MNRAS447,72–96(2015) doi:10.1093/mnras/stu2424 Revealing the X-ray variability of AGN with principal component analysis M. L. Parker,1‹ A. C. Fabian,1 G. Matt,2 K. I. I. Koljonen,3,4 E. Kara,1 W. Alston,1 D. J. Walton,5 A. Marinucci,2 L. Brenneman6 and G. Risaliti6,7 1InstituteofAstronomy,MadingleyRoad,CambridgeCB30HA,UK 2DipartimentodiMatematicaeFisica,Universita`degliStudiRomaTre,viadellaVascaNavale84,I-00146Roma,Italy 3AaltoUniversityMetsa¨hoviRadioObservatory,Metsa¨hovintie114,FIN-02540Kylma¨la¨,Finland 4NewYorkUniversityAbuDhabi,POBox129188,AbuDhabi,UAE 5CaliforniaInstituteofTechnology,1200EastCaliforniaBoulevard,Pasadena,CA91125,USA 6Harvard–SmithsonianCenterforAstrophysics,60GardenSt,Cambridge,MA02138,USA 7INAF–OsservatorioAstrofisicodiArcetri,LargoEnricoFermi5,I-50125Firenze,Italy D o Accepted2014November13.Received2014November13;inoriginalform2014May30 w n lo a d e d ABSTRACT fro m We analyse a sample of 26 active galactic nuclei (AGN) with deep XMM–Newton observa- h tions, using principal component analysis (PCA) to find model-independent spectra of the ttp differentvariablecomponents.Intotal,weidentifyatleast12qualitativelydifferentpatterns ://m n of spectral variability, involving several different mechanisms, including five sources which ras showevidenceofvariablerelativisticreflection(MCG–6-30-15,NGC4051,1H0707−495, .ox fo NGC3516andMrk766)andthreewhichshowevidenceofvaryingpartialcoveringneutral rd jo absorption(NGC4395,NGC1365andNGC4151).Inover halfofthesourcesstudied,the urn a variability is dominated by changes in a power-law continuum, both in terms of changes in ls .o flux and power-law index, which could be produced by propagating fluctuations within the rg a/ corona.Simulationsareusedtofinduniquepredictionsfordifferentphysicalmodels,andwe t C thenattempttoqualitativelymatchtheresultsfromthesimulationstothebehaviourobserved alifo in the real data. We are able to explain a large proportion of the variability in these sources rn ia usingsimplemodelsofspectralvariability,butmorecomplexmodelsmaybeneededforthe In s remainder. We have begun the process of building up a library of different principal com- titu ponents, so that spectral variability in AGN can quickly be matched to physical processes. te o WeshowthatPCAcanbeanextremelypowerfultoolfordistinguishingdifferentpatternsof f T e variability in AGN, and that it can be used effectively on the large amounts of high-quality ch n o archivaldataavailablefromthecurrentgenerationofX-raytelescopes.WewillmakeourPCA lo g codeavailableuponrequesttotheleadauthor. y o n A Keywords: galaxies:active–galaxies:nuclei–galaxies:Seyfert. p ril 2 , 2 0 1 5 viding acceptable fits to the same data set, meaning that spectral 1 INTRODUCTION fittingofintegrateddatasetsalonecannotsufficientlydistinguish Active galactic nuclei (AGN) can be extremely variable in the betweenalternativephysicalmodels.Byinvestigatingtheproperties X-ray band on time-scales as short as the light traveltime over a ofvariabilityinthesesources,wecanhopetoidentifythephysical few gravitational radii (R ), which can be from minutes to days, processes driving AGN, probing them at different distances from G dependingonthemassoftheblackhole(BH).Examinationofthe theeventhorizonbylookingatdifferenttime-scales. X-rayspectraofAGNrevealsalargevarietyofdifferentspectral Principalcomponentanalysis(PCA)isamethodofdecomposing shapes,producedbyvariousdifferentprocesses,mostnotablyab- adatasetintoasetoforthogonaleigenvectors,orprincipalcompo- sorption by intervening material (see review by Turner & Miller nents(PCs),whichdescribethevariabilityofthedataasefficiently 2009)andreflectionfromtheaccretiondiscandsurroundingma- aspossible(e.g.Kendall1975;Malzacetal.2006).Inpractice,when terial(seereviewsbyFabian&Ross2010;Reynolds2014).AGN appliedtoasetofspectra,thisproducesasetofvariablespectral spectracanbeverycomplex,withmultipledifferentmodelspro- componentswhichdescribethevariabilityofthesourcespectrum. Ifthespectrumismadeupofalinearsumofvariable,uncorrelated and spectrally distinct physical components, then PCA will, with (cid:2) Email:[email protected] sufficient data quality, return an exact description of the physical (cid:2)C 2014TheAuthors PublishedbyOxfordUniversityPressonbehalfoftheRoyalAstronomicalSociety RevealingAGNvariability 73 components.Theadvantageofthismethodisthatitproducesde- beyondthescopeofthispaper.Weselectonlysourceswithmore tailedspectraofeachvariablecomponent,inamodel-independent thanoneorbitofexposuretime.WeusedScienceAnalysisSoftware way. Calculating the rms spectrum (e.g. Edelson et al. 2002) can (SAS)version13.0.0foralldatareduction.Thedataarefilteredfor show the total variability as a function of energy, but cannot be background flares, and we use the EPPROC SAS task to reduce usedtodeterminehowmanyvariablecomponentscontributetothe the data. We use 40 arcsec circular regions for both source and variability or to isolate contributions from different mechanisms. backgroundspectraforallsources,selectingthebackgroundregion Components that are only weakly variable, such as variations in to avoid contaminating sources. Representative spectra for each absorptionorreflection,willusuallybedrownedoutbyvariations sourcecanbefoundinAppendixA. intheprimarycontinuum.Detailedspectralmodellingcanbeused ThelistofobservationIDsusedinthisanalysisforeverysource toovercomethislimitation,bycarefullyfittingthedatafordifferent