Mon.Not.R.Astron.Soc.000,??–??() Printed29January2010 (MNLATEXstylefilev2.2) The Building Up of the Black Hole Mass - Stellar Mass Relation A. Lamastra1(cid:63), N. Menci1, R. Maiolino1, F. Fiore1, A. Merloni2 1INAF-OsservatorioAstronomicodiRoma,viadiFrascati33,I-00040Monteporzio,Italy 2Max-Planck-Institutfu¨rExtraterrestrischePhysik,Giessenbachstrasse,D-85741Garching,Germany 0 1 0 Submitted: 2 n a ABSTRACT J WederivethegrowthofSMBHsrelativetothestellarcontentoftheirhostgalaxypredicted 9 undertheassumptionofBHaccretiontriggeredbygalaxyencountersoccurringduringtheir 2 merging histories. The latter are described through Monte Carlo realizations, and are con- nected to gas processes, star formation and BH accretion using a semi-analytic model of ] galaxyformationinacosmologicalframework.Thisallowsustoconnectthestarformation O processinthehostgalaxiestothegrowthofSupermassiveBlackHoles.Weshowthat,within C thisframework,theratioΓ ≡ (M /M )(z)/(M /M )(z = 0)betweentheBlackHolemass BH ∗ BH ∗ . and the galactic stellar mass (normalized to the local value) depends on both BH mass and h redshift.WhiletheaveragevalueandthespreadofΓ(z)increasewithz,suchaneffectislarger p for massive BHs, reaching values Γ ≈ 5 for massive Black Holes (M ≥ 109 M ) at z (cid:38) 4, - (cid:12) o inagreementwithrecentobservationsofhigh-redshiftQSOs;thisisduetothetheeffective- r nessofinteractionsintriggeringBHaccretioninhigh-densityenvironments(wheremassive t s haloesform)athighredshifts. a ToinvestigatehowdifferentobservationsofΓ(z)fitwithinourframework,weworkedout [ specificpredictionsforsub-samplesofthesimulatedgalaxiescorrespondingtothedifferent 1 observational samples for which measurements of Γ have been obtained. We found that for v BroadLineAGNsatintermediateredshifts1(cid:46)z(cid:46)2valuesofΓ≈2areexpected,withamild 7 trend toward larger value for increasing BH mass. Instead, when we select from our Monte 0 Carlosimulationsonlyextremelygasrich,rapidlystarforminggalaxiesattheepochofpeak 4 inthecosmicstarformation(2 ≤ z ≤ 3),wefindlowvalues0.3 ≤ Γ ≤ 1.5,consistentwith 5 recentobservationalfindingsonsamplesofsub-mmgalaxies;intheframeworkofourmodel, . 1 theseobjectsendupatz=0inlow-to-intermediatemassBHs(M ≤109 M ),andtheydonot (cid:12) 0 representtypicalpathsleadingtolocalmassivegalaxies.Thelatterhaveformedpreferentially 0 throughpaths(intheM −M plane)passingabovethelocalM −M relation.Wediscuss ∗ BH ∗ BH 1 how the global picture emerging from the model is consistent with a downsizing scenario, : v wheremassiveBHsaccretealargerfractionoftheirfinalmassathighredshiftsz≥4 i X Keywords: galaxies:active—galaxies:formation—galaxies:evolution r a 1 INTRODUCTION in the past, and that a strong physical connection exists between galaxyformationandthegrowthofSMBHs. ThediscoveryoftightlocalcorrelationsbetweenthemassM of BH Understanding the mechanisms responsible for such a con- SupermassiveBlackHoles(SMBHs)andglobalpropertiesoftheir nection, and establishing the relative time scales for star forma- hostgalaxies,likethestellarmassM orthevelocitydispersionσ , ∗ ∗ tion/assembly and for SMBHs growth, requires the measurement (Kormendy & Richstone 1995; Magorrian et al. 1998; Ho 1999; oftheabovecorrelationsathigherredshifts.Forsuchascope,the Gebhardtetal.2000;Ferrarese&Merritt2000;Marconi&Hunt moststraightforwardcorrelationtostudyistheM −M relation, 2003;Ha¨ring&Rix2004;Kormendy&Bender2009)constitutes BH ∗ sincetheotherfundamentalrelation M −σ wouldrequirethe an important breakthrough in the understanding of galaxy evolu- BH ∗ measurementofstellarvelocitydispersionswhichareverydifficult tion.IfthegrowthofSMBHsisconsiderablycontributedbyaccre- toachieveatz(cid:38)1innormalgalaxies. tionovercosmologicaltimes(Soltan1982;Yu&Tremaine2002; Marconietal.2004;Merloni&Heinz2008)thisimpliesthatmost, TheMBH−M∗relationinthelocalUniversehasbeenwidely ifnotall,galaxieshavehostedanActiveGalacticNucleus(AGN) investigatedintheliterature.Magorrianetal.(1998)werethefirst to find a correlation between the black hole mass and the stellar mass, however they found a very large scatter (∼0.51 dex). This relation was then re-examined using more reliable M and M BH ∗ (cid:63) E-mail:[email protected] measurements(Merritt&Ferrarese2001,Marconi&Hunt2003, (cid:13)c RAS 2 Lamastraetal. Ha¨ring & Rix 2004); these studies showed that the scatter in the thecorrelationbetweenstarformationrateandstellarmass,such M − M relation is comparable to the scatter in the M −σ observationstendtobebiasedtowardsmassivestellarhosts. BH ∗ BH ∗ relation(∼0.3dex,seeHa¨ring&Rix2004).Itisalsoworthnoting Attemptstoderivetheevolutionoftheglobal(cid:104)Γ(z)(cid:105)averaged that the relation is BH mass dependent (Ha¨ring & Rix 2004), overforwholeAGNpopulationrequiresadifferentapproach,based although the mean value of the MBH/M∗ ratio obtained from the onintegratedobservables.Bycomparingtheredshiftevolutionof Ha¨ring&Rix(2004)sample:(cid:104)log(MBH/M∗)(cid:105) (cid:39)-2.8,isconsistent theintegratedBHandofthestellarmassdensitiesMerloni,Rud- withthosederivedbyMerritt&Ferrarese(2001)andMarconi& nick&DiMatteo(2004)foundthat(cid:104)Γ(cid:105) ≈ (1+z)α withα (cid:39)0.5. Hunt(2003)whenthelatterisloweredbyafactorof5/3toaccount Sucharesultsisconsistentwiththeupperlimitsontheevolution formorerecentestimatesofthebulgevirialmasses(seeMaiolino of(cid:104)Γ(cid:105)derivedbyHopkinsetal.