Dust evolution with active galactic nucleus feedback in elliptical galaxies HiroyukiHirashitaa,TakayaNozawab aInstituteofAstronomyandAstrophysics,AcademiaSinica,P.O.Box23-141,Taipei10617,Taiwan bNationalAstronomicalObservatoryofJapan,Mitaka,Tokyo181-8588,Japan Abstract 7 Wehaverecentlysuggestedthatdustgrowthinthecoldgasphasedominatesthedustabundanceinellipticalgalaxieswhiledustis 1 efficientlydestroyedinthehotX-rayemittingplasma(hotgas). Inordertounderstandthedustevolutioninellipticalgalaxies,we 0 2constructasimplemodelthatincludesdustgrowthinthecoldgasanddustdestructioninthehotgas. We alsotakeintoaccount the effect of mass exchangebetween these two gas componentsinducedby active galactic nucleus(AGN) feedback. We survey n areasonable rangesof the relevantparametersin the modeland find that AGN feedbackcycles actually producea variety in cold Jgasmassanddust-to-gasratio.Bycomparingwithanobservationalsampleofnearbyellipticalgalaxies,wefindthat,althoughthe 5dust-to-gasratiovariesbyanorderofmagnitudeinourmodel,theentirerangeoftheobserveddust-to-gasratiosisdifficulttobe 2reproducedunderasingleparameterset. Variationofthedustgrowthefficiencyisthemostprobablesolutiontoexplainthelarge varietyindust-to-gasratiooftheobservationalsample. Therefore,dustgrowthcanplayacentralroleincreatingthevariationin ] Adust-to-gasratiothroughtheAGNfeedbackcycleandthroughthevariationindustgrowthefficiency. G Keywords: Activegalacticnuclei,Dust,Ellipticalgalaxies,Galaxyevolution,Interstellarmedium . h p -1. Introduction dustexistinginellipticalgalaxiesispossiblyinjectedfromout- o side via themergingoraccretionof externalgalaxies(Forbes, r t InthenearbyUniverse,ellipticalgalaxiesareknowntohave 1991;Temietal.,2004;Fujitaetal.,2013).Thelackofcorrela- s aless gas, dust, and ongoing star formation activity than spiral tionbetweendustFIRluminosityandstellarluminosityisalso [galaxies. Yet,theystillhavesomeamountofinterstellargasin takenasevidenceofexternaloriginofdust(Temietal.,2007); theformofhotX-ray-emittinghalogas(e.g.,O’Sullivanetal., however, this argument may not hold if dust is processed by 1 v2001) and cold gas (e.g., Wiklindetal., 1995). More- mechanismsunrelatedtostars. 0over, dust is detected for a significant fraction of elliptical Recently, Hirashitaetal. (2015) have proposed that the ex- 0galaxies by optical extinction (e.g., Goudfrooijetal., 1994; istence of dust in elliptical galaxies can be explained by dust 2 vanDokkumandFranx, 1995; Ferrarietal., 1999; Tranetal., 7 growth by the accretion of gas phase metals in the cold in- 2001) or far-infrared(FIR) emission (e.g., Knappetal., 1989; 0 terstellar medium (ISM). They also suggest that the presence .Smithetal.,2012;diSeregoAlighierietal.,2013). Dustmass of dust growth also explains the extinction curves observed 1 is estimated from the reddening in the optical or from FIR 0 in elliptical galaxies. The dust-to-gas ratio could become as 7dust emission, and ranges from ∼ 104 to ∼ 107 M⊙ (e.g., high as & 10−3 by accretion, explaining the high dust abun- 1Goudfrooijetal., 1994; Leeuwetal., 2004). Since the exis- danceinsomeellipticalgalaxies. However,dustgrowthinthe :tence of dust could affect the cooling and chemical processes v cold gas has not been considered as a major source of dust (Dwek, 1987; Fabianetal., 1994; VoitandDonahue, 1995; i in the context of dust evolution in elliptical galaxies. Dust X Seoketal., 2015; Hirashitaetal., 2015), the understandingof growth has already been noted as a major mechanism of dust rthe originandevolutionofdustinellipticalgalaxiesisimpor- a mass increase in a wide context of galaxy evolution (e.g., tantinclarifyingtheirevolution. Dwek, 1998; Hirashita, 1999; Inoue, 2003; Zhukovskaetal., Because the stellar population is dominated by old stars 2008; Valianteetal., 2011; MattssonandAndersen, 2012; whose ages are comparable to the cosmic age, the dust in el- Mancinietal., 2015; Poppingetal., 2016; Houetal., 2016; lipticalgalaxiesispredominantlysuppliedbyasymptoticgiant Zhukovskaetal., 2016), and some experimental studies have branch(AGB)starsratherthan bysupernovae. However,dust also shown that dust grains could grow by accreting the gas- destructionbysputteringintheX-rayemittingplasmaissoef- phase metals (Rouille´etal. 2014; but see Ferraraetal. 