The metallicity and elemental abundance maps of kinematically atypical galaxies for constraining minor merger and accretion histories
MMon. Not. R. Astron. Soc. , 1–10 () Printed 6 March 2019 (MN L A TEX style file v2.2)
The metallicity and elemental abundance maps of kinematicallyatypical galaxies for constraining minor merger and accretionhistories
Philip Taylor , (cid:63) , Chiaki Kobayashi , and Christoph Federrath Research School of Astronomy and Astrophysics, Australian National University, Canberra, ACT 2611, Australia ARC Centre of Excellence for All Sky Astrophysics in 3 Dimensions (ASTRO 3D), Australia Centre for Astrophysics Research, School of Physics, Astronomy and Mathematics, University of Hertfordshire, Hatfield, AL10 9AB, UK
Accepted Received ; in original form
ABSTRACT
Explaining the internal distribution and motions of stars and gas in galaxies is a key aspectin understanding their evolution. In previous work we identified five well resolved galaxieswith atypical kinematics from a cosmological simulation; two had kinematically distinct cores(KDCs), and three had counter-rotating gas and stars (CRGD). In this paper, we show thati) the KDC galaxies have flattening of stellar [O/Fe] at large galacto-centric radii due to theminor mergers that gave rise to the KDCs, and ii) the CRGD galaxies have an abrupt transitionin the gas metallicity maps, from high metallicity in the centre to very low metallicity furtherout. These galaxies are embedded in dark matter filaments where there is a ready supply ofnear-pristine gas to cause this effect. The non-linear increase in gas metallicity is also seen inthe radial profiles, but when the metallicity gradients are measured, the difference is buriedin the scatter of the relation. We also find that all five galaxies are fairly compact, with smalleffective radii given their stellar masses. This is because they have not experienced majormergers that kinematically heat the stars, and would have destroyed their unusual kinematics.In order to detect these signatures of minor mergers or accretion, the galaxy scaling relationsor radial metallicity profiles are not enough, and it is necessary to obtain the 2D maps withintegral field spectroscopy observations.
Key words: galaxies: kinematics and dynamics – galaxies: evolution – methods: numerical.
Galaxies evolve on timescales much longer than there have beenastronomers to observe them. Therefore, we see only a single in-stant in their evolution. By compiling catalogues of many galaxiesof different types – that is, at different stages of evolution – acrosscosmic time, we can hypothesise how galaxies grow and evolve.While gross properties, such as mass and morphology, are useful,the internal distribution and motion of stars and gas are vital tounderstand the history of an individual galaxy, and dynamical dis-turbances, which are believed to be formed by mergers, have beenused to understand the formation of early-type galaxies (e.g., Ko-rmendy & Djorgovski 1989; Schweizer et al. 1990; Schweizer &Seitzer 1992; Bender & Surma 1992).Just as important is the distribution of chemical elements;these provide a fossil record of the galaxy from when they arelocked up in star formation, and chemical abundance ratios provideinformation on the star formation history (e.g., Kobayashi 2016).Galactic archaeology surveys such as Gaia-ESO (Gilmore et al. (cid:63)
E-mail: [email protected] ∼ mil-lion stars in the Milky Way, offering unprecedented insight intothe history of our Galaxy. In more distant galaxies, such measure-ments of individual stars are not possible, but long-slit or integral-field spectroscopy (IFS) can provide the spatial distribution and ra-dial profiles of gas and stellar metallicity (e.g., Faber 1973; Davieset al. 1993; Spolaor et al. 2009, 2010; Kuntschner et al. 2010). Thegradient of these metallicity profiles provides information on themerging history and previous activity of the active galactic nucleus(AGN) of the galaxy (e.g., Kobayashi 2004; Taylor & Kobayashi2017).In recent years, IFS has superseded long-slit spectroscopy asthe standard tool for analysing the internal structure of galaxies inthe local Universe. Several surveys of hundreds or thousands ofgalaxies now exist, including SAURON (de Zeeuw et al. 2002),ATLAS (Cappellari et al. 2011), CALIFA (S´anchez et al. 2012),SAMI (Croom et al. 2012; Green et al. 2018), SLUGGS (Brodieet al. 2014), MASSIVE (Ma et al. 2014), S7 (Dopita et al. 2015;Thomas et al. 2017), and MaNGA (Yan et al. 2016), with the nextgeneration of IFS surveys such as Hector (Bland-Hawthorn 2015) c (cid:13) RAS a r X i v : . [ a s t r o - ph . GA ] M a r P. Taylor, C. Kobayashi, and C. Federrath set to observe 100,000 galaxies. By fitting single stellar population(SSP) models convolved with a line of sight velocity distribution toevery spectrum, maps of line of sight velocity, velocity dispersion,and SSP parameters such as metallicity, [ α /Fe], and age can be pro-duced. Furthermore, gas emission lines can also be fitted to yieldionised gas kinematics and abundance maps from nearby to distantgalaxies (e.g., Stott et al. 2014; Wisnioski et al. 2015).In most galaxies with large-scale, ordered motion, the starsand gas are co-planar and orbit in the same direction. A signifi-cant minority, however, are found to have misaligned gas and stel-lar kinematics due to the accretion of extra-galactic gas (e.g., Daviset al. 2011; Bryant et al. 2019). A still smaller fraction host a kine-matically distinct core (KDC); a change with radius of the positionangle of the kinematic axis in stellar velocity maps (e.g., Bender &Surma 1992; Krajnovi´c et al. 2008, 2011). There have been severaltheoretical studies attempting to replicate and explain the formationof KDCs, finding that gas-rich, equal-mass mergers are required toproduce a KDC (Jesseit et al. 2007; Hoffman et al. 2010; Bois et al.2011; Khochfar et al. 2011; Naab et al. 2014). Large-scale cosmo-logical simulations have recently been used to investigate galaxykinematics, but the detailed origin of KDCs has not yet been ad-dressed (Penoyre et al. 2017; Schulze et al. 2018; van de Sandeet al. 2018).In Taylor et al. (2018, hereafter Paper I), we presented fivegalaxies from our cosmological simulation (Taylor & Kobayashi2015a) that displayed atypical kinematics; three had counter-rotating gas and stars, and two had a KDC. These KDCs did notform due to major mergers, as in previous theoretical studies. TheKDC formation mechanism can be summarised as follows: i) thegalaxy forms in the field with co-rotating gas and stars; ii) thegalaxy falls into a dark matter filament, where the relative orien-tation of the stellar angular momentum and bulk gas flow leadsto the formation of a counter-rotating gas disc (CRGD); and iii)a minor merger, with a galaxy on a retrograde orbit compared tothe stellar angular momentum of the primary, deposits counter-rotating stars in the outskirts of the primary (we note that Bas-sett et al. 2017 found a similar result using idealised simulationsof disc galaxy mergers with a mass ratio of 1/10). The result-ing KDCs were larger than observed, and were not clearly dis-cernible in luminosity-weighted kinematic maps, which are a closeranalogue to kinematic maps generated from IFS data than mass-weighted maps. We suggest that this is a separate class of KDCs,which has not yet been identified in observations.The aim of this paper is to analyse the maps of stellar pop-ulations and gas and stellar chemistry that are comparable to IFUdata, and to determine the distinguishing features of these galaxies,beyond their kinematics, that set them apart from the wider galaxypopulation. In Section 2 we describe our simulation and analysistechniques, and introduce the galaxies with atypical kinematics inSection 2.3. We compare the properties of these galaxies to the fullpopulation of simulated galaxies in Section 3 before presenting ourconclusions in Section 4. The simulation used in this paper is a cosmological, chemodynami-cal simulation, first introduced in Taylor & Kobayashi (2015a). Oursimulation code is based on the smoothed particle hydrodynamics(SPH) code
