Table Of Content

1 Using archival observations recorded over a 5+ year timeframe with the NASA Galaxy Evolution Explorer (GALEX) satellite, we present a study of the ultraviolet (UV) variability of4360quasarsofredshiftsuptoz =2.5thathaveopticalcoun- terpartsintheSloanDigitalSkySurveyDR5spectroscopiccat- alog of Schneider et al. (2007). The observed changes in both the far UV (FUV: 1350 - 1785A˚) and near UV (NUV: 1770 - 2830A˚) AB magnitudes as a function of time may help differ- entiatebetweenmodelsoftheemissionmechanismsthoughtto operate in these active galaxies. A list of NUV and FUV variable quasars was derived from the UV light-curves of sources with5 ormoreobservationalvisitsbyGALEX thatspanneda time-frame> 3 months. By measuringthe errorin the derived mean UV magnitude from the series of GALEX observations foreachsource,quasarswhoseUVvariabilitywasgreaterthan the3-σvariancefromthemeanobservedvalueweredeemedto 1 be(intrinsically)UVvariable.Thisconservativeselectioncrite- 1 0 rion (which was applied to both FUV and NUV observations) 2 resultedinidentifying550NUVand371FUVquasarsasbeing statisticallysignificantUVvariableobjects. n a J 3 1 ] A G . h p - o r t s a [ 2 v 1 9 1 2 . 1 0 1 1 : v i X r a Astronomy&Astrophysicsmanuscriptno.welsh (cid:13)c ESO2011 January14,2011 GALEX observations of quasar variability in the ultraviolet BarryY.Welsh1,3,Jonathan M.Wheatley1,3 andJamesD.Neill2 1 SpaceSciencesLaboratory,UniversityofCalifornia,7GaussWay,Berkeley,CA94720 2 CaliforniaInstituteofTechnology,MC405-47,1200EastCaliforniaBoulevard,Pasadena,CA91125 3 EurekaScientificInc.,2452DelmerStreet,Oakland,CA94602 Preprintonlineversion:January14,2011 Keywords.Ultraviolet:Galaxies-Galaxies:quasars 1. Introduction are poweredby an underlyingstochastic process such as a tur- bulent magnetic field. Clearly there is still much debate as to ThefluxemittedfromquasarsandotherActiveGalacticNuclei theinterpretationoftheseopticalvariabilitystudies(Kozlowski (AGN) such as Seyferts, BL Lacs and Blazars has been found et al. 2010), and other emission mechanisms such as bursts of to varyaperiodicallyon time-scalesfromhoursto decadesand supernovaeevents close to the nucleus and extrinsic variations fromX-rayto radiowavelengths.Inthe case ofradio-loudand due to line-of-sight microlensing events cannot (as yet) be to- radio-quietquasarsopticalfluxchangesovertimescalesasshort tallydiscounted(Ulrichetal.1997).Itismostlyagreedthatthe asminutesandwithamplitudesrangingfromafewhundredths level of quasar variability increases with decreasing rest wave- to a few tenths of a magnitude (often refereed to as ‘optical length,whichisrelatedtotheobservationthattheirspectrabe- microvariability’),havebeenwidely reported(Miller,Carini& comebluerintheirbrightphases.Severalauthorshaveclaimed Goodrich 1989, Stalin et al. 2005). The mechanism(s) respon- a variability-redshiftcorrelationin the visible, althoughthe ac- sible for this short-term microvariability is still debated, with tualsignofthecorrelationinthestudiesisdebated(Giveonetal. muchofthevisiblespectralvariabilitybeingexplainedbyther- 1999,TreveseandVagnetti2002).Forexample,VandenBerket malprocessesassociatedwithablack-holeaccretiondiskand/or al.(2004)haveclaimedevidenceforredshiftevolutioninquasar non-thermal processes associated with the relativistic beaming variability,withgreatervariabilityoccurringathigherredshifts. ofajet(Ramirezetal.2010,Pereyraetal.2006).InfactSesaret However, such a trend is only revealed when an allowance is al.(2007)havefoundthatnearlyallquasarsvaryatsomelevel made for rest wavelength,absolute magnitudeandtime lag for when observed over long enough times scales, with ∼ 90% of each redshift interval. Such correctionsto the observed optical quasarsfoundin theSloan DigitalSkySurvey(SDSS) varying variabilitydataarecrucial,sincespectralfeaturesmaymovein by>0.03magoverasixyeartimescaleatvisiblewavelengths. andoutofthephotometricpass-bandswithchangesinthetarget Althoughtheactualcauseof(long-term)variabilityintheseob- redshift. jects is also still a matter of intense debate, variability studies QuasarsandAGNhavealsobeenwellstudiedatultraviolet of these sources are important since (in principal) they can set (UV) wavelengths, such that typically their far UV spectra ap- limitsonthesizeoftheemittingregionsandcanalsoconstrain pearasseveralstrongemissionlines(formedingaslocatedfur- modelsoftheemissionmechanismsthoughttooperateinthese ther away from a central black hole and excited by continuum activegalaxies. radiation)superposedonanunderlyingaccretion-drivencontin- Numerous variability studies have been aimed at establish- uum.Incontrast,theirnearUVspectracontainonlyaUVcon- ingcorrelationsbetweenquasarvisiblevariabilityandimportant tinuumwitha‘bump’at2800A˚ superposedwiththeMgIIemis- physicalparameterssuchassourceluminosity,redshift,time-lag sionlines(VandenBerketal.2001).TheUVvariabilityofthese (i.e.thetimeperiodbetweentwosetsofobservations)andradio objectshasbeenwidelyreported,particularlywithreferenceto loudness(Helfandetal.2001,VandenBerketal.2004,deVries reverberationmappingobservationsthatinprincipalallowmea- etal.2005,Ramirezetal.2009).Theestablishmentofsuchcor- surementof the size of the broad emission-line region and can relationsisimportantsincethereisincreasingevidencethatthe infer the mass of the alleged centralblack hole (Peterson et al. source of quasar variability is linked to the accretion disk sur- 1993, Kaspi et al. 2000, 2007). UV photons are generally be- rounding the central supermassive black hole engine (Peterson lieved to be radiated from the quasar supermassive black hole etal.2004,Pereyraetal.2006,Woldetal.2007).Forexample, accretion disk when its gravitational binding energy becomes Bachev (2009) has studied the observed time lags between the released, while X-rays are produced in the disk corona. A sta- continuumemissionfromseveralopticalbandsforalargesam- tistical survey of the UV variability of 93 quasar/AGN, as ob- ple of quasars. He demonstrates that the lags are broadly con- served with the I.U.E. satellite, was carried out by Paltani and sistent with an emission model in which the optical variations Courvoisier (1994) who found no strong evidence for correla- are largely due to the reprocessing of the central X-ray radia- tionsbetweenvariabilityandparameterssuchasredshift,lumi- tioninasurroundingthinaccretiondisk.However,bothKellyet nosityor UV spectra-index.However,changesand instabilities al.(2009)andMacLeodetal.