Mon.Not.R.Astron.Soc.000,000–000 (0000) Printed2February2008 (MNLATEXstylefilev2.2) High redshift quasars and the supermassive black hole mass budget: constraints on quasar formation models J. M. Bromley,1 R. S. Somerville2 and A. C. Fabian1 1 Institute of Astronomy, Universityof Cambridge, Madingley Rd., Cambridge CB3 0HA 2 Space Telescope Science Institute, 3700 San MartinDr., Baltimore, MD 21218 2February2008 4 0 ABSTRACT 0 We investigate the constraints on models of supermassive black hole (SMBH) and 2 quasar formation obtainable from two recent observational developments: the discov- n eryofluminousquasarsatz ∼6,andestimatesofthelocalmassdensityofSMBHs.If a ∼90percentofthismasswasaccretedatredshiftsz<∼3,assuggestedbytheobserved J quasarluminosityfunctions,thesejointconstraintsposeachallengeformodels,which 2 must account for the observed luminous quasar population at z ∼ 6 within a very 2 limited ‘mass budget’. We investigate a class of models based within the hierarchical structureformationscenario,inwhichmajormergersleadtoblackholeformationand 3 fuelling,andtheresultingquasarsshineattheirEddington-limitedrateuntiltheirfuel v isexhausted.Weshowthatthesimplestsuchmodel,inwhichaconstantfractionofthe 8 gas within the halo is accreted in each major merger, cannot satisfy both constraints 0 simultaneously. When this model is normalized to reproduce the number density of 0 luminous quasars at z ∼ 6, the mass budget is grossly exceeded due to an overabun- 1 1 danceoflowermassSMBHs.We explorearangeofmodificationsto thesimple model 3 designed to overcome this problem. We show that both constraints can be satisfied if 0 the gas accretion fraction scales as a function of the halo virial velocity. Similar scal- / ings have been proposed in order to reproduce the local M• −σ relation. Successful h modelscanalsobeconstructedbyrestrictingtheformationofseedblackholestored- p - shifts above zcrit ∼ 11.5 or to haloes above a velocity threshold vcrit ∼ 55 kms−1, or o assuming that only a fraction of major mergers result in formation of a seed SMBH. tr We also briefly discuss the issue of trying to assume a ‘universal M• −σ relation’ s withintheframeworkofsimplePress–Schechtermodels,andfurthershowthatafixed a universal relation between SMBH mass and host halo mass is unlikely. : v i Key words: accretion – galaxies: interactions – galaxies: active – quasars:general X r a 1 INTRODUCTION the numbers of observed quasars as well as in the range of luminosities and redshifts probed.The2dFQuasar redshift In recent times, a growing wealth of observations has re- survey(2QZ;Boyleetal.2000)andtheearlierLargeBright vealed a strong correlation between the mass of supermas- QuasarSurvey(LBQS;Hewett,Foltz& Chaffee1995) have sive black holes (SMBHs) and the luminosity of the stellar identified over 23000 quasars at redshifts 0.3<∼z<∼3.0. The bulges in their host galaxies, and an even stronger correla- Sloan Digital Sky Survey (SDSS; Gunn et al. 1998) has tion between SMBHmass andthebulgevelocitydispersion observed over 650 confirmed quasars in the redshift range (Magorrian et al. 1998; van der Marel 1999; Gebhardt et 3.0<∼z < 5.7 (Schneider et al. 2003) and a small num- al. 2000; Ferrarese & Merritt 2000; Tremaine et al. 2002). ber at z>∼5.7 which have been sufficient to constrain for Thishasinturnproducedincreasinglyaccurateestimatesof the first time the density of bright quasars at z ∼ 6 (Fan theoveralllocalSMBHmassdensity.IftheseSMBHsare,as et al. 2001a, 2003). These objects must host SMBHs with nowwidelybelieved,therelicsofearlierquasaractivity(e.g. masses on the order of several 109 M⊙ (Fan et al. 2001a, Rees 1984 and references therein) then these estimates can 2003; Willott, McLure & Jarvis 2003). Moreover, evidence place strong constraints on the nature and evolution of the has been found even in these highest redshift objects for quasarpopulationastheyeffectivelylimitthe‘massbudget’ molecular gas (Bertoldi et al. 2003a; Walter et al. 2003), available to account for quasar activity. dust (Bertoldi et al. 2003b), and metals (Freudling, Corbin & Korista 2003), underlining the fact that even at these The last few years have also seen dramatic progress in (cid:13)c 0000RAS 2 J. M. Bromley, R. S. Somerville and A. C. Fabian earlyepochsquasarsareassociated withgalaxies developed tailedmodellingofthestarformation,gascoolingandfeed- enough to haveexperienced significant star formation. backprocesseswhichalterandcompetefortheintergalactic In addition to the optical surveys already mentioned, gas thought to fuel quasars. In this Paper, we avoid these observations of quasars and other active galactic nuclei complications by focusing on constraints at very high red- (AGN) in other wavebands have also been steadily im- shift z ∼ 6, when cooling times are short and only a small proving. This includes substantial work done determining fraction of the baryons is locked up in stars, so these com- the soft X-ray quasar/AGN luminosity function (Miyaji, peting processes should be relatively unimportant. As well, Hasinger & Schmidt 2000, 2001) and also several imag- wedonotattempttomodelthecomplexinterconnectionof ing surveys of AGN selected in the hard X-ray band (e.g. galaxyandBH/quasarformationthroughfeedback.Instead, Mushotzky et al. 2000; Giacconi et al. 2001; Hasinger et we consider a suite of models with simple parametrized al. 2001; Alexander et al. 2003; Barger et al. 2003) where recipes for SMBH formation and fuelling, which point the samples are unbiased by any line-of-sight absorption which way for more sophisticated and physical modelling. As we may be present. Thus through multiwaveband observations discuss in the next section, attempts to combine data for progress is being made in piecing together the history of observed AGN in X-rayand optical wavebandsnow appear AGN and their SMBH. to be able to account for the observed present day SMBH Ithaslongbeenestablishedthatquasars,andindeedall mass density, and moreover seem to suggest that most of formsofAGN,aretheresultofgaseousmatteraccretingon it was assembled at fairly low redshifts z<∼3. This enables toaSMBHatthecentreofagalaxy(Zel’dovich &Novikov us to place rough limits on the ‘mass budget’ available at 1964; Salpeter 1964; Lynden-Bell 1969; Bardeen 1970) and higherredshifts.CombiningthiswiththeSDSSobservations hence must somehow be linked to galaxy formation. Since of the quasar numberdensity at z ∼6 allows us to identify that time a great number of models have been published successfulmodelswithinthebroadclasswehaveconsidered, exploring quasar formation and evolution within the con- and to rule out other model variants. text of the modern Cold Dark Matter (CDM) paradigm of structureformation (includinge.g.Efstathiou &Rees1988; Our results will in some sense reflect a ‘best case sce- Haehnelt & Rees 1993; Haehnelt, Natarajan & Rees 1998; nario’asinclusionofstarformationandfeedbackwouldpre- Cattaneo, Haehnelt & Rees 1999; Kauffmann & Haehnelt sumablyonlyacttomakefuellessplentifulforquasars,and 2000;Cavaliere&Vittorini2000;Monaco,Salucci&Danese thus strengthen our constraints. As a result however it is 2000; Nulsen&Fabian2000; Haiman &Loeb 2001; Wyithe difficulttoplacestrongconstraintsonthelowerredshiftbe- & Loeb 2002a, 2003; Hatziminaoglou et al. 2003; Granato haviour of the models as the characteristic upper peak and et al. 