GADGET -3 (Springel 2005), updated to include: star formation (Kobayashi et al. 2007), energy feedback and chemicalenrichment from supernovae (SNe II, Ibc, and Ia, Kobayashi 2004;Kobayashi & Nomoto 2009) and hypernovae (Kobayashi et al.2006; Kobayashi & Nakasato 2011), and asymptotic giant branch(AGB) stars (Kobayashi et al. 2011); heating from a uniform,evolving UV background (Haardt & Madau 1996); metallicity-dependent radiative gas cooling (Sutherland & Dopita 1993); and amodel for black hole (BH) formation, growth, and feedback (Tay-lor & Kobayashi 2014), described in more detail below. We usethe initial mass function (IMF) of stars from Kroupa (2008) in therange . −
120 M (cid:12) , with an upper mass limit for core-collapsesupernovae of
50 M (cid:12) .The initial conditions for the simulation consist of parti-cles of each of gas and dark matter in a periodic, cubic box h − Mpc on a side, giving spatial and mass resolutions of . h − kpc and M DM = 7 . × h − M (cid:12) , M gas = 1 . × h − M (cid:12) ,respectively. This resolution is sufficient to spatially resolve struc-ture within massive galaxies. We employ a WMAP-9 Λ CDM cos-mology (Hinshaw et al. 2013) with h = 0 . , Ω m = 0 . , Ω Λ =0 . , Ω b = 0 . , and σ = 0 . .BHs form from gas particles that are metal-free and denserthan a specified critical density, mimicking the most likely forma-tion channels in the early Universe as the remnant of Population IIIstars (e.g., Madau & Rees 2001; Bromm et al. 2002; Schneider et al.2002) or via direct collapse of a massive gas cloud (e.g., Bromm& Loeb 2003; Koushiappas et al. 2004; Agarwal et al. 2012; Be-cerra et al. 2015; Regan et al. 2016; Hosokawa et al. 2016). TheBHs grow through Eddington-limited Bondi-Hoyle gas accretionand mergers. Two BHs merge if their separation is less than thegravitational softening length and their relative speed is less thanthe local sound speed. A fraction of the energy liberated by gas ac-cretion is coupled to neighbouring gas particles in a purely thermalform.In previous works, we have compared the simulation used inthis paper with another having the same initial conditions, but with-out the inclusion of any BH physics. We showed that our model ofAGN feedback leads to simulated galaxies whose properties moreclosely match those of observed galaxies, both at the present dayand high redshift (Taylor & Kobayashi 2015a, 2016, 2017), andquantified the effects of AGN feedback on the host galaxy and itsimmediate environment (Taylor & Kobayashi 2015b; Taylor et al.2017). IFU observations of galaxies provide individual spectra across theface of the galaxy. From each spectrum, average gas and stellarmetallicities and chemical abundances can be extracted, as well asstellar age, and gas and stellar kinematics. In our simulation, theseproperties are tracked for each gas and star particle, and we producemaps of these quantities using the following method.Galaxies are rotated so that the net angular momentum of theirstars lies along the z -axis (i.e., along the line of sight), in keepingwith the similar analysis of Paper I. We smooth the properties ofindividual particles over pixels on a regular grid. Specifically, theaverage value of a quantity P in the j th pixel is given by (cid:104) P (cid:105) j = (cid:80) i P i f ij w i (cid:80) i f ij w i , (1)where the sum is over all particles of interest, typically the gas orstars of a galaxy, f ij denotes the fractional contribution of parti-cle i to pixel j , and the weights w i are either particle mass, or, for c (cid:13) RAS, MNRAS000
50 M (cid:12) .The initial conditions for the simulation consist of parti-cles of each of gas and dark matter in a periodic, cubic box h − Mpc on a side, giving spatial and mass resolutions of . h − kpc and M DM = 7 . × h − M (cid:12) , M gas = 1 . × h − M (cid:12) ,respectively. This resolution is sufficient to spatially resolve struc-ture within massive galaxies. We employ a WMAP-9 Λ CDM cos-mology (Hinshaw et al. 2013) with h = 0 . , Ω m = 0 . , Ω Λ =0 . , Ω b = 0 . , and σ = 0 . .BHs form from gas particles that are metal-free and denserthan a specified critical density, mimicking the most likely forma-tion channels in the early Universe as the remnant of Population IIIstars (e.g., Madau & Rees 2001; Bromm et al. 2002; Schneider et al.2002) or via direct collapse of a massive gas cloud (e.g., Bromm& Loeb 2003; Koushiappas et al. 2004; Agarwal et al. 2012; Be-cerra et al. 2015; Regan et al. 2016; Hosokawa et al. 2016). TheBHs grow through Eddington-limited Bondi-Hoyle gas accretionand mergers. Two BHs merge if their separation is less than thegravitational softening length and their relative speed is less thanthe local sound speed. A fraction of the energy liberated by gas ac-cretion is coupled to neighbouring gas particles in a purely thermalform.In previous works, we have compared the simulation used inthis paper with another having the same initial conditions, but with-out the inclusion of any BH physics. We showed that our model ofAGN feedback leads to simulated galaxies whose properties moreclosely match those of observed galaxies, both at the present dayand high redshift (Taylor & Kobayashi 2015a, 2016, 2017), andquantified the effects of AGN feedback on the host galaxy and itsimmediate environment (Taylor & Kobayashi 2015b; Taylor et al.2017). IFU observations of galaxies provide individual spectra across theface of the galaxy. From each spectrum, average gas and stellarmetallicities and chemical abundances can be extracted, as well asstellar age, and gas and stellar kinematics. In our simulation, theseproperties are tracked for each gas and star particle, and we producemaps of these quantities using the following method.Galaxies are rotated so that the net angular momentum of theirstars lies along the z -axis (i.e., along the line of sight), in keepingwith the similar analysis of Paper I. We smooth the properties ofindividual particles over pixels on a regular grid. Specifically, theaverage value of a quantity P in the j th pixel is given by (cid:104) P (cid:105) j = (cid:80) i P i f ij w i (cid:80) i f ij w i , (1)where the sum is over all particles of interest, typically the gas orstars of a galaxy, f ij denotes the fractional contribution of parti-cle i to pixel j , and the weights w i are either particle mass, or, for c (cid:13) RAS, MNRAS000 , 1–10 inematically atypical galaxies Table 1.