isshowninTable1(fullversionavailableonline).Forthreesources intervalsandidentifyingtheoriginofthevariability.However,this (NGC1365,MCG–6-30-15andArk120),wehavemadeuseofdata isbydefinitionnotmodelindependent,meaningthatverydifferent from joint XMM–Newton and NuSTAR (Harrison et al. 2013) ob- conclusionscanbedrawnfromthesamedataset(forexample,see servingcampaigns(Risalitietal.2013;Marinuccietal.2014;Matt the discussion on reflection/absorption models in MCG–6-30-15 etal.2014;Waltonetal.2014).Allotherdataispubliclyavailable, byMarinuccietal.2014).PCAcombinestheadvantagesofboth andwasdownloadedfromtheXMM–NewtonScienceArchive.To ofthesemethods,andcanbeusedtocalculatemodel-independent givesomeideaofthenatureofthevariabilityineachsourceand D o spectra of multiple variable spectral components. This technique howitchangesbetweenobservationsweshowcount–countplotsin w n has many applications both within astronomy and in other fields AppendixA.Inthemajorityofcases,theseplotsareapproximately lo a (Kendall1975),andhasbeenusedasapowerfultoolfortheanal- linearwithalargescatter,howeversomesources(e.g.NGC4051, de d y20si1s3o)f.X-raybinaryvariability(Malzacetal.2006;Koljonenetal. NUtGtlCey3&51M6)csHhaorwdya(d2o0w03n)t,uarnndatinloswomceoucnatssesas(ed.gis.cRuEsseJ1d0i3n4T+a3y9lo6r), from h Early attempts at using PCA to understand spectral variability thesoftandhardbandsseemtobeindependent. ttp in AGN were hampered by a lack of high-quality data. PCA has In general, we follow the methods discussed in Parker et al. ://m beenusedinaminorroleforexaminingX-rayspectralvariabilityin (2014a),hereafterP14a.Foreachsource,wecalculatethefractional n ra AGNforsometime(e.g.Vaughan&Fabian2004),butfrequently deviationsfromthemeanforasetofspectra(seeexamplespectra s.o atlowspectralresolutionandwithonlyonecomponentconfidently for NGC 4051 in Appendix A), extracted from 10 ks intervals xfo identified. The use of singular value decomposition (SVD; Press, (unless otherwise specified). These spectra are arranged into an rd Flannery&Teukolsky1986)allowsthefullspectralresolutionof n × m matrix M , wheren and m are the number of energy bins jou rn theinstrumenttoberetained,producingdetailedcomponentspectra andspectra,respectively.WethenuseSVD(Pressetal.1986)to als (Milleretal.2007).Morerecentwork(Parkeretal.2014a,b)has findasetofPCs(oreigenvectors)whichdescribethevariabilityof .org demonstrated that PCA can return multiple components from a thespectrumasefficientlyaspossible.SVDfactorizesthematrix a/ sufficiently large data set, effectively isolating different spectral M, such that M = UAV∗, where U is an n × n matrix, V is an t C a components.ThereisnowmorethanadecadeofarchivalXMM– m × m matrix and A is an n × m diagonal matrix. The matrices lifo Newton EPIC-pn (Stru¨der et al. 2001) data which can be used to UandVtheneachdescribeasetoforthogonaleigenvectorstothe rn examine the variability of AGN on long time-scales and at high matricesMM∗andM∗M,respectively.Theseeigenvectorsrepresent ia In s spectralresolution.Inthiswork,wepresentasystematicanalysis thespectralshapeofthevariablecomponents.Thecorresponding titu of 207 observations of 26 bright, variable AGN, using PCA to eigenvaluesaregivenbythediagonalvaluesofA,andareequalto te o revealhiddenpatternsofvariabilityandtorelatethesepatternsto thesquareofthevariabilityineachcomponent(inarbitraryunits). f T thephysicalprocessesinAGN.Thepaperisorganizedasfollows. Thefractionalvariabilityineachcomponentcanthenbefoundby ec h dividingthesquarerootsoftheeigenvectorsbythesumofallthe n o (i) InSection2,we describethedatausedinthisanalysis and squareroots. log howitwasprocessed,alongwithdetailsoftheanalysisitself.We y The resulting components show the strength of the correlation o include a demonstration of the method with a simple toy model, between energy bins, so a flat positive (or negative, the sign of n A showingthepotentialpowerofPCAasananalytictool. p PC(Aii)spInecStreac,taionndp3r,ewseentgtihveerdeseutaltislsooffouorusrimmueltahtoiodnosffosrimdiufflaetrienngt twahtheheiyrg-ehaaxseinsaeicrsgoaimerbspiroternaperrnyet)setchnoatmstaipsoppnivoeonstititnisvhgeoewaftfselcothtw.aTtehanlielsrbigsiinecssomvaanprdylicneaeqtgueadatliblvyye, ril 2, 201 physical models of AGN variability. These different spectra then 5 therequirementthattheeigenvectorsareorthogonal,i.e.theirdot- represent different predictions for the different spectral models, productmustbezero.Weexaminetheeffectofthisconstrainton whichwecanusetounderstandtheresultsfromrealdata. simulatedPCsinSection3.Wenotethatthemethodofpreparingthe (iii) WepresenttheresultsoftheanalysisinSection4,describ- spectraissuchthatconstantmultiplicativecomponentswillhaveno ingandshowingthePCsfoundforeachsource,alongwithsome effectonthePCAresults,astheywillnotaffectthefractionalresid- backgroundoneachobject.Wealsogivesomebasicinterpretation uals. However, a constant spectral component that changes with ofeachresult,attemptingtomatchthePCspectratothosefound energywillsuppressthespectralvariabilityofthevariablecompo- usingsimulations. nents.Thishasimportantimplicationsfordistinguishingbetween (iv) Finally,inSections5and6,wediscussoutmainresultsand absorptionandreflectioninAGNvariability. summarizeourconclusions. Thesignificanceofthecomponentsproducedisdeterminedus- ing the log-eigenvalue (LEV) diagram, an example of which (for Ark 564) is shown in Fig. 1. This shows the fraction of the to- 2 OBSERVATIONS, DATA REDUCTION AND