2016). ficientthattheobserveddustmasscannotbe explainedbythe Thus, constructing a dust evolution model that includes this balance between the supply from AGB stars and the destruc- dustgrowthmechanismwouldcontributetotheunderstanding tion(e.g.,Patiletal.,2007). Thus,someauthorsarguethatthe ofdustevolutioninellipticalgalaxies. Theoverallevolutionofdustshouldbeconsideredinrelation Emailaddress:[email protected](Hiroyuki tothegasevolution,especiallybecausegasanddustareusually Hirashita) dynamically coupled on galactic scales. Recent studies have PreprintsubmittedtoPlanetaryandSpaceScience January26,2017 proposedthat the gascoolingin galaxiesis stronglyregulated ofdustevolutioninanellipticalgalaxyinSection2. Weshow by the energy input from active galactic nuclei (AGNs). The theresultsinSection3. Wediscussthemodelpredictions,and energy input from AGN winds (SilkandRees, 1998; Fabian, comparethemwithobservationaldatainSection4. Finallywe 1999; King, 2005) or AGN jets (WagnerandBicknell, 2011; concludeinSection5. Mukherjeeetal., 2016) preventscoolingflowsfromoccurring and/or makes cold gas evaporate(see also CiottiandOstriker, 2. Model 2001; Fabian, 2012). These kinds of energy input are called AGN feedback, and are considered to play an important Weconstructamodelthatdescribesthedustevolutioninan role in galaxy formation and evolution (Crotonetal., 2006; elliptical galaxy. For dust sources, in addition to AGB stars BoothandSchaye,2009) considered in previousstudies (MathewsandBrighenti, 2003; Temietal. (2007) showed the existence of a dust emission Patiletal., 2007), we also consider dust supply from the cold componentextendedover5–10kpcinellipticalgalaxiesusing gaswheredustgrowsbytheaccretionofgas-phasemetals.The the Spitzer 70 µm band data. They suggestthatthis cold dust dustinthecoldgasisinjectedintothehotgasbyAGNfeedback componentwasoriginallycontainedinthecoldgas,whichwas togetherwiththecoldgas. Thedustinjectedintothehotgasis then heated by AGN feedback and eventually mixed with the destroyed by sputtering. The hotcomponent(∼ 107 K) is the hotgas. Intheirscenario,theheatedgasistransportedintothe gaswhosetemperatureiscomparabletothevirialtemperature hotgasbybuoyantforce,andthetime-scaleofthetransportis determinedbytheglobalgravitationalpotentialoftheelliptical around107yr. Kanedaetal.(2011)alsoshowedinanelliptical galaxywhile the cold componentis cold and denseenoughto galaxy (NGC 4125) a dust emission componentwhose exten- host dust growth (. 100 K; Hirashitaetal., 2015). To make sionissimilartothedistributionofthehotX-rayemittinghalo. the modelassimple as possible, we treateach gascomponent Themulti-phasestructuresandirregularmorphologiesofX-ray as a single zone and consider the mass exchange between the emittinghotplasmascouldbeexplainedbybubblescreatedby twocomponents. Dustgrowthanddestructionaretreatedcon- AGNfeedback(Buoteetal.,2003). sistentlywiththeevolutionofeachgascomponentasexplained Apartofthehotgasmaycooldowntoreformthecoldgas, below. which could contribute to the fueling of AGN (Werneretal., 2014). AGNfeedbackmaycompressthesurroundinggasand 2.1. Basicequations enhance gas cooling locally (ValentiniandBrighenti, 2015). Because dust grains can grow in the cold gas as mentioned We consider the mass exchange between the hot and cold above,thecoldgasinjectedintothehotgasbyAGNfeedback phases. The evolutionofthe coldgasmass Mg,c asa function would supply the dust to the hot gas. This dust supply could oftimetiswrittenas bedominantoverthedustproductionbyAGBstars. Ifso,for- dM mationofcoldgasandthesubsequentAGNfuelingleadingto g,c = M˙in−M˙ret, (1) dt AGN feedback play a dominant role in the dust evolution in ellipticalgalaxies. whereM˙inistheinfallrateofthecooledgasandM˙retisthegas If the dustevolutionis affected byepisodic AGN activities, return rate from the cold to the hot phase by AGN feedback. which appear as a result of a cycle of gas ejection and cool- Thesetwotermsaremodeledbelow. ing, the statistical properties of the dust abundances in ellip- Thehotgasistreatedasaconstantreservoirofgasforsim- tical galaxies are determined by the time-scales of dust pro- plicity;thatis,weassumethatthemassofhotgas,Mg,hiscon- cessing relative to the period of an AGN cycle. Laueretal. stant(dMg,h/dt = 0). Thistreatmentneglectsthecomplication (2005)inferredtheperiodofanAGNcycleinearly-typegalax- arisingfromthepossibilitythattherecouldbeasupply/lossof ies based onthe dustlifetime againstsputteringin the hotgas hotgas from/tooutside. Because the hot gasbasically acts as andthefractionofdustdetectionforasample,obtainingape- an efficient destroyer of dust regardlessof the choice of Mg,h, riod of ∼ 108 yr. The dust contained in the cold gas may be thevalueofMg,hhasaminorinfluenceontheresultscompared injected into the hot gas in an episodic way associated with withotherparameters. However,weshouldnotethatthedust- theAGNcycle,ifAGNfeedbackefficientlyheatsthecoldgas to-gas ratio in the hot gas is directly affected by Mg,h since it (Mathewsetal.,2013). directlyentersthe dust-to-gasratioin the hotgas. We discuss Inthispaper,wemakeatheoreticalmodelofdustevolution thevalueofMg,hagaininSection2.3. in AGN feedbackcycles by includingimportantphysicalpro- Thetimeevolutionofthedustmassinthecoldphase, Md,c, cessessuchasthemassexchangebetweenthehotandcoldgas