(2006)forz(cid:46)2. etal.2007). Thus,amajorchallengefornextstudiesofAGNandgalaxy formationisunderstandingwhichoftheaboveclassesofpathsΓ(z) TheSMBHmassesatz (cid:38) 0.5aregenerallyestimatedinlu- isthedominantoneinthecosmologicalevolutionofgalaxiesand minoustype1AGNslikequasars(QSOs)andradiogalaxies;for SMBHs, and what are the physical mechanisms determining the example, virial BH mass estimators are based on the assumption particularpathfollowedbyagalaxyasafunctionofitsproperties thatthebroad-lineregionoftheAGNsaredominatedbythegrav- in different cosmic epochs. While on the observational side such ityoftheSMBH(see,e.g.,Peterson&Wandel2000),andprovide a task would require a large, unbiased sample of AGNs, a com- massestimatesofSMBHsathigherredshifts(seeWandel,Peterson plementaryapproachistoconsidercosmologicalmodelsofgalaxy &Malkan1999;Kaspietal.2000;McLure&Dunlop2002,Mer- and SMBH evolution which assume a given physical mechanism lonietal.2009),extendingouttoz ≈ 6(e.g.,Willott,McLure& fortheAGNfeedingandforitsconnectionswiththeevolutionof Jarvis2003;McLure&Dunlop2004;Vestergaardetal.2004;Wal- the host galaxy; then, comparing the predictions of such models teretal.2004;Riechersetal.2008).TheevolutionoftheMBH/M∗ forspecificclassesofobjectswithdifferentexistingobservations ratio from the local value is generally quantified in terms of the would constitute a powerful probe for the proposed physical link parameter: betweengalaxiesandSMBHs. Indeed,intherecentyears,significantsstepsforwardhadbeen Γ≡(MBH/M∗)(z)/(MBH/M∗)(z=0). (1) taken toward the modelling of the cosmological co-evolution of galaxies and AGNs, starting from the seminal papers by Silk & AlthoughtheAGNactivitymakesitdifficulttoderivestellar Rees(1998);Fabian(1999);Cavaliere&Vittorini(2000);Wyithe masses for the host galaxies at z (cid:38) 1.5, several observational &Loeb(2003).Ontheonehand,recenthigh-resolutionhydrody- works suggest that at such redshifts the mean MBH/M∗ ratio is namicN-bodysimulations(see,e.g.,DiMatteo,Springel&Hern- significantlyhigherthanthepresentvalue,correspondingtoΓ>1. quist 2005, Springel 2005; Hopkins 2005, 2006) have shown the importanceofgalaxymergersastriggersforAGNaccretion,and Peng et al. (2006), applying the virial method on a sample theroleoftheAGNenergyfeedbackinthesubsequentevolution of 31 gravitationally lensed AGNs and 20 non-lensed AGN and of the host galaxy. However, such simulations are affected by a derivingthehostgalaxystellarmassfromtheR-bandluminosity, limitedexplorationofthesub-gridprescriptionsforthephysicsof found the M /M ratio to be higher than the present value by a SMBHs;mostimportant,theynecessarelyfocusonspecificgalaxy BH ∗ factorΓ≈2forAGNsat1≤z≤1.7andbyafactorofΓ≈4atz≥ systems,makingitdifficulttoassessthestatisticalrelevanceofa 1.7; analysing the hosts of 89 broad-line AGNs in the COSMOS particularpathinthebuildingupofthelocal M −M relation. BH ∗ surveyMerlonietal.(2009)deriveapositiveevolutionfittedwith Ontheotherhand,semi-analyticmodels(SAMs,Monaco,Salucci Γ ≈ (1+z)0.74 intheredshiftrange1 ≤ z ≤ 2.2;exploitingradio- &Danese2000,Kauffmann&Haenhelt2000;Volonteri,Haardt& loud unification McLure et al. (2006) obtained virial (linewidth) Madau2003;Granatoetal.2004;Mencietal.2006;Crotonetal. BHmassestimatesfromthe3CRRquasars,andthestellarmass 2006;Boweretal.2006;Marullietal.2008)canprobethewhole estimatesfromthe3CRRradiogalaxies,therebyprovidingblack setofpossiblepathsfollowedbygalaxiesinthe M −M plane BH ∗ holeandstellarmassestimatesforasinglepopulationofearly-type during their evolution. Since their results depend on the mecha- galaxies, finding Γ ≈ (1+z)2 for z (cid:46) 2; at even higher redshifts nisms assumed for the trigger of the AGN accretion and for the z≥4,measurementsofthedynamicalmassfromtheCOemission AGN energy feedback, SAMs are suitable to probe the statistical ofmoleculargasinthehostsofhigh-redshiftQSOs(Walteretal. outcomesprovidedbytheimplementedphysicalprocesseslinking 2004;Riechersetal.2008)yieldlargevaluesofΓ=5−10. SMBHswiththeirhostgalaxy. While the above studies indicate that the growth of SMBHs Here we adopt a state-of-the-art SAM (Menci et al. 2006, is faster than the stellar mass assembly, they all focus on lumi- 2008) which implements a specific, physical model for the BH nousAGNs,andarethusbiasedtowardsselectingthemostmas- accretionbasedoninteractionsastriggersforboththefuelingof sive SMBHs (see Lauer et al. 2007). In contrast to the above re- SMBHsandthestarbursts.Inturn,theinteractionrate,theamount sults, from the analysis of a sample of sub-mm selected galaxies ofgasavailableforBHaccretion,andthestellarcontentofgalax- (SMGs) Borys et al. (2005) find Γ to decrease with redshift for ies are computed from Monte Carlo realizations of galaxy merg- z(cid:46)2;morerecently,Alexanderetal.(2008)foundΓ(z=2)=0.3 ingtrees,enablingustofollowtheevolutionarypathsofgalaxies forSMGsatz ≈ 2,indicatingthatforsuchobjectsthegrowthof andoftheirSMBHsovercosmologicaltimes,fromthecollapseof SMBHsactuallylagsthatofthehoststellarmass.Theselatterstud- their progenitors from the primordial density field to the present. iesareaffectedbyadifferentbias,inthattheyselectultra-luminous Based on such a model we derive the evolutionary paths for the (L(cid:38)1012L ),gas-richgalaxies(withgasfractionrelativetobary- growthofSMBHsrelativetothatofthestellarmassoftheirhost (cid:12) onicmass f (cid:38) 0.6,seeGreveetal.2005;Tacconietal.2008), galaxies,toinvestigatewhicharethedominantpathsfollowedby gas characterizedbytheconversionofasignificantfractionoftheirini- differentgalaxypopulationsinbuildingupthelocalM −M re- BH ∗ tialgasreservoirof1010−1011 M intostarsinashorttimesscale lationinascenariocharacterizedbyAGNfuelingdrivenbygalaxy (cid:12) ofafew108 yrs(seeTacconi2008andreferencestherein);given interactions.Wethencomparedifferentmodelpredictionsforthe (cid:13)c RAS,MNRAS000,??–?? TheBuildingUpoftheBlackHoleMass-StellarMassRelation 3 M /M ratioofluminousintermediate-andhigh-redshiftgalax- inducinglossafangularmomentum.Therateofsuchinteractions BH ∗ ies,andforintermediateredshiftSMGswithexistingobservations, is anddiscussthephysicaloriginofthedifferentpathsΓ(z)followed τ−1=n Σ(r,v,V )V . (2) by galaxies and SMBHs during their evolution. Finally, we shall r T t rel rel providespecificpredictionstypicalofthescenariocharacterizedby Heren isthenumberdensityofgalaxiesinthesamehalo,V is T rel interaction-drivenAGNfueling,tobetestedwithnextexperiments. theirrelativevelocity,andΣisthecrosssectionforsuchencoun- terswhichisgivenbySaslaw(1985)intermsofthetidalradiusr t associatedtoagalaxywithgivencircularvelocityv(seeMenciet al.2003,2004). 