talvariabilitywhichcanbeassignedtoeachPC,soforArk564, ANALYSIS METHOD ∼90percent of the variability is in the first PC and so on. The WerestrictthisanalysistoXMM–NewtonEPIC-pndataonly.The componentsduetonoiseproducedbythePCAarepredictedtode- method used is also applicable to other instruments, but that is caygeometrically(seee.g.Jolliffe2002;Koljonenetal.2013),so MNRAS447,72–96(2015) 74 M.L.Parkeretal. Table1. Listofobservationsusedinthispaper.Sourcesareorderedbytheirfirstappearanceinthetext.Weshow 0.5–2and2–10keVcountratesforeachobservation,andtheratiobetweenthetwo,sothattheamplitudeofspectral variabilitycanbeestimated.Notethattheon-sourceexposuretimewillbesmallerthanthetotalduration,andtheactual usabletimeuseddependsonthesizeoftheintervalsweusetoextractspectra.Fulltableisavailableonline. Source ObservationID Date Duration 0.5–2keVrate 2–10keVrate Ratio (s) (s−1) (s−1) 2–10/0.5–2keV MCG–6-30-15 0111570101 2000-07-11 46453 7.67 2.90 0.38 0111570201 2000-07-11 66197 10.86 4.21 0.39 0029740101 2001-07-31 89432 15.05 4.82 0.32 0029740701 2001-08-01 129367 16.37 5.31 0.32 0029740801 2001-08-05 130487 14.98 4.72 0.32 0693781201 2013-01-31 134214 20.05 6.24 0.31 0693781301 2013-02-02 134214 11.60 3.86 0.33 0693781401 2013-02-03 48918 8.09 3.21 0.40 NGC4051 0157560101 2002-11-22 51866 2.82 0.61 0.22 D 0606320101 2009-05-03 45717 5.56 1.82 0.33 o w 0606320201 2009-05-05 45645 9.37 2.34 0.25 n lo 0606320301 2009-05-09 45584 11.40 2.40 0.21 a d 0606321401 2009-05-11 45447 8.47 1.69 0.20 ed 0606321501 2009-05-19 41843 8.56 2.09 0.24 fro 0606321601 2009-05-21 41936 17.58 3.21 0.18 m h 0606321701 2009-05-27 44919 3.51 1.30 0.37 ttp 0606321801 2009-05-29 43726 4.72 1.81 0.38 ://m 0606321901 2009-06-02 44891 2.15 0.51 0.24 n 0606322001 2009-06-04 39756 5.25 1.79 0.34 ras .o 0606322101 2009-06-08 43545 1.69 0.64 0.38 x 0606322201 2009-06-10 44453 4.43 1.46 0.33 ford 0606322301 2009-06-16 42717 6.19 1.37 0.22 jo u rn a ls AsimpletestcaseisshowninFig.2.Forthisexample,weadd .org ytog(xe)th=erstihnr(e2ex)f,uanlcotniognws:ityh1(rxa)nd=om0.n3o+isex./W8;eyc2r(exa)te=50si‘nsp(xec)traan’,d at C/ 3 a oftheformy(xi)=0.6a1y1(xi)+0.2a2y2(xi)+0.1a3y3(xi)+0.1a4, lifo wherea arerandomvalues,evenlydistributedbetween±0.5and rn xiarethej200valuesbetween0and6πoverwhichthefunctionsare ia In s calculated.Intheleft-handpanel,weshowthethreeinputspectra, titu minusnoise,inthemiddlepanelweshowasampleofthegenerated te o functionsandintheright-handpanelweshowthefunctionsrecov- f T eredusingourPCAcode.TheLEVdiagramforthistestisshownin ec h Fig.3,andclearlyshowsthatthreecomponentsaresignificant,with n o 47percentofthevariabilityinthefirstcomponent,7percentinthe log y secondand3percentinthethird.Alltheremainderisattributable o n tonoise.Thesevaluesarefunctionsoftheamplitudeandvariability A p ofPea1c4haccoamlcuploanteednte,xatnrdemthaelssipgencatrl-atoa-nndoiusseedractioomopfatrhiesodnastat.ospec- ril 2, 2 tralfittingtofindphysicalinterpretationsforthecomponentspro- 01 5 ducedinthatanalysis.However,thisisinefficientforlargesamples Figure1. LEVdiagramforArk564.Thisshowsthefractionalvariability of objects and could potentially compromise the model indepen- ineacheigenvectorobtainedfromthePCAofthissource.Theblackline denceoftheresults.Inthiswork,wecreatesimulatedspectra(see showsthebest-fittinggeometricprogression,fitfromcomponent4to50. Section3)basedonphysicalmodelsthatareallowedtovarywithin Theremainingthreecomponentsarefoundtobehighlysignificant,asthey givenparameterranges,thenusePCAtofindthecomponentspectra deviatefromthislinebymanystandarddeviations. foreachmodel.ThisproducesapredictedsetofPCspectraforeach model,whichwecanthenmatchtothePCspectrafoundfromthe deviation from a geometric progression can be used as a test of dataforeachsource. the significance of a component. In this case, three components deviatefromthebest-fittinggeometricprogression,andarehighly 3 SIMULATIONS significant.Forthesakeofbrevity,weonlyshowanexampleLEV diagram,ratherthanoneforeverysource.Thestrongeststatement Inthissection,weusesimulationstopredictthePCAspectrapro- aboutthesignificanceofthecomponentsweinvestigatecomessim- ducedbydifferentmodelsofAGNvariability.Thistechniquefor plyfromthestrongcorrelationbetweenpointsinadjacentbins.Any analysingPCAspectrawasusedbyKoljonenetal.(2013),andthe coherentcomponentsproducedareextremelyunlikelytobedueto methodweuseherewasintroducedinParkeretal.(2014b),here- randomnoise,whichisindependentbetweenbins. afterP14b,whereitisusedtodemonstratethedifferingpredictions MNRAS447,72–96(2015) RevealingAGNvariability 75 D o Figure2. Left:thethreefunctionsweaddtogether,alongwithnoise,asasimpletestofthePCAmethod.Middle:sampleof10‘spectra’createdbyadding w n togetherthethreefunctionsontheleftinrandomamountswithadditionalnoise.Right:thePCsreturnedbytheanalysis.Notethatthey-axisisarbitraryand lo a differentbetweentheleft-andright-handpanels.Errorbarsarenotplottedforclarity. de d fro m h ttp ://m n ra s .o x fo rd jo u rn a ls .o rg a/ t C a lifo rn ia In s Figure4. Thefirst(top,black)andsecond(bottom,red)PCsobtainedby titu usingPCAonsimulatedspectraofapowerlawwhichvariesinnormaliza- te o tionandphotonindex.Thefirstcomponentcorrespondstochangesinthe f T Figure3. LEVdiagramforthetestcaseshowninFig.2.Inthistest,three normalization,andthesecondtopivotingofthepowerlaw. ec h inputfunctionsaresummed,andnoiseisadded.TheLEVdiagramshows n o thatthreereturnedcomponentsarestatisticallysignificant,theremainder Allofthecomponentsproducedbythesesimulationsareequally lo g canbeattributedtonoise.Errorbarsarenotshown,butaresmallerthanthe validwiththey-axisinvertedastheyrepresentdeviationsfromthe y o pointsforthethreesignificanteigenvectors. mean, ratherthantheminimum,andwillthereforesometimes be n A p for intrinsic source variability and absorption variability in NGC pthoesictoivmepaonndensotsminettihmeemsannengeartiwveh.icInhgmeankeersalt,hwemeaotstteimntputittiovearsreannsgee. ril 2, 2 WewillinitiallyconsiderthePCsreturnedfromavariablepower 0 1365. 