Present-day properties of the galaxies presented in Section 2.3.Stellar mass, gas mass, BH mass, and effective radius are given. The kine-matic feature seen in each galaxy is also listed; counter-rotating gas disc(CRGD) or kinematically distinct core (KDC).Galaxy Kinematics log M ∗ log M gas log M BH R e [ M (cid:12) ] [ M (cid:12) ] [ M (cid:12) ] [kpc]ga0045 CRGD 10.8 9.8 6.2 3.2ga0064 CRGD 10.7 9.7 5.7 2.6ga0074 KDC 10.6 9.9 6.1 2.3ga0091 KDC 10.5 9.8 6.0 2.4ga0099 CRGD 10.5 9.9 5.9 2.0 Figure 1.
Star formation histories of the five galaxies with atypical kine-matics listed in Table 1. The time t denotes time since the Big Bang. closer comparison to observational data, V -band luminosity ( L V )for stars and star formation rate (SFR) for gas (see Paper I for de-tails). Note that we evaluate (O/Fe) j as < O > j / < Fe > j . This pro-duces qualitatively similar maps as < O/Fe > j , but smooths the con-tribution of individual particles with extreme values of [O/Fe]. Galaxies are identified using a parallel Friends-of-Friends (FoF)finder (based on a serial version provided by V. Springel). The FoFcode associates dark matter particles, separated by at most 0.02times the mean inter-particle separation, into groups. Gas, star, andBH particles are then joined to the group of their nearest dark mat-ter neighbour. Note that the ‘linking length’ of 0.02 used here issmaller than typically adopted in the literature (e.g., 0.2 in Springelet al. 2005; Kobayashi et al. 2007; Dolag et al. 2009; Vogelsbergeret al. 2014; Schaye et al. 2015). In these works, sub-halos are sepa-rated from the main FoF groups, whereas we use the smaller linkinglength to achieve the same result.Five of the 82 sufficiently well resolved galaxies identified inPaper I were found to have atypical kinematics; three had counter-rotating gas and stellar components, and two had a KDC. All ofthe five galaxies – denoted ga0045, ga0064, ga0074, ga0091, andga0099 – formed in the field, and later fell into dark matter fila-ments where their unusual kinematic signatures developed throughgas accretion and minor mergers. The basic galaxy properties aresummarised in Table 1, where the masses are measured for all par-ticles identified by the FoF code. The star formation histories ofthese galaxies are shown in Fig. 1. All but ga0099 experience a strong peak in star formation at early times, and the formation of aCRGD reduces their star formation rates at late times (see Paper Ifor more details).
The columns of Fig. 2 show maps of V -band surface brightness( µ V ), stellar metallicity log ( Z ∗ / Z (cid:12) ) , stellar [O/Fe], stellar age,gas metallicity log ( Z g / Z (cid:12) ) , and gas [O/Fe], respectively for thefive galaxies. The maps of stellar populations are qualitativelysimilar for all of the galaxies. They have negative metallicity gradi-ents, and most variation in the metallicity maps is radial. Similarly,[O/Fe] ∗ varies primarily radially, and all five galaxies have small,positive gradients. This is because the star formation in the cen-tre takes place over a longer timescale than the outskirts, and thusthere is more time for SNe Ia to contribute Fe enrichment. Thereis slightly more variety in the maps of stellar age, but in generalthe stars at the centre of the galaxies are younger, on average, thanthose further out.The maps of gas properties show greater structure than forstars, and can be separated by kinematic features. The galaxiesga0045, ga0064, and ga0099 have counter-rotating gas and stars asa result of prolonged accretion of gas from a dark matter filament(Paper I). These galaxies show abrupt changes in gas metallicity (at ∼ . , . , and . R e , respectively) from the enriched gas towardsthe centre, to the metal-poor gas recently accreted from the filament(see Section 3.1 for the radial profiles). Galaxy ga0045 also showsa metal-enhanced region above the centre, which is not caused bya satellite galaxy; from the (cid:104) v xy (cid:105) and φ maps of Paper I inflow andoutflow gas hit around this area. By contrast, ga0074 and ga0091,which host a KDC that formed following a minor merger, do nothave such an abrupt change in metallicity with radius. The materialaccreted from the minor merger orbits away from the centre of thegalaxies (Paper I), and the gas of the secondary is already enrichedcompared to the gas in the filaments, leading to the more extendedregions of high metallicity at ∼ R e .Among CRGD galaxies, ga0099 has a relatively feature-less [O/Fe] g profile with a shallow positive gradient, ga0045 andga0064 have a region of high [O/Fe] g in its centre. This is due to abrief but recent ( < Myr) episode of star formation in the centreof the galaxy, from which only core-collapse SNe have contributedto [O/Fe] g , raising the value locally. In KDC galaxies ga0074 andga0091, [O/Fe] g increases radially from the centre, before droppingagain at ∼ R e (see Section 3.1 for more discussion).Fig. 2 shows mass-weighted quantities, whereas observationsproduce luminosity-weighted quantities. When we produce thesame maps weighting by L V for stellar quantities and SFR for gas,they are qualitatively and, except for stellar age, quantitatively sim-ilar, and the arguments presented above hold. We show in Fig. 3 the metallicity (top row) and [O/Fe] (bottomrow) profiles for stars (left-hand panels) and gas (right-hand pan-els). The full galaxy population is shown in grey, and the galax-ies with atypical kinematics have blue triangle symbols for galax- These maps are not affected by the resolution of our simulation; we re-produced the maps with a random selection of half of the particles and findexcellent qualitative and quantitative agreement.c (cid:13)
RAS, MNRAS , 1–10
P. Taylor, C. Kobayashi, and C. Federrath
Figure 2.