2003; for a brief overview of several of these mod- subsequentdeclineofthequasarpopulationnumberdensity els see Haehnelt 2003). However, none of the above works atlowredshiftsislikelytobestronglyrelatedtofueldeple- have concentrated on successfully reproducing the highest tion due to star formation (see e.g. Kauffmann & Haehnelt redshiftpopulationspecificallywithintheconstraintsofthe 2000; Di Matteo et al. 2003). Nevertheless we can demand limited‘massbudget’impliedbytheestimatedlocalSMBH that the bright quasar population we track in our models mass density. never decreases in number below that of the observed pop- Any potential model of quasar formation finds itself ulation at lower redshifts. In practice this constitutes a rel- faced with three major unknowns. Firstly: where, how and atively weak constraint. with what mass do the initial ‘seed’ BHs form? Secondly: what events trigger their subsequent fuelling and growth, how much fuel do they supply and how efficiently is it con- InSection2wedetailourargumentslimitingtheSMBH verted into radiative energy? Thirdly: how does feedback mass density at high redshift and present the criteria we (from star formation or the AGN activity itself) regulate shallusetojudgewhichofourquasarmodelsaresuccessful. and possibly even check their growth? These uncertainties In Section 3 we present the basis for our family of quasar translate inevitably to a relatively large number of free pa- models and the methods of our implementation. We then rameters in models. In this Paper we aim to reduce this present in Section 4.1 a realization of the simplest member numberbyconcentrating on thehigh redshift quasarpopu- of this family of models. We show why such a model is not lation,allowingustoconcentrateonjustthefirsttwoissues: workableanduseittodemonstratethechallengesasuccess- formationandfuelling.Inordertoinvestigateveryluminous fulmodelmustfacebeforegoingontoexploreothermodels quasars such as those discovered at z ∼6, a very large vol- withinthefamilyinSection4.2.InSection5webrieflycom- ume must be simulated so as to capture these extremely pare our work to other recent results in the literature, and rare objects. For this reason, we rely, as in the majority of discusstheimplications ofimposingafixedscalingbetween theworkcitedabove,ontheanalyticPress–Schechter(Press black hole mass and halo virial mass or velocity at all red- & Schechter 1974) and extended Press–Schechter (Lacey & shifts. We summarize and conclude in Section 6. Through- Cole 1993) formalism to predict the merger history of dark out this paper we assume a ‘standard’ ΛCDM cosmology matter haloes. We consider a class of models based on the with matter density Ω0 = 0.3, vacuum density ΩΛ = 0.7, basic assumption that seed black holes are created in ma- Hubble parameter H0 = 70 kms−1Mpc−1, and normaliza- jor mergers, and that theseevents trigger an episode of gas tion σ8 = 1, although our conclusions are not significantly accretion in galaxies with apre-existingSMBH.Wefurther alteredforalowervalueofσ8=0.9(thepredictedhighred- assumethatSMBHsaccreteandradiateattheirEddington- shift SMBH mass densities would typically change by 5–10 limited rate untiltheir fuel supply is exhausted, and that a per cent for most of the models considered here, and in no fixedfractionoftheaccretedmassisconvertedtoradiation. cases would change so much as to alter our conclusions as To rigorously explore this problem would require de- to the‘success’ of a model). (cid:13)c 0000RAS,MNRAS000,000–000 High redshift quasars 3 2 OBSERVATIONAL CONSTRAINTS to a B-band magnitude limit of M ≤ −27.2. We then in- B tegrate the best-fitting luminosity function2 to obtain the Weshall make useof two primary constraints in evaluating numberdensityofquasarsabovethismagnitudelimit,which the models considered in this Paper: the number density at a redshift of z = 2 is equal to 2.86 × 10−8 Mpc−3 of luminous quasars at z ∼ 6 and an upper limit on the (where we have corrected for the lower Hubble parameter total mass density of SMBHs at z ∼6 (the ‘mass budget’). of H0 = 50 kms−1Mpc−1 used by Boyle et al. (2000) in We will also consider a lower limit on the quasar number theirΛCDM fit). density at lower redshift 2<z<6 as a weaker constraint. ∼ ∼ Boththelowandhighredshiftquasarnumberdensities In this section, we walk through the arguments we used to are based on results from optical surveys, and it is worth obtainthesequantitiesfromresultsintheliterature.Wealso noting that as a result there is always a danger that ob- describe several necessary conversions in detail. scured sources have been omitted from our tally. However The observed quasars at redshifts of z ∼ 6 lie at the thereismountingevidencethattheverybrightestmembers very limits of current detection and as such represent only ofthequasarpopulation,whichweareconcernedwithhere, theverybrightesttailofthetotalhighredshiftquasarpop- are preferentially unobscured (see e.g. Fabian 2003 and ref- ulation.Thusourmodelstoowillbeconcernedmostlywith erences therein). A further potential source of error is that these brightest, rarest objects. Our best information about thecorrectionsandtemplateswehaveusedfromElvisetal. the highest redshift quasar population comes from a sam- (1994)arebasedonasampleoflowredshift(z<2)quasars, ple of z>5.8 quasars identified by Fan et al. (2001a, 2003) ∼ andlittleisknownabouthowwellthesetemplatesapplyto basedoni-z colourselection fromSloanDigital SkySurvey high redshift objects, although an investigation byKuhnet (SDSS)imaging.Thesampleconsistsof6quasarsinthered- al. (2001) suggests that at least the rest frame optical and shift range 5.7<z <6.6 and is complete down to an abso- UV energy distributions change very little for quasars out lutemagnitudeofM1450 <−27.1(ABsystem;Oke&Gunn to redshifts as far as z∼3−4. 1983;Fukugitaetal.1996).Ithasameanredshiftofz¯=6.08 The final quantity we wish to determine is an esti- qanudasparrsedaitcttshtisherendusmhibftertodebnesi5ty.2o±f b2r.1ig×ht1(0M−11045M0 <pc−−327(.a1l)l mate of, or at least an upper limit on, the total mass density in SMBHs at z ∼ 6. The local SMBH mass den- thesefigurestakeintoaccountadjustmentsforourassumed sity may be estimated by two independent methods. One cosmology).Tocomparethiswiththeresultsfromourmod- method makes use of the M• − σ relation plus the ob- elsweneedtorelatethemagnitudelimit ofthesampletoa served galaxy luminosity/velocity function to directly sum minimum bolometric luminosity for thequasars. To dothis up the mass present in remnant SMBHs at the present westartwithagivenbolometricluminosityandconvertthis day (Aller & Richstone 2002; Yu & Tremaine 2002). These toarest frame 2500˚Amonochromatic luminosity usingthe estimates now seem to be converging on a value around bolometriccorrectionsofElvisetal.