1 law, and then investigate the effect of additional spectral compo- 5 nents. In each case, we will first look at the effect of including a constant component and varying the power law, then allowing 3.1 Method thenewcomponenttovary.Finally,wewillexamineaselectfew In general, we follow the method outlined in P14b and simulate exampleswithmorethanoneadditionalvariablecomponent. spectrausingtheXSPECcommand‘FAKEIT’,thenanalysetheresults Wherethecomponentsreturnedfromthesimulationcorrespond usingthesamemethodasweusefortherealdata.Theparameters directlytooneofthemodelcomponents,welabelthefigurewith ofinterestareselectedrandomlybetweenextremevaluesforeach therelevantsymbols:Npl,Nbb,Nref,NHandfcovcorrespondtopower spectrum.Forsimplicity,wesimulate10ksEPIC-pnspectra,tobe law,blackbodyandreflectionnormalizations,columndensityand assimilaraspossibletothosefoundfromrealdata.Wedonotmatch coveringfraction,respectively. themodelfluxtothedata,insteadexaggeratingthemodelfluxso thatthefeaturesaremoreprominent.Thisisequivalenttosimulating 3.2 Singleandmultiplepowerlaws longerexposuresatlowerfluxforthesimplemodelsdiscussedhere, butlesstimeconsumingtocalculate.Wedonotattempttoexactly Asabaselinemodel,weestablishthecomponentsexpectedfrom match the PCs produced by the data, instead looking to produce variabilityofapower-lawcontinuum.Fig.4showsthetwocom- generalpredictionsfordifferentvariabilitymechanisms. ponentsobtainedfromasimulationofasimplevariablepowerlaw, MNRAS447,72–96(2015) 76 M.L.Parkeretal. combinationoftwoormoresuchcomponentsthatchangeinrelative amplituderatherthanindex. 3.3 Blackbody/softexcesscomponents We now investigate the PCs produced from a constant or weakly variablesoftexcesscomponent.Foroursimulations,weuseablack- body for simplicity and brevity, but the resulting components are equivalent to those that would be produced by any other mod- els that explain the soft excess in terms of an additional compo- nentthatonlycontributesatlowenergies(e.g.Comptonizationand bremsstrahlungmodels). Initially, we consider the effects of a constant blackbody com- ponent on the PCs returned from a varying power law. For this Figure5. Left:thetwosignificantPCsfromasimulationofapowerlaw simulation,weincludeablackbodywithatemperatureof0.1keV varyinginnormalizationandphotonindex,inthepresenceofaconstant hardpowerlaw.Thisspectralcomponentsuppressesthevariabilityofthe and a normalization of 0.1, then vary a power law with a photon Do indexbetween1.9and2.1,andnormalizationsbetween0.5and1.0. w PCs at high energies. Right: the two PCs from a simulation where the n spectralpivotingarisesfromtwopowerlaws,onehard,onesoft,varyingin The effect of such a constant component is to suppress the vari- loa normalizationbutnotinindex.Thesecomponentsareverysimilartothose abilityseeninthevariablecomponents,pushingthebinswherethe de d producedbyapivotingpowerlaw,butwithadditionalcurvature. blackbodyisstrongesttowardszero.Thiscanbeseeninthetoprow fro ofFig.6,wheretheflatPCsproducedbyavaryingpowerlaware m h with no other spectral components. The photon index is allowed pushedtowardszeroatlowenergiesbytheblackbody. ttp to vary between 1.9 and 2.1 randomly, and the normalization is Fixing the photon index and allowing the blackbody to vary ://m allowedtochangebyafactorof2.Theresultantcomponentsare (betweennormalizationsof0.8and1.2)producesadifferentsecond n ra completelystraight,showingnofeaturesofanykind,althoughthere component,(topmiddlepanelofFig.6)whichhasthesameshape s.o isaslightincreasewithenergyintheprimarycomponent,duetoa astheblackbodyatlowenergies,thenisnegativebutclosetozeroat xfo correlationbetweenfluxandphotonindexinthemodel. highenergies.Thenegativevaluesarecausedbytheorthogonality rd jo Foragoodcomparisonwithrealdata,wereferthereaderto3C constraintofPCA,whichrequiresthatthedot-productofanytwo u rn 273inSection4.4.Thissourceisdominatedbyapowerlawfrom PCsbezero.Inpractice,thismeansthatiftheprimarycomponentis als arelativisticjet,andthefirsttwocomponentsfoundfromthedata 100percentpositive,approximately50percentoftheenergybins .org areanexcellentmatchtothepredictionsforavaryingpowerlaw ofallsubsequentPCsmustbebelowzero. a/ showninFig.4. Amorecomplexcomponentisproducedifwevarythephoton t C a Wenextinvestigatetheeffectofaddingasecondpowerlawto index of the power law as well. First, we vary the index weakly lifo the spectrum. Additional continuum components such as this are (between 1.95 and 2.05), and the results of this are shown in the rn ia hard to distinguish spectrally, but have been suggested by some thirdrowofFig.6.Thisproducesaminorchangeinthefirsttwo In s studies(e.g.Grupeetal.2008;Nodaetal.2013)andareanatural components–theybothshowaninclineathighenergies,ratherthan titu consequence of multizone Comptonization models. We therefore being completely flat, and produces a third significant PC (right- te o include a weaker second power law, with a harder photon index