2 THEMODEL ThefractionofcoldgasaccretedbytheBHinaninteraction Weusethesemi-analyticmodelasisdescribedindetailsinMenci eventiscomputedintermsthevariation∆jofthespecificangular et al. (2005, 2006, 2008); this connects, within a cosmological momentum j≈Gm/v ofthegastoread(Mencietal.2003) d framework,theaccretionontoSMBHsandtheensuingAGNactiv- (cid:12)(cid:12)∆j(cid:12)(cid:12) (cid:28)m(cid:48) r v (cid:29) itieswiththeevolutionofgalaxies.Herewerecallthebasicpoints. facc≈10−1(cid:12)(cid:12)(cid:12) j (cid:12)(cid:12)(cid:12)=10−1 m bd Vd . (3) rel Herebistheimpactparameter,evaluatedastheaveragedistanceof 2.1 HierarchicalGalaxyEvolution thegalaxiesinthehalo.Also,m(cid:48)isthemassofthepartnergalaxy intheinteraction,andtheaveragerunsovertheprobabilityoffind- Galaxy formation and evolution is driven by the collapse and ingsuchagalaxyinthesamehalowherethegalaxywithmassmis growth of dark matter (DM) haloes, which originate by gravita- located.Thevaluesofthequantitiesinvolvedintheaverageyield tionalinstabilityofoverdenseregionsintheprimordialDMden- values of f (cid:46) 10−2. For minor merging events and for the en- sityfield.Thisistakentobearandom,Gaussiandensityfieldwith acc countersamonggalaxieswithveryunequalmassratiosm(cid:48) (cid:28) m, ColdDarkMatter(CDM)powerspectrumwithinthe”concordance dominating the statistics in all hierarchical models of galaxy for- cosmology”(Spergeletal.2007)forwhichweadoptroundparam- etersΩΛ =0.7,Ω0 =0.3,baryonicdensityΩb =0.04andHubble mation,theaccretedfractiontakesvalues10−3(cid:46) facc(cid:46)10−2. constant(inunitsof100km/s/Mpc)h=0.7.Thenormalizationof Theaverageamountofcoldgasaccretedduringanaccretion thespectrumistakentobeσ =0.9intermsofthevarianceofthe episode is thus ∆macc = faccmc, and the duration of an accretion 8 fieldsmoothedoverregionsof8h−1Mpc. episode, i.e., the timescale for the QSO or AGN to shine, is as- sumedtobethecrossingtimeτ = r /v forthedestabilizedcold ThemergingratesoftheDMhaloesareprovidedbytheEx- d d gascomponent. tendedPress&Schechterformalism(seeBondetal.1991;Lacey The time-averaged bolometric luminosity so produced by a & Cole 1993). The clumps included into larger DM haloes may QSOhostedinagivengalaxyisthenprovidedby surviveassatellites,ormergetoformlargergalaxiesduetobinary aggregations,orcoalesceintothecentraldominantgalaxydueto ηc2∆m dynamicalfriction;theseprocessestakeplaceovertimescalesthat L= acc . (4) τ growlongerovercosmictime,sothenumberofsatellitegalaxies We adopt an energy-conversion efficiency η = 0.1 (see Yu & increasesastheDMhosthaloesgrowfromgroupstoclusters(see Tremaine 2002), and derive the X-ray luminosities L in the 2- Mencietal.2005,2006). X 10 keV band from the bolometric corrections given in Marconi The processes connecting the baryonic components to the etal.(2004).TheSMBHmassm growsmainlythroughaccre- growing DM haloes (with mass m) are described in our previous BH tion episodes as described above, besides coalescence with other papers(e.g.,Mencietal.2005).Thegasatvirialequilibriumwith SMBHs during galaxy merging. As initial condition, we assume theDMpotentialwellsundergoesradiativecooling.Thecooledgas smallseedBHsofmass102M (Madau&Rees2001)tobeini- mass m settles into a rotationally supported disc with radius r (cid:12) c d tiallypresentinallgalaxyprogenitors;ourresultsareinsensitiveto (typicallyrangingfrom1to5kpc),rotationvelocityv anddynam- icaltimet = r /v .Twochannelsofstarformationdmayconvert thespecificvalueaslongasitissmallerthansome105M(cid:12). d d d partofsuchagasintostars: i)quiescentstarformation,correspondingtothegradualconversion intostarsataratem˙ ∝m /t 2.3 AGNfeedback ∗ c d ii)starburststriggeredbyinteractions,whichdestabilizepartofthe Finally,ourSAMmodelincludesadetailedtreatmentofAGNfeed- cold gas available by inducing loss of angular momentum. Note back.Thisisassumedtostemfromthefastwindswithvelocityup thatgalaxyinteractions(amonggalaxieswithrelativevelocityV ) rel to10−1cobservedinthecentralregionsofAGNs(Weymann,Car- includenotonlymergingbutalsofly-byevents.Partofenergyre- swell&Smith1981;Turnsheketal.1988;Crenshawetal.2003; leasedbySNaefollowingstarformationisfedbackontothegalac- Chartasetal.2002;Poundsetal.2003,2006;Risalitietal.2005); ticgas,thusreturningpartofthecooledgastothehotgasphase theseareusuallythoughtooriginatefromtheaccelerationofdisc outflowsduetotheAGNradiationfield(Proga2007andreferences therein,Begelman2004).Thesesupersonicoutflowscompressthe 2.2 AccretionontoSMBHsandAGNemission gas into a blast wave terminated by a leading shock front, which ThemodelalsoincludesatreatmentofSMBHsgrowingatthecen- movesoutwardswithalowerbutstillsupersonicspeedandsweeps treofgalaxiesbyinteraction-triggeredinflowofcoldgas,follow- outthesurroundingmedium.Eventually,thismediumisexpelled ing the physical model proposed by Cavaliere & Vittorini (2000) fromthegalaxy. and implemented in Menci et al. (2006, 2008). The accretion of Quantitatively,theenergyinjectedintothegalacticgasinsuch cold gas is triggered by galaxy encounters (both of fly-by and of innerregionsistakentobeproportionaltotheenergyradiatedby mergingkind),whichdestabilizepartofthecoldgasavailableby theAGN,∆E =(cid:15) ηc2∆m .Thevalueoftheenergyfeedback AGN acc (cid:13)c RAS,MNRAS000,??–?? 