(1994).Usingthemean ρ•(z = 0) = 3−5×105 M⊙Mpc−3. Specifically, Yu & stpoeactluramlitneomsiptylaatettohfetrheestsafrmame aeuwtahvoerlsenwgeththoefn14co5n0v˚Aerwthtihcihs Tremaine(2002)findρ•(z=0)=2.9±0.5×105 M⊙Mpc−3 while an earlier calculation of Merritt & Ferrarese (2001a) aMll1o4w50s(uAsBtosycsrteeamte)afosrimouurlamteoddemlso1n.oWcherfiomndatic magnitude findsInρ•t(hze=ot0h)er∼ap4.p6ro×ac1h0,5fiMrs⊙taMttpecm−p3t.edbySol tan(1982), LBol =(1.0±0.075)×1036.66−0.4M1450,AB ergs−1 (1) the cumulative mass that must have been accreted by SMBHs in order to produce the observed quasar luminos- andhencethattheSDSSsamplemagnitudelimitofM1450 < ity function is summed up. In the past, the best available −27.1 corresponds to a minimum quasar luminosity of LBol =3.03×1047 ergs−1 in our models. desitffiemreantetsmoeftthhoedSsMhaBvHenmoatssaldweanyssitbyeρe•nfirnomgotohdesaegrteweomveenrty. To determine the number density of quasars as bright Improved estimates of the optical quasar luminosity func- as the SDSS z ∼ 6 sample at lower redshift, we use the tion (e.g. Boyle et al 2000; Fan et al. 2001b) are now avail- luminosity functions determined from the 2QZ and LBQS able,andextrapolating tohigherredshiftsandfaintermag- by Boyle et al. (2000). These results are quoted in absolute nitudeshasproducedamuchmorecompleteestimateofthe B-band Vega magnitudes so we first make use of the Elvis total contribution from accretion by unobscured AGN (e.g. et al. (1994) bolometric corrections to find, Chokshi & Turner 1992). The most recent calculations (Yu LBol =(1.0±0.06)×1036.59−0.4MB,VEGA ergs−1. (2) &preTsernemt daianyeis20a0r2o)unfidndρatchc(azt=th0e)t=ota1l.8c9on×tr1i0b5utMio⊙nMbypct−h3e where we have assumed an offset of 0.12 mag between the (assumingaccretedmatterisconvertedtoradiationwithan Vega and AB magnitude systems for quasar-like spectra efficiency of around 10 per cent) and indicate that around (Schmidt,Schneider&Gunn1995;thisoffsetismuchlarger 90percentofthistotalisduetoaccretionatredshiftsz<3. ∼ than that typically used for magnitude conversions of stel- However account must also be taken of the obscured lar sources because of the large difference between stellar population of quasars whose demographics we do not yet and quasar spectra). This implies that our minimum lumi- understandso well. Wedo knowtherearelarge numbersof nosity of interest, LBol = 3.03×1047 ergs−1, corresponds obscuredweakAGNsuchasSeyfertIIs,whichwillalsocon- 1 In fact the SDSS sample absolute magnitudes are originally calculated for the rest frame 1280˚A and then the authors make 2 Notethat the αand β parameters for the luminosityfunction theirowncorrectionstoarriveatthe1450˚Avalues,howeverthe Φ(MB,z) inBoyle et al. (2000) are both missingminus signs in effectsofthiscorrectiononourowncalculationswillbenegligible theforminwhichthearticlewasoriginallypublished(B.Boyle, comparedtoourinherentuncertainties. privatecommunication). (cid:13)c 0000RAS,MNRAS000,000–000 4 J. M. Bromley, R. S. Somerville and A. C. Fabian tributeto thetruevalueof ρacc (e.g. Alexanderet al. 2001; if theysatisfy theseconstraints subject to themore relaxed Brandt et al. 2001; Rosati et al. 2002), and indeed are re- condition105 M⊙Mpc−3 ≥ρ•(z∼6)>5×104 M⊙Mpc−3. quiredtoexplaintheX-raybackground(Fabian&Iwasawa 1999).Inadditiontheremayalsobeasignificantpopulation of faint highly reddened quasars (Barkhouse & Hall 2001; 3 MODEL FRAMEWORK Wilkes et al. 2002; Richards et al. 2003a), and there is also some evidence for a small population of totally obscured, In the hierarchical picture of structure formation, based Type II quasars (e.g. Crawford et al. 2002; Norman et al. on the CDM model, structures build up progressively over 2002;Wilmanetal.2003;Zakamskaetal.2003)althoughthe time, the smallest objects collapsing first and then gradu- contributionofthelattertotheSMBHmassdensityisprob- allymergingtogethertocreatelargerandlargerstructures. ably negligible. Attempts to include the effects of obscured Themuchlessabundantbaryonicmatterinitiallytracesthe sourcesincalculationsofρacc usingX-rayobservations(e.g. darkmatter.Butwhereasthedarkmatterisdissipationless Fabian & Iwasawa 1999; Elvis, Risaliti & Zamorani 2002) thebaryonsareable,throughshock-heatingandsubsequent or by combining multiwaveband observations (e.g. Salucci cooling, tocollapse down toform much moretightly bound et al. 1999; Barger et al. 2001) have in the past resulted in structures – the first (proto)galaxies – within the potential valuesofρacc(z=0) that can actually exceed theestimates wellofthecollapseddarkmatter‘haloes’.Thesegalaxiesare of the total local mass density ρ•(z=0) unless implausibly thenbroughttogetherbythesubsequenthierarchical merg- high radiation efficiencies are assumed. ers of their dark matter haloes where they can form early However,thisoverpredictionmayinparthavebeendue groups and clusters or perhaps merge together themselves. toincorrectlyassumingthatobscuredandunobscuredAGN Webaseourmodelonthepremisethatquasarfuellingis populations peak at the same redshift (e.g. Barger et al. accomplishedthroughtidalstrippingofangularmomentum 2001; Cowie et al. 2003), and a picture is now emerging from galactic gas during such major galaxy mergers. This inwhichtheaccretionfromthecombinedAGNpopulations popularpremisehassubstantial observational andtheoreti- wouldseemtoexactlyaccountforthedirectestimatesofthe calsupport(Negroponte&White1983;Barnes&Hernquist mass in SMBH remnants at the present day (Fabian 2003). 1991,1996;Bekki&Noguchi1994;Mihos&Hernquist1994, The high redshift evolution of the obscured AGN popula- 1996;Bahcall etal.1997; McLureetal.1999), althoughthe tion is much less well known than that of the unobscured mechanismresponsibleforthefinalinflowofthegasonceit sources(seee.g.Hasinger2002).Howeverbecausetheperiod has arrived within the last 100 pc from the galactic centre z = 0−3 represents around 84 per cent of the time since is still unknown (for an alternative to the merger hypoth- thebigbang,wemightneverthelessexpectthatthebulkof esis see e.g. Kawakatu & Umemura 2002; Granato et al. themass densityfrom obscured sources would bedeposited 2003). In order to track the mergers of galaxies within our intheepochz<∼3,asistruefortheunobscuredsources.In- modelweconstructMonteCarlorealizationsofdarkmatter deed the (already fairly high) estimate of ρacc(z =0) made ‘merger trees’ following the algorithm of Somerville & Ko- byBargeretal.(2001) frommultiwavebandobservationsof latt(1999).Thisalgorithmallowsustoreconstructatypical Chandra sources only considers the contribution made by evolutionary history for a given final dark matter halo (the accretion for z ≤3, so if this it to be consistent with other ‘root’ halo) and follow the hierarchical build up of its mass measurementstherecannotbeverymuchcontributionfrom overtimethroughaccretionandmergerofsmallerhaloesas higher redshifts. described by the extended Press–Schechter probability dis- It seems then that almost all of the observed local tribution(Lacey&Cole1993).Foraroothaloofgivenmass SMBH mass density is accounted for by radiative quasar andredshift thisprocess isaveraged overalarge numberof accretion, leaving very little room for any quiescent ac- different realizations, allowing us to calculate the average cretion modes. Moreover the vast majority of this mass contribution (and its time dependence) made by the pro- must be deposited at redshifts z<∼3, meaning that only a genitor haloes to all quantities of interest. The simulations very small fraction of the present day SMBH mass den- are repeated over a grid of different root halo masses and sity ρ•(z = 0) can be in place at high redshifts, i.e. the results are combined by weighting them according to ρ•(z>∼6) ≪ ρ•(z = 0). If we take the upper bound on theprobability of findinga halo of that mass at the chosen the current estimates on the local SMBH mass density, i.e. outputredshift. ρ•(z = 0) ∼ 5×105 