handpanel).Thiscomponentappearstoactasacorrectionfactorto f T (cid:3)=1andanormalizationof0.1timesthatoftheprimarypower thesecondcomponent,whichnolongerdescribesallofthepivoting ec h law.Whenwekeepthissecondcontinuumcomponentconstantand itself.Ifwedoubletherangethatthephotonindexvariesover,we n o varytheprimarypowerlawasbefore,theeffectissimplytolower seethatthesecondcomponentchangessignificantly(bottomrow). log y thefractionalvariabilityoftheprimarycomponentwithincreasing Thiscomponentismostsimilartothepivotingcomponentproduced o n energy.ThisisshowninthelefttwopanelsofFig.5. when the blackbody is constant (top row). The third component A p inWtheehpahvoetosnoifnadretxreoafteadpsrpiemcatrraylppoivwoetirn-lgawascboenintignuduume.toHcohwaenvgeers, cAhgaanigne,sthoentlhyirvdercyomslipgohntelyn,tahnedreshisowascothrreecstaimonefgaecntoerr,arlastthreurcttuhraen. ril 2, 2 itispossibletogenerateasimilareffectfromtheinterplaybetween a direct match to a single physical component. However, in this 01 5 two(ormore)continuumcomponentswithdifferentphotonindices case the third PC is used to make the second PC appear more changing in normalization. In the right two panels of Fig. 5 , we liketheblackbodyPCinrows2and3,ratherthantheotherway showthetwocomponentsproducedfromthesametwopower-law around.InFig.7,weshowtheeffectofaddingthe‘correctionfactor’ model,whenbothpower-lawcomponentschangeinnormalization componentontothesecond-orderblackbodycomponentfromthe but not index. The primary power law is varied in normalization weaklyvaryingpower-lawindexcase.Thisproducesacomponent between0.5and2,andthesecondarypowerlawbetween0.08and withthesamespectralshapeasthepivotingterm(topmiddlepanel 0.12.Asexpected,thefirstPCisverysimilarinthiscaseandthat ofFig.6). withnovariabilityinthesecondpowerlaw,buttherearequalitative The third components produced by these simulations demon- differencesinthesecondcomponent.Asthisnowcorrespondsto strate a key weakness of this kind of analysis – if two physical the spectral pivoting caused by changes in the flux of the second components have a similar effect on the spectrum, then they will continuumcomponent,itisenhancedathighenergies,ratherthan notbeexpressedastwoseparatePCs–rathertherewillbeonePC damped out as in the previous case. This may be relevant to the describing the average effect, and one describing the differences objects in Section 4.2, where the second component gets steeper betweenthetwo.Inthiscase,bothanincreaseintheblackbodyflux withenergy.Fortheremainderofthesesimulations,wewillonly andanincreaseinthephotonindexproduceasteeperspectrum,so consider the case of a single, pivoting power law, but the reader thesecondcomponentisanaverageofthesetwoeffects.Thiswas shouldbearinmindthatasimilareffectcouldbeachievedwitha also found to be the case in the absorption simulations shown in MNRAS447,72–96(2015) RevealingAGNvariability 77 D o w n lo a d e d fro m h ttp ://m n Figure6. PCsreturnedfromsimulationsofpowerlawandblackbodyvariability.Thetoprowshowsthecomponentsproducedwhenthepowerlawisallowed ra s tovaryinnormalizationandphotonindex,whiletheblackbodyiskeptconstant.Thesecondrowshowsthoseforafixedphotonindex,butwheretheblackbody .o x fluxisallowedtovary.Thethirdandfourthrowsshowthecomponentsproducedwhenallthreeparametersareallowedtovary,withtheblackbodyvariations fo rd beingstrongerinthethirdandthepowerlawpivotingdominatinginthefourth. jo u rn a ls binariesthanAGN),itisinstructivetonotethatvariationsinasoft .org spectral component can result in a PC that shows apparent hard a/ variability.Thismayberelevanttothefourth-orderPCinMCG–6- t C a 30-15andsimilarobjects(Section4.1). lifo rn ia In s 3.4 Distantreflection titu te o DistantorneutralreflectionisfoundinmanyAGN(e.g.Riccietal. f T 2014),andoccurswhenX-rayemissionfromthecoronaisscattered ec h andreprocessedbycoldmaterial,farfromtheBH.Becauseofthis n o muchlargerspatialscale,thevariabilityinthisspectralcomponent log y ismuchlowerthanthatfoundincomponentsthatoriginatefromthe o n innerdisc,orthoseduetointerveningcloudsorwinds.Itfollows A p tvhaartiatbhielitmyaiinnthefefeecnteorgfydbisatnandtsrweflheercetioitniswsitlrlobneg,topadratimcuplaorulyt tthhee ril 2, 2 6.4FeKαline. 01 5 WeshowinFig.8thetwoPCsreturnedfromasimulationofa Figure 7. The effect of subtracting the ‘correction factor’ PC from the varying power-law continuum and constant distant reflection. We second-orderblackbodyvariabilitytermforthesimulationsofpowerlawand modelthereflectioncomponentwiththeXILLVERmodel(Garc´ıaetal. blackbodyvariability.ThePCcorrespondingtotheblackbodyvariationsis 2013),withaninputpower-lawindexof2,anironabundanceof1, showninreddiamonds,andthethirdPCisshowninbluesquares.Whenone theionizationξ atthelowestallowedvalueof1andaninclination issubtractedfromtheother,theresultingcomponent(greencrosses)matches of30degrees.Thefluxofthereflectioncomponentisfixed,andis thepivotingPCproducedwhenthephotonindexvariationsdominate(see approximatelyequaltohalftheaveragefluxofthecontinuum.The thetoprowofFig.6). primarypowerlawisthenvariedasbefore.WhiletheresultingPCs doshowslightdifferencesincurvatureoverthewholeenergyrange, P14b,wherebecausethelow-energyspectrumofNGC1365was these are likely to be undetectable due to noise and the presence dominatedbydiffusethermalemissiontheabsorptionandintrinsic of other spectral components in the real data. By far the largest variabilitycomponentswereverysimilar,leadingtoanaveraging