(2010),usinga‘dampedrandom intheaccretiondiskmustsurelybelinkedtotheirobservedUV walk model’, have suggested that these optical emission varia- variability,butrecognizingandunderstandingtheemissionsig- tions result from thermalfluctuations in the accretion disk that naturesfromsuchfeaturesstillremainsproblematic. Welsh,Wheatley&Neil:QuasarUVvariability 3 ToproceedfurtherwithstudiesofUVvariabilitywerequire following criteria. Firstly, we only selected quasars with either accesstofarlargerdata setsthanpreviouslypossible,andwith (i) GALEX observationsin the All-skyImagingSurvey(AIS) thelaunchoftheNASAGalaxy Evolution Explorer (GALEX) that were brighter than NUV = 19.0 and with an exposure mag satellite (Martin et al. 2005) we now have a significantly large time of at least 100s, or (ii) those quasars with GALEX Deep database of far UV (FUV: 1350- 1785A˚) and near UV (NUV: Imaging Survey (DIS) observations of 1500s duration or more 1770 - 2830A˚) observationsof quasars and AGNs that covera (seeMorrisseyetal.2007foradetaileddiscussionoftheobser- timespanofover5years.Trammelletal.(2007)havepresented vational modes of the GALEX satellite). A further restriction apreliminaryanalysisofthebasicUVpropertiesof6371SDSS wasappliedsuchthatonlythosequasarsthathad5ormoreob- DataRelease3(DR3)selectedquasarsobservedduringthefirst servationalvisitsbyGALEX (andweredetectedinatleastone fewyearsoftheGALEX mission.Onemajorconclusionfrom ofthosevisits)wereselected.Wenotethatanobservationalvisit thisstudywasthat>80%oftheDR3listedquasarshadacor- byGALEX doesnotnecessarilyresultinadetection,sincean respondingGALEX NUVdetection,implyingthatmanymore intrinsically variable object may fall below the detection limit quasarsshouldbedetectedduringtheremainingyearsofopera- throughoutaparticularvisit.Wewereguidedinthesechoicesby tionoftheGALEX satellite.Bianchietal.(2009)haveshown Figure4 ofTrammeletal.(2007)whichshowsthetypicalrms thatGALEX isalsoidealfordetectingquasarswithblue(FUV- magnitudeerrorsforanAISandDISobservationwithGALEX. NUV) colorsthat are difficult to identifyfrom visible observa- Thoseresultsdemonstratethatthemeasurementerrorforatypi- tionsalone.Sincethepublicationoftheseinitialstudies,newer calAISobservationofarelativelyUVbright(NUV =19.0) mag andfarlargerversionsofboththeSDSSandGALEX catalogs object is approximately equal to the measurement error for a arenowin-hand.Inparticular,thecatalogofquasarscontained 1550sDIS observationof a faint (NUV > 21.0)source. In mag in the SDSS Data Release 5 (DR5) published by Schneider Figure1weshowaplotofthenumberoftheseselectedquasars et al.( 2007) lists some 77,000+ objects (with associated red- asafunctionofthenumberofobservationsintheGALEX NUV shiftandluminosityvalues),themajorityofwhichshouldhave andFUVchannels.Itcanbeseenthatthereisasignificantnum- UV counterpartsobservablewithGALEX. Furthermore,since ber of sources that have been visited between 5 and 20 times, manyquasarsareknowntobevariable,thispotentiallyenables withseveralsourcesbeingrepeatedlyobserved>100times.On ustostudytheUVvariabilityofasignificantlylargenumberof applicationoftheselectioncriteriaofbothexposuretimeand≥5 thesesources. visits,thisresultedinalistof4360quasarsthatweredetectedin The largest group of known variable sources listed in the atleastoneobservationalvisitbyGALEX. GALEX UVVariability-2catalogareactivegalaxies(Wheatley In Figure 2 we show a plot of these selected quasars as a etal.2008).Thissampleofsourceswasderivedsolelythrough function of the timespan covering all of their 4360 NUV and selection due to their observed UV variability, without prior 2709FUV GALEX observationalvisits(NB: we note thatthe knowledgethatsuchsourceswerein factAGN. Thesevariable sample of FUV selected quasars is smaller than the numberof sourcespossessedredshiftsashighasz ∼2.5andfluxchanges NUV selected ones due to operational constraints on the FUV aslargeas∼ 2.3ABmag.werefoundin bothof theGALEX detector that often precludedsimultaneous FUV and NUV ob- nearandfar UV bandsforAGN thatwere repeatedlyobserved servationsofasource).ThisFiguredemonstratesthatourquasar over a 3.5 year time period. Based on the apparent quality of sample is not biased towards any particular timeframe of vari- these data, we decided to search the entire GALEX archival ability,withthemajorityofselectedsourceshavingbeenrepeat- database(whichnowspans5+yearsofGALEX observations) edlysampledovera1to5yeartimespan.Wenotethatalthough forspatialmatchestoquasarspreviouslyidentifiedandlistedin thetimebetweenmanyobservationalvisitstoaparticularquasar theSDSSDR5catalogandthen,forthosesourcesthathadmul- mayspanseveralyears,thetargetmayberepeatedlyobservedon tiple visitmeasurements,determinewhichonesexhibitedmea- time-scales<1dayduringaseriesofobservationalvisitswith surable UV variability. This Paper discusses which GALEX GALEX. sources were identified with SDSS quasar counterparts, which AllthedatashowninFigures1and2clearlyindicatethatour onesshowedlong-termUV variability(asmeasuredovertime- sampleofquasarsisbothlargeandcontainssignificantnumbers frames>3months)andwhich,ifany,correlationsexistbetween of objects that have multiple (≥ 5) observational visits spread this UV variability and other astrophysicallyinteresting source overasignificantlylongtimespan.Inordertoensurethatquasars parameters. withpotentiallyshort-termvariationsofafewdayshad≥5ob- servations,wehavefurtherrestrictedourtargetselectiontothose ones whose visits spanned a time-frame > 3 months. This se- 2. Data Analysis lection criterion resulted in a final list of 3521 potentially UV Inordertoproducealistofquasarsthatwassuitableforanul- variable quasars, of which 1942 had 10 or more observational traviolet variability study we first searched the GALEX GR5 visits.ToensurethatalloftheUV-selectedsourceswerenotin dataarchivalrelease(attheMulti-MissionArchiveattheSpace any preferential direction on the sky, in Figure 3 we show the Telescope Institute, MAST) for sources that spatially matched galactic locations of our SDSS selected sample that have been optically identified quasars listed in the SDSS DR5 spectro- observed 5 or more times. Since our selection of targets was scopic catalog of Schneider et al. (2007). A spatial match was derived from the (northern hemisphere) SDSS survey catalog, deemedpositiveifthecentralpositionofthematchingGALEX clearly only quasars with declinations> -12◦ are available for source was within ± 8 arc sec of the listed SDSS position, present study. Although there are large areas of the sky where thisnumberbeingslightlylargerthanthepositionalerrorofthe we have no UV-selected quasars, Figure 3 shows no obvious GALEX observationsforapointsourcedetection.Thissearch sight-line preference apart from most sources being observed resultedin77,249quasarsthatwerecommontoboththeSDSS at high galactic latitude away from the effects of absorptionof andGALEX ultravioletsourcecatalogs. ultraviolet flux by gas and dust in the galactic plane. We then In order to produce a list of quasars whose UV