4 Lamastraetal. Figure1. LeftPanel:ThelocalpredictedMBH−M∗relationiscomparedwithdatabyHa¨ringandRix(2004,diamonds),andMarconi&Hunt(2003,squares, hereM∗isderivedusingthebest-fittingvirialrelationofCappellarietal.2006);thecolourcoderepresentsthedensity(perMpc3)ofBHsinagivenMBH−M∗ bin,asindicatedbytheuppercolourbar.Thecontourplotallowstoshowhowthepredictedbehaviourofthescatter(growingwithdecreasingstellarmass)is consistentwiththeobservedscatterofthedatapoints. CentralPanel:ThepredictedMBH−M∗relationatz=4(colourcodedasabove)iscomparedwithdatacorrespondingtoobservationsofbroad-lineAGNs andhostgalaxiesintherange4≤z≤6;detailsonthedataandcorrespondingreferencesareprovidedintheAppendixA.Wealsoshowasasolidlinethe predictedmedianlocalvalue,correspondingtothemedianlocalvalueofMBHformodelgalaxieswithgivenM∗. RightPanel:WeshowsomeofthepathsintheMBH(t)−M∗(t)planefollowed,duringtheirevolution,byBHs(andbytheirhostgalaxies)reachingafinal massofMBH(z=0)≥1010M(cid:12). efficiencyforcouplingwiththesurroundinggasistakenas(cid:15) = acterizesthemerginghistoriesofmassiveobjectsformedinbiased, AGN 510−2(seeMencietal.2008). highdensityenvironments(seeMencietal.2008). NotethattheM −M relationispredictedtoevolvewithz BH ∗ asshownbythecentralpaneloffig.1,withthemassiveendofthe distributionreachingtheregioncorrespondingtotheobservations ofluminous,broad-lineAGNsatz≥4(seeAppendixAfordetails 3 RESULTS:OVERVIEWOFSMBHGROWTH onthedata).Atypicalpredictionofourinteraction-drivenmodelis AsafirststepinthestudyoftherelativegrowthofBHsandhost thattheevolutionoftheMBH−M∗relationincreasesforincreasing galaxies, we show in fig. 1 the local (left panel) and the high- stellarorBHmass;indeed,whilelow-massBHsinmodelgalaxies redshift (z = 4, central panel) MBH − M∗ relation that we obtain athighredshiftsarecharacterizedbyMBH/M∗ratiosclosetotheir fromourmodel. localvalues(atleastforthemajorityofthem),forthemostmassive The predicted local relation is consistent with observations, simulatedBHsthe MBH/M∗ issubstantiallylargerthanthemodel andthedatalieonthepredictedconfidenceregionrepresentedby localvalues.Wecanquantifytheevolutionofthe MBH/M∗ inour thecontourplot,althoughthepredicteddistributionshowsasmall modelthroughthequantityΓdefinedinSect.1(eq.1);sinceour offsetfromtheobservedbest-fittingrelationsderivedbyMarconi model predicts a whole distribution of local MBH for each stellar &Hunt(2003)andHa¨ring&Rix(2004).Although,inprinciple, mass M∗ (not justa simple power-law,as is generallyadopted to the model parameters (like the AGN feedback efficiency (cid:15)AGN or fitthelocaldata),wecomputeΓasthedeviationofthe MBH/M∗ thenormalizationofthestarformationefficiency)couldbetuned ratioofeachmodelblackholefromitsownlocalvalue.Thenfor to optimize the fitting to the observed local MBH − M∗ relation, galactic hosts with stellar mass M∗ ≈ 1011M(cid:12) the BH masses at this would also affect the predicted properties of the galaxy pop- z ≥ 4 are typically MBH ≈ 109.6M(cid:12) (see central panel of fig. 1), ulation (e.g., the evolution of the luminosity function, the colour moremassivethanthepredictedlocalvaluebyafactorΓ≈5.These distributions,thelocalTully-Fisherrelation)atbothlowandhigh objects constitute the progenitors of the local extremely massive redshifts.Sincethemodelisintendedtoprovideaunifieddescrip- BHs,andtheirgrowthwithtimeisrepresentedbythepathsshown tion of AGNs and galaxy evolution, we chose to adopt the same intherightpaneloffig.1.Notehowsuchpathsarecharacterized fiducial model adopted in our previous works since it provides a byanassemblyofBHmassesextremelyrapidintheearlyphases, good match to a wide set of galactic properties (see Menci et al. which is faster than the stellar mass growth since they approach 2006,2008).Instead,arobustfeatureofthe MBH −M∗ isconsti- theirfinalpositioninthe MBH −M∗ planepassingabovethelocal tutedbythebehaviourofthescatter,whichisalmostindependentof MBH−M∗relation. themodelparametersandevenonthespecificmechanismassumed Thephysicaloriginoftheabovebehaviourcanbeunderstood to trigger the AGN feeding. In fact, its decrease with increasing as follows: such high-redshift massive BHs (the counterpart of M istypicalofhierarchicalscenarios,anditisduetotheearly thoseobservedbyWalteretal.2004;Maiolinoetal.2007;Riech- BH assemblyofprogenitorsintoauniquemainprogenitorwhichchar- ersetal.2008)areformedfromgalaxiescollapsedinbiased,high- (cid:13)c RAS,MNRAS000,??–?? TheBuildingUpoftheBlackHoleMass-StellarMassRelation 5 Figure3. ThepredictedevolutionoftheBHtostellarmassratioΓ(nor- malizedtothelocalvalue,seeeq.1),fortheevolutionofBHswithmasses MBH≥109M(cid:12)atz≥4.The6filledcontourscorrespondtoequallyspaced Figure2. TopPanel:Weshowthestarformationhistoriesforasubsetof valuesofthefractionofobjectswithagivenvalueofΓ(z)attheconsidered modelgalaxies.Allhistorieshavebeennormalizedtounitywhenintegrated redshift:from0.01forthelightestfilledregionto0.1forthedarkest.For overcosmictime,tomakeeasierthecomparison,suchanormalizationde- referencewedrawasadashedlinethelocalvalueΓ(0)=1.Thedatapoints finestheunitsofthey-axis. representtheexcessΓderivedfromtheobservedBHmassesofthehigh- BottomPanel:TheBHmassaccretionhistoriesforthesamemodelgalax- redshiftbroad-lineAGNfollowingtheproceduredescribedinAppendixA. ies,normalizedasabove.ThechosensubsetofgalaxiesinourMonteCarlo simulationisthesameadoptedtoillustratethepathsintheMBH−M∗plane infig.1. clumpsatlowz)characteristicofhierarchicalscenarios,asnotedby severalauthors(e.g.,Zhao,Jing&Bo¨rner2003;Diemand,Kuhlen densityregionsoftheprimordialdensityfieldwherethecollapse &Madau2007;Homanetal.2007;Ascasibar&Gottloeber2008); andgrowthofgalactichostsisaccelerated.Thestarformationand 2)theinteraction-drivenscenarioforthetriggeringofAGNactiv- BHaccretionhistoriesofsuchobjectsareshowninfig.2,andcan ity. beschematicallydividedintwophases. In the above picture, we expect the paths with Γ(z) ≥ 1 to At high redshifts z (cid:38) 3, the rapid interactions characteriz- dominatethegrowthhistoriesofmassiveobjects,formedinbiased ingthedenseenvironmentofsuchgalaxiesareeffectiveinrapidly regionsofthedensityfieldwherehigh-redshiftinteractionsareex- destabilizingthegalacticgasandinfeedingthecentralBHs.Infact, tremelyeffective,whileweexpectthehigh-redshiftvaluesofΓto insuchenvironmentsandcosmictimes,boththeinteractionratein progressively