iswrittenas componentsandthedustevolutioninthosegases.Fordustpro- dM M cessing, we consider dust destruction by sputtering in the hot dtd,c =DhM˙in−DcM˙ret+ τ d,c, (2) grow gas and dust growth by the accretion of gas-phase metals in the cold gas. This modelingenablesus to examinehowAGN whereD ≡ M /M andD ≡ M /M arethedust-to-gas h d,h g,h c d,c g,c feedback cycles affect the dust abundancesin elliptical galax- ratiosinthehotandcoldphases,respectively(M andM are d,h d,c ies. Wecanalsoexaminetheeffectofdustgrowthonthedust the dust mass in the hotand cold gas, respectively), and τ grow abundanceinellipticalgalaxiesforthefirsttime. is the time-scale of dust growth by the accretion of gas-phase The paper is organizedas follows: we formulatethe model metalsinthecoldgas.Thedustgrowthtimescaleτ isgiven grow 2 where [x] is the floor function, which indicates the largest in- Table1:Parametersinthemodel. teger that satisfies [x] ≤ x. This equation indicates that the Parameter Unit Fiduciala Rangeb outflowisperiodicallyturnedoneveryτAGN withadurationof M˙in,0 M⊙ yr−1 3 1–10 fretτAGN. τ yr 107 (0.3–3)×107 Theinflowofcooledhotgasisassumedtooccuronlywhen tr τAGN yr 108 fixed thereisnooutflowactivity(i.e.,whenMret =0): f — 0.3 0.1–0.5 ret 0 if0≤t/τ −[t/τ ]≤ f , τ yr 107 106–108 M˙ (t)= AGN AGN ret (7) τacc yr 107 106–108 in ( M˙in,0 otherwise, sput Z — 0.02 fixed wherewetreatM˙ asaconstantfreeparameter. in,0 DAGB — 0.01 fixed Based on the inferred dust lifetime and dynamical time in α yr−1 1.6×10−12 fixed thecenterofanellipticalgalaxyandtherateofdustdetection M∗ M⊙ 1010.5 1010–1011 forellipticalgalaxies,Laueretal.(2005)suggestedthatthecy- cle ofdustappearanceanddisappearanceis∼ 108 yr. We can aFiducialvalue. regard this time as τ if that dust cycle is induced by the bRangeofvaluesinvestigatedunlesstheparametervalue AGNfeedbackcycle.ATGhNus,weadoptτ =108yr. Sincethe AGN isfixed. timeevolutionofgasanddustisdeterminedbythetime-scale relative to τ , we fix τ and change other time-scales as AGN AGN in Section2.3(equation9), andistreatedasafunctionofD . describedbelow. Themaximumcoldgasmassisroughlyesti- c Thetimeevolutionofthedustmassinthehotphase, Mh,c, on matedasM˙in,0τAGN(1−fret),whichshouldbecomparabletothe theotherhand,iswrittenas observedgasmass∼afew×108M⊙(e.g.,Wiklindetal.,1995). dMdtd,h =−DhM˙in+DcM˙ret− Mτ d,h +DAGBαM∗, (3) TinhviessmtigeaatnesathraatnMg˙eino,0f∼∼a1f–e1w0×M(⊙1−yr−fr1e.t)−S1inMce⊙cyor−ld1.gTahsuiss,dwee- sput tectedinasignificantfractionofellipticalgalaxies, f cannot ret beverynearto1. Thus,wefocusonarangeof0< f ≤0.5. where τ is the dust destruction time-scale by sputtering in ret sput thehotgas,D isthedust-to-gasratioinAGBstarwinds,α For τtr, it may be reasonable to adopt the dynamical time. AGB AsmentionedintheIntroduction,Temietal.(2007)suggested isthemasslossratesofAGBstarsperstellarmass,and M∗ is dusttransportbybuoyancyactingontheheatedcoldgasinor- the total stellar mass (αM∗ is the total mass loss rate of AGB stars). UsingM =D M , M =D M ,andequation(1), der to explain the origin of extended dust FIR emission com- d,c c g,c d,h h g,h ponent. Theyestimatedthetime-scaleoftransportas∼ 107 yr we rewite equations(2)and(3)as thefollowingequationsfor based on the dynamicaltime-scale. Thus, we adoptτ ∼ 107 thedust-to-gasratios: tr yr,andalsoexamineanorder-of-magnituderangeforτ . tr dD D M˙ c = c +(D −D ) in , (4) dt τ h c M 2.3. Parametersfordustevolution grow g,c We have introduced two time-scale parameters that govern dDh =−D M˙in − Dh +D M˙ret +D αM∗. (5) thedustevolution: dustgrowthtime-scale(τgrow)anddustde- dt hM τ cM AGBM struction(sputtering)time-scale(τsput). g,h sput g,h g,h We adopt the following estimate of the time-scale of grain We solve equations (1), (4) and (5). Below we formulate growth(Asanoetal.,2013): some undeterminedtermsand estimate a reasonablerangefor eachparameter(Table1). τ = 2×106 a nH,c −1 Tgas −1/2 acc 0.1µm!(cid:18)103cm−3(cid:19) 50K! 2.2. Massexchangebetweenthephases Z −1 × yr, (8) The mass exchangebetween the cold and hot phases is de- (cid:18)0.02(cid:19) scribed by M˙ (inflow of cooled hot gas to the cold gas) and in where a is the grain radius, n is the hydrogennumber den- M˙ (returnofcoldgastothehotphasebyAGNfeedback).For H,c ret sity in the cold gas, T is the gas temperature, and Z is the simplicity, we assume that AGN feedbackoccurs periodically gas metallicity (e.g.,Inoue, 2011; Asanoetal., 2013). Because of with a period of τ . During an episode of AGN feedback, AGN somepoorlyconstrainedparameterssuchasaandn ,wetreat the cold gas is transported to the hot gas, and this transporta- H,c τ as a given constant parameter for simplicity, but investi- tion time-scale is parameterized by τ . This transport (or the acc tr gateawiderangeinτ . Moreover,sinceaccretionisefficient