variabil- searchedthislistof3521sourcesforthe oneswhichpossessed ity could be accuratelyassessed as a functionof time, we then statisticallysignificantUVvariabilitybetweeneachoftheirob- down-selected the large list of UV-selected quasars using the servationalvisits.Althoughthequasarnatureofthesesourcesat 4 Welsh,Wheatley&Neil:QuasarUVvariability visiblewavelengthswasalreadyknownfromtheSDSSDR5cat- variationis0.11mag,whichwasdeterminedfora16.5ABmag alog,wedecidedthatstatisticalconfirmationoftheirassociated quasarofredshiftz =0.4. UVvariabilitywasalsorequiredforthisstudy.Duetothelimited The3-σthresholdcutisshownsuperposedonbothFigures4 sensitivityoftheGALEX observationsthisnaturallyrestricted and5respectivelyfortheNUVandFUVobservedmagnitudes. ourstudytothemoreUVvariablequasarsinthissample. Applicationofthisstatisticalsignificanceselectioncriterionre- sultedintheidentificationof550quasarsasbeingNUVvariable and 371 quasarsas being FUV variable, with the vast majority 2.1. UVVariability Selection oftheFUVvariableobjectsbeingcommonwiththoseidentified In orderto determine which of the 3521 selected quasarswere as variable in the NUV. For simplicity, from now on we term variableatleastonceduringtheirseriesofobservations,were- this sub-set of selected variableobjects as UVQs (i.e. ultravio- quireknowledgeofthesize oftheerrorassociatedwiththe es- letvariablequasarstellarobjects).Thetypicallevelsofobserved timation of each of the NUV and FUV magnitudes measured UVvariabilityforeachoftheseUVQs,canbebestviewedwhen foreachvisitto a particularsource.Thiserrorin thismeasure- shown as NUV and FUV light-curves (i.e. the AB magnitude ment is affected by the length of the observation by GALEX, for each visit versus observation date). In Figure 6 we show thepositionofthesourceontheUVdetectorandtheassociated the FUV and NUV light-curves for a relatively bright (SDSS varyingskybackgroundsignalwhicharebothdependentonthe J163915.80+412833.6)andafaint(SDSSJ221629.33+002340) (changing)orbitalviewingcharacteristicsoftheGALEX satel- quasar,thatwerebothrepeatedlyobservedovera4+yeartime- lite. Thus, in order to progress further in our study we clearly framewith GALEX. Itshouldbe notedthat faintandvariable require a more consistent estimation of the measurement error objectscanfallbelowtheGALEX thresholdofdetectionduring fortheUVmagnitudesderivedfromeachobservationalvisitby aseriesofobservations,andhencethederivedmeanmagnitude theGALEX satellite.Fortunately,Gezarietal.(2008)havees- (whichis computedfromonlyactualdetectionswithina series tablished a robustnumericalmethodof assessing the statistical ofvisits)willbeofahighervaluethanhadtheobjectbeende- significance of the variability of a series of GALEX UV ob- tectedduringallvisits.Thus,thecomputedchangeinmagnitude, servations of the nuclei of normal quiescent galaxies, which is whichisameasureoftheobject’svariability,willbeunderesti- directlyapplicabletoourpresentQSOdataset.Themethodin- mated.FinallywenotethatwedidnotdiscoveranyUVQsthat volves deriving the quantity, σ(<m>), which is defined as the werehighlyvariableintheFUVbutnotvariableintheNUV. standarddeviation(σ)fromtheweightedmean(<m>)ofase- ries of observed UV magnitudes (m). For a series of observa- 2.2. TheStructureFunction tionsofthesameobject,thequantityofσ(<m>)willvarysuch thatabovesomestatisticallevelanyobservedchangewillbedue ThemeanvariabilityofUVQsisoftendescribedbyaStructure tointrinsicsourcevariability.Gezarietal.(2008)estimatedthe Function,whichisthecorrelationbetweentwopointsinasetof PoissonerrorsforaseriesofUVmagnitudedeterminationsus- measurementsofanobject’slightcurve(Simonettietal.1985). ing GALEX (see equation (2) in that paper). They found that Thus, in its simplest form, it provides a statistical measure of themeasuredstandarddeviationfromameanUVsourcemagni- variability as a function of the time-lag between many sets of tudewasdominatedbysystematicerrorsthatwerenotaccounted observations.Howeverwenote,asseveralauthorshavepointed purelybythe photometricerror.We note thatthe majorcontri- out,therearelimitationsinitsapplicabilitytoderivetimescales bution to the error in the estimation of a UV magnitude using and amplitudes of quasar variability (Vio et al. 1992). In addi- GALEX is thoughtto be due to the (often varying) sky back- tion, quasar variabilityis most probablya complexfunctionof groundlevel.Thustobeconservative,Gezarietal.(2008)used several(possiblyinterdependent)physicalparameterssuchaslu- acutof5-σabovethemeanmagnitudeleveltodetermineintrin- minosity, rest-frame wavelength, redshift, black hole mass and sicsourcevariability. time-lag. Helfand et al. (2001) have pointed outthat ‘the com- Valuesofσ(<m>)forourquasarsamplewerealsoderived plexinterdependenceoftheobservedvariability(ofquasars)on fromequation(2) of Gezariet al. (2008) foreach set of obser- wavelength,redshift,bolometricluminosity,radioloudnessand vational visits. The values of NUV and FUV AB magnitudes possibly other parameters makes it difficult to draw definitive recordedfor each visit to a selected quasar,together with their conclusionsregardingthephysicsofthecentralengineorquasar measurementerrors,wereretrievedfromtheGALEX GR5data evolution’.Thus,‘caveatemptor’shouldbethewatchwordsfor archivalreleaseattheMAST.Weusedcircularaperturemagni- any discussion of quasar variabilitydependencies,and thus we tudesofa6arcsecradius,aslistedintheMASTcatalog,which employamethodbywhichthesedependenciesofvariabilitycan wereobtainedusingtheGALEX pipe-linegeneratedSExtractor beplausiblydisentangled. magnitude estimator (Bertin & Arnouts 1996). In Figure 4 we Forourpresentstudy,inwhichmostoftheidentifiedUVQs showaplotofσ(<m>)versus<m>forthemultipleNUVob- each have > 5 measurements, we start by using the standard servationsof3521oftheselectedquasars.Asonemightexpect, formulationoftheStructureFunction(SF)givenbydiClemente thevalueofσ(<m>),whichisessentiallya measureoftheer- etal.(1996): rorinthederivedmeanABmagnitudeofaquasar,increaseswith thefaintnessoftheobjectunderstudy.Allofthese3521quasars SF=pπ/2∗h|∆m(∆τ)|i2−hσ2i n weredetectedasbeingoptically variableintheSDSSDR5cat- alog,andalthoughtheyalsoallmaybeUVvariable(thoughthis where ∆m(∆τ) is the difference in UV magnitude be- hasneverbeenactuallyproveninanylargestudy),wehavede- tween any two observationsof a UVQ separated by ∆τ in the cidedtotakeaconservativeapproachinselectingonlythemore quasarresttimeframe,andσ 2 isthesquareoftheuncertainty n UV variable objectsfor furtherstudy in this paper. To this end in thatdifference(beingequalto thesum ofthetwo individual wehavedecidedtouseathresholdcutusingastandarddeviation observations’ errors in quadrature). The final SF for all of our of3-σ abovethelevelofσ(<m>)foragivenmeanmagnitude UVQ observations is just the rms of the summation for all (<m>) to reveal significant UV variability. On application of possible combinationsof magnitude measurements for a given thisthresholdcutwefindthatthesmallestdetectablemagnitude valueoftime-lag. Welsh,Wheatley&Neil:QuasarUVvariability 5 In Figure 7 we show log-log plots of the SF (i.e. the mea- rest-framefrequencyandtotheincreaseofrest-framefrequency sure of UV variability)versusthe rest frametime lag (in days) scanned at high redshift for a given frequency(di Clemente et forboththeNUVandFUVobservationsofeachofthestatisti- al. 1996). The Structure Function derived in the previous sec- cally significantvariable UVQs selected from Figures4 and 5. tion,whichdescribedUVvariabilityasafunctionofrest-frame FollowingtheworkofVandenBerketal.