Maps of mass-weighted stellar and gas properties for our five simulated galaxies with atypical kinematics. The columns show V -band surfacebrightness (mag arcsec − ), stellar metallicity log ( Z ∗ / Z (cid:12) ) , stellar [O/Fe], stellar age, gas metallicity log ( Z g / Z (cid:12) ) , and gas [O/Fe], respectively. The rangeof values in each panel is shown in brackets (low values are blue, high are red). Each panel is R e on a side. ies with a CRGD, and red stars for KDCs. Both mass- and L V -weighted stellar metallicity profiles mirror the full population, andthere are no statistically significant distinguishing features. In gen-eral, both stellar metallicity and [O/Fe] gradients become steeperwhen weighted by luminosity. This is because Fe-enhanced popu-lations are younger and brighter. A similar effect is seen for the ob-served metallically gradients in early-type galaxies (Goddard et al.2017). Note that in this work we do not split our simulated galax-ies by type, but see Fig. 3 and the associated discussion of Taylor& Kobayashi (2017).For all of the KDC galaxies, however, the [O/Fe] ∗ profiles be-come shallower at r (cid:38) . R e . As described in detail in Paper I,these galaxies acquire a KDC due to a minor merger with a sec-ondary galaxy that deposits its stars at large radii. The low-masssecondary galaxies formed stars over a longer timescale than theold stellar populations at the outskirts of most galaxies (see Fig. 4),and had correspondingly more time for enrichment of Fe by SNe Ia, leading to lower average stellar [O/Fe] in the outskirts of theKDC galaxies compared to the full galaxy population. There is nosuch distinguishing feature for the stellar metallicity. Therefore, itis very important to measure elemental abundances at outskirts ofgalaxies in order to find this type of KCD in observations, which isnot possible for stellar populations.The gas-phase metallicity profiles show greater variation onthe whole than the stellar metallicity profiles. The five galaxies withatypical kinematics (red and blue lines) lie within the envelope ofthe profiles of the full population (grey). However, as discussed forFig. 2, the three CRGD galaxies (blue triangles) show a rapid in-crease of Z g towards the centre, and the profiles are not well fit bya straight line, which is also the case for luminosity-weighted Z ∗ profiles; the slope changes appear at ∼ . , . , and . R e , re-spectively. When weighted by SFR, one of the highlighted profiles(ga0045) has an upturn to high metallicity at (cid:38) R e ; this is because c (cid:13) RAS, MNRAS000
Maps of mass-weighted stellar and gas properties for our five simulated galaxies with atypical kinematics. The columns show V -band surfacebrightness (mag arcsec − ), stellar metallicity log ( Z ∗ / Z (cid:12) ) , stellar [O/Fe], stellar age, gas metallicity log ( Z g / Z (cid:12) ) , and gas [O/Fe], respectively. The rangeof values in each panel is shown in brackets (low values are blue, high are red). Each panel is R e on a side. ies with a CRGD, and red stars for KDCs. Both mass- and L V -weighted stellar metallicity profiles mirror the full population, andthere are no statistically significant distinguishing features. In gen-eral, both stellar metallicity and [O/Fe] gradients become steeperwhen weighted by luminosity. This is because Fe-enhanced popu-lations are younger and brighter. A similar effect is seen for the ob-served metallically gradients in early-type galaxies (Goddard et al.2017). Note that in this work we do not split our simulated galax-ies by type, but see Fig. 3 and the associated discussion of Taylor& Kobayashi (2017).For all of the KDC galaxies, however, the [O/Fe] ∗ profiles be-come shallower at r (cid:38) . R e . As described in detail in Paper I,these galaxies acquire a KDC due to a minor merger with a sec-ondary galaxy that deposits its stars at large radii. The low-masssecondary galaxies formed stars over a longer timescale than theold stellar populations at the outskirts of most galaxies (see Fig. 4),and had correspondingly more time for enrichment of Fe by SNe Ia, leading to lower average stellar [O/Fe] in the outskirts of theKDC galaxies compared to the full galaxy population. There is nosuch distinguishing feature for the stellar metallicity. Therefore, itis very important to measure elemental abundances at outskirts ofgalaxies in order to find this type of KCD in observations, which isnot possible for stellar populations.The gas-phase metallicity profiles show greater variation onthe whole than the stellar metallicity profiles. The five galaxies withatypical kinematics (red and blue lines) lie within the envelope ofthe profiles of the full population (grey). However, as discussed forFig. 2, the three CRGD galaxies (blue triangles) show a rapid in-crease of Z g towards the centre, and the profiles are not well fit bya straight line, which is also the case for luminosity-weighted Z ∗ profiles; the slope changes appear at ∼ . , . , and . R e , re-spectively. When weighted by SFR, one of the highlighted profiles(ga0045) has an upturn to high metallicity at (cid:38) R e ; this is because c (cid:13) RAS, MNRAS000 , 1–10 inematically atypical galaxies Figure 3.