return of cold gas to the hot gas) lasts for a fraction of τ , acc AGN only in the dense medium, it depends on the mass fraction of andthisfractionisparameterizedby f . Thatis, theduration ret densegasinthecoldgas,whichisdenotedas f . Thedust of an AGN feedbackepisode is f τ . We assume that the dense ret AGN growthtime-scaleiseffectivelyestimatedasτ /f . Based outflow(M˙ )ofthecoldgasisonduringtheAGNfeedbackas acc dense ret on the aboveestimate, we examineτ = 106–108 yr (107 yr acc M˙ (t)= Mg,c/τtr if0≤t/τAGN−[t/τAGN]≤ fret, (6) for the fiducial case), allowing for a small value for fdense or ret ( 0 otherwise, variationinaandnH,c(Kuoetal.,2013;Schneideretal.,2016; 3 Aoyamaetal.,2017). Inreality,theaccretiontime-scaleisalso sizedistributionisconsistentwiththeopticalextinctioncurves affected by grain size distribution (KuoandHirashita, 2012); observedbyPatiletal.(2007),althoughtheydidnotconsidera thus,theabovearepresentstheappropriateaverageofthegrain fullcycleofAGNfeedback.Atthesametime,theyalsoshowed size.1 Considering the uncertainties in grain size distribution, thattheresultinggrainsizedistributiondependsonthesizedis- τ maybeoutoftherangeabove(106–108 yr). Ifτ . 106 tributionofdustgrainsproducedbyAGBstars. Toavoidsuch acc acc yr, the grain growth is so efficientthat the dust-to-gasratio in acomplication,wetreatτ andτ asgivenconstantparam- acc sput the cold gasstays at a maximumvaluegiven below(D ) for eters and concentrate on the total dust mass. The grain size sat most of the time regardless of the value of τ (. 106 yr). If distributionrealizedafterAGNfeedbackcycleswillbeinvesti- acc τ &108yr,incontrast,dustgrowthisnotefficientenoughto gatedinthefuturework. acc raise the dust-to-gasratio above10−4. Therefore,the rangeof We adopt DAGB = 0.01, α = 1.6× 10−12 yr−1, and M∗ = τacc = 106–108 yr coversall the resulting behavior of interest 1010–1011 M⊙(TsaiandMathews,1995). However,theprecise for D , i.e., coversallthe interestingrangeof D aswe show valuesofthesequantitiesdonotaffectthedustabundanceinthe c c below. coldgasaslongasthe dustformedin AGBstarsis efficiently Using a given value of τ , the dust growth time-scale is destroyedby thehotgas. We fix D andα, andonlyfocus acc AGB givenby onthevariationofM∗. τgrow = 1−Dτac/cD , (9) forAssimepxlpiclaitiyn.edThinehSoetcgtiaosnm2a.1ss,iwnethfiexcetnhetrahloatfgeawskmpacs,swMhicg,hh c sat coulddirectlyexchangethegasmasswiththecentralcoldcom- where the denominator on the right-hand side means that ac- ponent,isestimatedas108–109 M⊙ (TsaiandMathews,1995). cretionis saturatedas Dc approachesDsat, which isthe abun- Thus, we adopt Mg,h = 3×108 M⊙. The dust-to-gasratio in dance of the metals available for dust formation. We assume thehotgas(D )isinverselyproportionalto M bydefinition; h g,h Dsat = 0.01 in this paper, since we are considering solar however,becausedustinthehotgasisefficientlydestroyedby metallicity environment (i.e., Dsat ∼ Z⊙). The abundance sputtering,asshownbelow,Dh ≪Dc issatisfiedwhenthehot of available metals for dust formationmay be correlated with gasis actively accretedon to the cold gas, exceptfor the case the stellar mass because of the stellar mass–metallicity rela- where the sputtering time-scale is as long as the AGN feed- tion (Gallazzietal., 2005). However, this relation has a large backcycle(τ &τ ). Therefore,theinflowispracticallya sput AGN scatter, anditisalsosuggestedthatthevelocitydispersion(or “dust-free”gasforthecoldgas,sothatthevalueofD doesnot h the virial mass) is more tightly related to the stellar metallic- influenceD . Thismeansthatthechoiceof M ,whichcould c g,h ity (Smithetal., 2009). Moreover, it is not clear how tightly affectD ,doesnothaveasignificantinfluenceonD . h c the stellar metallicity is related to the gas metallicity because We assume the following initial conditions: D = D = c h of accretion of external gas (SuandIrwin, 2013). Thus, we 10−6, and Mg,c = 107 M⊙. As seen later, the initial condition simply adopt a constant Dsat, but we also note that increas- isnotimportant,sincethesystem“forgets”theinitialcondition ing/decreasing Dsat has broadly the same effect of short/long in∼τAGN. τ . acc The time-scale of dust destruction by sputtering is es- timated as (TsaiandMathews, 1995; Nozawaetal., 2006; 3. Results Hirashitaetal.,2015) 3.1. Fiducialcase a n −1 τsput =7.1×106 0.1µm!