(2004)wehaveplot- time-lag, took no account of any of the other possible inter- tedthedataaveragesin10time-lagbins,chosenwithequalin- dependenciessuchasluminosityorredshift.Torevealsuchmulti tervalsinthelogarithmicrest-frametimelag.Notethatthedata inter-dependeniceswouldrequire4-dimensionalplots,andthus as presentedin Figure 7 makeno accountforanyotherplausi- weneedanalternativemethodtoexploreotherpossiblecorrela- blevariabilitydependenciessuchasluminosity,redshiftorradio tionswiththeUVSF.Therefore,wecloselyfollowtheworkof loudness. In this form, the variability function is often termed VandenBerketal.(2004)whohavedescribedamethodthatat- an ‘ensemble’ Structure Function. Error bars were determined temptstodisentanglethesepossiblephysicalinterdependencies. bypropagatingthermserrorsderivedfortheaveragemagnitude Sinceeachofthe10time-lagbinsoftheUVQsplottedinFigure differenceand measurementuncertaintyin quadraturefor each 7 contains objects with a wide-rangeof other (differing)phys- observation. ical parameters, we start by subdividing the targets contained ThedatashowninFigure7revealtwoimportantfacetsofthe within each time-lag bin into separate (and assumed to be de- UV variability of quasars. Firstly, the amplitudes of variability generate) physical parameters. This then allows us to obtain a inboth theNUVandFUVarefarlargerthanthoseobservedat set of (quasi-independent) variability relations with respect to visiblewavelengths(whicharetypicallyonly∼0.03magrms), the SF for a particular time-lag, UVQ luminosity, redshift and andsecondly,thelevelsofFUVvariabilityaregreaterthanthose radio loudness. In the following subsections we present these observedintheNUV. variability relationships, and then briefly discuss their possible The NUV SF shown in Figure 7 increases linearly (with a astrophysicalmeaninginSection4. slope of 0.439) from a rest frame time-lag of 1 to ∼100 days, withadefiniteturn-overandsubsequentflatteningfortime-lags >∼ 300 days. The actual best-fit to all of the NUV data is a quadratic in log-log form. A similar correlation of variability 3.1. TimeLag-UVVariability Dependence with time-lag has also been found in many other studies, and the flattening of the curve at long time-lags has been reported by both Vanden Berk et al.( 2004) and Cristie et al. (1997). FirstlyweexplorethedependenceofUVQNUVandFUVvari- HoweverwenotethatdeVriesetal.(2005),inanopticalsurvey ability on the rest frame time-lag, in a form that is (quasi) in- of long-termquasar variabilityup to timescales of ∼ 40 years, dependentof the parameters of luminosity and radio loudness. hasobservedthatthisapparentflatteningoftheSFonlylastsfor WedividealloftheUVQsinourUVvariabilitysample,whose a few yearsbefore continuingits linear increase with time-lag. redshift (z) values span the range of 0 to 2.5, into three red- Inessence,theyclaimnosignificantturn-overoftheirSF,such shiftbinsofsize0<z <0.47,0.47<z <1.33andz >1.33. that visible quasar variability increases monotonically with in- The physical relevance of these 3 redshift bins with respect to creasingly long time-lags. Although our sampling of UVQs is observationsthroughthe two GALEX pass-bandswill be dis- limitedtotime-scalesof<5years,inspectionofthebehaviorof cussed in Section 4. In Figures 8 and 9 we show log-log plots theNUVSFcurvesuggeststhattherateofincreaseinUVvari- oftherespectiveNUVandFUVStructureFunctionsversusrest abilityasafunctionoftimedecreaseswhenobservedovertime- frametime-lag,with the data averagesbeingplotted separately lags greater than ∼ 2 years. The FUV Structure Function, also foreachofthesethreeredshiftbinsatanygivenrestframetime- shownin Figure 7,followsthe same generaltrendas the NUV lagvalue.Bothoftheseplots,inagreementwiththesimpleren- curve,exceptthat we have observeda greater level of baseline sembleSFsshowninFigure7,showthattheamplitudesofboth variability of UVQs in the FUV. However, the change in FUV the NUV and FUV variability of UVQs increase as a function variabilityasafunctionoftime-lagislesspronouncedthanthat of rest frame time-lag, irrespective of the value of quasar red- intheNUV,asindicatedbythebest-fitlineofslope0.292.The shiftfortime-lags< 200days.Fortime-lags> 300daysthere FUVdata,likethoserecordedintheNUV,alsoshowaflattening isa pronouncedrolloverin theNUV SF forallredshiftvalues, inUVvariabilityfortime-lags>∼300days. which is also observed in the lower S/N FUV data. To further demonstratethisroll-overeffectwehavealsoincludedseparate 3. Variability CorrelationPlots linearbest-fitsforeachofthe3redshiftbindatasetsinFigures 8and9.Thesestraight-linefitsshowthatrateofincreaseofthe Previous studies of quasar variability in the rest-frame of the NUVSFuptoa time-lagof∼200daysisgreatestforsources optical/UVregimehaverevealedseveralapparentphysicalcor- ofredshift0.47<z <1.33,aneffectthatisalsomirroredinthe relations. For example, trends of variability with wavelength FUV data set of Figure 9. For time-lags > 300 days it is clear andluminosity(firstdiscoveredbydiClementeetal.1996and thatthesamelinearfitsareinappropriateforalloftheNUVand Hooketal.1994,respectively)havebeenconfirmedbyVanden FUVdata,suchthatafarslowerincreaseintherateoftheNUV Berk et al. (2004) who have found that quasars are more vari- andFUVSFsoccurs. able at shorter rather than longer wavelengths and the ampli- tude of variability decreases with increasing luminosity (i.e. intrinsically brighter objects vary less). These results, and the Figures8and9bothshowthatfortime-lags<20daysthere SF method used to obtain them, have recently been verified is a definite trendforthe higher-z UVQsto be moreNUV and for a sample of quasars observed over a 10 year period by FUVvariablethantheirlow-z counterparts.Thistrendhasalso MacLeodetal.(2010).Furthermore,radio-loudquasarsappear beenseenforvisiblequasarsbydeVriesetal.