Metallicity (top row) and [O/Fe] profiles (bottom row) as a function of galacto-centric radius (in units of effective radius, R e ) for the full galaxypopulation (grey) and those galaxies with atypical kinematics. The two leftmost columns are for L V - and mass-weighted stellar profiles, respectively, and thetwo rightmost are for SFR- and mass-weighted gas quantities. Blue triangles are for galaxies with a CRGD, and red stars are KDCs. star formation is more likely to take place in metal-rich gas that cancool efficiently (see Section 3.3 for more discussion).The [O/Fe] g profiles are shown in the lower-right panels ofFig. 3. The two KDC galaxies (red stars) have very similar profileswith a peak at r ∼ . R e . Gas-phase [O/Fe] g can vary due to i) re-cent star formation and O production, ii) delayed Fe enrichment bySNe Ia, and iii) mass-loss from stellar populations having [O/Fe] ∗ .For these galaxies, we found no evidence for enhanced star forma-tion following the mergers that gave rise to the KDCs in Paper I,so it is unlikely to have the option i) at ∼ . R e . Since [O/Fe] g is roughly the same as [O/Fe] ∗ there, this gas should mainly comefrom mass-loss. . R e is also roughly the same as the radius of theKDC (Paper I), which might block SN Ia enriched gas falling in. Itis also clear that two of the CRGD galaxies (ga0045 and ga0064;blue triangles) have profiles that differ significantly from a straightline; these are the galaxies identified in Section 3.3 as having un-usual values of the [O/Fe] g gradient. In these galaxies, SN Ia en-riched gas falls in near the centre, and the central [O/Fe] g is highbecause of the option i), as discussed for Fig. 2. We estimate ages using equation (1) with P = age and w = L V or mass; this is directly equivalent to the method used in observa-tions where linear combinations of SSP models are used to repro-duce the observed spectrum in each spaxel, and the average ageof that spaxel is then the weighted mean age of the SSP models(e.g., Zheng et al. 2017). In order to produce radial profiles fromthe maps of Fig. 2, we treat each Voronoi cell as a single point lo-cated at the generating point of that cell. Finally, we bin the data bygalacto-centric distance, and adopt the median value of each bin;the resulting age profiles are shown in Fig. 4.In the left-hand panel we show mass-weighted age profiles as a function of radius, expressed in units of R e , and in the right-handpanel are L V -weighted age profiles. L V traces young stellar popu-lations, causing many of the profiles in the right-hand panel to havelower ages, especially at small radii. In either weighting scheme,there are no distinguishing features in the age profiles of the kine-matically interesting galaxies that would set them apart from thefull population.From the profiles of Fig. 4 we can derive age gradients, ∇ Age , for our galaxies. Throughout this paper, we use ∇ to de-note dd( r/R e ) . We fit a linear function of the form Age = Age + ∇ Age × r/R e to each profile, and estimate errors on the fittedparameters using a bootstrapping technique. The gradients of themass- and L V -weighted age profiles are shown in the top and bot-tom panels of Fig. 5, respectively. Most galaxies have small posi-tive age gradients with mass-weighted values ∼ Gyr R − , exceptthe most massive galaxies where the gradient is closer to 0. Someintermediate-mass galaxies ( (cid:46) M ∗ / M (cid:12) (cid:46) ) have nega-tive gradients, indicating that star formation occurred most recentlyaway from the centre of the galaxies. In these galaxies, the blackhole has grown sufficiently massive that AGN feedback is strongenough to quench central star formation (Taylor et al. 2017), lead-ing to an older central stellar population and negative age gradients.In more massive galaxies, star formation was quenched earlier, andan old stellar population exists throughout these galaxies, givingrise to shallower gradients. This is likely to be exacerbated by theeffect of major mergers mixing stars of different ages, as also hap-pens to the metallicity gradients of massive galaxies (Kobayashi2004; Taylor & Kobayashi 2017). L V -weighted age gradients arequalitatively similar, but tend to be larger in magnitude for bothpositive and negative gradients. L V traces young stars, and thesteeper gradients are due to the fact that recent star formation hasnot taken place uniformly across the galaxy. This effect is also seen c (cid:13) RAS, MNRAS , 1–10
P. Taylor, C. Kobayashi, and C. Federrath
Figure 4.
Profiles of average stellar age with galacto-centric radius. Grey lines are for the full galaxy population, and red stars and blue triangles are for thegalaxies with KDCs and counter-rotating gas and stars, respectively. The left-hand panel shows mass-weighted ages, and the right-hand L V -weighted ages. Figure 5.
Age gradients as a function of stellar mass. The top panel is formass-weighted ages, the lower for L V -weighted. Red star and blue trianglesymbols show galaxies with KDCs and CRGDs, respectively. in observations of both early- and late-type galaxies (Goddard et al.2017).The blue triangle and red star symbols in Fig. 5 show thegalaxies with CRGDs and KDCs, respectively. These galaxies donot have preferentially steep or shallow gradients, and KDC galax-ies tend to have flatter gradients (luminosity weighted in partic-ular), though the scatter is large in this mass range. The atypicalkinematics seen in these galaxies form at late times (Paper I), andthe galaxies do not experience major mergers or late bursts of starformation. This means that the processes that gave rise to the agegradients in these galaxies likely happened before the CRGDs andKDCs formed. Although it is likely that the radial age profile ofgalaxies hosting a KDC changes when the KDC forms, due to theaccretion of stars from the secondary galaxy, this effect will de-pend on the age of the accreted galaxy and cannot be used to helpidentify a galaxy with a KDC. Fig. 6 shows stellar (left columns) and gas-phase (right columns)scaling relations for our population of simulated galaxies, as wellas the two KDC galaxies (red stars) and the three CRGD galax-ies (blue