differenceisthestrong,narrowironlinevisibleat6.4keV,which effect.However,thedominantdriverofthespectralvariabilitycan suppresses the variability of the power-law components, pushing stillbeidentified,asshownhereandinP14b. themtowards0.Wenotethatinthecasewheretheprimarypower Whilewedonotfindanysourcesthatshowsuchsimpleblack- law is heavily obscured, then it is probable that distant reflection body variability (which is far more likely to be visible in X-ray willleavemoresignaturesinthePCspectraatlowenergies. MNRAS447,72–96(2015) 78 M.L.Parkeretal. asubstantialcontributiontothetotalflux.Thesecondcomponent is similar, although starting from a diagonal rather than flat line, itislikewisepushedtowardszerowherethereflectioncomponent dominates. Fig. 10 shows the three significant components obtained from simulationsofamodelwithavaryingpowerlawandvaryingrel- ativisticreflection,forarangeofionizationparameters.Thefluxes ofthepower-lawandreflectioncomponentsarekeptapproximately equal, but the reflection component is only allowed to vary by a factor of 2, compared to 5 for the power law. The first two PCs areequivalenttothoseshowninFig.9,andarelargelyduetothe power-lawvariability.TheadditionalthirdPCrepresentsallofthe reflectionvariabilitythatcannotbeadequatelydescribedbythefirst two PCs. This component displays the correlated soft excess and Figure8. Thefirst(top)andsecond(bottom)PCsfromasimulationof broadironlinetypicalofrelativisticreflection.Unlikethedistant apower-lawcontinuumvaryinginnormalizationandspectralindexinthe reflection discussed in Section 3.4, relativistic reflection makes a D presenceofaconstantneutralreflectioncomponent.Themaindifference strong contribution at soft energies, and has a much broader iron ow causedtotheshapeofthecomponentsreturnedisthestrongnarrowiron line. nlo linefeatureat6.4keV. These simulations are particularly relevant to the sources dis- ade d cussedinSection4.1,fiveofwhichshowboththelow-andhigh- fro energybreaksinthefirstandsecondcomponentsandahigherorder m h PC,verysimilartothosepresentedinFig.10,withacorrelatedsoft ttp excessandbroadironline. ://m n ra s 3.6 Neutralabsorption .o x fo We are also interested in the effects of absorption variability in rd jo AGN spectra, although there are problems with using PCA in an u rn absorption-dominatedvariabilityregime.Oneofthekeyassump- als tionsofPCAisthatthedatasetcanbeexpressedasalinearsum .org of PCs. This assumption is reasonable when applied to additive a/ componentssuchasareflectionspectrum,butisnotvalidwhenwe t C a consider variable multiplicative components applied to a variable lifo spectrum, potentially leading to spurious terms being produced. rn ia Constantabsorptionproducesnosuchproblem,asaconstantmul- In Figure9. Thefirsttwo(topandbottom,respectively)PCsrecoveredfrom s asimulationofapowerlawvaryinginnormalizationandphotonindex,in tiplicative factor makes no difference to the fractional deviations titu weusetocalculatethePCs.Nevertheless,asshownbyP14band te thepresenceofaconstantrelativisticallyblurredreflectioncomponent.The o effectofthiscomponentistosuppressthespectralvariabilityattheenergies aswedemonstratehere,itispossibletofindphysicallymeaningful f T oftheironlineandsoftexcess. componentsfromsuchananalysis.Westressthatinallcases,the ec h shapeoftheunderlyingspectrumisunimportant,providedthatit n o isrelativelyconstantcomparedtotheabsorptionvariability.Again, log y thisisduetothePCsbeingcalculatedfromthefractionalresiduals o 3.5 Blurredreflection n ratherthanthetotalspectrum. A p Wsmeenaroewdprereflseecnttiotnhefrroemsultthseoifnsniemrualcactiroentisoinncdliuscd.inPg14reblaptirveissetnictatlhlye parFtiiagl.1c1ovsehroinwgsathbesothrpretieonPCcosmpepcotrnaenptro(dmuocdeedllwedhewniwtheZcPonCsFidAeBrSa ril 2, 2 twocomponentsproducedfromasimulationofavaryingpowerlaw, in XSPEC) applied to a (cid:3) = 2 power law. The covering fraction 01 5 andthethreecomponentsfoundwhenaweaklyvariable(approxi- is allowed to vary randomly between 0 and 1, and the column mately0.4timesasvariableasthepowerlaw)relativisticreflection density is allowed to vary by a factor of 3. The first component componentisadded.Wereproduceandexpandupontheseresults correlateswellwiththecoveringfraction;however,theothertwo here, investigating the effects of a strongly blurred and ionized componentsreturnedbytheanalysisdonotcorresponddirectlyto reflectioncomponent. a single parameter and represent changes in the column density In Fig. 9, we show the two components returned from a simu- at different covering fractions. The first component matches well lationofavaryingpowerlawandconstantreflection.Thereflec- withthefirstcomponentofNGC4395(Section4.3.1),whichshows tion parameters are the same as those in Section 3.4, except the strongabsorptionvariability(Nardini&Risaliti2011).Indeed,ifwe normalizationofthereflectioncomponentisincreasedsothatthe simulateasourcewherethevariabilityisdominatedbychangesin 0.3–10keVfluxis approximately equaltothatofthepower law. thecoveringfractionofaneutralabsorber,allowingsomevariability The reflection spectrum is then convolved with the KDBLUR, with of the underlying power law, we obtain two components which the inner and outer radii set at 1.235 and 400 gravitational radii, are in excellent agreement to those shown found for NGC 4395. respectively,andanemissivityindexof3.ThefirstPC,shownin ThissimulationisshowninFig.12,forapartialcoveringabsorber thetoppanel,canessentiallybethoughtofasaflatlinewiththe that