(2005).However, morevariablethantheirradio-quietcounterparts(Helfandetal. fortime-lags>100daysthistrendmostlydisappearssuchthat 2001), and the apparent increase in variability for higher red- boththeFUVandNUVSFvariabilityamplitudesaresimilarfor shift quasars is mainly due to the increase of variability with allvaluesofUVQredshiftwhenobservedoverlongtime-lags. 6 Welsh,Wheatley&Neil:QuasarUVvariability 3.2. Luminosity-UVVariability Dependence 20cmradiofluxdensityasmeasuredintheVLAFIRSTsurvey andusingthe absoluteSDSS i magnitudesanddistanceslisted Unlikeredshift,absoluteluminosityisanintrinsicphysicalprop- in Schneider et al. (2007) to derive a measure of their optical ertyofaquasarandthusaninvestigationofUVvariability(i.e. luminosity, we compute values of the ratio of radio to optical SF) as a function of this parameter should be less affected by luminosity in the rest frame (R∗). Using equations (1) and (2) otherphysicalinterdependencies.Wemakeuseofthevaluesof ofSramek&Weedman(1980)toderiveR∗,weusethewidely luminositygivenin Schneideretal. (2007) foreach ofourUV accepteddefinitionofR∗ <1forradioquietsourcesandR∗ > variablesources,andnotethatthesevalueswerecalculatedus- 1forradioloudsources.Sincemanyofourvariablequasarsdo ingextinctioncorrectedSDSS(absolute)i magnitudesandwith nothavelistedvaluesfortheirradiofluxes,ourNUVsampleis theassumptionthatthequasarspectralenergydistributioninthe restricted to 484 radioloud and 66 radio quietsources and our ultraviolet-opticalcanberepresentedbyapowerlaw. FUVsamplecontains327radioloudand44radioquietsources. Westartbyseparatingthepossibledependencyeffectsofrest Hence, the reader is reminded that we are dealing with small frametime-lagbydividingthedataintofourtime-lagbinsof1- numberstatisticsfortheradioquietsourcedatainbothFigures 10days,11-100days,101-500daysand>500days.InFigure 12and13. 10weshowlog-logplotsofboththeNUVandFUVvariability Firstly, inspection of both Figures clearly shows that the amplitudes (i.e. the SF’s) as a function of (averaged) absolute overalllevelofSFvariabilityisgreaterintheFUVthantheNUV magnitude in bins of 2 mag width for each of the aforemen- for bothradioloud andradio quietUVQs. Thisis merely a re- tionedtime-lagbins. Thebest-fitstothese data showthatthere statement of the behaviorof the entire UVQ sample, as shown is a slight decrease in both the NUV and FUV SFs as a func- in Figure 7. However Figure 12 also indicates that radio loud tionofincreasingluminosityfortime-lags>100days.Similar sources are ∼ a factor 2 more NUV variable as a function of (visible)luminosityplotshavealsobeenconstructedbyVanden time-lagthanradioquietsources,andtheplotalsoshowsalarger Berk et al. (2004) for a sample of 25,000quasars in the SDSS scatterintheshapeoftheNUVSFforradioloudthanforradio survey,in which they similarly foundthatintrinsically brighter quiet UVQs. However, there is only a marginal difference be- objects vary less than intrinsically fainter ones. However, for tween the best-fit slopes to these NUV SF data for both types shorter time-lags we find the opposite behavior in which the ofradiosource.Therefore,ourdatawouldappeartosupportthe NUV and FUV SFs increase as a function of increasing lumi- notion that NUV variability is not strongly linked to the radio nosity(wenotetheapparentexceptionoftheNUVSFfortime- properties of UVQs, and as such, the physical mechanism re- lags < 10 days, but this may be due to small numberstatistics sponsiblefortheNUVvariabilityofradiodetectableUVQsop- with associated larger errors). The NUV and FUV data shown eratesirrespectiveoftheirradioflux. inFigure10alsoshowthatvariabilityamplitudesaregreaterat TheFUV datashowninFigure13havefarfewermeasure- longertime-lagsthanatshortertime-lagsforallvaluesofUVQ mentsandlargererrorsthanthe NUV data, butagainthe radio luminosity.ThisbehaviorisalsorevealedintheSFsofFigures loudUVQsappeartobemorevariableasafunctionoftime-lag 7-9,whichwerederivedindependentlyofluminosity.Thisdif- (especiallyforperiods>100days)thantheradioquietsources. ferencein variabilityforbothshortandlongtime-lagvaluesis The best-fit slope of 0.16to the FUV radio loud data is signif- discussedfurtherinSection4.1withrespecttotheworkofKelly icantly differentfrom that of the best-fit slope of 0.059 for the etal.(2009). radio quiet data. This FUV variability difference is most pro- We notethattherearenolow-luminosityUVQsdetectedat nounced for time-lags > 100 days. However, we note that this highredshiftinoursample,suchthatthelowluminositybinsin conclusionisbasedononlyasmallnumberofUVQs.Wenote bothplotsof Figure 10 are populatedmostly bylow-z objects. that several authors have claimed that radio-loud quasars are Therefore, since the data in Figure 10 may still contain some (marginally) more variable than their radio-quiet counterparts hidden dependencyon quasar redshift, in Figure 11 we re-plot (Eggers et al. 2000, Helfand et al. 2001), whereas Bauer et al. theNUVandFUVvariabilitydatabutnowsubdividedintobins (2009) have found no evidence for any change in the optical ofredshift.Forconvenienceofplottingwehaveselectedrepre- variabilityamplitude of quasars with increasing radio loudness sentativeredshiftbinsof0.5<z <0.895andz >2.028forthe (luminosity).Severalstudieshavefoundthatopticalmicrovari- NUVdata,andbinsof0.5<z <0.895and1.4<z <2.03for ability,operatingovertime-scalesofminutestohours,occursin the FUV data,all plottedforthe sametwo representativetime- quasarsregardlessoftheirradioproperties(Ramirezetal.2009). lag bins as used in Figure 10. For the NUV data in Figure 11 OurpresentNUV(andFUV)observationssupportthenotionin the4curvesofSFversusluminositydonotpreciselyfollowthe whichradioloudUVQsare onlymarginallymoreUV variable sametrendsasshowninFigure10,althoughthereisanoverall thanradioquietsources,irrespectiveofthetime-frameofobser- generaltrendofdecreasingSFvaluewithincreasingluminosity. vation. Thisgeneraltrendofdecreasingvariabilityamplitudeisalsoob- servedintheFUVdata,exceptforthecaseofhighz andsmall 4. SummaryDiscussion time-lags where an increase of SF variability with luminosity is observed. Since the general trends seen in both Figures 10 Beforebrieflydiscussinganyphysicalinterpretationofthecor- and11aresimilar,wecandeducethatredshiftdoesnotseemto relationsshownin Figures7-13ofSection3,we firstprovide haveaprofoundeffectontheUVvariability-luminositynature a short summary of the underlying UV properties of the UVQ of these objects, and also that the shape of the UV SFs is only sample.We alsoremindthe readerthatdueto ourconservative weaklydependentonquasarluminosity. approachintheselectionofUVvariablequasars,thesefindings mayonlyapplytothemoreUVvariableofthisclassofobject.In Figure14weshowatypicalquasarrest-framespectrum(adapted 3.3. RadioLoudness-Variability Dependence fromVandenBerketal.2001),whichshowstheprominentLyα, In Figures 12 and 13 we plot the SFs for the NUV and FUV CIV, CIII and MgII emission lines superposed on a significant variabilityofourquasarsampleagainsttheirrestframetime-lag underlying UV continuum level, which for wavelengths blue- for both radio loud and radio faint UVQs. Based on their peak wardofLyαisheavilyextinguishedbytheLyαforestofabsorp- Welsh,Wheatley&Neil:QuasarUVvariability 7 tionlines.TheentireUVenergydistributionisoftenreferredto All these results are therefore consistent with the increasingly as the ‘big blue bump’, and is thoughtto be dominated by the widely-held belief that the majority of a quasar’s UV-optical 104−5K thermal emission from a central black hole accretion spectrumarisesdirectlyfromtheaccretiondiskandthatchanges disk. Based on the sensitivity and observationalstrategy of the intheleveloftheunderlyingUVcontinuaofquasarsareproba- GALEX satellite,thegreatmajorityofquasarsthatcanbede- blythemaindriveroftheobservedvariabilityintheseobjects. tectedpossessredshiftsintherange0≤z ≤2.5(Bianchietal. 