lower when galaxies formed in less dense environ- eq. (2) and the fraction of destabilized gas (eq. 3) are large; the ment(andhencewithalowermassonaverageatanygivenz)are first,duetothelargedensities,andthesecondduetothelargeratio considered,duetotheprogressivelylowerefficiencyofinteractions m(cid:48)/m≈1(i.e.,amassm(cid:48)ofthemergingpartnerclosetothemass intriggeringBHaccretionathigh-z.Indeed,thedependenceofthe mofthemainprogenitor)characteristicofthisearlyphase(z(cid:38)3) aboveeffectonBHmassandonredshiftconstitutesatypicalsig- whengalaxyinteractionsmainlyinvolvepartnerswithcomparable natureoftheaboveinteraction-drivenmodelforthegrowthofBHs mass.Thefrequentstarburstsinstarformationandthecorrespond- ingalaxies. ingepisodesofBHmassaccretionatthisearlycosmictimesare showninfig.2. Atlowerz,thedeclineoftheinteractionrateandofthedesta- bilized fraction facc suppresses the growth of BHs which in our 4 RESULTS:THEEVOLUTIONOFMBH−M∗FOR model is only due to galaxy interactions (see the drop in the BH DIFFERENTAGNPOPULATIONS accretionepisodesinthebottompaneloffig.2),whilequiescent 4.1 TheEarlyBHGrowthandHigh-redshiftQSO star formation still proceeds, as is shown by the smooth compo- nentinthestarformationhistoriesinthetoppaneloffig.2.The Anefficientwaytotesttheabovescenarioistoinvestigatehowthe latter,quiescentcomponentofstarformationcontinuestobuildup pathsfollowedbyBHsandbytheirhostgalaxiestoreachthelocal stellar mass at z (cid:46) 2 (though at a milder rate), thus lowering the M −M relationdependontheirproperties(mass,gasfraction, BH ∗ M /M ratio.Theoverallresultisthat,whentheM /M ratiois starformationrate),andtocomparesuchspecificpredictionswith BH ∗ BH ∗ normalizedtothefinallocalvalue,theexcessΓ(z)increaseswith thecorrespondingavailableobservationsforwhichwewillusethe redshift. meanlocalvalueoftheM /M ratioobtainedfromtheHa¨ring& BH ∗ Notethattheaboveresultistheoutcomeoftwophysicalpro- Rix(2004)sampletocalculateΓ.Weshalldevotetothisthenext cesses:1)theapproximativelytwo-phasegrowthofcosmicstruc- subsections. tures(withmajormergingathighredshiftsandaccretionofsmall ToexploreindetailthegrowthofBHsingalaxieswithdiffer- (cid:13)c RAS,MNRAS000,??–?? 6 Lamastraetal. Figure4. LeftPanel:ThepredictedevolutionoftheBHtostellarmassratioΓ(normalizedtothelocalvalue,seeeq.1),forAGNsselectedastohavea bolometricluminosityL≥1044.5erg/sat1≤z≤2.The4filledcontourscorrespondtoequallyspacedvaluesofthefractionofobjectswithagivenvalueof Γ(z)attheconsideredredshift:from0.01forthelightestfilledregionto0.1forthedarkest.ThedatapointsrepresenttheexcessΓderivedfromtheBHmasses oftheCOSMOSbroad-lineAGNmeasuredbyMerlonietal.(2009).NotethatthebulkoftheBHbuildingupoccurredadhigherredshift,whenmostofthe galaxyinteractionstriggeringtheBHaccretiontakeplace:theslowevolutionΓ(z)for1≤z≤2isdrivenbythegradualandmuchmildergrowthofthestellar mass. RightPanel:ThedistributionoflogΓforallgalaxiesintheredshiftrange1 ≤ z ≤ 2formodelgalaxies(solidline)iscomparedwiththecorresponding observeddistribution(Merlonietal.2009,filledhistogram):thelatterhasbeenobtainedincludingthelowerlimitsshownintheleftpanel.Theaveragevalue andthedispersionforlogΓshownonthetoprefertotheobserveddistribution,thesevaluesarenearlyinsensitivetotheexclusionofthelowerlimits. entproperties,westartwiththemostextremeobjects,namely,the agreement with the observed distribution (see right panel in fig. massiveBHalreadyinplaceathighredshifts. 4).Thisindeedconstituteanextremelyimportanttestformodels Thetimeevolutionofthe M /M ratioofsuchextremeob- basedonthehierarchicalscenarios,sincethescatterisdirectlyre- BH ∗ jectsisshowninfig.3intermsoftheexcessΓ(z)overtheirlocal latedbythespreadinthemerginghistoriesofthehostgalaxies,a value for model BHs with masses M ≥ 109M at z ≥ 4. For specificpredictionofthesemodelswhichcannotbetunedthough BH (cid:12) such objects Γ(z) ≥ 1 holds at any time between z = 6 and the adjustablefreeparameters. present, with Γ ≈ 5 holding at at z = 5−6. These Γ values are Note that the model predictions shown in fig. 4 refer to all consistentwiththeobservationsofluminousQSOat3.6< z <6.4 AGNsintheredshiftandluminosityrangesspecifiedabove,while (seeAppendixA). thedatawecomparewithincludeonlyunobscuredobjects.How- ever,inourmodeltheobscurationpropertiesofAGNsdependonly on the amount of gas swept by the blast wave originated by the 4.2 TheBHgrowthatIntermediate-redshiftandBLAGNs AGN at the time of observation (see Sect. 2.3), i.e., on the de- tailed hydrodynamic of the interstellar medium during the last, Now, we focus on BL AGNs at intermediate redshift 1 ≤ z ≤ 2 short (∆t (cid:46) 5107 yrs) AGN active phase. Since the Γ ratios of and AGN bolometric luminosity L ≥ 1044.5 erg/s, for which the modelBHsbasicallydependontheintegrated,pasthistoryofthe modelresultscanbecomparedwithdetailedexistingobservations. simulated BHs and host galaxies, we expect distribution of Γ for Suchobjectsareexpectedtoforminlessbiasedregionsofthepri- obscured and unobscured AGNs to be similar. While we cannot mordialdensityfieldcomparedtotheclassofhigh-redshiftQSO performanexactinvestigationofthisissue(wedonotmodelthe discussedintheprevioussection;infact,theyhaveacomparable nuclearobscurationofAGNs),afirst-orderestimateoftheeffectof AGNluminositybutarefoundatalowerredshifts,andthusareex- selecting only obscured objects can be performed by considering pectedtooriginatefrommerginghistoriescharacterizedbyalower inourmodelonlyAGNshostedingalaxieswithhighgassurface rate of high-redshift encounters. Therefore, according to our pic- densities Σ ≥ 30M pc−2: evenin this casethe mean valueof turediscussedattheendofSect.3,weexpectsuchobjectstoshow Γinthedisgtarsibutionin(cid:12)fig.4isonlydecreasedbyafactor≈ 1.2, a lower excess Γ(z). This is indeed the case, as illustrated in fig. leavingthescatterunchanged. 4,wherewecomparethemodelpredictionswithobservationsby Merlonietal.