M⊙Mpc−3, and then as a first ap- We track the history of each halo back in time until proximation assume that around 90 per cent of this comes we find a progenitor halo with a circular velocity below fromaccretioneventsbelowaredshiftofz∼6(certainlywe 30 kms−1 or a mass below 1.7×108 M⊙; the latter mass knowfortheunobscuredsourcesthiswouldbetrueevenfor corresponds to acircular velocity of 30kms−1 at a redshift z<3) thenwearriveatan upperlimit onthehighredshift ofz ∼20,soweeffectivelytrackallstructureswithcircular SMBH mass density of ρ•(z∼6)∼5×104 M⊙Mpc−3. velocities as small as 30kms−1 back to redshifts of z ∼20 Thus in this paper we shall consider a model ‘success- and only progressively larger structures beyond. Structures ful’ if it both reproduces the SDSS z ∼ 6 population of smallerthan30kms−1 areunlikelytobeabletoaccretegas bright quasars with a SMBH mass density ρ•(z ∼ 6) ≤ efficientlyinthepresenceofaphotoionizingbackground(see 5×104 M⊙Mpc−3 and continues to grow this population e.g.Gnedin2000),andsincetherecentresultsfromWMAP of brightquasars sothat theydonotdrop below thevalues indicatethat reionization islikely tohaveoccurred as early predicted bytheBoyle et al.(2000) luminosity functionsat as z = 14−20 (Kogut et al. 2003) our chosen resolution z =2. Because of the uncertainties involved and the some- should be valid out to similar redshifts. whatadhoc approachwehavetakentoreachourmassden- Oncewehavecreatedamergertree,weassociate bary- sity limit we shall consider models ‘marginally successful’ onic material with each newly formed dark matter halo. (cid:13)c 0000RAS,MNRAS000,000–000 High redshift quasars 5 This is assumed to fall towards the centre of the resulting diation pressure exactly balances the gravitational pull on potential well and reside there as a self-supported gaseous infalling matter, (proto)galaxy. The initial amount of baryonic matter avail- 4πGcm able to a pristine dark matter halo is set by the universal L=1.15 pM• (4) σ e baryon fraction (here we use f = 0.13, corresponding to Ω =0.019h−2,which isconsistebnt with theobservations of wheremp isthemassoftheproton,σetheThompsoncross- b sectionoftheelectron,M•thecurrentmassoftheblackhole Levshakov et al. 2002). This ratio remains fairly constant and the factor of 1.15 accounts for themean atomic weight throughout the history of a halo since the only mechanism perelectron foratypicalhydrogenandheliumgasmixture. for baryon loss in our models is the conversion of accreted Setting these two expressions equal yields an ordinary dif- matter to quasar radiation, which is fairly negligible. We ferential equation with thesolution defertheconsideration ofprocesses thatmay competewith quasars for the gas in these galaxies (e.g. star formation, M•(t)=M•(0)et/κ (5) feedback and consequent heating) to a subsequent paper whereM•(0)istheinitialblackholemassandthetime-scale (Bromley, Somerville & Fabian 2004, in preparation) so as κ is given by to minimize the number of parameters in our model – al- though we note that this means our results will only apply κ=(1.15)−1 cσe ε . (6) 4πGm 1−ε to high redshifts where gas is plentiful. p Whether such galaxies subsequently merge when their We can derive a similar expression for the amount respectivedarkmatterhaloesmergewilldependontherate of unaccreted gas remaining mfuel(t), and by setting at which they can lose their orbital energy via dynamical mfuel(tExh) = 0 find the amount of time tExh for which a friction against the background dark matter. When track- quasarshinesbeforeallofthegasavailabletoitisexhausted: ing halo-halo mergers within the merging tree we label the 1 central galaxy of the most massive dark matter halo as the mfuel(t)=mfuel(0)− 1−ε[M•(t)−M•(0)] (7) new central galaxy of the system and all the other galax- ies become ‘satellites’. We compute the dynamical friction tExh=κln 1+(1−ε)mfuel(0) . (8) time-scaleforthese‘satellite’ galaxies usingtheapproxima- (cid:20) M•(0) (cid:21) tion of Binney & Tremaine (1987), modified to account for For computational simplicity, we only record activity from non-circular orbits (Lacey & Cole 1993); see Somerville & accretion on to central galaxies. Thus if a quasar is ac- Primack (1999) fordetails. Ifthistime-scale isshorterthan creted into a new halo, becoming a satellite galaxy, it will thetime tothenext‘branch’of thetree(i.e.thenexthalo- be‘turnedoff’afterasmaller time-scalethantheonegiven halo merger), then the satellite is merged with the central above. We find that this approximation has a negligible ef- galaxy. Otherwise the satellite remains until the next halo- fect on ourresults. halo merger, when its ‘merger clock’ is reset and the calcu- The exact value of the parameter ε in all these equa- lation is repeated. tions will depend on the nature of the accretion disc and Whenever two galaxies merge, any central black holes thespin of theblack hole. Taking intoaccount photon cap- they contain are also assumed to merge instantly. In addi- turebytheblackholeitself,theorysuggestsεcannotexceed tion, if a ‘major’ galaxy merger occurs (defined here as a 0.057 for a static black hole and 0.4 for a maximally rotat- mergerinwhichthesatellite galaxyaccountsforatleast 30 ing black hole, although in fact accretion will act to bring per cent of the total system mass), then some fraction fm the black hole spin to its ‘canonical’ value which reduces of the gas is stripped of its angular momentum and falls to this maximum to ε < 0.3 (see Thorne 1974; in principle thecentre,whereit can fuela centralSMBH. Thequantity theeffective efficiency could exceed this limit since the disc fm is left as a free parameter of themodel. luminosity could be augmented by magnetohydrodynamic Once accretion on to a particular black hole has been processes such as magnetic coupling with the SMBH, mag- ‘turnedon’bysuchagalaxy-galaxymerger,weassumethat netic stress in the disc or torque on its inner boundary e.g. the accretion proceeds at the Eddington limit until the fu- Gammie1999;Krolik2001;Li2002;Wang,Lei&Ma2003). ellinggassuppliedbythemergerisexhausted.Theluminos- MostSMBHmodelsassumetheefficiencyofatypicalaccre- ity of an accreting black hole L is governed by the rate at tiondisctobewellbelowthismaximum,ataroundε≈0.1, which accreting matter is supplied m˙fuel, and the efficiency and weshall follow this convention,although in Section 3.2 with which this accreted matter is converted to radiation – wedodiscusshowthebehaviourofourmodelisaffectedby typically quoted in terms of the fraction ε of its total rest takingdifferent values. mass energy that is liberated. Thus L = εm˙fuelc2, where c Clearly if one quasar differs from another only in that is the speed of light. That fraction of the accreted matter itshinesbrightlyenoughtobevisiblefortwiceaslongthen which is not converted to radiation adds tothe mass of the it is twice as likely to be observed. Thus since we expect black hole, and so the black hole growth rate M˙• is just the bright lifetime of a typical quasar to be much smaller M˙• =(1−ε)m˙fuel.Combiningthesetwoexpressionswefind thanourmodel’sintrinsictimesteps(thetimesbetweenhalo the quasar luminosity is related to the black hole growth and galaxy mergers) we group our results intoredshift bins rate by of width dz = 0.5 and calculate the fraction of the time spanned by the bin that each quasar spends shining above ε L= M˙•c2. (3) our magnitude limit. When calculating number densities 1−ε from our results we then weight each quasar by this fac- However, for Eddington limited accretion the black hole tor, in addition to the probabilistic weight of its root halo must be shining at the Eddington luminosity where the ra- discussed earlier. (cid:13)c 0000RAS,MNRAS000,000–000 6 J. M. Bromley, R. S. Somerville and A. C. Fabian The question remains as to how the initial ‘seed’ black 3.2 The radiative efficiency of quasars holes are formed. Several previous models have consid- The efficiency of converting accreted material to radiation, ered the possibility that SMBHs form from much smaller (∼ 10−100 M⊙) black holes, remnants from either stan- given by ε, is the controlling parameter in the equation for Eddington limited black hole accretion. A high value of ε dard galactic star formation (e.g. Haiman & Loeb 2001) means that most of the accreting matter is turned to radi- or from the very first stars – the Population III stars (e.g. ation, and so only a small mass accretion rate is possible Volonteri,Haardt&Madau2003).Howevertherearevarious before the limiting Eddington luminosity is achieved. Con- problems associated with such scenarios (see e.g. Haehnelt versely,alowervalueofεmeansmoremassmustbeaccreted 2003). In particular some mechanism is required to facili- to produce the same amount of energy in radiation and so tate their migration to the centre of their host galaxy. We themassgrowthrateforablackholeshiningattheEdding- instead focus on the possibility that SMBHs form directly ton luminosity is higher. from the collapse of a large gaseous cloud in the centre of a (proto)galaxy. This was first discussed many years ago The probability of observing an accreting black hole (Rees 1984; Haehnelt & Rees 1993; Silk & Rees 1998); and depends upon the brightness of the accretion event and its although it has obvious problems with the need to avoid duration.IfblackholesaccreteattheEddingtonluminosity, fragmentation during the collapse, it nevertheless remains then their brightness is a function only of their mass. Thus a possibility and a candidate mechanism has been sketched the number of accreting black holes above some luminosity out byHaehnelt & Rees(1993). Weshall furtherassume in depends on the number of black holes above a particular our model that this collapse is also connected to the major mass limit – which will depend on the rate at which black merger of galaxies – that is, the same tidally stripped gas holes have been able to grow prior to the epoch of obser- which falls into the centre of a post-merger galaxy and can vation – while the duration of a given accretion event will fuel quasar accretion will also be viewed as the source of depend inversely on the black hole accretion rate if fuel is the initial SMBH’s formation. Precisely how effective (how limited. This means the overall effect of the value of ε on likely it is to happen in a given major merger) this is and observations can becomplicated. how much of the gas this initial formation consumes (how In our models we find that an increased ε tends to de- efficient it is) will be left as free parameters of the model. creasethenumberdensityofquasars(aboveagivenbright- We shall however assume that the process is quiescent – nesslimit)athighredshifts,butincreasethenumberdensity thatis, that noradiation is producedin thisinitial collapse atlowerredshifts.Thusεaffectsthesteepnessoftherisein – although at high redshifts this is unlikely to make much quasar numberswith time. differencetoourpredictionsasallbutthebrightestofthese If fuel is plentiful then a lower efficiency model should ‘formation flashes’ would betoofaint tobeobservableany- grow faster. However one would expect that once the rate way. atwhichmajormergerscanresupplyaSMBHwithfuelbe- comesthelimitingfactorthenthenumberofbrightquasars in a low and a high efficiency model would become equal. In the slower accreting (high efficiency) model they would shineforlonger howeverand sohaveahigherchanceof be- 3.1 Model summary ing observed. This indeed would seem to be the reason for the increased numbers of bright quasars at low redshift in A brief summary of the steps involved in our modelling is ourhigher efficiency models. as follows: Howeverat high redshifts thesituation becomes rather • Create hierarchical merger trees describing the forma- complex,andtherearethreemain factors whichneedtobe tion history of dark matter haloes. take into account. Firstly the equalizing between high and • Associate baryons with each newly formed halo, form- low efficiency models discussed above may be less effective, ing (proto)galaxies. since even though at redshifts z ∼ 6 the vast majority of • Ateachlevelinthehierarchy,representingthemergerof quasar growth is still limited by therate of major mergers, progenitor dark matter haloes, compute whether and when thetimerequiredtoaccretethefuelsuppliedjustbyasingle each of the constituent galaxies will merge (based on dy- major merger can in fact represent a very large fraction of namical friction). thetotalobservedredshiftbinatsuchredshifts.Secondlyit • IneverygalaxymergeranyexistingSMBHsaremerged. isthecasethatinalowefficiencymodelthetotalamountof • Ineverymajorgalaxymergersomefractionoftheavail- matteraccretedbyaSMBH(andhenceitsmaximumlumi- able baryons is assumed to be stripped of its angular mo- nosity)willbeslightlylargerthaninahighefficiencymodel mentum and fall to the centre, where it can either form a sincelessofthemass-energyofthefuelisconvertedtoradi- new ‘seed’ SMBH or fuel an already existing SMBH. ation. While this is a small effect, it nevertheless can make • OnceprovidedwithfuelaSMBHshinesattheEdding- a difference when looking (as we do) at the very brightest tonluminosityandaccretesatthecorrespondingEddington endofthequasarpopulation.Lastlyanadditionalcomplica- rate untilthis fuel is exhausted. tion arises if an accreting SMBH becomes a satellite before exhausting its fuel and then merges with the SMBH of a The mass ratio that constitutes a major merger is fixed at new central galaxy. In this case the mass contribution will 0.3 and the radiation efficiency of accretion is fixed at 0.1. belarger in thefaster growing (low efficiency) models since The efficiency and effectiveness of SMBH formation, and the satellite SMBH will have reached a larger mass. These the form of the accreted fraction of baryons are left as free factors would seem to combine to create the drop in high parameters. redshift quasars we see in our higherefficiency models. (cid:13)c 0000RAS,MNRAS000,000–000 High redshift quasars 7 Indeed the last of these complications would seem to boundary can also be made successful (by a suitable choice representapossiblebarriertoaccuratelyfollowinghigheffi- of the free parameters) for values in the range 20 – 40 per ciency models within our current framework. Although our cent(althoughthisisoftenattheexpenseofsteepeningthe standardchoiceofε=0.1ishardlyaffectedbythesimplified riseinquasarnumberswithtimewhichcouldhaveimplica- treatment in our models whereby only accretion in central tionsforthelowredshiftquasarpopulationnotinvestigated galaxies is followed, it would appear that once efficiencies here)3. as high as ε = 0.2 are considered SMBH growth becomes Concerning the assumptions made in our model, there so slow that significant numbers of bright quasars become arethreemainpointswhichneedtobeaddressed.Firstlywe satellitesbeforeexhaustingtheirfuelandsothissimplifying note that our assumption that quasars always shine at the assumption begins to break down. Eddingtonluminosity rightuptothepointwheretheirfuel Thussincea higherefficiency modelboth increases the isexhaustedissomewhatunphysical.Realisticallyaquasar’s low redshift quasar numbers, where our model is already lightcurveislikelytorisesteeplyasitisfirstactivated,ap- poorlyconstrained,andalsomayneedamoredetailedtreat- proach the Eddington luminosity and then decay away as mentofsatellitegalaxies,wewilldelayfurtherinvestigation fuel becomes scarce; however the exact form such a curve of such models to asubsequent paper(Bromley et al. 2004, would takeisnot knownandwill affect theoverall shapeof in preparation). thequasarluminosityfunction(seee.g.discussioninCatta- neo 2001). Nevertheless the effects of a decaying tail in the quasarlight curvearelikely to bemost important in repro- 3.3 Caveats and assumptions ducingthefaintendoftheluminosityfunction,andsincewe Inthis section we brieflydiscuss some caveats in ourmodel areconcentratinginthispaperonthebrightquasarsvisible and how we have tried to deal with them, and also outline at high redshift, the simple approximation we use is prob- thepossibleconsequencesoftheassumptionswehavemade. ably sufficient. Were we to include such an effect then we Webeginbyexaminingsomedetailsinthealgorithmsofour suspectitwouldrequireifanythingslightly morenumerous simulation. or more massive quasars to still account for the observed It is usual in Monte Carlo based merger tree models high redshift quasar numbers, making the constraints from to weight each root halo by the standard Press–Schechter theSMBHmassdensitypresentedinthispapereventighter. distribution (Press & Schechter 1974) which provides the Secondly we need to consider the effects of our forma- expected number density of haloes of a given mass at a tion and accretion scenarios. We have assumed that forma- given epoch. However this distribution has several known tion of an initial seed SMBH requires a major merger, as discrepancies when compared to N-body simulations (see doessubsequentaccretion.Aconsequenceofthisisthatun- e.g. Somerville et al. 2000). In an attempt to correct some- lessahaloacquiresitsSMBHfrom aminormerger,itmust whatforthese,weusethecorrectedmassfunctionofSheth experience at least two major mergers in order to activate & Tormen (1999) to perform this weighting in our models a quasar. Thus at high redshifts, when few SMBHs have (other‘improved’mass functions, e.g. Jenkinset al. (2001), beenformed,quasaractivitywill alreadybebiasedtowards would yield similar results). Unfortunately,although asim- haloes that are in ‘merger-rich’ environments. This is in ilar correction can be applied to the conditional mass func- contrast both to previous models which have seeded haloes tion used to generate the merger histories, the resulting with black holes completely independently of mergers, and probability functions violate the Markov condition and are tothosewhichhavenotdifferentiatedbetweenaccretionand nolongersuitableforgeneratingself-consistentmergertrees. formation but treated them as the same process (examples Therefore,tominimizetheeffectsofsuchinaccuraciesinthe ofbothtypesofmodelarediscussedinSection5).Whenwe rest of the merger tree, we resimulate each tree over a grid examinemodelvariantswithareducedSMBHformationef- of output redshifts, placing a root halo at each redshift bin ficiency(i.e.wherenotallthegasfreedinthemajormerger wewish torecord dataover(asusualaveraging each overa is usedin creating theinitial SMBH),we havecontinuedin grid of halo masses). the same spirit by assuming that the gas not consumed by Similarly, on the topic of the merger tree algorithms, the initial formation is expelled from the galaxy by some it should be noted that due to the way our algorithm is outflowprocess(ofcourseitmaybethatthefraction ofgas implemented (see Somerville & Kolatt 1999 for details) it able to reach the galactic centre is slightly less anyway in is sometimes thecase that thevery earliest progenitor on a thecasewherenopriorSMBHexists).Forreasonably large given‘branch’ofthemergertreemayhaveacircularvelocity efficienciesthisdoesnotseemtoounlikely;mostofthemod- belowourresolutionlimit.Sinceourchoiceofresolutionwas els we shall show to be successful have initial SMBHs with based on thenotion that smaller haloes would beunableto meanmassesintherange∼5×103−5×105 M⊙ (atz=6) acquirebaryonsinaphotoionizingbackground,westripthe whichwouldindeedreleaseasubstantialamountofgravita- progenitor of all its baryons in such cases so as to remain tionalbindingenergyuponformationandcouldconceivably consistent with this premise. expeltheremaininggasbackoutofthecentralinnerparsec Finally wenotethatsincesmallmassratiomergersare oftheirgalaxy.Howeverastheformationefficiencybecomes much more common, the results of the model can be fairly sensitive to the exact definition of a ‘major’ merger. In our 3 Itshouldbenotedthatwhileitappearsusualinquasarmodels model we have chosen to set the boundary of the minor- toconsider thefractionofthetotal systemmassasthebasisfor major divide as the point where a merging satellite galaxy the minor-majordivide (as we do here), many galaxy formation accounts for 30 per cent of the total mass in the system. models (which use the divide in determining starbursts) instead However we have found that all of the models we inves- usetheratioofbaryonicmasses toclassifymergersas major vs. tigate in this paper which prove successful for our chosen minor;e.g.Somerville&Primack(1999); Coleetal.(2000). (cid:13)c 0000RAS,MNRAS000,000–000 8 J. M. Bromley, R. S. Somerville and A. C. Fabian smaller this rapidly becomes less and less likely, so clearly thenthiswouldweakenourconstraintsonthehighredshift we should treat as suspect a model which relies on partic- ‘mass budget’. ularly small efficiencies (which it would be hard to justify physicallyanyway).AttheendofSection4.3webrieflydis- cuss the effects on the model if we remove this assumption andinsteadallowgasunusedinSMBHformationtoaccrete 4 MODEL RESULTS on to the newly formed SMBH in the next timestep. How- We now examine several different variants of the models ever, since our timesteps are the intervals between galaxy withintheframeworkjustdescribed.Webeginwiththesim- and halo mergers they are not evenly spaced, this is only a plest case, which turnsout tobe unableto satisfy our joint very crude treatment and we do not spend much time on constraints, and then add additional parameters in an at- it. A better approach would have been to allow accretion tempt to findone or more successful variants. of leftover gas directly after the formation; but this would require a sufficient understanding of the formation process to be able to estimate the time-scales involved and with- 4.1 The simplest case outtheinclusion ofstarformation andotherprocesses that wouldcompeteforthisgasthetreatmentislikelytoremain Torealizeaparticularmodelweneedtospecifythefraction unrealistic anyway. of gas accreted on to central SMBHs in major mergers, f m Lastly we examine the assumptions we have made to (which may beafunction of other variables), and theeffec- arrive at theconstraints used to judge our models. We first tiveness and efficiency of the formation of the initial seed consider the number density of quasars at z ∼6 calculated SMBH. Perhaps the simplest scenario is to take f to be m fromtheSDSSdata.Whilewehaveassumedthatthevalue equaltoafixedvalueandassumeacompletelyeffectiveand found by Fan et al. (2003) represents the true density of efficient SMBH formation process – that is, in every major the population, our models are not in fact very sensitive to mergerin which theremnant doesnot contain aSMBH,all changes in this value by a factor of a few. Although none the tidally stripped gas falling to the centre is assumed to of the six quasars in the SDSS z ∼6 sample show multiple formanewSMBH.Thesinglefreeparameter,f ,maythen m images(Fanetal.2003;seealsoRichardsetal.2003b),nev- be fixed by requiring the model to reproduce the observed erthelessthediscoveryofgalaxiesneartothelineofsightof numberdensityof bright quasars at redshift z ∼6. Wefind twoofthesequasarssuggeststhattwoofthesamplemayin that a value f =0.0095 achieves this. m factbemagnifiedbylensing(Shioyaetal.2002;Whiteetal. We show several results from this model in Figs. 1a–d. 2003; Wyithe2003). If the intrinsic luminosity of these two In the top panels we plot the number density of luminous quasarsbeforemagnificationisinfactbelowtheSDSSmag- (M1450 < −27.1) quasars and of SMBHs as a function of nitude cut, this would reduce the number density of bright redshift. In Fig. 1a, the vertical arrows indicate the typical (M1450 <−27.1) quasars at z ∼6 by a factor of ∼ 32. This uncertainty on the mean, calculated from both the scatter figureisconsistent withanearlier studybyWyithe&Loeb over different realizations of halo merger histories and the (2002b) which concluded lensing of the SDSSsample is un- errors in our luminosity binning due to the uncertainties in likely to have increased the true number density at z = 6 thequasarbolometriccorrection.Becauseitisnotclearthat by much more than a factor of ∼1.5. Changes at this level thesamplingerrorsofdifferentluminositybinsareindepen- wouldnothoweveraffectourmodelssignificantly,andsowe dent the errors are not combined in quadrature but added do not feel that the possible presence of lensing will affect directly (reflecting the worst case scenario). Although the our conclusions. errors arebased on theassumption that thevariance of the In constraining our high redshift SMBH ‘mass budget’ underlying distribution can be approximated by the stan- we have further assumed that all the accreted mass from dard estimator for a Gaussian, comparing several different quasars combined with the mass of their initial SMBHs is runs of the model suggest they provide a fairly good indi- accountedforbytheobservedcentralSMBHsinpresentday cation of thetrueerrors. In thebottom panels we show the galaxies – that is, that the evolution of SMBHs is lossless. contribution to the mass density of SMBHs at z = 6 as a Howevertherearetwomechanismsthatmightlead tomass functionofSMBHmass,andtheaveragemassofthecentral loss intheevolution of centralgalactic SMBHs.Firstly it is SMBH as a function of host halo mass, also at z = 6 and possiblethatthemergerofSMBHsisalossyprocess,aspo- withan indication ofhowmuchofthismassisduetoblack tentiallyasignificantamountofthemass-energyofabinary hole mergers and formation (seeds). The latter plot is cre- SMBHsystemcouldbereleasedasgravitationalradiationin ated by averaging the properties of the SMBH in final root thelast stages of itsmerger (seee.g. Yu& Tremaine 2002). haloesofdifferentmassesatz∼6(actuallyz =5.75)overa Additionally, it is possible that the time-scale for SMBH large number of realizations (note that by construction the merging is sufficiently long that there is a high probability SMBH of a ‘root halo’ always corresponds to the SMBH in of accreting a third SMBH, which is likely to lead to the thecentral galaxy). ejection of the smallest SMBH or in some cases even the Wecanseethat,whennormalizedatz=6asdescribed ejection of all three SMBHs via the ’slingshot’ mechanism above, the model overpredicts the number density of lumi- (seeHaehnelt&Kauffmann2002;Volonterietal.2003,and nous quasars NQSO at lower redshift, by a factor of ∼ 25 references therein). Any such ejected SMBHs that escape by z = 2. As we have discussed, this is likely to be due to agglomeration inthenucleusoftheirhostgalaxywould not our neglect of star formation and feedback. However, the beincluded in thekind of censusused toestimate thelocal more serious problem is that when this model is normal- SMBHmassdensity,andthuseffectivelyconstituteanother izedtoreproduceNQSO atz =6,itinevitablyandseriously mode of mass loss. If either of these processes is significant overproducestheSMBHmassbudget,alreadyproducingby (cid:13)c 0000RAS,MNRAS000,000–000 High redshift quasars 9 z = 6 a total mass density of 3.6×105M⊙Mpc−3, in ex- Spread of Bright (M1450 < -27.1) Quasar Number Densities Across Realizations cess of the observationally derived value of Yu & Tremaine 4.5 d (2002) for z =0. an 89.75% of realizations produced no bright quasars B 4 Part of the problem is that only a small fraction of en SMBHsare‘activated’bymajormergersatagiventime.To o Giv 3.5 abnelEumddininogutsonenloiumgihtetdoqmuaaksearitminutsottphoesSseDssSSazS∼M6BHsamofplaet, buting t 3 least M• = 2×109 M⊙. Only about 1/100 of the galaxies Contri 2.5 withsuchmassiveSMBHsareactiveatagiventimeatsuch ons 2 redshifts.Howeverthemainreasonthatthemodelsobadly alizati 1.5 exceedstheSMBHmassbudgetcanbeseeninFig.1c:most Re ofthetotalmassdensityiscontributedbylowmassSMBHs, e of 1 g objects which even when accreting at the Eddington rate enta 0.5 would not be luminous enough to be detected by the SDSS erc P at this redshift. 0 10-11 10-10 10-9 10-8 10-7 10-6 Theshort-dashedlineinFig.1dshowsthemeaninitial Bright (M1450 < -27.1) Quasar Comoving Number Density at z = 6 (in Mpc-3) massforSMBHs(i.e.themasstheycontainaftertheirinitial gaseouscollapse) asafunctionofthemassofthedarkmat- Figure 2. The distribution of the number density of bright terhalotheSMBHfinallyendsupinatz∼6.Interestingly (M1450 < −27.1) quasars at a redshift of z = 6 obtained in this appears to become almost flat as the host halo mass individualrealizationsofthemodel,groupedintermsoftheper- increases, although the mechanism for this is not obvious. centage of realizations producing values within the range of a ToafirstapproximationtheinitialSMBHmasswilljustbe given logarithmic bin. The model uses the same parameters as a function of the mass of the halo it first formed in. Thus that shown in Fig. 1, and once all the realizations are averaged onepossibility isthatthisflatteningofthecurverepresents togethermatchestheobservednumberdensityatz=6.Notethat anupperlimittothelikelymassofthosehaloeswhichform only∼10percentoftherealizationsproducednon-zeronumber their own SMBHs, larger haloes being very likely to sim- densities. Had our simulation included rarer, larger mass haloes plyacquiretheirSMBHfrom suchsmallerhaloes whichare thenthelowerdensitybinswouldlikelybecomemorepopulated. However the mean isdominated by the2–4 per cent of the real- incorporated into their collapse (due to the nature of hier- izationsproducingvaluesintherange10−7−10−9 (whichwould archicalstructureformation).Howeverthismeanbehaviour notbeaffectedbyconsideringlargerhaloes).Evensignificantin- does