2007).Forquasarswith redshiftsinthe rangefromz =0.11to 0.47thestrongLyαemissionlineispresentwithintheGALEX 4.1. EnsembleStructureFunctionDependence FUVpassband,andforredshiftsofz =0.46to1.33theLyαline TheensembleStructureFunctionsshowninFigure7clearlyre- ispresentwithintheNUVpassband.Inaddition,theLyαforest vealthatboththe FUV andNUV variabilityof UVQsare well ofabsorptionlines(whichconsiderablyreducestheUVfluxina quasarspectrum)affectstheentireGALEX passbandsforred- correlatedwithrestframetime-lagsof<∼300days,suchthatthe UVvariabilityamplitudeincreasesalmostlinearlywithincreas- shiftsofz >1.33.Therefore,anyobservedUVvariabilitywithin ingtime-lag.However,forlongertime-lagsthereisarolloverin either of the GALEX FUV or NUV passbands will be highly both SF curves, which is also observed at visible wavelengths. dependentonwhichoftheemission/absorption/continuumfea- It is therefore initially tempting to attribute this rollover to a turespresentinthequasarspectralenergydistributionhavebeen change in the underlying physical process that cause the ob- redshiftedinto,oroutof,thosetwopassbandsandalsowhichof served UV variability. However, a rollover in the SF at large thesefeaturesaredeemedtobemostresponsibleforthephysical time-lagshasrecentlybeenfullyexplainedbythe‘dampedran- originoftheobservedUVvariability.Therefore,ourselectionof domwalk’modelofquasarvariabilitypresentedbyKellyetal. 3redshiftbins(0.11<z <0.47,0.47<z <1.33andz >1.33) (2009)andMacLeodetal.(2010).Theseauthorsmodelquasar for the plotting of data in Figures 8 to 9 represent the appear- light curvesas a continuoustime stochastic process, which re- anceanddisappearanceofthestrongLy-αlineintheFUVand sultsintheestimationofacharacteristictimescaleandamplitude NUVbands.Additionally,becauseoftimedilationeffects,high forvisiblefluxvariations.Inthevisible,thesefluxvariationsare redshift objects are potentially observed for less time than low thoughttoresultfromthermalfluctuationsthataredrivenbyan redshiftones.Thiswillpossiblyaffectanystudyofthetime-lag underlyingstochasticprocesssuchasaturbulentmagneticfield onUVQvariability,whichiswhywehavebinnedourtime-lag associatedwiththeblackholeaccretiondisk.Theshapeoftheir dataintherestframe. powerspectrum(i.e.SF)ispredictedtobe1/f2,whichthenflat- Presently,explanationsoftheprimarycauseofquasarvari- tenstowhitenoiseattimescaleslongcomparedtotherelaxation ability can be broadly split into 3 classes: (i) gravitational mi- time. This relaxation time is interpreted (for quasar flux varia- crolensing, (ii) discrete-event (Poissonian) processes and (iii) tions) as the time requiredforthe time series of fluctuationsto accretion disk instability models. Vanden Berk et al. (2004) becomeroughlyuncorrelated.FromourpresentNUVandFUV havediscussedthesevariousprocesses(seethemanyreferences datathisrelaxationtimeis∼400days. therein)andtheirvisibledatafavoranaccretiondiskinstability Thereadershouldbereminded,however,thatourGALEX (orjet)causeoftheobservedopticalvariability.Sincemanyof observations are biased towards the detection of brighter and ourpresentresults(showninFigures7-13)generallyagreewith lower-redshift objects which could potentially be making a thevisibleobservationsofquasarvariabilitybyVandenBerket larger contribution to the ensemble SF at long time lags com- al.(2004),weshallonlyaddresswhetheramodelinvolvingac- pared with the higher redshift (and fainter) UVQs. Hence, al- cretion disk instabilities may explain our UV variability data. though the ensemble SFs shown in Figure 7 are important in- However,thereadershouldnotethatnotheoreticalmodel,irre- dicators of quasar variability, only by discussing the potential spective of its underlyingphysicalmechanism,currentlyexists dependenciesshownFigures8-13canwebetterinterpretthese thatcanbedirectlycomparedwiththeir(orourownUV)obser- data. vations. Long-term(7.5yr)changesintheUVcontinuumlevelofor- der25%-125%havebeenreportedforlowluminosityquasars 4.2. TimeLagDependence (Kaspietal.2000),whichisabouttwicethecontinuumvariabil- ity observedover a similar period for high luminosity (high-z) TheNUVandFUVSFsshowninFigures8and9asafunction objects(Kaspietal.2007).Interestinglyforthislattersampleof oftime-lagfollowthesame generaltrendastheirensembleSF (11) quasars, none showed Ly-α emission variations to a level counterpartsshowninFigure7,inthattheamplitudeofUVvari- of<7%overthe6-yearperiodofobservations.A similarcase ability increases (linearly) as a function of rest frame time-lag ofnon-variabilityoftheLy-αlinehasalsobeenreportedforob- up to ∼ 200 daysand thereafterthere is a pronouncedrollover servationsof3C273recordedovera15yrtimeframebyUlrich in the NUV and FUV SFs for all redshift values. This overall etal(1993).Severalauthorshaveclaimedthatquasarcontinuum trend is similar, irrespective of the value of UVQ redshift. For variationsaredominatedbyprocessesthatareintrinsictotheac- short time-lags< 20 daysthere is a definite trend for higher-z cretiondisk(Petersonetal.2004),andTrevese,Kron&Bunone UVQstobemoreNUVandFUVvariablethantheirlow-z coun- (2001) have suggested that the observed spectral changes are terparts,butthistrendallbutdisappearsfortime-lagsinexcess consistentwith temperaturevariationsof a blackbodyof ∼ 2.5 of100days.Ifweassumethattheobservedvariabilityismainly x104KthatemitsintheUV.MorerecentlyPereyraetal.(2006) due to changes in UV line emission, then we might expectthe havealsofoundthatmostoftheoptical-UVvariabilityobserved NUV band to exhibit greater variability than that recorded in in quasars is due to changes in the UV continuumlevel which theFUVbandforobjectswithz =0.46to1.33(i.e.therangein can be directlyattributedto processesinvolvingchangesin the whichthestrongLy-αlineenterstheNUVbandpass).However, disk mass accretionrates. Repeatedlyvisible spectroscopicob- ourobservationsdonotstronglysupportthisnotionsincetheSF servationsof2,500quasarsover>50daytime-framebyWilhite variabilitylevelsoftheFUVdatafortheseobjectsaresimilarto etal.