(cid:18)10−2Hc,hm−3(cid:19) yr, (10) Firstofall,weshowthetimeevolutionofrelevantquantities underthefiducialparametervalueslistedinTable1. InFig.1, wheren isthenumberdensityofhydrogennucleiinthehot weshowtheevolutionsofthedust-to-gasratiosinthehotand H,h gas. Aswedidforτ above,wetreatτ asagivenconstant coldgas,thecoldgasmass,andthedustmassesinthecoldand acc sput parameter. Based on this estimation, we adopt τ = 107 yr hotgas. Herewedescribehoweachofthesequantitiesevolves sput forthefiducialcaseandexaminearangeof106–108yrconsid- asafunctionoftime. ering the variety in a and n . Note that fixing τ and τ Allthequantitieshavealmostperiodicbehaviorswithape- H,h acc sput implicitlyassumesafixedgrainsize(orgrainsizedistribution). riodgivenbyτ . InassociationwiththeonsetofAGNfeed- AGN In reality, the grain size distribution itself is determined as a back,thedustgrowninthecoldgasisinjectedintothehotgas, result of the cycle of dust destruction and growth. Therefore, sothatthedust-to-gasratiointhehotgasincreases. However, a consistent treatment of those time-scales and the grain size the dust supplied to the hot gas is quickly destroyed by sput- distributionwouldbenecessary. Hirashitaetal.(2015)consid- tering, and the dust-to-gasin the hot gas tends to convergeto eredtheevolutionofgrainsizedistributionbydustdestruction thevaluedeterminedbythebalancebetweendustsupplyfrom andsubsequentdustgrowthandshowedthattheresultinggrain stars and dustdestructionby sputtering. After AGN feedback stops,thecoldgasincreasesitsmassbygasinfall.Inthisphase, becausethedust-to-gasratiointheinfallinggas,whichisequal 1More precisely, a is the ratio of the mean a3 to the mean a2 to the dust-to-gas ratio in the hot gas, is smaller than that in (HirashitaandKuo,2011). As shownin HirashitaandKuo(2011), the time thecoldgas,thedust-to-gasratiointhecoldgasdecreasesasa variationofacouldbeeffectivelyincorporatedinthedefinitionofτaccsothat τacccouldstillbetreatedasconstant. resultofdilution. 4 Figure1: Timeevolutions ofrelevant quantities. (a)Dust-to-gas ratiointhe Figure2:Timeevolutionsofrelevantquantities.(a)Dust-to-gasratiointhehot cold(Dc)andhotgas(Dh)(solidanddottedlines,respectively). (b)Coldgas gas(Dh).(b)Dust-to-gasratiointhecoldgas(Dc).(c)Coldgasmass(Mg,c). mass(Mg,c)(solidline),anddustmassesinthecoldgas(Md,c)(dashedline) Thesolid,dotted,anddashedlinesshowtheresultsforM˙in,0 =3(fiducial),1, andinthehotgas(Mh,c)(dottedline). and10M⊙,respectively,inallthepanels. The above general behavior does not change by the choice expected, the cold gas mass is larger if M˙ is larger, but the in,0 of parameter values. Thus, for the purpose of simplification, logarithmicamplitudeofthecoldgasmassdoesnotdependon we hereafter concentrate on the dust-to-gas ratios in the cold M˙ , sincewefixthetime-scalesofAGNfeedbackandmass in,0 and hotgas (D and D ), and the cold gas mass (M ), since transport. c h g,c thedustmassesinthehotandcoldgasjusttracetheevolution of the corresponding dust-to-gas ratios. Below we show the 3.3. Masstransporttime-scaleinAGNfeedback dependenceoneachparameterlistedinTable1. The mass transport time-scale of AGN feedback, τ , regu- tr lates the mass loss rate of the cold gas when AGN feedback 3.2. Massinflowrate ison(equation6). InFig.3,weshowthetimeevolutionofthe WeconsiderthemassaccretionontothecoldgaswhenAGN quantitiesofinterestforτ =107(fiducial),3×106,and3×107 tr feedbackisoff.Themassinflowrateisregulatedbytheparam- yr. eter M˙ . In Fig. 2, we show the time evolution of D , D , in,0 c h Theincreaseanddecreaseofthedust-to-gasratiointhehot and Mg,c forM˙in,0 =3(fiducial),1,and10 M⊙ yr−1. Belowwe gas is sharper if τ is shorter just because the effect of mass describehowtheresultsdependonM˙ . tr in,0 transportfromthecoldtohotgasoccursonashortertime-scale. The dust-to-gas ratio in the hot gas has a larger amplitude Theamplitudesofallthe quantitiesshowninFig. 3 arelarger foralarger M˙ . Thisisbecausemoredustisaccumulatedin in,0 for shorter τ because larger mass is exchanged between the tr the coldgas. Whenthisdustycoldgasisinjectedintothehot coldandhotphases.Inparticular,theamplitudeofthecoldgas gas by AGN feedback, the hot gasis enrichedwith dust. The massissensitivetoτ . Accordingly,thedust-to-gasratiointhe tr dustinjectedintothehotgasisreadilydestroyedbysputtering; coldgasdropsmoreforshorterτ becausetheeffectofdilution tr thus, the hot gas is dust-poorfor most of the time. The dust- by the inflow is more significant. In contrast, the dust-to-gas to-gasratio in the hotgas dropsslightly morein the case of a ratiointhehotgasdropstothevalueinsensitivetoτ whenthe tr higherMin,0becausetheinflowtransportsthedustfromthehot AGNfeedbackisoff(i.e.,whenthereisnosupplyofdustfrom gastothecoldgasinourmodel. However,thelowestlevelof thecoldgas)becauseofsputtering. ThelevelofthelowestD h thedust-to-gasratiointhehotgasisnotverysensitiveto M in,0 is determinedbythe balancebetweendustformationbyAGB sinceitisdeterminedbythebalancebetweendustsupplyfrom starsanddustdestructionbysputtering. AGB stars and dust destructionby sputtering. In terms of the dust-to-gas ratio in the cold gas, the periodic decrease is due 3.4. DurationofAGNfeedback to the inflow of dust-poorgas. However, since dust growthis