triangles). These datasets have been described in detailbefore (mass-metallicity relations and R e - M ∗ relation in Taylor &Kobayashi 2015a, 2016, and metallicity and elemental abundanceratio radial gradients in Taylor & Kobayashi 2017). Note, however,that these stellar metallicity and abundance gradients shown in thispaper are with respect to linear distance (expressed in units of R e ),rather than log distance as in Taylor & Kobayashi (2017). Never-theless, the results are qualitatively similar, and the minimum in ∇ log Z ∗ at ∼ M (cid:12) is visible in both datasets (Spolaor et al.2009, 2010; Kuntschner et al. 2010; Taylor & Kobayashi 2017).This mass is consistent with the transition mass for star formationin Taylor et al. (2017). Here we are concerned with the positionof the five kinematically atypical galaxies within the distributions;these are shown by the blue triangle symbols (CRGDs) and redstars (KDCs). L V - or SFR-weighted values are shown by fadedblack points, and the median of these and mass-weighted relationsare shown by the black and green lines, respectively.The stellar metallicity, [O/Fe], and their gradients, of the kine-matically atypical galaxies are near the median relations of the fullpopulation, which is expected from the fact that these galaxies donot undergo major mergers. The stellar mass–metallicity relation isset during the peak of star formation ( z ∼ , Taylor & Kobayashi2016), and gradients are strongly affected by major mergers (Tay-lor & Kobayashi 2017). Although the five galaxies of note lie veryclose to the median [O/Fe] ∗ – M ∗ relation, we remark that the dif-ference in the L V - and mass-weighted median relationships is dueto the fact that the young stars traced by L V formed from gas thathad more time to be enriched by Fe from SNe Ia than on aver-age, causing the distribution to shift down. In contrast, the effectiveradii ( R e ) of these galaxies are smaller than average, given theirmasses (bottom panels). This is because the galaxies do not un-dergo major mergers that can kinematically heat the stars, whichis also required to allow the atypical kinematic features to surviveto the present day. The R e dependence on the merging history isfound in Kobayashi (2005), who suggested that this is the origin ofthe scatter in the fundamental plane.The galaxies with CRGDs have stellar metallicities and metal- c (cid:13) RAS, MNRAS000
Age gradients as a function of stellar mass. The top panel is formass-weighted ages, the lower for L V -weighted. Red star and blue trianglesymbols show galaxies with KDCs and CRGDs, respectively. in observations of both early- and late-type galaxies (Goddard et al.2017).The blue triangle and red star symbols in Fig. 5 show thegalaxies with CRGDs and KDCs, respectively. These galaxies donot have preferentially steep or shallow gradients, and KDC galax-ies tend to have flatter gradients (luminosity weighted in partic-ular), though the scatter is large in this mass range. The atypicalkinematics seen in these galaxies form at late times (Paper I), andthe galaxies do not experience major mergers or late bursts of starformation. This means that the processes that gave rise to the agegradients in these galaxies likely happened before the CRGDs andKDCs formed. Although it is likely that the radial age profile ofgalaxies hosting a KDC changes when the KDC forms, due to theaccretion of stars from the secondary galaxy, this effect will de-pend on the age of the accreted galaxy and cannot be used to helpidentify a galaxy with a KDC. Fig. 6 shows stellar (left columns) and gas-phase (right columns)scaling relations for our population of simulated galaxies, as wellas the two KDC galaxies (red stars) and the three CRGD galax-ies (blue triangles). These datasets have been described in detailbefore (mass-metallicity relations and R e - M ∗ relation in Taylor &Kobayashi 2015a, 2016, and metallicity and elemental abundanceratio radial gradients in Taylor & Kobayashi 2017). Note, however,that these stellar metallicity and abundance gradients shown in thispaper are with respect to linear distance (expressed in units of R e ),rather than log distance as in Taylor & Kobayashi (2017). Never-theless, the results are qualitatively similar, and the minimum in ∇ log Z ∗ at ∼ M (cid:12) is visible in both datasets (Spolaor et al.2009, 2010; Kuntschner et al. 2010; Taylor & Kobayashi 2017).This mass is consistent with the transition mass for star formationin Taylor et al. (2017). Here we are concerned with the positionof the five kinematically atypical galaxies within the distributions;these are shown by the blue triangle symbols (CRGDs) and redstars (KDCs). L V - or SFR-weighted values are shown by fadedblack points, and the median of these and mass-weighted relationsare shown by the black and green lines, respectively.The stellar metallicity, [O/Fe], and their gradients, of the kine-matically atypical galaxies are near the median relations of the fullpopulation, which is expected from the fact that these galaxies donot undergo major mergers. The stellar mass–metallicity relation isset during the peak of star formation ( z ∼ , Taylor & Kobayashi2016), and gradients are strongly affected by major mergers (Tay-lor & Kobayashi 2017). Although the five galaxies of note lie veryclose to the median [O/Fe] ∗ – M ∗ relation, we remark that the dif-ference in the L V - and mass-weighted median relationships is dueto the fact that the young stars traced by L V formed from gas thathad more time to be enriched by Fe from SNe Ia than on aver-age, causing the distribution to shift down. In contrast, the effectiveradii ( R e ) of these galaxies are smaller than average, given theirmasses (bottom panels). This is because the galaxies do not un-dergo major mergers that can kinematically heat the stars, whichis also required to allow the atypical kinematic features to surviveto the present day. The R e dependence on the merging history isfound in Kobayashi (2005), who suggested that this is the origin ofthe scatter in the fundamental plane.The galaxies with CRGDs have stellar metallicities and metal- c (cid:13) RAS, MNRAS000 , 1–10 inematically atypical galaxies Figure 6.