(2005)haveshownthatthestrongestquasaremissionlines thoseoftheirNUVvariability.Thus,itseemslikelythattheob- vary by only 30% as much as the underlying continuum level. servedFUVvariabilityfortheUVQsinthisredshiftrangemay 8 Welsh,Wheatley&Neil:QuasarUVvariability bebetterexplainedthrough(stochastic)changesintheunderly- sponsibleforthisbehavioressentiallydisappearsfromourdata ingUVcontinuumlevel. fortime-lags>100days. 4.3. LuminosityDependence 4.4. RadioLoudness-UVvariability Although it is generally agreed that the intrinsic luminosity of AlthoughthedatapresentedinbothFigures12and13havebeen a quasar may be linked to matter accreting from a disk onto a constructedwithrelativelyfewUVQs,wenotethenearsimilar- black hole, the physical mechanism(s) responsible for the ob- ity betweenthe shape of the NUV SFs for bothradio loud and servedchangesinvariabilityarestillwidelydebated.Aspointed radio quiet objects. In addition, we note that both Figures also out in the Introduction, possible variability mechanisms from show that radio loud UVQs are marginally more variable than extrinsic gravitational sight-line microlensing to intrinsic ac- radio-quietsources(afactor2intheNUV andafactor1-3in cretion disk instabilities have been forwarded to explain the the FUV, depending on time-lag), in agreement with the work observations. Using I.U.E. satellite observations, Paltani & ofbothEggersetal.(2000)andHelfandetal.(2001).Similarly, Courvoisier (1997) found that quasar variability and luminos- severalothervisiblevariabilitystudies(VandenBerketal.2004, ity (absolute magnitude) were related by a power law slope of Ramirez et al. 2009) have revealeda correlationbetween radio -0.08,suchthatrelativevariabilitydecreasesasthequasarlumi- loudnessandUV/opticalvariabilityamplitudethatwassugges- nosity increases. This anticorrelation between quasar variabil- tive but not conclusive.The source of quasar radio emission is ity and luminosity has also been confirmed at optical wave- thoughttooriginateeitherinanaccretiondiskaroundthecentral lengths using large homogeneous data sets (Bauer et al. 2009, blackholeor in shockwavesina relativisticjetejected froma VandenBerketal.2004).However,wenotethatAietal.(2010) regionaroundthe centralblack hole. Since the processes asso- claim that the inverse dependence of variability on luminosity ciated with a powerfulnuclear radio source probablyinfluence is not significant for a sample of SDSS selected AGN (mostly UVvariabilityovertime-lags<∼100days,wemightexpectthe quasars).Theyclaimthattheiranalysiswascapableofrecover- greatest difference in variability between radio loud and radio ing AGN with a small variability in the low luminosity regime quiet objects to appear in this time-regime. Both of our FUV which were largely missed in most of the previous studies. and NUV data do not supportsuch a scenario, and to the con- Furthermore,previousrelationshipsbetween quasar luminosity trary the (sparsely sampled) FUV SF suggeststhat the greatest andotherparameterssuchasblackholemassandspectralline- variabilitydifferenceoccursfortimelags>100daysforradio widths have mostly been derived from single-epoch estimates sources.Thus,althoughourdatadorevealmarginaldifferences andusing(mostlyuntested)extrapolationsfromlowtohighred- betweenthebehaviorofradioloudandradioquietsources,we shiftobjects(Vestergaard&Peterson2006). believethattheradiopropertiesofUVQsappearonlyplayami- OurpresentfindingsshowninFigure10indicatethatthereis norroleinthegrosslevelsofobservedUVvariability. ageneraldecreaseinbothNUVandFUVvariabilityasafunc- tionofincreasingabsolutemagnitudefortime-lags>100d.The opposite effect is observed for time-lags < 100d (apart from 5. Conclusion the weak exception in the NUV for time-lags < 10d). Figure 11showsthatthisgeneraldecreaseinUVvariabilityasafunc- Using archival observations recorded at NUV and FUV wave- tionofluminosityisnotstronglycoupledtotheUVQredshift,in lengths over a 5+ year timeframe with the NASA Galaxy conflictwithother(visible)studiesthathaveclaimedaredshift- Evolution Explorer(GALEX) satellite, we have performedan variability relationship (Trevese & Vagnetti 2002). Both of the ultravioletvariabilityanalysisofquasarsthathaveopticalcoun- NUVandFUVSFsshowninFigure10revealasimilarbehav- terparts in the Sloan Digital Sky Survey (SDSS) DR5 spectro- iortothoseshowninFigures7-9(whichwerederivedindepen- scopic catalog of Schneider et al. (2007). The application of a dentlyofluminosity),inthatvariabilityamplitudesaregreaterat set of (conservative) selection criteria has resulted in the iden- longertime-lagsthanforshortertime-lagsforallvaluesofQSO tification of 550 quasars in the NUV and 371 in the FUV as luminosity.Therefore,itwouldseemfromourpresentdatathat being statistically significantly variable using data recorded in theintrinsicluminosityofaUVQmayonlybeweaklyconnected 5 or moreGALEX observationsspreadovertime periods> 3 totheoriginofthephysicalmechanismcausingtheUVvariabil- months.TypicallywehaveobservedawidevarietyofUVvari- ity, and that this mechanism operates over most time-lags and abilitylevelsrangingfrom0.11to3.0ABmagnitudesarisingin isnotstronglydependentofthe UVQredshift.Asdiscussedin quasarswithredshiftsof0<z <2.5. Section4.1,althoughtheactualunderlyingphysicalcauseshave In order to explore possible correlations within these vari- yet to be proven, the application of a ‘damped random walk’ ability data we have used a Structure Function analysis that modelofstochasticthermalfluctuationsfromanaccretiondisk closely follows the work of Vanden Berk et al. 2004, who canstatisticallyexplainthegeneralshapeoftheensemblevari- performed a similar analysis on visible observations of SDSS ability of quasars, and in particular the shape of their SFs as a quasars.Bysubdividingthevariablequasarsintotime-lagand/or function of time-lag (Kelly et al. 2009, MacLeod et al. 2010). redshift bins we have been able to obtain a set of (quasi- Therefore,inordertoexplainthemajorityofourUVvariability independent) variability relations with respect to the Structure datawedonothavetoinvokevariousphysicalmechanismsthat Function(SF)forvaluesofrestframetime-lag,luminosityand operateoverdifferenttime-lagsthatdependonvariousphysical radio loudness. Since our selection criteria for UV variability parameters,butinsteadthegrossbehavioroftheUVvariability wasquiteconservative,ouranalysisoftheSFsmayonlybeap- of quasars can essentially be described by a 1 / f2 dependence plicabletothemoreUVvariableQSOs. thatflattenstoanearconstantatlowfrequencies(i.e.longtime- TheSF-time-lagplotsclearlyrevealthattheamplitudesof lags).ThisseemstobetrueapartfromtheeffectnotedinSection variabilityinboththeNUVandFUVarefarlargerthanthoseob- 4.2(andshowninFigures8and9)inwhichhigher-z UVQsap- servedinquasarsatvisiblewavelengths.Additionally,thelevels peartobemorevariablethantheirlow-z counterpartsfortime- ofFUVvariabilityaregreaterthanthoseobservedintheNUVat lags<20days.Whateverphysical(orobservational)effectisre- anygivenvalueoftime-lag.Wealsofindthattheamplitudesof Welsh,Wheatley&Neil:QuasarUVvariability 9 boththeNUVandFUVvariabilityofUVQsincreaseasafunc- Cristiani,S.,Trentini,S.,LaFranca,F.