efficientenough,theoriginalhighdust-to-gasratioisrecovered ThedurationofAGNfeedbackrelativetotheAGNdutycy- as soonasthe coldgasattainsenoughmassbythe inflow. As cleisparametrizedby f ,sothatanepisodeofAGNfeedback ret 5 Figure5:SameasFig.2butforvarioustime-scalesofdustgrowthbyaccretion (τacc). The solid, dotted, dashed, and dot-dashed lines show the results for Figure3: SameasFig.2butforvariousmasstransporttimes(τtr). Thesolid, τacc=107(fiducial),106,3×107and108yr,respectively,inallthepanels. dotted,anddashedlinesshowtheresultsforτtr =107(fiducial),3×106,and 3×107M⊙,respectively,inallthepanels. lastsfor f τ . InFig.4,weshowthetimeevolutionofthe ret AGN quantitiesofinterestfor f =0.3(fiducial),0.1,and0.5. ret ComparingFigs. 3and4, we findthatthedurationofAGN feedback has an effect similar to the mass transport time in terms of the amplitudes: a larger f makes the amplitudesof ret thecoldgasmassandthedust-to-gasratiointhecoldgaslarger. The cold gas naturally decreases more if AGN feedback lasts longer, and the subsequent dilution of dust abundance by in- flowbecomesmoresignificant. Thedust-to-gasratiointhehot gashasa largeramplitudefor a smaller f since thecold gas ret massislarger(i.e.,thedustmasssuppliedfromthecoldgasis larger). 3.5. Dustgrowthtime-scale Theaccretiontime-scaleτ regulatesthedustgrowthinthe acc coldgas.InFig.5,weshowthetimeevolutionofthequantities ofinterestforτ =107(fiducial),106,3×107,and108yr. acc Naturally,thecoldgasmass,whichisnotrelatedtodust,is not affected by τ . As expected, the dust-to-gasratio in the acc cold gas becomes higher for shorter τ . Moreover, if τ is acc acc as long as τ , the overall level of dust-to-gas ratio is sup- AGN pressedbecausedustdoesnotgrowsufficientlywithinanAGN feedbackcycle. Thedust-to-gasratioin thehotgasisalsoaf- fected,sincethemajorsourceofdustinthehotgasisthedust growninthecoldgas. WeobserveasecularincreaseofD for c τ = 3× 107 yr. This indicates that, if τ is smaller than acc acc Figure4: SameasFig.2butforvariousratiosoftheAGNfeedbackduration τ ,thedust-to-gasratiointhecoldgastendstoincreaseun- totheAGNfeedbackdutycycletime(fret).Thesolid,dotted,anddashedlines leAsGsNitreachesthesaturationlimit(D ).Therefore,weobserve showtheresultsfor fret = 0.3(fiducial), 0.1,and0.5,respectively, inallthe sat panels. abifurcationforthevalueofDc:ifτacc <τAGN,thedust-to-gas ratioincreaseswhileifτ >τ ,thedust-to-gasratioissup- acc AGN pressedandD ∼ D . Inthelattercase,dustgrowthhaslittle c h 6 Figure 6: Same as Fig. 2 but for various time-scales of dust destruction by Figure7:SameasFig.2butforvariousstellarmasses(M∗).Thesolid,dotted, sputtering (τsput). The solid, dotted, and dashed lines show the results for anddashedlinesshowtheresultsforM∗=1010.5(fiducial),1010,and1011M⊙, τsput=107(fiducial),106,and108yr,respectively,inallthepanels. respectively,inallthepanels. influenceonthedustbudgetandthedust-to-gasratioissimply isdeterminedbythebalancebetweendustsupplyanddustde- determinedbythebalancebetweendustsupplyfromAGBstars structionisproportionaltothestellarmass.Themaximumlevel anddustdestructioninthehotgas. ofthedust-to-gasratiointhehotgasisinsensitivetothestellar mass, because it is determined by the dust injection from the coldgas.Thedust-to-gasratiointhecoldgasisnotsensitiveto 3.6. Dustdestructiontime-scale thestellarmassaslongasdustgrowthisefficientenough. Dustinthe hotgasisdestroyedbysputtering. Thedestruc- tiontime-scaleisregulatedbyτ . InFig.6,weshowthetime sput evolution of the relevant quantities for τ = 107 (fiducial), 4. Discussion sput 106,and108yr. In this section, we mainly compare the theoretical calcula- Naturally, the cold gas mass, which is not related to dust, tionswithobservationaldata.Hereweutilizetheobservational is not affected by τ . As expected, the dust-to-gas ratio in sput datafordustandgasinellipticalgalaxies. Becauseofthegen- thehotgasislargelyaffectedbythesputteringtime-scale,with eraldeficiencyofgasanddustinellipticalgalaxies,dataislim- longer τ predictinghigherD . The dust-to-gasratio in the sput h ited, while the stellar mass is relativelyeasy to be determined hot gas is roughly proportionalto τ (i.e., inversely propor- sput inthesesystemsusingopticalphotometricobservations. Thus, tionalto thedustdestructionefficiency). Thedust-to-gasratio wealsoadoptthestellarmassandutilizeit(mainlyfornormal- inthecoldgasisalsokepthighforτ =108yr:ifD remains sput h ization). ashighasD ,thedust-to-gasratiointhecoldgasiskepthigh c becausethedust-to-gasratiointheinfallinggasishigh(i.e.,the 4.1. Sample dustabundanceinthecoldgasisnotdilutedbytheinfall).This Oneofthenewestsystematicdustobservationsinnearbyel- effectissignificantonlyifτ &τ . sput AGN lipticalgalaxiesiscarriedoutbyHerschel.Weadoptthesample from Smithetal. (2012) and diSeregoAlighierietal. (2013). 