varies in covering fraction from 0.5 to 1, with N fixed at H reflectioncomponentsubtracted.Thisthenrepresentsthesuppres- 1022 cm−2, applied to a power law with (cid:3) = 2 that varies by a sionofvariabilityatenergieswheretherelativisticreflectionmakes factorof∼20percent. MNRAS447,72–96(2015) RevealingAGNvariability 79 D o w n lo a d e d fro m h ttp ://m n ra s .o x fo rd jo u rn a ls .o rg a/ Figure10. Thefirst(left,black),second(middle,red)andthird(right,blue)PCsobtainedbyusingPCAonsimulatedspectraofapower-lawplusrelativistic t C a reflectionmodel,wherethepower-lawfluxvariesbyafactorof4andthereflectedfluxbyafactorof1.5.Thesecomponentsareshownforfourdifferent lifo ionizationparameters,andthefractionalvariabilityattributedtoeachcomponentisshownineachplot.Blackdashedlinesindicatezeroonthey-axisforthe rn secondtwoplots(axislabelsarenotshown,asthescaleisarbitrary). ia In s titu te o f T e c h n o lo g y o n A p ril 2 , 2 0 1 5 Figure12. ThetwoPCsfoundfromasimulationofpartialcoveringab- sorptionandintrinsicvariability. 3.7 Ionizedabsorption Figure11. ThethreePCsproducedbyasimulationofvaryingabsorption. In Fig. 13 (reproduced from P14b), we show the PCs produced Apowerlawisconvolvedwithaneutralpartialcoveringabsorptionmodel, byvariationsinanionizedpartialcoveringabsorber.Forthissim- andthecolumndensityandcoveringfractionoftheabsorberareallowedto change.Thefirsttermshownherecorrespondswelltothecoveringfraction, ulation, we use the ZXIPCF model (Reeves et al. 2008) and allow andthesecondtwotermsrepresentcorrectionfactorstothis,whichdepend the covering fraction to vary randomly between zero and one for onthecolumndensity. differentvaluesoftheionizationξ.ThisproducesasinglePCin allcases,thespectralshapeofwhichdependsstronglyontheion- The marginally more complex simulations from P14b showed ization. We also investigate allowing the column density to vary, the effects of diffuse thermal emission on the PCs returned from keeping the covering fraction fixed at 0.5. This produces almost partialcovering.Thisaddition,whichcharacterizesthePCspectra identicalprimarycomponents,andinthecaseofthelowestioniza- ofNGC1365,dampsoutthevariabilityatlowenergies. tionsimulationasecondcomponent,similartotheonefoundinthe MNRAS447,72–96(2015) 80 M.L.Parkeretal. differentpatternofvariability.Thesesourcesshowsimilarsuppres- sionoftheprimarypowerlawbyareflectioncomponent,butthe pivotingtermsteepenssharplywithenergy. (iii) In Section 4.3, we discuss the three sources (NGC 1365, NGC 4151 and NGC 4395) that show good evidence of variable partialcoveringabsorption.Thehigherordertermsdifferbetween these objects, and may indicate the presence or lack of intrinsic variability. (iv) Finally,weincludethePCspectrafortheninesourceswhich donotappeartofitintothegroupingsdiscussedsofar.Theseobjects arepresumablyexhibitingdifferentvariabilitymechanisms,andwe discussthemonanindividualbasis. 4.1 Group1:MCG–6-30-15analogues Within the sample of objects we have analysed, four additional D o NLS1 sources that show the same four variable components as w n MCG–06-30-15havebeenidentified.ThesesourcesareNGC4051, lo 1H0707−495, NGC 3516 and Mrk 766. We also found several ade d sourceswhichcouldbedisplayingthesamevariabilitypattern,but fro havelowerdataquality,makingitimpossibletobecertain. m h ttp Figure13. PCsproducedbysimulationsofpartiallycoveringionizedab- 4.1.1 MCG–6-30-15 ://m n sorption.Thecoveringfractionisallowedtovaryrandomlybetween0and ra 1,withallotherparametersfixed,producingasinglesignificantcomponent. MCG–6-30-15isverywell-studied,brightandhighlyvariablenar- s.o x The four panels correspond to four different ionization parameters. This rowlineSeyfert1(NLS1)galaxy.ItwasthefirstAGNinwhicha fo figureisreproducedfromP14b. relativisticallybroadenedironlinewasfound(Tanakaetal.1995), rdjo andalsoshowsthecharacteristicfeaturesofwarmabsorption(Otani urn neutralcase(Fig.11),isalsoreturned.Wefindthatastheionization etal.1996). als parameterisincreased,thestrengthofthecomponentreturned(in Fig.14showsthethreePCspectrapresentedbyP14a,alongwith .org termsofthefractionofthespectralvariabilityattributabletothis theweakfourthcomponentnotdiscussedinthatwork.Thefirstthree a/ component,andnottonoise)lowers.Thisisduetothedecreased t Ca effect of the absorption, and means that we are most likely to be lifo abletodetectneutralabsorptionvariabilityusingthismethod. rnia In s titu 4 PCA RESULTS FROM THE DEEP te XMM–NEWTON SAMPLE of T e Inthissection,wepresenttheresultsfromPCAofthe26sources ch n in our sample. For each source, we show all the significant PCs olo returnedandthefractionofthetotalvariabilityattributabletoeach gy o component. As in the case of the simulations presented earlier, n A the resulting component spectra are equally valid with the y-axis p inverted.BasedonthePCAresultsobtained,weattempttoarrange ril 2 and categorize the sources analysed in a logical manner, through , 2 0 comparisonwiththesimulationspresentedintheprevioussection, 15 andbreakthesampleintofoursubgroups. (i) ThefirstsourceanalysedusingthismethodwasMCG–6-30- 15, which was presented in P14a. This was found to have four significantPCs,althoughthefourthwasweakandwasnotinvesti- gatedindetail.HavinganalysedoursampleofAGN,wefindfour moresources(NGC4051,1H0707−495,Mrk766andNGC3516) whichdisplaythesamepatternofvariabilitywithfourcomponents, andseveralotherswhicharelimitedbyfluxorlackofvariability, but which show at least the first two components, with