponentinadditiontoahardpowerlawcon- This is the most luminous ULX in the sample, with its observed tinuum. We show such a fit, with the very soft component mod- luminosityat,orabove,1040ergs−1.ItsX-rayspectrumwasob- elledbyaMCDBB,inTable5.Thisfitisstatisticallyacceptable served to soften between the observations, with the X-ray emis- forthedata,andanimprovementoverthesinglepowerlawcontin- sion in thesecond, more X-ray luminous epoch represented by a uummodelat99.97%(> 3σ)confidence.Note,however,thatthe significantlysofter powerlaw photon indexthan thefirst(∼ 2.16 additionofaclassicalblackbodyemissionmodel(withkT = 0.1 against ∼ 1.91). However, its X-ray spectrum is not well-fit by keV)ratherthanaMCDBBmodelprovidesanequallyacceptable eithersimplemodel,withtheMCDBBmodelrejectedathighsta- Chandraobservationsoffive ultraluminousX-raysources innearbygalaxies 9 Figure4.AsforFigure3,exceptforthefirstepochspectrumofCXOUJ132938.6+582506.Hereweplotthepile-upcorrectedspectralmodelovertheactual data,todemonstratethedegreeofpile-upintheobservedX-rayspectrum. tisticalsignificance. Two-component modelsincorporating avery best-fitwasobtained,withthecontributionoftheputativethermal softMCDBBplusapowerlawcontinuumofferedimprovementsto componentlimitedtonomorethan3%ofthe0.5–8keVflux. thefit,asshowninTable5,thoughonlyatthe69%and94%signif- However,theMEKALfitreliesheavilyonthefactthatablend icancelevelsinthefirstandsecondepochsrespectively.Aclassic oflinesbetween0.5–0.75keV(predominantlyOVIIandOVIII), blackbody component offeredasuperior improvement inthesec- whencombinedwithafactor∼2increaseinthemodelNH,gives ondepoch,atthe96%significancelevelaccordingtotheF-statistic. agoodmatchtotheadditionalsoftemissionapparentinthesecond- epochspectrum.Isthisthermalcomponentrealorsimplyanarte- Curiously, asubstantial improvement inthefittothesecond factoffittingafairlycomplexmodeltodatawithlimitedspectral epoch data was obtained when the MCDBB component was re- resolutionandpoorstatistics(N.B.∼600countsoutof∼2000in placedwithaMEKALsolarabundancethermalplasmawithkT ∼ thefullspectrumoriginatefromtheMEKALcomponent)?Clearly 0.18keV(Table5).Fortwoadditionalfreeparameters,thereduc- thereisnodirectevidenceinthesecond-epochspectrumforindi- tioninχ2of23.3impliesa99.99%significancelevel(i.e.>3.5σ) viduallines(thoughthisisperhapsconsistentwithlowtemperature intermsoftheF-test.Figure5showstheresultingbest-fitspectrum oftheplasmaandthe>100eVspectralresolutionoftheACIS-S andthecontributionoftheMEKALcomponent.Forcomparison,a detectorbelow1keV).Ontheotherhandwhenweallowtheabun- similar model (powerlaw plus kT ∼ 0.18 keV MEKAL) was fit- dance of O, Ne and Fe (which produce the most significant line ted to the first-epoch data and no significant improvement to the features in the kT ∼ 0.18 keV plasma) to vary in the VMEKAL 10 T.P.Robertsetal. apile-upfractionofinexcessof10%ofallevents(seeabove).The XSPECv.11.2parameterisationoftheDavis(2001)pile-upcorrec- tionmodelwasusedtocorrectforthiseffect,theresultsofwhich are shown in Table 5. This resulted in a much improved fit to a powerlawcontinuumspectrum,withtheinferredphotonindexsoft- eningconsiderably.Hence,whilstthetwouncorrectedspectraap- peartohaveverydifferenthardnesses,afterthecorrectionforpile- uptheyarebothrevealedtobeintrinsicallyverysoftX-rayspec- tra.Thephotonindicesofthepowerlawcontinuumfitsareconsis- tentwithintheerrorsafterpile-upcorrection,albeitwithaslightly hardervalueinthefarmoreluminousfirstobservation. 4.3 Temporalproperties Figure 5. The second epoch count rate spectrum of CXOU J123551.7+275604. Thebest-fit ofthe powerlaw plus MEKALmodelto 4.3.1 Short-termvariability theoverallspectrumisshownasthedottedlinewiththecontributionofthe We derived short-term lightcurves for seven of the eight ULXs MEKALcomponenthighlighted(solidline). in both of their observation epochs, using the CIAO routine LIGHTCURVE. The exception was CXOU J120922.6+295551, model, weobtain abest-fitof 1.3 solar and a90%lower-limitof which was too faint for this analysis with only 59 and 90 counts 0.33solar,implyingthatthereisastrongpreferenceforasubstan- detectedintotal perepoch. Eachlightcurve wasbinned toanav- tialcontributionfromemissionlines.However,thisisonthebasis erage(overtheobservation)of25countsperbin,givingtemporal of a spectral model which may not be correct. We conclude that resolutionsrangingfrom∼60secondsinthebestcase,to∼2500 aninterpretationof theadditional softfluxpresentinthesecond- seconds.Theresultingdatawastestedforgrossvariabilityusinga epochintermsoftheemergenceofabrightthermalcomponentis χ2 testagainstthehypothesisofaconstantcountrate.Fiveofthe potentiallyveryinteresting(seeSection5.3),albeithighlyspecula- ULXshowednostrongevidenceforshort-termtemporalvariabil- tive. ityineitherepoch,withareduced-χ2statisticof1.25orless.How- Finally,wenotethatasthispaper wasundergoing therefer- ever,twoULXshowedstrongevidenceforshort-termvariabilityin eeingprocesstheseChandraspectrawerepublishedbyCropperet oneoftheirtwoobservationepochs.Theselightcurvesareplotted al.(2003)aspartofanXMM-Newtonstudyofthissource.Theydo in Figure 6. The appropriate χ2/degrees of freedom statistics are not reportthepossibledetectionof aMEKALcomponent, though 39.5/14forthe2002JuneobservationofCXOJ120922.2+295600 theydidnotexplicitlylookforit.Insteadtheyfitaparticularmodel (which had a 1407 second resolution), and 190/24 for CXOU totheChandradata,basedonfitstothesuperiorquality(andlater J123558.6+275742 in its 2001 January observation (at 404 sec- epoch) XMM-Newton spectra, comprising a sub-solar abundance ondresolution).Theotherobservationsofbothsourceswerecon- absorber (TBVARABS, set at 0.31 times solar) applied to pow- sistentwithaconstant flux,implyingthattheshort-termvariabil- erlaw plusblackbody emission components. Wehaveapplied the ity states are themselves a transient phenomenon. These highly- Cropperetal.model,plusvariantsincorporatingsoftMCDBBand variablestatesbothoccurwhentheULXinquestionis(onaverage) MEKAL components, to the second-epoch spectrum and do get a atthelowerofitstwoobservedX-rayfluxes. marginalimprovementineachcasewithrespecttoouroriginalfits WealsoperformedaKolmogorov-Smirnovtestcomparingthe (∆χ2 ∼2−4);neverthelessthepowerlawplusMEKALmodelstill cumulativephotonarrivaltimeswiththeexpectedarrivaltimes,as- clearlyprovidesthebestresultintermsoftheminimumχ2. suming each source flux was invariant with time. This provides a separate indicator of short-term variability, free of any possi- ble gross binning effects, and sensitive to lower-amplitude (10 – 4.2.5 CXOUJ123558.6+275742(NGC4559X-4) 20%) sustained changes inthe ULXcount ratesthan theχ2 test. We performed this test on all eight ULXs. It confirmed that the ThisobjectwastheonlyULXinthesamplewithanX-rayspec- 2002 June observation of CXO J120922.2+295600 was variable trumclearlybest-fitwithaMCDBBmodel.Theinferredinnerac- at > 99% probability. Curiously, the 2001 January observation cretiondisctemperatureswerebothgreaterthan1keV,consistent of CXOU J123558.6+275742 was only variable at the ∼ 90% withpreviousASCA,andsomeChandraobservationsofULXwell- level according to the Kolmogorov-Smirnov test. Again, none of fitbythismodel(e.g.Colbert&Mushotzky1999;Makishimaetal. theotherULXappearedvariableaccordingtothistest. 2000;Robertsetal.2002).TheluminosityoftheULXmorethan IntheJanuary2001lightcurveofCXOUJ123558.6+275742 doublesinthefivemonthsbetweenobservations,andthisisassoci- therearefour“peaks”intheX-rayemission,separatedbyregular atedwithaslighthardeningoftheX-rayspectrum,withthebest-fit intervalsof6–7bins,duringwhichtheX-rayfluxis4–5times temperatureoftheinneraccretiondiscrisingfrom∼1.1to∼1.3 brighterthaninthelowestfluxbins.Thishintsatapossiblechar- keV. acteristictimescale for the variability of ∼ 2500 seconds for the peaks.Itisalsonoticeablethatthelightcurvebehavessimilarlybe- tweenthepeaks(thisquasi-repetitivebehaviouristhelikelyreason 4.2.6 CXOUJ132938.6+582506(NGC5204X-1) for the lower probability of this source being variable according TheX-rayspectrumofthisULXwasfarbetterfitbyapowerlaw totheKolmogorov-Smirnovtest).Similarregularvariabilityinthe continuum model than a MCDBB in both epochs, with it consti- formof“quasi-periodicflaring”hasrecentlybeenrevealedinthree tuting a statistically-acceptable fit in the second epoch. It was a separateobservationsofanextremelyvariableULXinM74,CXO poorerfitinthefirstepoch,thoughthisobservationwasaffectedby J013651.1+154547(Kraussetal.2003).However,thecharacteris-