3.7. Stellarmass ThesamplegalaxiesadoptedarelistedinTable2. Weselected Thetotalstellarmassaffectsthedustevolutionthroughdust galaxieswithamorphologicaltypeofellipticalgalaxies(E)and productionbyAGBstars. InFig.7,weshowthetimeevolution excludedobjectswithstellarmasswith<109.5 M⊙. NGC4374 of the relevantquantitiesfor M∗ = 1010.5 (fiducial), 1010, and is common for both samples. The stellar mass is taken from 1011 M⊙. Corteseetal.(2012). ThegasmassesaretracedbyHiandH2 The stellar mass does not affect the cold gas mass in our (CO);therefore,theobservedgasmassesarecomparedwiththe modelsinceitonlycontributestothedustproduction.Because cold gas mass in our models. We only compare the observed the dust supply from AGB stars to the hot gas is proportional dust mass with the dust mass in the cold gas in our models, to the stellar mass, the lowest levelof dust-to-gasratio which because the observationalsensitivity to the diffuse component 7 in the hot gas depends on the spatial extension, which is not which affect the relative abundance of dust to gas (i.e., dust- known.Inanycase,sincethedustmassinthecoldgasismuch to-gasratio). Thoseparametersareτ , f , τ andτ . The tr ret acc sput larger than that in the hot gas in most of the time, adding the samerangesfortheparametersasabove(i.e.,therangesinTa- dustmassinthehotgasdoesnotsignificantlychangethecom- ble1)areinvestigated.Weonlyplottherelationatt>2×108yr, parisonbelow. whenthetrajectoryalmostconvergestoalimitcycle(by“for- getting the initial condition”) except for the case with τ = acc 4.2. Coldgasmass 3 × 107 yr in Panel (c). In this case, there is a secular drift towardhighdustabundanceasshowninFig.5. ThevariationofthecoldgasmassinourmodelsinSection3 Overall,weobserveinFig.8thatthevariationisdominated is broadly in the range of the observed gas masses shown in Table 2 (∼ 109 M⊙ down to ∼ 107 M⊙ or less). Therefore, bythechangeofthegasmassbecausethevariationismostlyin thediagonaldirectiononthediagram.Ifthegasmasschanges, we confirm that the choices of parameter values in Section 2 are reasonable as far as the cold gas mass is concerned. The both Md,c/M∗ and Mg,c/M∗ moves on the diagonal line under a constantD . Inotherwords, thevariationofthedust-to-gas largestcoldgasmassisrealizedifwe adoptthelargestinflow c rate (M˙in,0 = 10 M⊙ yr−1; Fig. 2). Therefore,we confirmthat ratio can be seen in off-diagonal behavior. Indeed, such off- diagonalbehavioris seen clearly for a short τ in Fig. 8a and themaximuminflowratebytheaccretionofthecooledgasis. tr 10M⊙yr−1. Theamplitudeofthecoldgasmassis,ontheother a large fret in Fig. 8b. In bothcases, the cold gasmass is lost drastically in the epochof AGN feedback,so thatthe dilution hand, determined by the time-scale of gas transport in AGN of dust abundance by the inflow of dust-poor gas has a large feedback (Fig. 3) and the duration of AGN feedback(Fig. 4). impactonthedust-to-gasratiointhecoldgas(Sections3.2and If the variationof the gasmass in the sample is interpretedas due to time evolution, τ & 3 × 107 yr or f . 0.1 is not 3.4). tr ret We observe in Fig. 8 that a single model doesnot cover an favoredbecausethevariationofthecoldgasissmallerthanan order of magnitude. The non-detection of Hi and H gas for area wide enoughto explain all the data points. In particular, 2 a large part of the sample galaxies is consistent with even the thedatapointsathighMg,c/M∗andlowMd,c/M∗aredifficultto mostextremecaseofτ =3×106yror f =0.5. Thismeans bereproduced.Asshownabove,thedilutioneffectappearsjust tr ret after the epoch of AGN feedback, when the cold gas mass is that it is difficult to constrain the lower bound for τ and the tr low. Therefore,itisdifficulttoreproducethelowdustcontent upperboundfor f . ret at the high gas mass. In our scenario, the only way to repro- ducesucha dust-poorandgas-richdata pointis to changethe 4.3. Dustmass efficiencyofaccretionasshowninFig.8c. Thesameeffectis The dust mass in the cold gas varies between ∼ 105 and ∼ alsorealizedbyloweringD (equation9).Therefore,thelarge 106M⊙inthefiducialmodel(Fig.1).Themaximumofthedust dispersionofdustmassathsaitghgasmasscanbeinterpretedas mass is achieved when the cold gas mass becomes the largest variousdustgrowthefficiencies. Avarietyindustgrowtheffi- (i.e., just before AGN feedback). In this phase, the dust-to- ciencycouldbecausedbyavarietyindensegasfraction,cold gasratioisalmostsaturatedifτ issignificantlyshorterthan acc gasdensity,metallicity,orgrainsizedistribution(Section2.3). τ . Ifwe adoptthemaximumvalueofthedust-to-gasratio AGN Somestudiesarguethatthedust-to-stellarmassratiochanges (D =0.01),weobtainthemaximumdustmassinthecaseof sat as the system is enriched with dust (e.g., Re´my-Ruyeretal., thelargestcoldgasmass(109M⊙;seeabove)as∼107M⊙.The 2015). Dustenrichmentisindeedthemostimportantfactorof