a similar spectral shape. The key features of this group of objects are the suppressionofthecontinuumvariabilityaroundtheenergiesofthe iron line and soft excess and a PC showing a strong correlation betweentheseenergybands. Figure14. PCspectrafoundinMCG–6-30-15,orderedfromtoptobottom (ii) Wefindasecondgroupoffourobjects(Ark564,PKS0558- bythefractionofvariabilityineachcomponent,aspresentedinP14a.The 504,Mrk335and1H0419−577)withasimilarbutqualitatively percentageofvariabilityineachcomponentisalsoshown. MNRAS447,72–96(2015) RevealingAGNvariability 81 PCsfoundinthisobjectwereanalysedindetailbyP14a,whofound thattheywerewellexplainedbytheeffectsofapowerlawvarying innormalizationandphotonindex,anduncorrelatedvariationsina relativelyconstantreflectionspectrum.Thesefindingsareconsistent withthelight-bendinginterpretationofthevariabilityinthisAGN, in which the height of the primary X-ray source above the disc changes (Martocchia, Karas & Matt 2000; Miniutti et al. 2003; Miniutti&Fabian2004)leadingtomoreextremevariationsinthe primaryemissionthaninthereflectedemission. Thesuppressionoftheprimarycomponentattheenergiesofthe softexcessandironlineindicatethepresenceofastrong,relatively constantspectralcomponentattheseenergies.Likewise,thebreaks inthesecondcomponentcorrespondtothesameenergies,wherethe primarypowerlaw,andhencevariationsfromthechangesinphoton index,aresuppressed.Thisisbestexplainedbyastrongrelativis- tically blurred reflection component, which is relatively constant D o whencomparedtothepowerlawduetolightbending.Apartially w n coveredpowerlawcanreproducethespectrumofthesource,butnot lo a thespectralshapeoftheobservedPCs(P14a,b).InP14b,weshowed de d thatthefirstthreecomponentscouldbeproducedbythevariable fro continuumandreflectionmodel,andthatthepredictionsforeither m h ionizedorneutralpartialcoveringabsorptionvariabilityarecom- ttp pletelydifferentfromthoseduetointrinsicvariability(Figs11and ://m 13) and therefore cannot explain the observed variability without n ra extremefine-tuningofmultiplespectralcomponents. s.o Itisinterestingthatthereflectioncomponentshowsaturnover xfo below∼0.7keV(thirdpanel),whereasthesuppressionofthecon- rd jo tinuumvariability(firstpanel)showsnosuchbreak.Thissuggests u rn thatthevariablereflectioncomponentisolatedhereisnotthesole als origin of the soft excess in this source. The fourth-order compo- .org nent also appears to contribute to the soft excess, although it is a/ stronglysuppressedbythereflectioncomponent.Theoriginofthis t C a ccoomnvpinocnienngtlyrermepariondsuacemiytsutseirnyg, saismwuelathioavnes.so far been unable to Figure15. PCspectrafoundinNGC4051,orderedfromtoptobottomby liforn thefractionofvariabilityineachcomponent.PCs1,2,3and4correspond ia to one, two and four and three from MCG–06-30-15, and the additional In s component5appearstobeattributabletoavariableabsorptionfeature. titu te o 4.1.2 NGC4051 Wenotethatthevariabilityisdominatedbythefirst(power-law) f T e component,thespectrumofwhichisqualitativelyverysimilarto ch NGC 4051 is an NLS1 which is extremely variable on all time- n the rms spectrum of NGC 4051 presented by Ponti et al. This is o scales.Pontietal.(2006)showedthatthespectrumofthesource unsurprising,giventhatbothmethodsshouldproduceaspectrum logy invariousfluxstatescouldbewelldescribedbyapower-lawplus of the relative strength of variability (which is largely due to the on relativistic reflection model, like that used in MCG–6-30-15. It A variablepowerlawinbothcases)butitisaninterestingconfirmation p a&lsUotetlxehyib2i0t1s3s)t,rownhgilcyhflfauvxo-udrepmeonddeelnstintivmoelvilnaggsin(tArilnsstoicn,vaVraiaubgihliatyn ofTouhremmeutchhodh.ighersignaltonoiseinthethirdcomponentrelative ril 2, 2 andrelativisticreflectionoverreprocessingbydistantmaterial. tothefourthcomponentinMCG–06-30-15givesustheopportunity 015 OuranalysisofNGC4051revealsfivesignificantPCs,shownin toinvestigatethiscomponentinmoredetail,andhopefullytoun- Fig.15,fourofwhichcorrespondwelltothosefoundinMCG–6- derstanditsorigin.InSection3,weshowedthatacomponentwith 30-15.PCs1,2and4matchcomponentsone,twoandthreefrom thisspectralshapecanbeproducedbyaddingasecondreflection MCG–06-30-15andthesimulationshowninFig.10.Thesecompo- componenttothespectrum,withahigherionizationparameterand nentsshowthesamebreaksanddips,anddifferonlyquantitatively. moreextremerelativisticblurring. The third PC from NGC 4051 appears to match the weak fourth componentfoundinMCG–6-30-15.Finally,thefifthcomponent, which has no analogue in MCG–6-30-15, shows what appears to 4.1.3 1H0707−495,NGC3516andMrk766 beanabsorptionedgeatanenergyof∼1keV,withnootherstrong features visible in the spectrum. We suggest that this component 1H0707−495,NGC3516andMrk766alldisplaythesamefour corresponds to a change in the properties of an ionized absorber, variablecomponentsasMCG–06-30-15andNGC4051,againwith asdescribedin(Ogleetal.2004),howeverthisconclusionisex- the only large difference being the order of the third and fourth tremelytentativeandmustbetreatedwithcautionbecauseofthe component,whichisreversedwithrespecttoMCG–06-30-15in1H highorderofthisPC.Weconclude,basedonthealmostidentical 0707−495andMrk766.Thecomponentspectraoftheseobjectsare components produced, that the variability in this source is driven shown in Fig.16, with the third and fourth components swapped bythesameprocessesasinMCG–6-30-15. in Mrk 766 and 1H0707−495 for ease of comparison. All four MNRAS447,72–96(2015)
Description: