MNRAS000,000–000(0000) Preprint20January2017 CompiledusingMNRASLATEXstylefilev3.0 On the Dynamics of Supermassive Black Holes in Gas-Rich, Star-Forming Galaxies: the Case for Nuclear Star Cluster Coevolution Pawel Biernacki(cid:63), Romain Teyssier and Andreas Bleuler CenterforTheoreticalAstrophysicsandCosmology,InstituteforComputationalScience,UniversityofZurich, Winterthurerstrasse190,8057Zurich,Switzerland 7 1 0 2 20January2017 n a J ABSTRACT 8 We introduce a new model for the formation and evolution of supermassive black holes 1 (SMBHs) in the ramses code using sink particles, improving over previous work the treat- ment of gas accretion and dynamical evolution. This new model is tested against a suite of ] A high-resolutionsimulationsofanisolated,gas-rich,coolinghalo.Westudytheeffectofvari- G ousfeedbackmodelsontheSMBHgrowthanditsdynamicswithinthegalaxy. Inrunswithoutanyfeedback,theSMBHistrappedwithinamassivebulgeandisthere- . h foreabletogrowquickly,butonlyiftheseedmassischosenlargerthantheminimumJeans p mass resolved by the simulation. We demonstrate that, in the absence of supernovae (SN) - feedback,themaximumSMBHmassisreachedwhenActiveGalacticNucleus(AGN)heat- o ingbalancesgascoolinginthenuclearregion. r t WhenourefficientSNfeedbackisincluded,itcompletelypreventsbulgeformation,so s a that massive gas clumps can perturb the SMBH orbit, and reduce the accretion rate signifi- [ cantly.Toovercomethisissue,weproposeanobservationallymotivatedmodelforthejoint evolutionoftheSMBHandaparentnuclearstarcluster(NSC),whichallowstheSMBHto 1 remaininthenuclearregion,growfastandresistexternalperturbations.Inthisscenario,how- v ever, SN feedback controls the gas supply and the maximum SMBH mass now depends on 0 9 thebalancebetweenAGNheatingandgravity.WeconcludethatSMBH/NSCco-evolutionis 1 crucialforthegrowthofSMBHinhigh-zgalaxies,theprogenitorsofmassiveellipticaltoday. 5 Keywords: methods:numerical-galaxies:evolution-galaxies:active-quasars:supermas- 0 siveblackholes-galaxies:starclusters:general . 1 0 7 1 : 1 INTRODUCTION 2007; Nandra et al. 2007; Fabian 2012; Yesuf et al. 2014; Che- v ungetal.2016).TheformationoftheSMBHthemselvesremains i Supermassive Black Holes (SMBH) are found in the central re- X amystery.Twomainscenariosareconsideredleadingtomassive gionofmassivegalaxiesatallredshifts,mostlyintheformofAc- r enoughSMBH:1)directcollapseofmassiveclumpsofpristinegas a tive Galactic Nuclei (AGN). There is accumulating evidence that (Loeb&Rasio1994;Bromm&Loeb2003)or2)mergersofstellar SMBHaretightlylinkedtotheevolutionoftheirhostgalaxy(Rich- remnantsindensestellarclusters(Quinlan&Shapiro1990;Porte- stoneetal.1998;Ferrarese&Merritt2000;Gebhardtetal.2000; giesZwartetal.1999;Devecchi&Volonteri2009),eachscenario Marconi&Hunt2003;Ha¨ring&Rix2004;Kormendy&Ho2013), havingclearstrengthsandweaknesses,asexplainedinthereviews puttingAGNphysicsatthecentreofourunderstandingofgalaxy ofBegelmanetal.(2006)andVolonteri(2010). evolution.ThestrongcorrelationofSMBHmassesandstellarve- locitydispersion,forexample,suggestsapossibleco-evolutionof Motivatedbytheseobservationalhints,theoreticalmodelsof thecentralSMBHanditshostgalaxy(Magorrianetal.1998;Laor SMBH growth and their associated feedback (mostly based on 2001; McLure & Dunlop 2002;Ha¨ring & Rix 2004). AGNfeed- complexnumericalsimulations)becameinrecentyearsmoreand back is also often invoked as one of the possible origins of the moresophisticated,withmixedsuccesseswhencomparedagainst quenchingofstarformationinellipticalgalaxies(Schawinskietal. observational data (Springel et al. 2005; Di Matteo et al. 2005; Boweretal.2006;Crotonetal.2006;Hopkins&Hernquist2006; Ciottietal.2010;Teyssieretal.2011;Duboisetal.2011).AGN (cid:63) [email protected] feedbackintheoreticalmodelsofgalaxyformationhasprovenvery (cid:13)c 0000TheAuthors 2 P.Biernackietal. efficient at regulating the Star Formation Rate (SFR) in massive, automaticallyextractedfromthesimulationusingtherecentlyde- red and dead galaxies, but the X-ray properties of the intergalac- velopedclumpfinderonboardtheramsescode(Bleuler&Teyssier tic gas are very difficult to reproduce. One natural explanation to 2014).Wealsoimprovedthedynamicalintegratorofthesinkparti- the difficulties of these models is the formidable range of scales cle,allowingustoperformdetaileddynamicalstudies.Finally,we one has to capture, in order to resolve numerically the entire ac- addedtwo newingredientsto themodel, namelyafully momen- cretionflow fromparsecscalestowards thelaststableorbit (typ- tum conserving drag force and a model for SMBH and Nuclear ically 10−5 pc). Numerical implementation of SMBH formation, Star Cluster (NSC) co-evolution. Our goal is to apply these var- theiraccretionflowsandassociatedenergeticoutflows,havetorely ious ingredients to model simulations featuring a cooling, Milky onstrongapproximations,usuallyreferredtoas“subgridmodels”. Way-sizedhalo(SeeSection3),leadingtotheformationofagas- Note that the same technique is applied to star formation recipes rich,clumpyandviolentlyturbulentdisc,reminiscentofthehigh- in galaxy formation simulations, making the whole endeavour of redshift galaxies population detected in deep Hubble Space Tele- simulatinggalaxiesverychallenging. scopeimages.InSection4,weoutlinethefactthatSMBHdynam- Astheresolutionofgalaxyformationsimulationsisincreas- ics in this turbulent environment is extremely chaotic, leading to ing,fromthousandsofpcinlarge-scalecosmologicalsimulations theejectionoftheSMBHfromthecentralregionofthegalaxy,un- (Duboisetal.2014;Vogelsbergeretal.2014;Schayeetal.2015; lessoneconsidersveryspecificdynamicalmodels.Realisticstellar Dubois et al. 2016), to hundreds of pc in cosmological zoom-in andAGNfeedbackmodelsmakethesituationevenmorecritical. simulations of galaxy formation (e.g. Kim et al. 2011; Angle´s- InSection5,wefinallydiscussamodelwhereSMBHsareeither Alca´zaretal.2014;Duboisetal.2015),ultimatelyreachingafew hostedandprotectedbyaparentNSC,ormassiveenoughtosus- pcinisolateddiscssimulations(Gabor&Bournaud2013;Hopkins taintheviolentperturbationsfromtheirhostgalaxy.InSection6, et al. 2014), these subgrid models need to be tuned and adapted wediscussvariousobservationalargumentsinfavourofthisnew totheincreasinglybetterresolvedinterstellarmedium(ISM)struc- scenario. ture,withanincreasinglystrongersupersonicturbulence. The goal of this paper is precisely to study such a model of SMBHformation,growthandfeedbackinhighlyresolved,turbu- lent and clumpy galactic discs, typical of high redshift, gas-rich galaxies(Elmegreenetal.2008a;Dekeletal.2009;Bournaudetal. 2 ANEWMODELFORSMBHFORMATIONAND 2012).ThisenvironmentisparticularlyrelevanttoSMBHphysics, EVOLUTION astheseclumpydiscsarebelievedtobetheprogenitorsofthegiant ellipticalshostingthemostmassiveSMBHsinourpresentepoch The first generation of SMBH models was developed in the con- (Kormendy&Ho2013;McConnell&Ma2013). textofcosmologicalsimulations,withresolutionaround1kpcor NumericalmodelsofSMBHformationandevolutionareall more(Bellovaryetal.2010;Vogelsbergeretal.2014;Schayeetal. basedontheso-called“sinkparticle”technique.TheSMBHisrep- 2015;Duboisetal.2016)orforrelativelysmoothgalaxymodels, resented by a point mass, moving through the fluid and interact- using either a pressurised ISM equation of state (Truelove et al. ingwithitthroughaccretionandejectionofmass,energyandmo- 1997;vandeVoortetal.2011)oralowgasfractionrelevantfor mentum. Sink particles were first implemented in simulations of low-redshiftgalaxyevolution.Thesinkparticlewasnotallowedto star-forming turbulent molecular clouds (Bate et al. 1995), using moveawayfromthegalaxycentre,byeitherforcingittoremain aSmoothedParticleHydrodynamics(SPH)code.Krumholzetal. close to the gravitational potential minimum, or by using various (2004)wasthefirstonetoproposeasinkparticleimplementation drag forces (Springel et al. 2005; Okamoto et al. 2008; Gabor & for grid-based codes, using Adaptive Mesh Refinement (AMR). Bournaud2013).ThenextgenerationofSMBHmodelsneedtobe The sink particle technique was then adapted to the SMBH for- abletoresolvetheSMBHdynamicswithinthegalaxy,andmore mationandevolution,hereagainfirstinSPHcodes(Springeletal. importantly,tofollowitsevolutionwithinhighlyturbulent,gas-rich 2005;DiMatteoetal.2005)andthenlaterinAMRcodes(Dubois environments typical of galaxy evolution at high redshift. In this et al. 2010; Kim et al. 2011). The key ingredients in our SMBH section,wepresentthenew-generationSMBHmodelimplemented formationandevolutionmodelsarethefollowings:a)theforma- intheramsescode.Itisheavilybasedontheoldmodelpresentedin tionoftheSMBHparticleandinparticularthechoicetotheinitial Duboisetal.(2010)andTeyssieretal.(2011),andcapitalisesover seedmass(e.g.Begelmanetal.2006;Volonteri2010),b)thedy- the new sink particle implementation we have developed within namicsoftheSMBHparticle,withthepossibleinclusionofadrag the context of star-forming molecular clouds (Bleuler & Teyssier force(seetherecentworkofTremmeletal.2015),c)thegrowthof 2014). theSMBHparticlemassasafunctionoftime,withtwofundamen- AlthoughwemodelSMBHsascollisionlessparticles,wedo talingredientsbeingtheBondi-Hoyle-Lyttleton(Hoyle&Lyttleton notusetheParticleMeshsolverdesignedforthedarkmattercom- 1939;Bondi&Hoyle1944;Bondi1952)accretionrate,limitedto ponent.Instead,weplacearoundeachsinkasphericaluniformdis- theEddingtonaccretionrate(forobservationalconstrainsseee.g. tributionoftestparticles(wecallthem“cloudparticles”)ofradius Kollmeieretal.2006;Steinhardt&Elvis2010),andfinally,d)the r = 4∆x , where ∆x is the size of a cell at the highest sink min min feedbackfromtheSMBHparticlethataffectsthesurroundinggas refinement level. These cloud particles are evenly spaced within (Ostrikeretal.2010;Angle´s-Alca´zaretal.2013;Choietal.2014, thesphere(withroughly8cloudparticlespergridcell)andfollow 2015; Nayakshin 2014; Costa et al. 2014), and therefore couples thesinkparticleasarigidbody.Thesecloudparticlesareusedto backtoallthepreviousingredientsofthemodel. probethegasdistributionaroundthesinkandtodistributetheac- Inthiswork,wepresentanewimplementationoftheSMBH cretionandtheejectionofmass,momentumandenergy.Notethat formationandevolutionmodelintheramsesAMRcode(Teyssier the value for the sink sphere radius can be modified by the user, 2002),inheritedfromtheearlierworkofDuboisetal.(2010)and withrecommendedvaluesrangingfrom1to4∆xmin. Teyssieretal.(2011),butsignificantlyimprovedinmanyaspects Inthefollowingsubsectionswegivemoredetailsontheim- (seeSection2).Forexample,oursinkparticleformationsitesare provementsofourSMBHsinkparticleimplementation. MNRAS000,000–000(0000) SMBHsdynamicsingas-richgalaxies 3 2.1 SMBHformation follows GM ThelifeoftheSMBHinoursimulationsbeginswiththeformation r = sink, (4) ofthesinkparticle.Itisaproblemwhichdeservesitsowncareful Bondi v2 Bondi consideration,butherewereduceittotheidentificationofapossi- (cid:113) v = c2+v2 , (5) bleformationsiteandtothechoiceoftheinitialmass Mseed.The Bondi s rel twomainscenariosforSMBHformationare1)directgascollapse wherecs isthelocalsoundspeedofthegasandvrel istherelative or2)formationthroughstellarremnantscollisionsinadensestellar velocitybetweenthesinkvelocityv andthegasaveragevelocity sink system.Inbothcase,SMBHformationisassociatedtoexception- withinthesinkspherev¯ allydenseregions,probablyatveryhighredshift,withproperties leadingfirsttotheformationofanintermediatemassblackhole, vrel=vsink−v¯ (6) whichwillaccretegasandgrowevenmoreintotheSMBHregime. Onecandefinethefree-fallvelocityontothesinkparticleas Modellingtheseprocessesisclearlyoutofthescopeofthis (cid:115) paper,asitwouldrequiremuchhigherresolutionandtheaddition GM ofphysicalingredientsthatareabsentfromoursimulations,orthat vff,sink= r sink (7) arenotevenreallyunderstoodtoday.Wethereforedirectlycreate sink ourfirstandonlySMBHwhenthefirstdenseclumpofgasforms. Thedimensionlessradiuscanbewrittenas Thisallowsthesinkparticletoevolveinadenseenvironment,mim- ickingtheearlyphaseofSMBHgrowth.Forthis,weuseourbuilt- xsink=v2Bondi/v2ff,sink (8) inclumpfinderphew(Bleuleretal.2015)andformthesinkparticle andobviouslyindicateswhethertheaccretionflowaroundthesink inthemostmassivegasclumpatachosentime(seeSection3).It issupersonicforx <1orsubsonicforx >1. sink sink isworthemphasisingthatinthisformationscenarioseedSMBHis Inthestrongsupersonicregimewhere x (cid:28)1,thedimen- sink trappedinnucleargasclump;iftheSNfeedbackisincluded,then sionlessdensityprofileoftheBondisolutionasymptotestoα(x)(cid:39) theinitialhostclumpisquicklydestroyed. x−3/2.Onecanre-writetheaccretionrateinthestrongsupersonic Thevalueoftheinitialseedmassisratherarbitrary.Atypical limitas dvaylnuaemoifcaMlsseimedu=la1ti0o5nsM(e(cid:12).,g.isBuosoutahll&ySadchoapyteed2i0n09la)r.gDe-irseccatlecohlyladprsoe- M˙ (cid:39)4πρ¯r3/2 (cid:112)GM =3 Mgas (9) scenariosofSMBHformationdopredictseedmassesofthismag- Bondi sink sink tff,sink nitude(e.g.Begelmanetal.2006).Inthispaper,weprefertoadopt wherethesinkfree-falltimeisdefinedas amorepragmaticapproachandconsidertheseedmassasafreepa- (cid:115) rameter.TheBondiaccretionmodelwedescribeinthenextsection r3 r isbasedonthestrongassumptionthatthesinkparticlegravityfield tff,sink= GMsinskink = vffs,isninkk (10) dominatesoverthegasself-gravity.Aminimumseedmassequalto andtheavailablegasmasswithinthesinksphereradiusis thesimulationminimumJeansmassappearstobetherightchoice, asournumericalexperimentsinSection4indicate. 4π Mgas= 3 ρ¯rs3ink (11) Oneconcludesthatinthestrongsupersoniclimit,theaccretionrate does not depends on the gas properties anymore, but only on the 2.2 SMBHaccretion availablegasmassandthesinkfree-falltime.Thiscorrespondsto OncetheSMBHhasformed,itcontinuestogrowinmassviaaccre- amaximumphysicallymotivatedaccretionrateontothesink. tionofgasfromitssurroundings.Themostpopularandphysically Inthestrongsubsoniclimit,wherex (cid:29)1,onehasα(x)(cid:39)1, sink motivatedapproachtocomputetheaccretionrateontotheSMBH andtheaccretionratecanbewrittenas particleistousetheBondi-Hoyle-Lyttletonformulae(laterBondi for short; Hoyle & Lyttleton 1939; Bondi & Hoyle 1944; Bondi M˙Bondi(cid:39)4πρ¯rB2ondivBondi (12) 1952): Thisisthisformulathatisusedinmostsinkparticleimplementa- M˙Bondi=4πρ∞rB2ondivBondi, (1) ttihoins,laasntdfworemwuolaueldisliokneltyovsatrleidssi,nasthiensKurbusmonhioclzreegtiaml.e(,2w00h4e)r,eththaet where Bondi radius is much smaller than the sink radius. Manipulating slightly the previous equation, one can rewrite the accretion rate ρ¯ ρ∞= (2) formulaeas α(x ) sink M˙ (cid:39)3 Mgas 1 (13) αisthedimensionlessdensityprofileoftheBondiself-similarso- Bondi tff,sink x3/2 lution(seee.g.Chapter6ofShu1992),ρ¯istheaveragegasdensity sink withinthesinksphere,and This shows explicitly that the subsonic accretion rate is much smaller than the supersonic one. This also shows that if the sim- xsink=rsink/rBondi (3) ulationtimestepiscontrolledbythesinkCourantcondition isthedimensionlessradiusevaluatedatthesinksphereradius.This ∆t(cid:54) tff,sink (14) functionα,firstintroducedbyKrumholzetal.(2004)intabulated 3 form,isacrucialingredienttodescribetheaccretionflow,andisof- onecannotremovemorethantheavailablegasmasswithinthesink tenmissinginmanysinkparticlealgorithmimplementations.The sphereinonetimestep. Bondi radius r and the Bondi velocity v are defined as It has been proposed by Springel et al. (2005) and Booth & Bondi Bondi MNRAS000,000–000(0000) 4 P.Biernackietal. Schaye(2009)toboostthepreviousaccretionrateformula,toac- oftheaccretedgaswithinthesinkspheredoesnotcoincidewiththe countforunresolveddensityandtemperaturefluctuationsatscales centreofthespherex . sink lowerthanthecellsize.Inthispaper,wefollowthesameidea,al- lowingthesoundspeedofthegastobereduced,owingtosmaller unresolvedtemperaturefluctuations.Thisboilsdowntoreplacing inthepreviousformulaethesoundspeedby 2.3 SMBHdynamics cs→cs/βboost(ρ¯) (15) The next fundamental requirement of our sink particle algorithm wheretheboostfactorisdefinedas is to model properly the dynamics of the SMBH. The sink parti- cletrajectoryfollowsfromthedynamicalevolutionofapointmass βboost(ρ)=max[(ρ/ρ∗)2/3,1.0], (16) particle,subjecttothegravitationalforceofthegas,starsanddark Incaseofzerorelativevelocity,thisformulacorrespondsexactlyto matterparticles,andalsosubjecttoadragforceduetoatightcou- themodelproposedbyBoothandSchaye.Wewouldliketostress plingbetweentheaccretedgasandthesink. thattheonlyeffectofthisboostistochangethetransitionfromsu- Notethatthelatterhasbeenofteninvokedintheliteratureto personictosubsonicaccretion,butthestrongsupersonicaccretion justify why one could artificially locked the sink particle coordi- rate will not be modified from its maximally physically allowed natestotheminimumofthepotentialwell(e.g.Sijackietal.2007; valuederivedabove.Wewouldlikealsotostressthatonecannot Costaetal.2014),orartificiallypushedinthedirectionofthehalo modifiedtherelativevelocityv fromphysicalgrounds.Sinkpar- centre(Gabor&Bournaud2013).Thereisnophysicalmotivation rel ticles with very high relative velocities are therefore likely to ac- forthesemodels.LowermassSMBHcanbeexpectedtogetscat- creteverylittlemass,astheyshould.Reducingtherelativevelocity teredbymassivegasclumps(e.g.Gabor&Bournaud2013).Other artificiallyhasbeenalsousedinthepasttoboosttheaccretionrate, physicallymotivatedmodelsdoexistintheliterature,thatcanhelp withoutanyphysicalmotivation. preventingthesinkparticlefromwanderingaroundthegalaxy.For An important ingredient specific to SMBH accretion is the example, Tremmel et al. (2015) proposed to estimate the amount maximal allowed accretion rate onto the black hole, namely the ofdynamicalfrictionthatismissingduetopoorresolution,which Eddingtonrate, consistsinasub-gridmodelforadragforcebetweenthesinkand the collisionless component. Similar sub-grid model can be con- M˙ = 4πGMsinkmp = Msink (17) structedforthepotentiallymissingdragforcebetweenthesinkand Edd (cid:15)σ c t r T S thesurroundinggasmedium(Chandrasekhar1943;Ostriker1999; wherempistheprotonmass,σTistheThomsoncrosssectionand Chaponetal.2013).Wewillproposehereanotherphysicallymo- (cid:15)ristheShakura&Sunyaev(1973)radiativeefficiencyforaSMBH tivatedmodelbasedontheEddingtonlimitedaccretion. accretion;(cid:15)r=0.1.TheseconstantsarecombinedintotheSalpeter First, the gravitational interaction between the sink and the time,astS(cid:39)45Myr.Finally,theaccretionrateontotheSMBHis matterdistribution,aswellasbetweenthesinkandpossibleother computedusing sinksinthecomputationalbox,arebothtreatedusingadirectsum- M˙acc=min(M˙Bondi,M˙Edd). (18) mationmethodofasoftened1/r2Newtonianacceleration.Wepre- fer this new approach than using the Particle Mesh method, as it We would like to stress that the Eddington rate comes from the givesmoreaccuratetrajectories,especiallyiftheSMBHmassdom- followingpicture:gasisaccretedusingtheBondiratetowardsthe inates the local potential. The softening radius used in the force SMBH accretion disc, and the accretion energy is converted into calculationsissetto2∆x ,asinBleuler&Teyssier(2014). min accretion luminosity, which in turn will remove the fully ionised Whenthesinkaccretesgasfromwithinthesinksphere,italso gasinthevicinityoftheSMBH,ifitexceedstheEddingtonlumi- accretesthecorrespondingmomentum,whichtranslatesintoanef- nosity.Sinceouraccretionmodelisappliedtoverylargegalactic fectivedragforcebetweenthegasandthesink.Whentheaccretion ofISMscales(saybetween10pcto1000pc),wedonotresolve rateontotheSMBHisEddington-limited,thesituationishowever theregionwhereradiationpressurewillremovethegasandcontrol morecomplicated.Asdescribedbefore,theEddingtonlimitforthe the accretion onto the SMBH. Eddington limited accretion there- radiationisenforcedinthevicinityoftheSMBH,wherethegasis foremeansthatgasisaccretedattheBondirate,andthendecreted fullyionisedandhasreachedtheSMBHaccretiondisc.Wecon- ataslightlysmallerrate,thenetbudgetbeingthe(small)Eddington sider in this paper that the gas accretion rate towards the SMBH rate.Thispictureisquitedifferentfromwhatisconsideredusually accretiondiscissetbytheBondiformula,andcorrespondstothe and will be used later in the paper to introduce an additional gas large scale flow, while the gas accretion rate onto the SMBH is dragforceonthesinkparticle. setbytheEddingtonlimit.Thedifferencebetweenthetworates, We discuss finally one important technical detail: once we Bondi minus Eddington, corresponds to gas being decreted from knowthesinkparticle’scurrentaccretionrate,weremovegasfrom the accretion disc region and redistributed on large scale, in our thesinkspherebyintegratingthepreviousaccretionrateoverthe casewithinthesinksphere. timestep. ∆Mgas=−M˙acc∆t (19) M˙dec=M˙Bondi−M˙acc (21) In order to avoid emptying very low density gas cells in the sink This process of accretion and ejection will lead to an additional sphere,weremovefromeachcell(labelledi)thefollowingmass- exchange of momentum between the gas and the sink, hence an weightedcontribution, additionaldragforce. ∆M Wemodelthisadditionaldragforcebyrequiringthatthecen- ∆ρi=−ρi acc (20) treofmassofthejointgas+sinksystemremainfixedduringthe M gas accretion,andthatitstotalmomentumisconserved.Ifwenotethe Animportantconsequenceofthisstrategyisthatthecentreofmass gascentreofmasswithinthesinksphereasxgas,thistranslatesinto MNRAS000,000–000(0000) SMBHsdynamicsingas-richgalaxies 5 ashiftinthesinkcoordinatesgivenby First,wehavethecoldaccretionregime,forwhichcoolingdom- inates over heating. The Bondi accretion rate is so high that we Mgasdxdgtas = M˙decxsink−M˙Bondixgas, considertheaccretiontobeEddingtonlimited, Msinkdxdstink = M˙Bondixgas−M˙decxsink, (22) M˙acc= MtsSink. (27) andasimilarmomentumtransferbetweenthesinkandthegas(in WeconsiderforthecoolingfunctiononlyBremsstrahlungsothat otherwordsadragforce)givenby: Λ(T)=Λ T1/2 (28) 0 Mgasdvdgtas = M˙decvsink−M˙Bondivgas, wmhaetiroenΛfo0r(cid:39)h1ig.2h×te1m0p−e27raeturgres−an1dcmlo3wKm−e0.t5a.llTichiitsyigsaas.gWooedcaopnpcrlouxdie- Msinkdvdsitnk = M˙Bondivgas−M˙decvsink, (23) ismphmereed,iactoeolylinthgawt,ilfloarlawagyivsewnianvoevraegrehegaatsindge,nasnitdytwheitshiinnkthweilslirnek- Theseequationsaresolvedforeachtimestep,andareusedtomod- maininthecoldaccretionregime,unlesstheSMBHmassbecomes ifythesinkpositionandvelocity,butalsothegasdensity,momen- largeenough,sothat tum and total energy within the sink sphere. More details on the t M >n2Λ T1/2 S V . (29) numericalimplementationaregivenintheAppendix.Notethatin sink H 0 (cid:15) (cid:15)c2 sink c r case of zero decreted mass (pure unlimited Bondi accretion), the Becausethesinkisnowmassiveenough,heatingdominatesover momentum transfer only comes from the accreted gas mass onto cooling,andthesinksphereentersthesecondphase,namelythehot thesink,asitshould.Intheoppositecase,whentheaccretionrate accretionregime.Forthis,wenowassumethatthegastemperature isstronglyEddingtonlimited,themassdecretionrateismaximal isalwayslargeenoughthattheaccretionrateisequaltotheBondi and almost equal to the Bondi rate. This results in a strong drag rate.WealsoconsidertheSMBHtobeatrestinthecentreofthe forcebetweenthesinkandthegas. galaxy.Wethenobtainfortheaccretionrate (GM )2 2.4 SMBHfeedback M˙acc(cid:39)4πρ c3s(itn)k (30) s Inthispaper,weonlyconsideramodelforwhichthermalenergy Wecannowsolvetheenergyequation,ignoringthecoolingterm, isinjectedwithinthesinksphere,usingfortheSMBHluminosity and obtain the time evolution of the sound speed within the sink thefollowingformula sphere where (cid:15)r =0.1 is the aLccArGeNtio=n(cid:15)dciMs˙cacrca(cid:15)drcia2t,ive efficiency and (cid:15)(c24is) cs(t)=125(cid:15)c(cid:15)rc2(cid:32)GRMsisniknk(cid:33)2Rstink1/5 (31) afreeparameterrepresentingthecouplingefficiencybetweenthe Obviously, the temperature in the sink region will not grow in- blastwaveenergyatsmallscaleandtheresultingthermalenergy depositedatlargescale.Basedonpreviousworkusingtheramses definitely. As soon as it reaches a high enough value, the gas in the vicinity of the SMBH will expand and cool adiabatically. code(Teyssieretal.2011;Duboisetal.2012),wefixeditsvalue to(cid:15)c=0.15,whichisquitetypicalofthecorrespondingliterature, We consider that we have reached the maximum temperature af- with values ranging from 0.05 (Springel et al. 2005; Wurster & terone soundcrossingtimeof thesinksphere,namely tcross(t)= Thacker2013)to0.15(Booth&Schaye2009;Gabor&Bournaud Rsink/cs(t)=t.Combiningthiswiththepreviousequationgivesus themaximumpossiblesoundspeedinthehotaccretionphase 2013). impleAmnenimtaptioorntanistitmhaptrowveemdeenptosciotmnpoawretdhetormthael perneevrigoyusatraemvesreys cs,max=125(cid:15)c(cid:15)rc2(cid:32)GRMsink(cid:33)21/6 (32) finetimestep(i.e.thetimestepofthemaximumlevelofrefinement sink (cid:96)max), and not only at main coarse time steps as before. We also Itcanbecomparedtothegalaxyescapevelocitytoassessthepos- donotconsideraminimuminjectiontemperature,asinBooth& sibilityfortheSMBHtounbindthegasfromthenuclearregion. Schaye(2009)orTeyssieretal.(2011).Moreover,thethermalen- Besides various constants that we set to our fiducial values ergyisdistributedineverygascellwithinthesinkspherepropor- ((cid:15)c=0.15and(cid:15)r=0.1),weseethattheonlyvariablesenteringthe- tionallytothegasdensity.Thismass-weighteddepositionscheme ses various formulae are the SMBH mass, Msink, the sink sphere preventstheapparitionofunrealisticallylargegastemperature,as radius Rsink, and finally the average gas density within the sink opposedtothevolume-weighteddepositionscheme. spherenH.Insertingtypicalvaluesforourpresentsimulation,we Theseimportantchangesnowallowustomodelthecompe- can compute first the critical SMBH mass beyond which heating titionbetweenheatingandcoolingwithinthesinksphere.Indeed, dominates over cooling, so that the sink sphere can exit the cold onecanwriteanenergyequationfortheaveragegasspecificinter- accretionregimeandactuallyheatsthegasaroundtheSMBH nalenergywithinthesinksphereas (cid:32) n (cid:33)2(cid:32) R (cid:33)3 ρD(cid:15) = LAGN−n2Λ(T) (25) Msink,crit(cid:39)8×104 M(cid:12) 100HH/cc 10s0inpkc , (33) Dt V H sink whereweassumedthegastemperaturetobefixedat106 Kinthe wherethespecificinternalenergyisrelatedtothetemperatureand coolingfunction.Ifthisisthecase,thenthetemperaturewithinthe thesoundspeedby sinkspherewillsteadilyincreaseaccordingtoEq.(32)andreach k T themaximumsoundspeed (cid:15)(cid:39) B (cid:39)c2(t) (26) µmH s (cid:32) M (cid:33)1/3(cid:32) R (cid:33)−1/3 Wewantnowtodistinguishtworegimesofaccretiononthesink. cs,max(cid:39)750km/s 108siMnk(cid:12) 10s0inpkc (34) MNRAS000,000–000(0000) 6 P.Biernackietal. Insummary,ifenoughgasmakesitintothesinksphere,thedensity Table1.SummaryoffiducialparametersrelatedtoSMBHsinkparticlesin willbehighandcoolingwilldominate,maintainingthegastemper- ramsessimulations. aturetorelativelylowvaluesandtheaccretionratetotheEddington limit.If,ontheotherhand,thegasdensitywithinthesinksphereis Parameter Fiducial Description toolow,orifthesinkmassistoolarge,weenterthehot,adiabatic value regimeforwhichthegastemperatureisquicklyrisingtoitsmax- imumvalue.Unfortunately,allthesequantitiesdependsensitively Mseed 106M(cid:12) Sinkseedmass ontheadoptedresolution.Abetterspatialresolution,resultingin Mclump 108M(cid:12) Massoftheclumpinwhichweseedthe asmallersinkradius,canreducethecriticalSMBHmass,butcan sink alsoincreaseitbyallowingforlargergasdensities.Betterspatial resolutioncanalsoincreasethegastemperatureinthehotaccretion Direct yes ThedirectN-bodysolverusedtoevolve solver thetrajectoryofasink regimesignificantly.Ifonewantstoalleviatethissensitivitytores- olution,onecanrelatethegasdensityinthecriticalmasstotheav- Drag yes Gasdragforcefromaccretion eragedensityoftypicalgascloudsthatarebombardingtheSMBH inthenuclearregion,andonecanchoosetodepositthefeedback αboost Eq.(16) BoostfactorfortheBondivelocity energywithinafixedradius,invokingotherphysicalprocessesto setthisenergydepositionscale.Inwhatfollows,wewillusethis simpleanalyticalargumentstointerpretournumericalresults. length 3WeNuUseMthEeRAICMARLcSoEdTeUrPamses (Teyssier 2002) and its second- λJ=cs(cid:114)Gπρ = (cid:115)Γmπ2HkGBTn∗∗ (cid:39)332pc(cid:39)4∆xmin (36) order, unsplit Godunov scheme to solve the Euler equations. The andintheminimumJeansmass evolutionofdarkmatterandstarsisperformedwiththeAdaptive 4π (cid:18)λ (cid:19)3 Particle-Meshsolverwithcloud-in-cellinterpolation.Thedynam- MJ= n∗mH J (cid:39)4×106M(cid:12) (37) ical evolution ofthe sink particle isperformed the direct gravita- 3 2 tionalacceleration(seeSubsection2.3). Themeshrefinementstrategywehaveadoptedforalloursimula- Ourinitialconditionsfeatureanisolated,gas-rich,slowlyro- tionsisaquasi-Lagrangianapproach,wherecellsarerefinedonce tating (spin parameter of 0.04) dark matter halo of 2×1012 M(cid:12) their mass exceed 8×mres, where our mass resolution is set to sampled using one million dark matter particles. The halo has a mres (cid:39)1.5×105M(cid:12), so that our minimum Jeans mass is always truncatedNFW(Navarroetal.1997)profilewithaconcentration sampledbyatleast32resolutionelements.Inallsimulations,star parameterc=10andwiththecircularvelocityV200=160kms−1, formation is modelled with a Schmidt law with a rather low ef- whichresultsintheradiusR200=230kpc,whilethehaloistrun- ficiency (cid:15)∗ =0.01 coming from observations of local molecular catedat514kpc.Initially,thegaseoushaloisinhydrostaticequi- clouds(Krumholz&Tan2007).Collisionlessstarparticlesoffixed libriumandhastheuniversalgasfractionof fgas=15%.Theini- mass1.3×105 M(cid:12) arespawnedstochasticallywithaPoissondis- tialisationfollowsthesetupofTeyssieretal.(2013).Ourfiducial tributionifthegasdensityinthecellislargerthann∗mH(Rasera& runhasaspatialresolutionof∆xmin=78pc. Teyssier2006).Feedbackfromsupernovae,ifconsidered,ismod- Using an isolated cooling halo is dictated by a compromise elled with a non-thermal energy injection with efficiency of 10% betweenrealisticbutexpensivecosmologicalsimulationsandide- and yield of 10% (1 M(cid:12) of metals for each 10 M(cid:12) of ejected alisedbuthighlyresolvedisolateddiscsimulations.Sinceweare material). The non-thermal energy dissipation timescale is set to usingarealisticinitialangularmomentumprofileinspiredfromthe 10Myr.Weboosttheefficiencyofoursupernovaefeedbackrecipe average angular momentum distribution from N body simulation by grouping stochastically multiple star particles into one single (Bullock et al. 2001), gas will be continuously accreted from the starclusterofmass108M(cid:12). halo into the disc, with the right amount of angular momentum, AsitwasalreadymentionedinSection2.1,weallowonlyone giving us the possibility to feed the nuclear region, and possibly sink to form in our galaxy. We form the sink at around 200 Myr thecentralSMBH. after the start of the simulation. We use the phew clump finder We use the Sutherland & Dopita (1993) model for radiative (Bleuleretal.2015)toidentifythemostmassivegasclumpasthe coolingofgasforH,Heandmetallinesforgashotterthan104K formation site for the SMBH, and let the sink evolve from there. andfrommetalfine-structurecoolingprocessesatlowertempera- AllfiducialparametersofourSMBHmodelarelistedinTable1. tures.Weadvectthemetallicityintheformofapassivescalarand we choosethe initialmetallicity tobe Zini=0.05 Z(cid:12). Apressure floorisintroducedathighdensityandlowtemperature,toprevent theuncontrolledfragmentationofgasbeyondthespatialresolution, possiblyleadingtotheformationofnumericalsingularities(espe- 4 RESULTS cially because we are using a low star formation efficiency). The We now present our simulation results, including each important temperaturecorrespondingtothepressurefloorissetto processonebyone,inordertocomparethem,andgaugetheirrel- (cid:32)n (cid:33)Γ−1 ative importance. These processes are listed in Table 2. For each Tfloor=T∗ nH∗ (35) feedbackprocess,weusetheparametersdescribedintheprevious section.WehoweverconsidertheSMBHseedmassasafreepa- withacriticalgasnumberdensityn∗=9cm−3,acriticaltempera- rameter, and we explore values ranging from 105 to 109 M(cid:12), as tureT∗=2×103K,andΓ=2.ThisresultsintheminimumJeans listedinTable2. MNRAS000,000–000(0000) SMBHsdynamicsingas-richgalaxies 7 enoughtohaveasustained,largerthanEddington,Bondiaccretion Table2.Summaryofsimulationrunsandparametersusedinthisstudy.Pa- rametersvariedwithrespecttothefiducialrunarehighlightedinboldprint. rate. Columns:(1)subsectioninwhichthesimulationsareanalysed(withexcep- Wearguethatthecriticalmassforthesinkparticletoaccrete tionforfiducialrun);(2)maximumallowedrefinementlevel;(3)fractionof fastenoughistheminimumJeansmassassociatedtoouradopted SNenergydepositedinthegas;(4)dragforcemodelled(orinclusionofa mesh resolution. Indeed, assuming that v =0, we can re-write rel nuclearstarcluster);(5)initialseedmassinlog10M(cid:12);(6)AGNfeedback. theparameterthatcontrolswhetherBondiaccretionissubsonicor supersonic(seeEq.8)as Se(c1ti)on l(m2a)x (cid:15)(S3N) d(r4a)g m(s5e)ed AG(N6)fbk. x = c2sλJ (cid:39) MJ (38) sink GM M sink sink 4.1 14 0.0 yes 5 no where we used the fact that r =4∆x =λ . For our fiducial sink min J 44..11 1144 00..00 yyeess 76 nnoo resolution,theJeansmass MJ is4×106 M(cid:12).Forthelowestseed mass,whichisbelowtheJeansmass,accretionfollowstheBondi 4.1 14 0.0 yes 8 no rate,andisratherlow,becausetheaccretionissubsonic.Notethat 4.1 14 0.0 yes 9 no inthisregime,becausetheaccretionrateislow,thedynamicalcou- 4.2 14 0.0 yes 5 yes plingbetweenthesinkandthegasisweak,makingthesinkvery 4.2 14 0.0 yes 6 yes sensitive to external perturbations. One can see in Figure 1 that 4.2 14 0.0 yes 7 yes the trajectory of the sink particle is quite perturbed, with visible 4.2 14 0.0 yes 8 yes oscillationsaroundthecentreofthegalaxy.Theseoscillationsin- 4.2 14 0.0 yes 9 yes crease the relative velocity between the gas and the sink, further 4.3 14 0.1 yes 5 no contributing to the low accretion rate. Once the sink mass grows 4.3 14 0.1 yes 6 no beyond106M(cid:12),about800Myrafterthestartofthesimulationfor 4.3 14 0.1 yes 7 no thesmallseedmassorimmediatelyaftercreationfortheotherseed 4.3 14 0.1 yes 8 no masses,theBondiaccretionevolvesfromsubsonictosupersonic.A 4.3 14 0.1 yes 9 no muchmorerapid,Eddington-limitedexponentialgrowthfollows. 4.4 14 0.1 yes 5 yes Afterthisphase,theSMBHmassseemstosaturate,andgrows 4.4 14 0.1 yes 6 yes onlymildly,mostlybecauseoftheslowaccretionoffreshgasinto 4.4 14 0.1 yes 7 yes the nuclear region. Indeed, since we didn’t include any feedback 4.4 14 0.1 yes 8 yes processesinthisfirstexperiment,thefinalSMBHmassisregulated 4.4 14 0.1 yes 9 yes bytheavailablegasmasswithinthenuclearregion.Thisregime, 4.5 14 0.1 NSC 5 yes calledhereaccretion-limitedgrowth,wasfirstdiscussedinBour- 4.5 14 0.1 NSC 6 yes naud et al. (2011). The late accretion phase is controlled by an- 4.5 14 0.1 NSC 7 yes gularmomentumtransferinthegalacticdisk,triggeredbyvarious 4.5 14 0.1 NSC 8 yes instabilitiesandslowlyfeedingtheSMBHwithfreshgas.Inthis 4.5 14 0.1 NSC 9 yes case,theSMBHtrajectoryremainswellwithinthenuclearregion, adenseandmassivestellarbulgethatprovidesaverystableenvi- 4.6 15 0.0 yes 6 yes 4.6 15 0.1 yes 6 yes ronmentfortheSMBH.Asaresult,thesinkparticleneverleaves 4.6 15 0.1 NSC 6 yes thenuclearregion. 4.2 AGNfeedback-limitedgrowth Wehaverepeatedthesamesimulationsasintheprevioussection, 4.1 Accretion-limitedgrowth but this time with AGN feedback. The resulting dynamical and Ourfirstsuiteofsimulationshasbeenperformedwithoutanyfeed- mass evolutions of the SMBH are shown in Figure 2a and Fig- backprocessesandwithonlyonesinkparticleseededinthefirst, ure2b.Theonlydifferencewiththeprevioussetupisthefinalmass massiveenough,nucleargasclump,growingviaEddington-limited ofthesink,whichisnowregulatedbyAGNfeedback. Bondiaccretion.Becauseoftherelativelylowangularmomentum TheinitialgrowthoftheSMBHinoursimulationswithAGN inourcoolinghalo,mimickingwhatweexpectfromcosmological feedbackisverysimilartotherunswithoutfeedback.Duetothe simulations,thesesimulationswithoutfeedbackleadtotheforma- largeamountsofgasinthenuclearregion,feedbackheatingdoes tion of a gas-rich, clumpy and bulge-dominated galaxy (see also no affect the gas surrounding the sink, as cooling dominates. As Teyssieretal.2013;Duboisetal.2016)thatresemblesmanyob- soonastheaccretionrateishighenough,heatingdominatesover served high-z galaxies, in particular the so-called “blue nuggets” coolingandtheSMBHquicklyreachesitsmaximummass,which (seee.g.Damjanovetal.2009). inourcaseisaround2×108M(cid:12). ThetrajectoryandthemassgrowthoftheSMBHareshownin Themaximum,self-regulatedmassisrelatedtotheheating- Figure1.Forallouradoptedseedmasses,theSMBHremainswell coolingbalancewehavediscussedinSubsection2.4(seeEq.(32) withinthenuclearregion(definedhereasthecentralkiloparsec),in and(34)).Usingthesimulationwith Mseed=106M(cid:12) asanexam- whichtheywereborn.Interestingly,thelowestseedmass105M(cid:12) ple,weseethatataround420Myr,thesink’sgrowthisterminated. showsaverydifferentbehaviourthantheother,largerseedmasses. Initially, the gas density in the sink sphere is quite large, around Itsgrowthisveryslow,foralmost1Gyr,andonlywhenitreaches ∼600H/cm3,sothatclearlycoolingdominatestheenergybudget 106 M(cid:12) does it have a high enough accretion rate and grows ex- inthesinksphere.Gradually,astheSMBHmassgrows,feedback ponentially.Theotherseedmassesstartgrowingexponentiallyim- isabletoheatthegasmoreandmoreinthesinksphere,untilthe mediatelyaftertheircreation,whichmeansthattheyaremassive SMBHmassreachesthecriticalvalueforwhichheatingdominates. MNRAS000,000–000(0000) 8 P.Biernackietal. c]6 p 1e5 1e8 k e [5 1e6 1e9 1010 r 1e7 t n 109 e4 c ] o fl l M 108 ha3 [ m BH 107 o2 M 1e5 r f 1e6 nce 1 106 11ee78 a st 105 1e9 Di0 0 200 400 600 800 1000 1200 1400 0 200 400 600 800 1000 1200 1400 Time [Myr] Time [Myr] (a)Distancebetweenthesinkandthecentreofthehalo (b)SMBHmassgrowth Figure1.EvolutionofdistancetothecentreofhaloandsinkmassfortherunswithoutneitherSNandAGNfeedbacksforfivedifferentseedmasses:105M(cid:12) -red,106M(cid:12)-blue,107M(cid:12)-green,108M(cid:12)-purple,and109M(cid:12)-orange.Thesinkparticleoccupiespositioninthecentreofthehaloanditsgrowthis limitedfirstbyEddingtonrateandlaterbyangularmomentumlossinthegas. c]6 p 1e5 1e8 k e [5 1e6 1e9 1010 r 1e7 t n 109 e4 c ] o fl l M 108 ha3 [ m BH 107 o2 M 1e5 r f 1e6 nce 1 106 11ee78 a st 105 1e9 Di0 0 200 400 600 800 1000 1200 1400 0 200 400 600 800 1000 1200 1400 Time [Myr] Time [Myr] (a)Distancebetweenthesinkandthecentreofthehalo (b)SMBHmassgrowth Figure2.EvolutionofdistancetothecentreofhaloandsinkmassfortherunswithoutSNfeedbackbutwithAGNfeedbackforfivedifferentseedmasses: 105M(cid:12)-red,106M(cid:12)-blue,107M(cid:12)-green,108M(cid:12)-purple,and109M(cid:12)-orange.Thesinkparticleresidesinthecentreofthehalotravellingwithmost massiveclumpanditsgrowthislimitedfirstbyEddingtonrateandlaterterminatedatself-regulationscaleduetoitsfeedbackheating. Veryquicklythegassoundspeedwithinthesinksphererises,until right),afterwhichtheaveragesoundspeedquicklyexceedsvesc, itexceedstheescapevelocityofthehalo.Whenthishappens,gas whichthenmarkstheendofthecoldaccretionregime(bottomleft) isremovedfromthenuclearregionbyablastwave,whichreduces and the beginning of the hot mode of accretion. For comparison, theaveragegasdensitydowntoorevenbelowρ¯(cid:39)10H/cm3,and wehaveplottedinFigure3oursimpleanalyticalpredictionsfrom makesfeedbackevenmoreefficient.Feedbackisabletomaintain Subsection 2.4. We can predict quite nicely the onset of efficient thesoundspeedtoahighvalue(cs(cid:39)400km/s),stronglyreducing heating, when the SMBH mass reaches its critical value, as well theaccretionrate(seeEq.(5)andbottomleftpanelofFigure3). as the end of the mass growth, when the maximum sound speed This is only when feedback processes are able to accelerate reachestheescapevelocityofthehalo. gastotheescapevelocitythatthegrowthofSMBHishalted(see also Silk & Rees 1998; Fabian 1999). This can be seen on Fig- ure3,whereweplotvariousaveragequantitiesmeasuredinthesink The case with Mseed =109M(cid:12) is very different than all the sphere.Thegasdensity(topleft)dropsbytwoordersofmagnitude othercases.Here,theinitialseedmassisalreadyabovethemaxi- assoonasthemaximumsoundspeedsignificantlyexceedsthehalo mum,self-regulatedmass.AGNfeedbackimmediatelyblowsaway escapevelocity(vesc,topright).ThecriticalSMBHmassMsink,crit, thegasfromthenuclearregion.Asaresult,thegasinthevicinity forwhichheatingbalancescooling,isreachedat420Myr(bottom ofthesinkremainsveryhotandtheaccretionrateverylow. MNRAS000,000–000(0000) SMBHsdynamicsingas-richgalaxies 9 104 2500 cs cs,max cs,theo 2000 103 c] vesc /s] c m H/ 1500k [ [ y 102 y t t si 1000ci n o e l e D V 101 500 ngas 100 0 103 10 17 Bondi r] 102 Eddington 10 18 y / ] ) fl 10 19 3 M 101 m [ c e 10 20 s at 100 g/( n r 10 21 er o [ i 10 1 e et 10 22 at cr R c 10 2 A heating 10 23 cooling 10 3 10 24 300 350 400 450 500 300 350 400 450 500 Time [Myr] Time [Myr] Figure3.Timeevolutionof1)averagegasdensitywithinthesinksphere(topleft);2)average,mass-weighted,soundspeed(blue)andmaximumsoundspeed (red)withinthesinksphere(topright).Wehavealsorepresentedoursimpletheoreticalmodel(Eq.32and34)(green)comparedtothehaloescapevelocity (orangedashed);3)Bondi(red)andEddington(blue)accretionrates(bottomleft)and4)averageheating(red)andcooling(blue)rateswithinthesinksphere (bottomright)forsimulationwithAGNfeedbackonlyandMseed=106M(cid:12). 4.3 Supernovaefeedback-limitedgrowth proceedsmuchmorerapidlyandthesinkmasscangrowuptoits accretion-limitedvalue,asinSection4.1. We now remove AGN feedback from the picture, but include in- In order to estimate the mass of the typical clumps that stead supernova feedback from dying massive stars. We use the will perturb the trajectory of the SMBH, we use the classical samesimulationsuitethanbefore,withseedmassesfromM = seed Toomre analysis of gas fragmentation in an idealised razor thin 105M(cid:12)toMseed=109M(cid:12).OnFigure4,weagainplotthetimeevo- disc(Toomre1964).ThelargestunstablewavelengthistheToomre lutionofthedistanceofthesinkparticletothehalocentreandofits length mass.Hereagain,wecanseetwodifferentregimes.Lowandinter- mediateseedmassesarequicklyremovedfromthecentralkilopar- sec.There,supernovaefeedbackisefficientenoughtodestroythe GΣ f gas gas parent clump, and the sink particles are perturbed by interaction λ (cid:39) (cid:39) R . (39) T κ2 π gal withnearbyclumps.Asaconsequence,thetrajectoryofthesinks becomes more complicated and eccentric, and the relative veloc- itybetweenthesinkandthegaswithinthesinkspheregrowssig- In this approximate formula G is gravitational constant, nificantly,reducingtheaccretionrateandthecorrespondingdrag Σgas=Mgas/πR2gal is the gas surface density, κ(cid:39)Vgal/Rgal is the forceaccordingly.ForseedmasseslargerthanMseed=108M(cid:12),the epicyclicfrequency,Vgal= (cid:112)GMtot/Rgalisthegalaxycircularve- sinktrajectoryappearsasmuchlessperturbedandthesinkmanage locityand fgasisthegas-to-totalmassfractioninthedisc.Inorder to remain within the nuclear region. As a consequence, accretion