(2009). ThedistributionofΓforsuchintermediateobjectsisstilldom- 4.3 TheGrowthofBlackHolesinGas-RichGalaxies: inatedbyΓ ≥ 1atanyrefdshift;themodelpredictstypicalvalues ComparisonwithSubmillimiterEmittingGalaxies Γ ≈ 1.5−2,smallerthanthoseattainedbythehigh-redshiftQSO shown in fig. 3, and consistent with the observational range. Be- Inourmodel,acompletelydifferentsetofBHaccretionhistories sidesprovidinganaverageΓclosetotheobservationalvalue,itis isthatcorrespondingtoAGNsinsub-mmgalaxies(SMGs).Their interestingtonotehowthemodelpredictsascatterinremarkable largegasfractions(f (cid:38) 0.6relativetothetotalbaryonicmass) gas (cid:13)c RAS,MNRAS000,??–?? TheBuildingUpoftheBlackHoleMass-StellarMassRelation 7 Figure5.ThepredictedevolutionoftheBHtostellarmassratioΓ(normalizedtothelocalvalue,seeeq.1),isshowninPanelsAandBforAGNsinSMG-like galaxiesselectedfromourMonteCarlosimulationaccordingtothecriteriai)-iii)(seetext).The5filledcontourscorrespondtoequallyspacedvaluesofthe fractionofobjectswithagivenvalueofΓ(z)attheconsideredredshift:from0.01forthelightestfilledregionto0.1forthedarkest.PanelsAandBreferto thedifferentselectioninX-rayluminosityoftheAGNsshownonthetopofthePanels.WecomparewithdataforSMGswithX-rayobscuredAGNs(Panel A)andwithBLAGNs(PanelB)inthesameluminosityrangeadoptedfortheselectionofmodelgalaxies;thedatacorrespondtoindividualgalaxiesfor whichbothBHandstellarordynamicalmassesmeasurementswereavailableintheliterature(seeAppendixB).TheerrorbarsinthePanelAshowonlythe uncertaintiesinthedataduetotheadoptedEddingtonratioλ,whileerrorbarsonPanelBincludetheobservationaluncertaintiesinboththeBHandthestellar massmeasurements.Wealsoshowasashadedareathe1−σuncertaintyregionaroundthevalueofΓobtainedbyAlexanderetal.(2008)fromaverage BH(adoptingλ=0.2)andstellarmasses(obtainedfromnearIRluminosityandCOlinewidths)takenfromvarioussamples(seeAlexanderetal.2008and referencestherein).Notethattheextremelygas-richgalaxiesselectedasSMGsrepresentrareevolutionarypaths(seetextforthecomputedandtheobserved numberdensities)comparedtotypicalgalaxiesatsuchredshifts:thestatisticalfluctuationsassociatedwiththelownumberofselectedSMGsinourmodel reflectsintheirregularcontoursandtheisolatedbinsshowninthefigure. andstarformationrate(m˙ =100−1000M /yr,seeIntroduction) tiveburstsandBHaccretionepisodesattheseearlyepochs.Since ∗ (cid:12) indicatethatthesegalaxiesoriginatefrommerginghistorieschar- inourmodelthehigh-redshiftinteractionsaretheonlytriggerfor acterizedbylessprominenthigh-redshft(z(cid:38)4)starburstsandBH earlyBHgrowth,thisresultsinlowervaluesofΓ(z)atlatertimes accretionepisodes,sothatalargefractionofgasisleftavailableat (z = 2−3). Such results do not depend on the details of the se- lowerredshiftsz ≈ 2−3.Tocomputequantitativelythepredicted lectioncriteriai)-ii)thatweadoptedtoextractSMG-likegalaxies M and M for such galaxies and to compare with existing ob- from our Monte Carlo simulations. Indeed, we have verified that BH ∗ servations,weselectedfromourMonteCarlosimulationsgalaxies adoptingdifferentthresholdsforthegasfractions(f ≥ 0.6)and gas withi)gasfractions f ≥ 0.7,ii)starformationratesm˙ ≥ 100 starformationrates(upto400M /yr)thefractionofgalaxieswith gas ∗ (cid:12) M /yr;iii)AGNswithX-rayluminositiesL ≤ L ≤ L ,whereL Γ<1isalwaysdominant. (cid:12) 1 2 1 andL havebeenchosenastomatchtheselectioncriteriaadopted It must be noted that the observed SMG at z=2 in the right 2 fortheobservationswecomparewith(Borysetal.2005;Alexander paneloffig.5hasaΓvaluemuchlowerthanthatpredictedbythe etal.2008);theX-rayluminosities(intheband2−10keV)have modelatthesameredshift.Inviewoftheunavoidableuncertainties beencomputedfromthebolometricluminositiesineq.(4)adopting affectingthepresentobservationaldata(discussedinAppendixB), thebolometriccorrectionbyMarconietal.(2004). itwillbeimportanttoconfirmthereliabilityandthestatisticalsig- Theresultingpredicteddistributionofthe M /M ratioare nificanceofsuchadatapointasamarkeroftheunobscuredAGNs BH ∗ shown in figs. 5a and 5b for SMG with AGN luminosity in two inSMGs,sinceitwouldconstituteaseveretestforourmodelwhich different ranges (see caption) as to compare with the Γ estimates predictsabasicallysimilarbehaviourofΓ(z)forobscuredandun- ofsixX-rayobscuredandthreebroadline(BL)SMGsforwhich obscuredAGNs.Ontheotherhand,thepredicteddistributionofΓ individual estimates of M and M were available in the litera- forSMG-likegalaxiesshowsamilddependencewiththeAGNlu- BH ∗ ture(seetheAppendixBfordetailsontheadopteddata).Although minosity,beinglargerforluminousAGN.Althoughtheuncertain- the comparison with observations results are still indicative (due tiesinpresentdatadonotallowyettomakeprecisecomparisons, tothelargeuncertaintiesassociatedwiththeEddingtonratiosand thistrendconstitutesamodelpredictionwhichisatleastconsistent with the measurements of both the BH and the stellar mass, see with estimates of Γ based on average values for M and M in BH ∗ AppendixB),themodelpredictsforbulkoftheSMGpopulation SMGgalaxies(seetheshadedarea). values of Γ ≤ 1, due to the slower growth of BHs at high red- Aninterestingfeatureofourinteraction-drivenmodelforBH shifts.Thisisduetotheparticularmerginghistoriescorresponding accretionisthattheSMGsproducedbythemodelendupinlow- toourselectioncriteria;largeresidualgasfractionsatz ≈ 2.5im- redshiftdescendantswhichliebelowthelocal M −M relation BH ∗ plylessfrequenthigh-redshiftencountersatz ≥ 4andlesseffec- asshowninfig.6.Inotherwords,thelow-redshift(z≤2)descen- (cid:13)c RAS,MNRAS000,??–?? 8 Lamastraetal. whichcontinuestobuildupstellarmassatz(cid:46)2,thoughatalower rate(seefig.2a).IntermsoftheratioΓ(z)suchaglobalbehaviour isrepresentedbyfig.7,wherethedistributionofΓisshownbythe colouredcontoursasafunctionofredshiftfortheentire(i.e.,forall galaxiescontainingBHwithmassesM ≥105M ).Astrikingfea- (cid:12) tureofsuchadistributionisthemodestincreaseofΓ(z)forz (cid:46) 3 asopposedtoitsrapidupturnathigherredshiftsz(cid:38)4;suchamild increase of the global Γ at low-intermediate redshifts is in agree- mentwiththatderivedbyMerlonietal.