Scaling relations for our simulated galaxies. Metallicity and [O/Fe], and the gradients of these quantities, are shown for both stars (left columns)and gas (right column) as functions of stellar mass. Effective radius, R e , and gas fraction, f g , are also shown on the bottom row. Faded black points showquantities weighted by L V (for stars) or SFR (for gas), and faded green points are for mass-weighted quantities. Solid black and green lines denote the medianlight- or mass-weighted relations, and dashed lines show th and th percentiles. Blue triangles are for the galaxies with a CRGD, and red stars are forKDCs; filled stars and triangles are for L V - or SFR-weighted quantities, open symbols are mass-weighted. licity gradients that can be ∼ σ from the median trend when L V weighted. To understand the reason, in Fig. 7 we show the distri-bution of stellar metallicity with deprojected (i.e. 3-dimensional)radius for these galaxies; the full distribution in shown in outline,and those stars younger than 1 Gyr are shown as full points. Inga0045 and ga0064, there is a concentration of young, metal-richstars in the central ∼ kpc. These galaxies transitioned from co- tocounter-rotating gas approximately 1 Gyr before the present (PaperI), with the bulk of these young stars forming from the last of theco-rotating gas. These young populations comprise only ∼ percent of the total stellar mass of these galaxies, but are bright in L V ,leading to the ∼ σ effect seen. Galaxy ga0099 developed a CRGDearlier, about 6 Gyr before the present, and there is no concentrationof recent star formation that can have the same effect.The five galaxies have below-average gas metallicity whenweighted by mass due to their accretion at late times of near-pristine gas from filaments. However, when weighted by SFR, threeof the galaxies appear more metal-rich than average, which is be-cause star formation is more likely to occur in high-metallicity gasthat can cool efficiently. The gas metallicity gradients are consis-tent with those of the full sample. The gas-phase metallicity gra-dients therefore develop normally, even though their total metallic-ity is lower. The values and gradients of [O/Fe] g for these galax-ies, and the galaxy population as a whole, are similar regardless ofthe weighting used. Two of the galaxies with counter-rotating gas(triangle symbols) have [O/Fe] g gradients away from the averagetrend, but this is because their [O/Fe] g radial profiles are not wellfit by a straight line; see Section 3.1 for more details. In the bottom-right panel of Fig. 6, we show the gas fractionof our simulated galaxies as a function of stellar mass (gas in anyphase throughout the galaxy is included). Three of the five kine-matically atypical galaxies lie slightly above the median relation,but they are not the most gas-rich galaxies at a given mass. Theirhigher-than-average gas content is likely a consequence of their lo-cation within dark matter filaments where gas is readily available,and not a clear indication of the presence of kinematic features. In this section, we separate the components of the CRGDs andKDCs in order to understand the differences in the radial profilesfurther. This is only possible with simulations, which have the fullkinematic information of individual star particles. We divide thestars based on their angular momentum, j ∗ ; stars with j ∗ · J ∗ < make up the KDC, the others orbit away from the centre of thegalaxy, in the same direction as the gas (see Paper I for more de-tails). Here, J ∗ denotes the total stellar angular momentum of thegalaxy, J ∗ = (cid:80) i j ∗ ,i .Table 2 lists the stellar metallicities and [O/Fe] ratios, bothmass- and light-weighted, for the components of the two galax-ies. In general, the co-rotating stellar components (in the outskirtsof the galaxies) have much lower metallicity than the cores; thesestars were almost entirely accreted from the merged galaxy. Oursimulated galaxies follow the mass-metallicity relation (Taylor &Kobayashi 2016), and the low metallicities of the co-rotating com- c (cid:13) RAS, MNRAS , 1–10
P. Taylor, C. Kobayashi, and C. Federrath
Table 2.
Stellar metallicity and [O/Fe] for the KDC galaxies ga0074 and ga0091. The stellar components that co- and counter-rotate compared to the gas (inthe outskirts and the core of the galaxy, respectively), as well as the values for the full galaxy are given.Galaxy Mass-Weighted L V -Weighted log Z ∗ / Z (cid:12) [O/Fe] ∗ log Z ∗ / Z (cid:12) [O/Fe] ∗ ga0074 co-rotating .
17 0 .
04 0 . − . counter-rotating .
21 0 .
02 0 . − . all .
19 0 .
03 0 . − . ga0091 co-rotating .
16 0 .
04 0 . − . counter-rotating .
18 0 .
03 0 . − . all .
17 0 .
04 0 . − . Figure 7.
Distribution of stellar metallicity with deprojected (i.e. 3-dimensional) radius for the three galaxies with a CRGD. The full distri-bution is shown in outline; those stars younger than 1 Gyr are shown assolid points. ponents reflect the fact that the mergers had small mass ratios( ∼ / ).Although the co-rotating components are in the outskirts, the[O/Fe] ∗ ratios are only marginally higher when weighted by mass,and are rather lower when weighted by luminosity. This is due tothe longer star formation timescale as discussed for Fig. 2. Thisis not seen in the observation of NGC 4365 (Davies et al. 2001),where the [O/Fe] ∗ is the same between the core and the main body.Our KDC is a different type and might be found in the next gener-ation of IFU survey with a large field of view such as with Hector(see Paper I for more discussion). Here we presented 2D maps of stellar populations and gas-phasechemical abundances of the simulated galaxies with atypical kine-matics shown in Paper I. Two of the five galaxies host a minor-merger-originated KDC, and the other three have a CRGD com-pared to the angular momentum of the stars, caused by delayedfilament accretion. The maps of gas-phase metallicity (Fig. 2) showed distinct ar-eas of high and low metallicity separated by a sharp boundary forthe galaxies with a CRGD. The non-linear increase is also seen inthe radial profiles (Fig. 3) but when the gradients are measured,the difference is buried in the large scatter of the gradient-mass re-lation (Fig. 6). The radial gradients are useful to flag the galaxiesformed by major mergers, and the signatures of minor mergers orfilament accretion can be detected only with IFU mapping. Gas-phase metallicity maps are easily obtained from IFU data since theemission lines from ionised gas typically have high signal-to-noise.We predict that all galaxies that fall from the field into a filamentshould have gas metallicity maps with a sharp boundary, regardlessof the relative orientation of their gas and stellar angular momenta.Galaxies that form CRGDs via this mechanism are expected to havesuch metallicity maps, but not all galaxies with such metallicitymaps will have a CRGD. In the galaxies with a CRGD, we alsofind some non-linear gradients in stellar metallicity; this becomesmuch clearer when we separate the stars that are formed from theCRGD (Fig. 7), which is possible only in simulations with full 6Dkinematics.For the KDC galaxies, we divide the stars into co- and counter-rotating components using 6D kinematics, and show that the KDCcomponent has lower metallicity than the main component of thegalaxy (Table 2). We also found that they could be distinguishedby their stellar [O/Fe] profiles (Fig. 