forthiswavelengthtobetrulyunstable,theToomreparametermust MNRAS000,000–000(0000) 10 P.Biernackietal. satisfy In a previous section, we have seen that a seed mass lower Q= πGcΣsκ <1. (40) WthaenseMeJneoanws trheastulatsseiendamnaasrstilfiocwiaelrlythalonwM, sTurbessounltiscianctchreetsioinnkrpaater-. gas ticlebeingscatteredoutofthenuclearregionbylargegasclumps. wherecscanbetakenaseitherthesoundspeedorthevelocitydis- Similarly, a large seed mass with a dynamical friction time scale persionofthegas.Undersuchconditions,onecanthenestimatethe comparableto(orshorterthan)theorbitaltimet (cid:39)200Myrwill orb massofthemostmassiveclumpsastheToomremassMT defined helpmaintainingthesinkparticlewithinthenuclearregion. by Insummary,largeinitialseedmasses(108and109M(cid:12))have a larger accretion rate, as M˙ ∝M2 , so they can grow fast, MT=Σgasπ(cid:18)λ2T(cid:19)2(cid:39) Mt4oπtf2g3as. (41) atottohrebiirtaEldpdeirntugrtobnat-iloimnsi.teFdurrathtBeeo,rnmadniodreb,eMcBoHmiescqoumicpkalyralbelsestsoenMsitive, T seed ForaMilkyWay-likegalaxy,onehas Mtot(cid:39)1011 M(cid:12) inthedisc thussinkparticlesdonotsufferfromencounterswithlargermass (not to be confused with the total mass in the halo, which is one perturbers.Also,itsdynamicalfrictiontimescaleisrelativelyshort, orderofmagnitudelarger).Atlowredshift,ingalaxiessimilarto helpingtheSMBHtoremaininthecentre. ourownMilkyWay,onefinds fgas(cid:39)0.1,whichresultsinatypi- calclumpmassof MT (cid:39)2.5×106 M(cid:12).Athighredshift,however, 4.4 AGNfeedback-limitedgrowthwithsupernovaefeedback likethecoolinghaloset-upweareadoptinginthispaper,thegas fraction is much higher, fgas(cid:39)0.5, for a similar total mass. This WenowcombinesupernovaandAGNfeedback,repeatingthesame leads to much bigger clumps, with MT (cid:39)3×108 M(cid:12). This value numerical experiments. As before, the low and intermediate seed istypicalformassiveandgasrichgalaxies(see.e.g.Genzeletal. masses Mseed=105M(cid:12), Mseed=106M(cid:12) and Mseed=107M(cid:12) do 2008,2011;Guoetal.2012;Tacconietal.2013;Tamburelloetal. not really grow, as can be seen in Figure 5 for the red, blue and 2015,forin-depthdiscussion).Inconclusion,asinkparticlewith greenlines,andasitwasalreadythecaseforoursupernova-only massM (cid:54)M willhaveitstrajectoryeasilydisruptedbyclumps feedback model. The large seed mass, on the other hand, are al- sink T inthedisc.Largersinkmasses,ontheotherhand,willresultina readytoocloseorevenlargerthantheirmaximum,self-regulated muchmorestableorbitalevolution(seebelow). SMBHmass,asitwasalreadythecaseforourAGN-onlyfeedback In order for the sink particle to reach (or remain in) the nu- model.Soeventheselargeseedmassesdonotfavourafastgrowth clearregionofthegalaxy,weneedtoestimatethedynamicalfric- ofthesinkparticles,whicharecontinuouslyperturbedbyclumps tion timescale as introduced by Chandrasekhar (1943). Although withmasscomparableorsmallerthantheToomremass.Moreover, theoriginalformulawasderivedforacollisionlessfluid(darkmat- sincethesinkmassisnotgrowingmuchbeyond109 M(cid:12),thedy- terandstars),averysimilarformulacanbeusedtocomputethe namical friction time scale remains longer or comparable to the dynamical friction on the gas (Ostriker 1999). For the gas drag, orbitaltimescaleandthesinkparticleskeepmovingaroundwith a correction factor must be introduced, compared to the original eccentricorbitsandlargepericentreradii(seeFigure5withviolet collisionless case, but only for transonic relative velocities. For a andorangelines;alsoFigure10,leftcolumn). SMBHwithatypicalorbitalvelocityof200km/s,thedragforceis SNandAGNfeedbacksworkhandinhandtocompletelypre- likelytobeinthestrongsupersonicregime,forwhichnocorrection vent SMBH growth in this gas rich, highly turbulent and clumpy isrequired. environment. We argue that only SMBH already as massive as Using Chandrasekhar’s formula, we compute the dynamical 1010 M(cid:12) can survive in the nuclear region of such a galactic en- frictiontimescalet (e.g.Eq.8.12ofBinney&Tremaine2008) vironment,becausetheyresisttheperturbationsfromclumpsand df because they have a short-enough dynamical friction time scale. t = 1.65 R2orbσ (42) This conclusion is of course valid only if one considers that our df lnΛGM twofeedbackmodelsarerealisticenough,whichisofcoursehighly BH speculative,sincetheyrelyonsub-gridphysics.Thesemodelsare wheretheCoulomblogarithmisgivenby neverthelessquitestate-of-the-art,andarerequiredtoexplainthe lnΛ=lnRGgaMlVo2rb (43) lfoowrmsattairofnoqrmueanticohninegffiincimenacsysi(vfeorgSalNaxfieeesd(bfoacrkA)GanNdfteoedebxapclaki)n.star BH ThefactthatSMBHcannotgrowatall(excepttheextremely Rorb and Vorb are the orbital radius and orbital velocity of the massiveones)ifonecombinesthetwosourcesoffeedbackenergy SMBH.Assumingthatthevelocitydispersionofthecollisionless isthereforeafundamentalprobleminthetheoryofSMBHgrowth componentsσ(cid:39)Vgal,theorbitalradiusandvelocityoftheSMBH andco-evolutionwithgalaxies.Thisalsoexplainswhymanyau- to be of the order of the galaxy radius Rgal(cid:39)5 kpc and circular thorshavetorelyonartificialtrickstomaintaintheSMBHwithin velocityVgal(cid:39)200kms−1,andfinallyusinglnΛ(cid:39)6.9asatypical thenuclearregionsofgalaxies,especiallywhenperforminghigh- valueforourpurposes,wefindthedynamicalfrictiontimescaleto resolutionsimulations. be 108M(cid:12) 4.5 GrowthwithinaNuclearStarCluster t (cid:39)2.7Gyr . (44) df M BH Oneofthekeydifferencebetweenthesimulationwithsupernova- OnlySMBHwithmassesgreaterthan109M(cid:12)willbeabletodecay feedback and the simulations without supernova-feedback is the quickly enough to the centre of the galaxy, as they will have an presenceofamassivebulge,orinotherwords,amassivenuclear orbitaldecayratecomparableorfasterthantheirrotationrate.Itis concentrationofstars(seeFigure8).Indeed,inthenosupernova interestingtoseethattheToomremassandthecriticaldynamical feedbackcases,wedoformmassiveclumpsofgasandstarswith frictionmassarebothcomparableto109M(cid:12)inhigh-redshiftMilky massesoftheorderof(orsmallerthan)theToomremass,thatap- Wayanalogues(seealsoBournaudetal.2014). pear as bumps in the stellar surface density profile in Figure 8. MNRAS000,000–000(0000)