observeddustmassisindeedsmallerthanthisvalue. Because changingthedust-to-stellarmassratioingas-richstar-forming thelogarithmicvariationofthedust-to-gasratioissmallerthan galaxies,sincedustproductionbystarsanddustgrowthinthe thatofthecoldgasmass,thetimeevolutionofthedustmassis cold denseISM are activelyoccurring. In thispaper, we have dominatedbythechangeofthecoldgasmassinourmodels. shownthat, inellipticalgalaxies, thedust-to-stellarmassratio is still varied by dust growth, although it is driven by a cycle 4.4. Dust-to-stellarmassratio ofAGNfeedback,notbychemicalenrichment. Therefore,our Tocanceloutthesizeeffectsofellipticalgalaxies,itisuseful results imply that, if AGN feedback affects the galaxy evolu- to normalizethe dustmass to an indicatorof totalmass scale. tion, we also needto considerthevariationofdustabundance Herewe usethestellarmassforthenormalization;thatis, we byAGNfeedback. usethedust-to-stellarmassratiofortheindicatoroftherichness of dust. As an indicator of AGN feedback, we also show the 4.5. Dust-to-gasratio ratioofthecoldgasmasstothestellar mass, andexaminethe As shown above, the behavioron the Md,c/M∗–Mg,c/M∗ di- relation between Md,c/M∗ and Mg,c/M∗. For the data points, agram is dominated by the overall oscillation of the cold gas weexcludetheobjectswithoutanyconstraintontheHimass, mass. Toexaminetheeffectofdustabundancevariationmore sinceitisimpossibletoconstrainthegasmassinsuchacase.If clearly,itisusefultoconsideraquantitynormalizedtothegas Hiisdetected,wesumtheHimassandtheH masstoobtain mass. In this way, we can cancel the effect of oscillatory be- 2 thetotalcoldgasmass(ifH isnotdetected,weaddtheupper havior of the cold gas mass, and concentrate on the variation 2 limitofH gasmasstotheHigasmass). of dust abundance. We still use the gas-to-stellar mass ratio, 2 In Fig. 8, we show the relation between Md,c/M∗ and Mg,c/M∗, for the horizontal axis to show the cycle of AGN Mg,c/M∗. We only show the dependence on the parameters feedback. Thus, we consider the Dc–Mg,c/M∗ diagram here. 8 Table2:EllipticalgalaxysampleadoptedfromSmithetal.(2012)anddiSeregoAlighierietal.(2013). Name Othername logM∗ logMdust logMHI logMH2 logD Ref.a (M⊙) (M⊙) (M⊙) (M⊙) VCC763 NGC4374,M84 11.18 5.05b 8.96 <7.23 −3.91c 1,2,3 HRS150 NGC4406,M86 11.22 6.63 7.95 <7.4 −1.43c 1,3 HRS186 NGC4494 10.88 5.08 8.26 <7.35 −3.23c 1,3 HRS241 NGC4636 10.98 5.06 9.0 <7.02 −3.94c 1,3 VCC345 NGC4261 11.32 5.81 <8.45 <7.70 >−2.71d 2,3 VCC881C NGC4406 11.22 5.47 7.95 — −2.48e 2,3 VCC1226 NGC4472,M49 11.39 5.49 <7.90 <7.26 >−2.50d 2,3 VCC1619 NGC4550 10.03 5.41 <7.90 7.20 >−2.57d 2,3 HRS3 NGC3226 10.21 5.96 — <7.13 — 1,3 HRS43 NGC3608 10.27 <4.88 — <7.28 — 1,3 HRS49 NGC3640 10.70 <5.20 — <7.25 — 1,3 HRS125 NGC4339 10.30 <5.17 <7.84 <7.2 — 1,3 HRS135 NGC4365 11.48 <6.17 <8.18 <7.62 — 1,3 HRS179 NGC4473 10.71 <5.19 <7.92 <7.16 — 1,3 HRS181 NGC4478 10.09 <4.85 — <7.31 — 1,3 HRS211 NGC4552,M89 10.80 <5.67 <7.92 <7.36 — 1,3 HRS214 NGC4564 10.25 <5.30 <7.79 <7.33 — 1,3 HRS218 NGC4570 10.48 <5.67 <7.31 <7.47 — 1,3 HRS236 NGC4621,M59 10.99 <5.76 <7.92 <7.24 — 1,3 HRS245 NGC4649,M60 11.34 <5.40 <7.92 <7.59 — 1,3 HRS248 NGC4660 10.05 <4.85 <7.92 <7.30 — 1,3 HRS258 NGC4697 11.10 5.46 — <7.25 — 1,3 HRS312 NGC5576 10.60 <5.51 — <7.46 — 1,3 HRS316 NGC5638 10.52 <5.23 — <7.54 — 1,3 VCC1316 NGC4486,M87 11.24 5.34 — <7.18 — 2,3 VCC1327 NGC4486A 10.05 4.92 — — — 2,3 aReferences:1)Smithetal.(2012);2)diSeregoAlighierietal.(2013);3)Corteseetal.(2012). bAdoptedfromRef.1. Ref.2gives5.30withthesamemassabsorptioncoefficient. cTheupperlimitofH massisused,butthisdoesnotaffecttheestimateddust-to-gasratiosincethegasmassisdominatedby 2 Higasmass. dWeusetheupperlimitsofgasmasstoderivethelowerlimitsofdust-to-gasratio. eWeneglecttheH massforthegasmass. 2 9 Figure8:Relationbetweendust-to-stellarmassratio(Md,c/M∗)andgas-to-stellarmassratio(Mg,c/M∗).(a)Dependenceonτtr.Thethick(green),medium(blue) andthin(red)linesrepresentthetrajectoryforτtr=107(fiducial),3×106,and3×107yr,respectively. (b)Dependenceon fret.Thethick(green),medium(blue), andthin(red)linesshowthetrajectoryfor fret=0.3(fiducial),0.5,and0.1,respectively. (c)Dependenceonτacc.Thethick(green),mediumthick(blue),medium thin(red),andthin(brown)linesrepresentthetrajectoryforτacc=107(fiducial)106,3×107,and108yr,respectively.(d)Dependenceonτsput.Thethick(green), medium(blue),andthin(red)linesshowthetrajectoryforτsput =107(fiducial),106,and108yr,respectively. Thesethreelinesarealmostoverlapped. Wealso showobservationaldatapointsinallthepanelslistedinTable2: Thediamondsshowthegalaxiesforwhichbothdustmassandgasmassaredetected,whilethe crosseswitharrowspresentthedatapointsforwhicheitherdustmassorgasmassisnotdetected(onlyanupperlimitisobtained). Thecrossontherightlower cornershowsthetypicalerrorfortheobservationaldatapoints. 10