&Andreani,P.,1997,A&A,321,123 tion of rest frame time-lag, irrespective of the value of quasar deVries,W.,Becker,R.,White,R.andLoomis,C.,2005,AJ,129,615 redshift, for time-lags < 200 days. For time-lags > 300 days diClemente,A.,Giallongo,E.,Natali,G.etal.,1996,ApJ,463,466 Eggers,D.,Sahffer,D.andWeistrop,D.,2000,AJ,119,460 there is a pronounced rollover in the NUV SF for all redshift Gezari,S.,Basa,S.,Martin,D.C.etal.,2008,ApJ,676,944 values,whichisalsoobserved(withalowersignificance)inthe Giveon,U.,Maoz,D.,Kaspi,S.etal.,1999,MNRAS,306,637 FUVvariabilitydata.Thesedataalsoshowthatfortime-lags< Helfand,D.,Stone,R.,Willman,B.etal.,2001,AJ,121,1872 20daysthereisanapparenttrendforhigher-z UVQstobemore Hook,I.,McMahon,R.,Boyle,B.&Irwin,M.,1994,MNRAS,268,305 Kaspi,S.,Smith,P.,Netzer,H.etal.,2000,ApJ,533,631 NUVandFUVvariablethantheirlow-z counterparts.However, Kaspi,S.,Brandt,W.,Maoz,D.etal.,2007,ApJ,659,997 none of of other plots of SF dependence show any strong de- Kelly,B.,Bechtold,J.andSiemiginowska,A.,2009,ApJ,698,895 pendenceonredshift,andthiseffectmaybedueto someother Kozlowski,S.,Kochanek,C.,Udalski,A.etal.,2010,ApJ,708,927 physicaldependency.Overall, these results are best interpreted MacLeod,C.,Ivezic,Z.,Kochanek,C.etal.,2010,ApJ,721,1014 underthe assumptionthattheUV variabilityislinkedto varia- Martin,D.C.,Fanson,J.,Schiminovich,D.,etal.2005,ApJ,619,L1 Miller,H.R.,Carini,M.&Goodrich,B.,1989,Nature,337,627 tionsinthelevelofcontinuum,andnotline,emission. Morrissey,P.,Conrow,T.,Barlow,T.etal.,2007,ApJS,173,682 Our data also indicate that for time-lags > 100 days the Paltani,S.andCourvoisier,T.,1994,A&A,291,74 more luminousUVQs tend to be less NUV and FUV variable, Paltani,S.andCourvoisier,T.,1997,A&A,323,717 whereastheoppositeeffectismostlyfoundforshortertime-lags Pereyra,N.,VandenBerk,D.,Turnshek,D.etal.,2006,ApJ,642,87 Peterson,B.,1993,PASP,105,247 (with the weak exception in the NUV for lags of < 10 days). Peterson,B.M.,Ferrarese,L.,Gilbert,K.etal.,2004,ApJ,613,682 This differencein the level of SF variability between long and Ramirez,A.,deDiego,J.,Dultzin,D.andGonzalez-Perez,J-N.,2009,AJ,138, shorttime-lagsisalsoseenintheensembleSF-timelagplots, 991 which were derived independently of luminosity. We also find Ramirez,A.,Dultzin,D.anddeDiego,J.,2010,ApJ,714,605 that UVQ redshift does not seem to have a profound effect on Schneider,D.,Hall,P.,Richards,G.etal.,2007,AJ,134,102 Sesar,B.,Ivezic,Z.,Lupton,R.etal.,2007,AJ,134,2236 anyofthevariability-luminosityrelations(alhtoughwenotethe Simonetti,J.,Cordes,J.andHeeschen,D.,1985,ApJ,296,46 previouslymentionedtrendforhigher-z UVQstobemoreNUV Sramek,R.&Weedman,D.,1980,ApJ,238,435 and FUV variable than their low-z counterparts as a function Stalin,C.,Gupta,A.,Gopal-Krishna,Wiita,P.&Sagar,R.,2005,MNRAS,356, ofshorttime-lags).Theapparentweakeffectofredshiftonthe 607 Trammell,G.,VandenBerk,D.,Schneider,D.etal.,2007,AJ,133,1780 UV variability-luminosityis perhapsnotsurprisingsince lumi- Trevese,D.,Kron,R.&Bunone,A.,2001,ApJ,551,103 nosity is an intrinsic physical propertywhereas redshift is not. Trevese,S.andVagnetti,F.,2002,ApJ,564,624 These data can possibly be interpreted through the application Udalski,A.,Wyrzykowski,L.,Soszynski,I.etal.,2009,astro-ph,0909.1326 ofa‘dampedrandomwalk’modelofstochasticthermalfluctu- Ulrich,M.-H.,Courvoisier,T.&Wamsteker,W.,1993,ApJ,411,125 ationsfromanaccretiondiskwhichcanstatisticallyexplainthe Ulrich,M.-H.,Maraschi,L.anhdUrry,C.M.,1997,ARA&A,35,445 VandenBerk,D.,Richards,G.,Bauer,A.etal.,2001,AJ,122,549 generalshapeofourVUVSFs((Kellyetal.2009,MacLeodet VandenBerk,D.,Wilhite,B.,Richard,G.etal.,2004,ApJ,601,692 al.2010). Vestergaard,M.&Peterson,B.,2006,ApJ,641,689 Although our data set is small, it seems that radio loud Vio,C.,Christiani,S.,Lessi,O.andProvenzale,A.,1992,ApJ,391,518 sourcesaremarginallymoreNUVvariableasafunctionoftime- Wheatley,J.,Welsh,B.Y.andBrowne,S.,2008,AJ,136,259 Wilhite,B.,VandenBerk,D.,Kron,R.etal.,2005,ApJ,633,638 lagthanradioquietUVQs.Thereforeitwouldseemthatwhat- Wilhite,B.,Brunner,R.,Grier,C.etal.,2008,MNRAS,383,1232 everthe mainmechanismisresponsibleforthe observedNUV Wold,M.,Brotherton,M.andShang,Z.,2007,MNRAS,375,989 variability in quasars, it seems to operate irrespective of their radioflux.Suchbehaviorcanalso be explainedbythe damped randomwalkmodel.Ourdataalsohintsthatfortime-lags>100 days radio loud sources may be more FUV variable than radio quietsources,althoughthisconclusionisbasedonsmallnumber statistics. In summation,our analysisfavorsa quasarmodelin which UV variabilityis mainly due to (stochastic) changesin the un- derlyingcontinuumlevel,ratherthanmodelsthatfavorgravita- tionalmicrolensingordiscrete-eventprocesses.Finally,wenote that as yet, no detailed quasar model currently exists that can be directly compared with our observations of NUV and FUV variability. Acknowledgements. We particularly acknowledge the dedicated work of the GALEXmissionoperationssupportstaffatJPL/CaltechinPasadena.Financial support for this research was provided by the NASA GALEX Guest Investigator program and NASA grant NAS5-98034 to UC Berkeley. We are indebtedtoDr.SuviGezariwhomademanysuggestionsthathavegreatlyim- provedthispaper. References Ai,Y.,Yuan,W.,Zhou,H.etal.,2010,ApJ(inpress) Bachev,R.S.,2009,A&A,493,907 Bauer,A.,Baltay,C.,Coppi,P.etal.,2009,ApJ,696,1241 Bertin,E.&Arnouts,S.,1996,A&AS,117,393 Bianchi,L.,Rodriguez-Merino,L.,Viton,M.etal.,2007,ApJS,173,659 Bianchi,L.,Hutchings,J.,Efremova,B.etal.,2009,AJ,137,3761 Courvoisier,T.&Ulrich,M.,1985,Nature,316,524 10 Welsh,Wheatley&Neil:QuasarUVvariability Fig.1.PlotofthenumberofGALEX repeatedvisitsforquasars with DIS integrationtimesof > 1000sand brightquasarswith AISobservationof> 100s(see textforselectioncriteria).Full Fig.3.GalacticdistributionoftheUVselectedquasarswith≥5 lineisforquasarswithGALEX NUVobservationsanddashed observationalvisitsbyGALEX.Filledcirclesarequasarswith lineisforquasarswithFUVobservations. bothNUVandFUVobservations,whileopencirclesarequasars withNUVonlyobservationsbyGALEX. Fig.2.PlotofthenumberUVselectedquasarsandthetimepe- riodspanningalloftheirGALEX observations.Fulllineisfor quasars observedin the GALEX NUV pass-band, dashed line isforquasarsobservedintheFUVpass-band.