(2004)andHopkinsetal. (2006)bycomparingtheobservedintegratedBHandstellarmass densitiesatsuchredshifts. Note that the overall predicted trend for larger values of Γ with increasing redshift is associated to an increasing the spread inthedistribution.Thisconstitutesaquantitativepredictionforthe z-dependence of the Lauer et al. (2007) bias. In this respect, the Γ distributions shown in fig. 3, 4, and 5 can be interpreted as a Figure6.TypicalgrowthpathsforBHandstellarmassesforourmodel quantitativeevaluationoftheincidenceofsuchabiasindifferently SMG-likegalaxies.Thedotsrepresentthefinallocation(atz = 0)ofthe selectedsamplesatdifferentredshifts.Infact,fig.7showsthatthe descendentsofSMG-likegalaxiesselectedaccordingtoourcriteriai)-iii) (seetext)intheredshiftrange2 ≤ z ≤ 3.Thedashedlinerepresentthe differentobservationswecomparedwithinthepreviousSections medianvalueofthelocalMBH−M∗relationforallmodelgalaxiesatz=0. arenotfullyrepresentativeoftheglobalbehavioursincetheyre- fer to specific populations selected according to different criteria whichsampleonlyaportion(inthecaseofSMGactuallyaminor dants of SMGs are predicted to have BHs with low-intermediate fraction)ofthewholegalaxypopulation. masses M = 108 −109 M (see fig. 6), and the BHs in SMG- BH (cid:12) likegalaxiesat2 (cid:46) z (cid:46) 3areinanactivegrowingphase,asitis showninfig.6bytheirsteeplyrisingpathsintheM −M plane. 5.2 Modelpredictionsanddownsizing ∗ BH Conversely,wecomputedthatafraction≈ 10%oflocalgalaxies AmajorresultofourprevioussectionsisthattheevolutionofΓ(z) withBHmassesintherangeM =108−109M hadaprogenitor BH (cid:12) isastrongfunctionoftheBHmass.Suchastrongmassdependence whichpassedthroughanSMG-phaseatsomeredshiftintherange ofΓ(z)isillustratedbyfig.8(leftpanel),whereweshowwiththe 2≤z≤3. contourstheaveragevaluesΓ(z)asafunctionofthefinal(z=0)BH Suchapictureisinagreementwiththeindependentfindings mass(representedinthey-axis)andofredshift;theaverageruns ofAlexanderetal.(2008),basedonthelargernumberdensityof overallpathsleadingtothefinalBHmass(andhenceoverallthe SMGscomparedtothatoflocalgalaxieshostingBHswithmasses possiblemainprogenitorsatagivenredshift).Thefigureillustrates exceeding109 M .Tocheckforthefullconsistencyofourresults (cid:12) that indeed for massive objects M ≥ 109M at high redshifts BH (cid:12) andofourinterpretationwiththeobservationsofSMGs,wehave z (cid:38) 4 large values of Γ > 3 are expected, since such BH form compared the number densities of the model SMG galaxies with inthemostbiasedregionsofthedensityfieldwherehigh-redshift thenumberdensitiesρSMG (cid:39)2.5×10−5Mpc−3 measuredbySwin- interactionswerefavoured;valuesofΓ (cid:38) 1.5−2arealsonatural bank et al. (2006) based on the redshift distribution obtained by forBHatintermediateredshifts1 ≤ z ≤ 2,whilethelowvalues Chapmanetal.(2005)forz=1-3.5SMGswithS850µm >5mJy.To ofΓobservedinSMGat2 ≤ z ≤ 3arenotrepresentativeofthe performsuchacomparisonwefirstconvertedtheabovefluxlimit wholeAGNpopulationatsuchredhifts(theaverageΓrepresented intermsofstarformationrate(usingtherelationsinSwinbanket infig.8is≈1.5),butinsteadonlyapplytoapopulationoriginated al.2008),sinceoursemi-analyticmodeldoesnotincludethesub- bypeculiarstarformationhistories(seesection4.3). mm emission from dust; for the model SMG galaxies above the Animmediateimplicationoftheaboveisthatmassivelocal resultingthresholdinstarformationratethepredictednumberden- galaxies and their BHs have formed preferentially through paths sityinthesameredshiftintervalisρ = 4.4×10−5Mpc−3 when SMG (in the M − M plane) passing above the local M − M re- SMGswith f ≥0.6areconsidered,andρ =1.9×10−5Mpc−3 ∗ BH ∗ BH gas SMG lation;thisisillustratedinfig.8(rightpanel)whereweshowthe forSMGswith fgas ≥ 0.7,fullyconsistentwiththeobservational fractionofpathscharacterizedbyagivenΓ(intworedshiftbins) estimates. leadingtolocalgalaxieswithmassM ≥1012M .Thepathswith ∗ (cid:12) Γ>1dominatethestatisticsnotonlyathigh-redshiftz=4−5but also-thoughtoalowerextent-atz = 1−3.Asaconsequence,in 5 DISCUSSION themerging-drivenscenarioSMGsdonotrepresenttypicalpaths leadingtolocalmassivegalaxies,butrathercorrespondtopeculiar 5.1 ThegrowthofBHsfordifferentAGNpopulations pathsinthetailofthedistribution,namelythoseselectedastolead Theresultspresentedintheprevioussectionsleadtothefollowing to gas-rich galaxies at z = 1−3; this interpretation is supported globalpictureofourcosmological,interaction-drivenmodelforthe bytheconsistencyofthepredictednumberdensityofsuchpecu- BHaccretion.TherelativegrowthspeedofSMBHscomparedto liarpathswiththeobservednumberdensitiesofSMGs,shownand thestellarmassischaracterizedbyalargespreadinthegalaxypop- discussedattheendofsect.4.3. ulation,withanoveralltrendforafasterBHgrowthcomparedto TheabovemassdependenceofΓ(z)hasstraightforwardimpli- galaxiesathighredshiftsz(cid:38)3;thisisbasicallyduetothefactthat cationsfortheso-calledAGNdownsizing,i.e.,thefasterbuilding- BHgrowonlywheninteractionsareeffective,i.e.,athighredshifts upofluminousAGNscomparedwiththosewithlowerluminosities z (cid:38) 3, while star formation proceeds not only through impulsive whichisindicatedbyobservations(see,e.g.,Marconietal.2004); bursts(athighredshifts)butalsothroughquiescentstarformation suchadownsizingeffectisalsocharacterizingthefasterdrop(2.5 (cid:13)c RAS,MNRAS000,??