not preclude the possibility of significant outliers, and creasesinthelowerdensitybinsfromrunningthesimulationover it is another less apparent failure of this simple model that largerhaloeswouldnotchangethis(andsowouldnotaffectour someoftheinitialseedSMBHshaveextremelylargemasses results). – in some cases as large as ∼109 M⊙. While we are assuming that fragmentation is somehow sufficientlysuppressedtoallowtheinitialSMBHtocollapse, the formation efficiency to fseed ∼ 1.8×10−4 in order to neverthelessitseemsunlikelythatsuchaprocesscouldcon- reducetheSMBHmassdensityat z∼6toourtargetvalue tinuetoscaleuptosuchlargemasses.Possiblytheformation (the model is ‘marginally successful’ with fseed ∼ 0.036). process would in fact be rather inefficient in its use of gas. Such a low efficiency is probably unlikely in any case (as Alternativelyitmaybethat,becausewehaveneglectedthe discussed in Section 3.3), but also the required increase in star formation and feedback which probably will consume f is sufficiently large that the models even more severely m muchofthegasinthehalo,theamountofgasthatisavail- overpredict the number density of luminous quasars at low able at thefirst major merger is unrealistically large. luminosity (by a factor of ∼ 220). It remains to be seen Asimplewaytocorrectthisproblemwouldbetomake whether the competing effects of star formation and feed- themorephysicalassumption that thecollapse process will back can account for these discrepancies. For the moment not be 100 per cent efficient, but rather only some fraction weproceedundertheassumptionthatthiscanonlybepart of thetotal gas involvedin thecollapse (fseed) will actually of thesolution. end up forming a SMBH. A hard upper limit to the initial It is worth mentioning that because quasars such as mass(Mseed,max)couldalso beimposedtocoverthecaseof those found in the SDSS z > 5.8 sample (which are scat- formation in extremelylarge early galaxies. Wewill assume tered over 2870 deg2) are so intrinsically rare, we find we the gas not used in the collapse is lost in an outflow (or must average over around 700 Monte Carlo realizations of in any case does not become available for later accretion), our model before we get good convergence. For example, in this case one might hope that this correction would also a single realization of the model is the result of a Press– help to correct the other problem – the excess total SMBH Schechter weighted average of a grid of root haloes of 50 mass density. This is because haloes which form SMBHs different masses, producing ∼ 6.17 ×105 separate galaxy but then experience little or no subsequent accretion will merger events between redshifts z = 5.75−6.25, of which make a smaller contribution to the total SMBH mass den- on average ∼26000 are major and of these only ∼380 will sity.Howeverinpracticethisiscounterbalancedtoastrong generate quasars with M1450 <−27.1. In fact about 90 per degree by the need to increase f in order to maintain the centofthe700realizationsatz=6didnotproduceasingle m agreement with the observed bright quasar numberdensity quasar above this magnitude limit. The distribution of the at z =6, because the SMBHs responsible for these quasars quasarnumberdensitiesfoundintheother∼10percentof must now be grown from smaller seeds. As a result we find therealizations is shown in Fig. 2. that with a cap of Mseed,max = 106M⊙, we must reduce It is natural to ask whether the rather strong conclu- (cid:13)c 0000RAS,MNRAS000,000–000 10 J. M. Bromley, R. S. Somerville and A. C. Fabian Comoving Number Density for QSOs M1450 < -27.1 Comoving Number Density of SMBHs 10-5 10-1 10-6 10-2 10-3 > 107 -3Mpc 10-7 Boyle et al. fit -3Mpc 10-4 nsity / 10-8 Fan et al. fit nsity / 10-5 > 108 e e mber D 10-9 SDSS z > 5.8 sample mber D 10-6 u u N 10-10 N 10-7 10-11 10-8 > 109 10-12 10-9 0 1 2 3 4 5 6 7 2 3 4 5 6 7 Redshift Redshift Distribution of SMBH masses at z = 6 Average SMBH Properties vs Final Host Halo Mass, z = 6 106 109 Total SMBH Mass 3er Mpc 105 Total: 3.6 x 105 110078 SMBHI nMitaiasls C Aoqlluaipresde bMya BssH o Mf SerMgeBrHs p nsity / Solar Units 110034 Mass (solar units) 111000456 g Mass De 102 Average 110023 movin 101 101 Co 100 100 10-1 105.5 106 106.5 107 107.5 108 108.5 109 109.5 1010 1010.5 109 1010 1011 1012 1013 1014 SMBH Mass / Solar Units Host DM Halo Final Mass (solar units) Figure1.Resultsfromthesimplemodel,withaconstantaccretionfractionfm=0.0095andcompletelyefficientandeffectiveformation of seed SMBHs. (a) Upper Left Panel: The comoving number density of bright (M1450 < −27.1) quasars (dashed line) as a function of redshift, shown withthe observational estimates (solid lines; see text for details). (b) Upper Right Panel: The cumulative comoving number density of SMBHs larger than 107, 108 and 109 M⊙ as a function of redshift. (c) Lower Left Panel: The mass distribution of SMBHsintermsoftheircontributiontothetotal SMBHmassdensityatz=6.(d)Lower Right Panel: AverageSMBHmassatz=6 as afunction of host halomass (solid)and the contribution to this frommergers (longdashed) and initialcollapse(short dashed), the remainingcontributionbeingfromaccretion. sions reached above are specific to the particular approach lowered and the seed mass capped, the number density of wehavetakentoform seed SMBH.Clearly whencompared luminous quasars increases too rapidly, overproducing the to ‘seeding’ mechanisms which use small seeds, our method observed values at lower redshift much more severely than would place more BH mass in galaxies which acquire an our original model (increasing the seed mass reduces this initial black hole but then see little accretion. However, we problem but exacerbates the overproduction of the SMBH have seen that the end result was not very different when mass density). Thus, although we restrict our attention to we reduced the mass of the seeds, as more accretion was the ‘gaseous collapse’ model for seed BH formation in the required to continue to match the observed quasar num- rest of this paper, it is likely that our conclusions will also ber density at z = 6. One could argue that for this reason bepertinent to models considering other mechanisms. our result should not be very sensitive to the details of the recipeforcreatingseedBHs.Toconfirmthiswehaverunour simple model with an alternative scenario in which haloes 4.2 Model variants are seeded with 10 M⊙ BHs as soon as the virial velocity reaches some critical value. The results are shown in Fig. 3 We have shown that the model with the minimal number forvcrit =40kms−1,illustratingthataslongaswenormal- of free parameters cannot satisfy our requirements. In this ize our model to NQSO at z = 6, the problem of exceeding section we explore other models with a minimal number of the SMBH mass budget remains. This is true regardless of additionalparameters.Oursimplemodelhadtwomainfail- the choice of the critical velocity. Although the total mass ings: firstly, it overproduced the total mass in SMBHs, de- densityisindeedreduced,itstilljustexceedsthevaluethat positing too much mass in small SMBHs; and secondly the we deem ‘marginally successful’. Also, as with the previous masses of the seed SMBHs could be excessively large. As experiment in which the efficiency of seed formation was already mentioned, the latter can be easily resolved by de- (cid:13)c 0000RAS,MNRAS000,000–000