–?? TheBuildingUpoftheBlackHoleMass-StellarMassRelation 9 Figure7.ThepredictedevolutionoftheBHtostellarmassratioΓforthewholegalaxypopulationinourMonteCarlosimulation;thecolourcoderepresents, atanyredshiftz,thefractionofobjectswithexcessΓ(z).TheaveragevaluesofΓderivedfromobservationsofSMGs(Alexanderetal.2008),ofBLAGNs (Merlonietal.2009)andhigh-redshiftQSOs(seeMaiolinoetal.2007andAppendixA)areshownasshadedareascoveringtheirobservedredshiftrangeand observationaluncertainties. Figure8.LeftPanel:TheaveragevaluesofΓatdifferentredshifts(x-axis)correspondingtoBHwithagivenfinalmass(y-axis)arerepresentedascontours. TheaverageistakenoverallmainprogenitorsinourMonteCarlomerginghistoriesleadingtotheagivenfinalmass. RightPanel:ThefractionofpathsleadingtolocalmassivegalaxieswithM∗≥1012M(cid:12)asafunctionoflogΓisshownfortworedshiftbins:4≤z≤5(shaded region),and1≤z≤3(solidline). dex)fromz≈2tothepresentofluminous(L ≥1045erg/s)AGNs Thisisillustratedbyfig.9(leftpanel),wherethecontoursrepresent X compared to the low-luminosity population (L ≤ 1043 erg/s), thefractionofBHmassformedatagivenredshift(inthex-axis), X which is observed to follow only a gradual decrease (less than 1 fordifferentvaluesoftheBHfinalmassinourmodel;anygiven dexinthenumberdensity)fromz = 2toz = 0(see,e.g.,Fioreet fractionofthefinalmassisassembledatlowerredshiftfortheless al.2003;Hasinger,Miyaji&Schmidt2005;LaFrancaetal.2005). massiveobjectswithM (cid:46)108M withrespecttomoremassive BH (cid:12) Infact,themassdependenceofΓ(z)representedinfig.8implies objects.Wealsoreproduceasasequenceofpointsasimilarcon- thatthemassiveBHactuallygrowfasterthanthelow-massBHs. tourplotproducedbyMarconietal.(2004)basedontheobserved (cid:13)c RAS,MNRAS000,??–?? 10 Lamastraetal. Figure9.LeftPanel:ThepredictedgrowthhistoriesofBHwithdifferentfinalmass.ForanyfinalBHmass(representedinthey-axis),the10filledcontours correspondtodifferentvaluesoftheaveragefractionalmassMBH(z)/MBH(z=0)oftheprogenitorBH:MBH(z)/MBH(z=0)=0.05,0.1,0.2,0.3,0.4,0.5,0.6, 0.7,0.8,0.9,fromthelightesttothedarkestcontour.Wehavehighlightenedwiththickblackcontoursthelevelscorrespondingtovalues0.9,0.5,and0.05, tocomparewithpointsobtainedbyMarconietal.(2004).ThesewereobtainedfromtheanalysisoftheevolutionoftheobservedAGNluminosityfunction (Uedaetal.2003),undertheassumptionofconstantEddingtonratioλ=1.Forsuchobservationalpoints,thesymbolssquares,diamondsandcircles,mark thecontourlevelscorrespondingtothegrowthof90%,50%,and5%ofthefinalBHmass,respectively. RightPanel:WeplotascontinuouscontoursthesamegrowthhistoriesfromtheanalysisbyMerloni(2009,seetext),assumingthataccretionatlowm˙ tobe radiativelyinefficient(RIAF).ThedashedcontoursshowthesamegrowthhistorybutassumingforallAGNsthesameradiativeefficiency(8%)independently ofλ. massandluminosityevolutionofAGNs,undertheassumptionof nosity functions. Thus, the discrepancy is likely to be due to the fixedEddingtonratioλ=1,andthatgalaxymergingdoesnotaffect assumption(commontoalltheapproachesbasedontheobserved thegrowthofSMBHs;althoughbothassumptionsnotholdinour luminosityfunctions)thattheBHgrowthisalwaysdominatedby model,itisinterestingtonotethattherepresentationofdownsizing accretion,andthattheBHgrowthduetomergingisnegligibleat naturallyarisingfromtheinteraction-drivenmodelresemblesthat anyz. extrapolatedfromobservations. Thisconclusionisnotunexpectedwhenoneconsiderstheba- sicfeaturesofthehierarchicalclusteringpicture.Indeed,inallap- The general observed trend of larger assembled fraction for proachedbasedonhierarchicalscenarios,themergingsectorofthe increasingly massive BH is quite well accounted for by our models predicts massive objects to form later; such a trend is in- interaction-driven model, although the model seems to show a bending in correspondence of the 50% contour level at z ≈ 1.5 vertedwhenoneconsiderhowandwhenbaryonsareconvertedinto for masses above 109 M which is not present in the data; this starsoraccretedintoSMBHs.Theoveralldownsizingeffectinhi- (cid:12) erarchical models thus results from the balance between the two might constitute a true model inadequacy, or simply result from effects above; the downsizing trend resulting at low-intermediate thesimpleassumptionsofconstantλadoptedinthederivationof massesinourmodelmeansthataccretionlargelydominatesover the data points by Marconi et al. (2004) from the analysis of the merginguptoverylargefinalBHmassesM ≥109M . local BH mass function and of the evolution of the AGN lumi- BH (cid:12) WhilethecomparisonperformedinFig.9isjustafirststep nosity functions. To clarify this issue we reproduce in fig. 9b the toward a detailed observational testing of the downsizing proper- results from an improved version (Merloni 2009, in prep.) of the tiesofcosmologicalmodelsforthegrowthofSMBHs,theglobal Merloni&Heinz(2008)analysis,basedonupdatedAGNluminos- pictureemergingfromtheinteraction-drivenmodelforthegrowth ityfunctions.Suchananalysisissimilarinmethodologytothatby Marconietal.(2004),butavoidstheassumptionλ=1byderiving ofSMBHsseemstoconsistentlyaccountforawidesetofindepen- dent observations concerning the growth of SMBHs; these range phenomenological,physicallymotivatedrelationsbetweentheac- fromthedistantQSOs,totheintermediate-zBLAGNs,totheex- cretionrate,theBHmassandtheAGNluminosity,andsolvingfor tremepopulationconstitutedbytheSMGsatz ≈2,totheaverage acontinuityequationfortheSMBHmassfunctionevolution.Be- growthhistoriesestimatedfromtheobservedluminosityfunctions, yondthedetailsconcerningtheexactvaluesoftheassembledmass extending over a wide range of redshifts 0 (cid:46) z (cid:46) 4 and of BH fractionderivedfromtheobservedluminosityfunctions(whichis stilllargelyaffectedbyuncertainties,asshownbythedifferences masses107(cid:46)MBH/M(cid:12)(cid:46)109. among the results obtained under different assumptions, see cap- tion to fig. 9), the comparison with the Merloni & Heinz (2008) approach shows i) that the BH growth inferred from the evolu- 6 CONCLUSIONS tion of the AGN luminosity functions generally yields a stronger downsizing effect at large masses compared with our model, and We have investigated the growth of SMBHs relative to the stel- ii)thatsuchaneffectisprobablynotduetotheassumptionsonλ larcontentofthehostgalaxyintheframeworkofaninteraction- adoptedtoderivetheBHgrowthhistoriesfromtheobservedlumi- driven model for the feeding of BHs during the merging history (cid:13)c RAS,MNRAS000,??–??