3). At large radii ( r (cid:38) R e ),the [O/Fe] profiles of these galaxies flattened, and had lower valuesthan the full galaxy population. This region of the KDC galaxiesis made primarily of stars accreted from a minor merger, whichformed on a longer timescale than the in situ star formation of theprimary. This is a clear prediction for KDCs formed via the mech-anism presented in Paper I, and it is very important to measure[O/Fe] in stelar populations at larger radii ( > R e ). For the samereason, the age gradients tend to be flatter (Figs. 4 and 5).None of these kinematically atypical galaxies experiences amajor merger, and so there are no significant differences in the scal-ing relations, except for the size-mass relation. All of the galaxieswith atypical kinematics were found to have lower-than-averagevalues of R e , given their masses (Section 3.3). With major mergers,the stellar distribution becomes kinematically hot, which increases R e (this is also found in the simulations of early-type galaxies inKobayashi 2005). Major mergers would also destroy the interestingkinematic features of these galaxies, and therefore galaxies withsuch kinematics should always be compact.We should note that these features were clear in mass-weighted kinematic maps, but much less discernible in light-weighted maps. Using stellar populations synthesis models, it ispossible to measure both maps with IFU data (e.g., Goddardet al. 2017). In any case, future IFU surveys like Hector (Bland- c (cid:13) RAS, MNRAS000
Distribution of stellar metallicity with deprojected (i.e. 3-dimensional) radius for the three galaxies with a CRGD. The full distri-bution is shown in outline; those stars younger than 1 Gyr are shown assolid points. ponents reflect the fact that the mergers had small mass ratios( ∼ / ).Although the co-rotating components are in the outskirts, the[O/Fe] ∗ ratios are only marginally higher when weighted by mass,and are rather lower when weighted by luminosity. This is due tothe longer star formation timescale as discussed for Fig. 2. Thisis not seen in the observation of NGC 4365 (Davies et al. 2001),where the [O/Fe] ∗ is the same between the core and the main body.Our KDC is a different type and might be found in the next gener-ation of IFU survey with a large field of view such as with Hector(see Paper I for more discussion). Here we presented 2D maps of stellar populations and gas-phasechemical abundances of the simulated galaxies with atypical kine-matics shown in Paper I. Two of the five galaxies host a minor-merger-originated KDC, and the other three have a CRGD com-pared to the angular momentum of the stars, caused by delayedfilament accretion. The maps of gas-phase metallicity (Fig. 2) showed distinct ar-eas of high and low metallicity separated by a sharp boundary forthe galaxies with a CRGD. The non-linear increase is also seen inthe radial profiles (Fig. 3) but when the gradients are measured,the difference is buried in the large scatter of the gradient-mass re-lation (Fig. 6). The radial gradients are useful to flag the galaxiesformed by major mergers, and the signatures of minor mergers orfilament accretion can be detected only with IFU mapping. Gas-phase metallicity maps are easily obtained from IFU data since theemission lines from ionised gas typically have high signal-to-noise.We predict that all galaxies that fall from the field into a filamentshould have gas metallicity maps with a sharp boundary, regardlessof the relative orientation of their gas and stellar angular momenta.Galaxies that form CRGDs via this mechanism are expected to havesuch metallicity maps, but not all galaxies with such metallicitymaps will have a CRGD. In the galaxies with a CRGD, we alsofind some non-linear gradients in stellar metallicity; this becomesmuch clearer when we separate the stars that are formed from theCRGD (Fig. 7), which is possible only in simulations with full 6Dkinematics.For the KDC galaxies, we divide the stars into co- and counter-rotating components using 6D kinematics, and show that the KDCcomponent has lower metallicity than the main component of thegalaxy (Table 2). We also found that they could be distinguishedby their stellar [O/Fe] profiles (Fig. 3). At large radii ( r (cid:38) R e ),the [O/Fe] profiles of these galaxies flattened, and had lower valuesthan the full galaxy population. This region of the KDC galaxiesis made primarily of stars accreted from a minor merger, whichformed on a longer timescale than the in situ star formation of theprimary. This is a clear prediction for KDCs formed via the mech-anism presented in Paper I, and it is very important to measure[O/Fe] in stelar populations at larger radii ( > R e ). For the samereason, the age gradients tend to be flatter (Figs. 4 and 5).None of these kinematically atypical galaxies experiences amajor merger, and so there are no significant differences in the scal-ing relations, except for the size-mass relation. All of the galaxieswith atypical kinematics were found to have lower-than-averagevalues of R e , given their masses (Section 3.3). With major mergers,the stellar distribution becomes kinematically hot, which increases R e (this is also found in the simulations of early-type galaxies inKobayashi 2005). Major mergers would also destroy the interestingkinematic features of these galaxies, and therefore galaxies withsuch kinematics should always be compact.We should note that these features were clear in mass-weighted kinematic maps, but much less discernible in light-weighted maps. Using stellar populations synthesis models, it ispossible to measure both maps with IFU data (e.g., Goddardet al. 2017). In any case, future IFU surveys like Hector (Bland- c (cid:13) RAS, MNRAS000 , 1–10 inematically atypical galaxies Hawthorn 2015) will need a wide field of view and high sensitivityin order to find these features.
ACKNOWLEDGEMENTS
Parts of this research were supported by the Australian ResearchCouncil Centre of Excellence for All Sky Astrophysics in 3 Di-mensions (ASTRO 3D), through project number CE170100013.C.F. gratefully acknowledges funding provided by the Aus-tralian Research Council (Discovery Projects DP150104329 andDP170100603, and Future Fellowship FT180100495) and bythe Australia-Germany Joint Research Cooperation Scheme (UA-DAAD). CK acknowledges support from the UK’s Science andTechnology Facilities Council (grant ST/R000905/1). The simu-lations presented in this work used high performance computingresources provided by the Leibniz Rechenzentrum and the GaussCentre for Supercomputing (grants pr32lo, pr48pi and GCS Large-scale project 10391), the Partnership for Advanced Computing inEurope (PRACE grant pr89mu), the Australian National Computa-tional Infrastructure (grant ek9), and the Pawsey SupercomputingCentre with funding from the Australian Government and the Gov-ernment of Western Australia, in the framework of the NationalComputational Merit Allocation Scheme and the ANU AllocationScheme. Finally, we thank V. Springel for providing GADGET-3.
REFERENCES
Agarwal B., Khochfar S., Johnson J. L., Neistein E., Dalla VecchiaC., Livio M., 2012, MNRAS, 425, 2854Bassett R., Bekki K., Cortese L., Couch W., 2017, MNRAS, 471,1892Becerra F., Greif T. H., Springel V., Hernquist L. E., 2015, MN-RAS, 446, 2380Bender R., Surma P., 1992, A&A, 258, 250Bland-Hawthorn J., 2015, in Ziegler B. L., Combes F., Danner-bauer H., Verdugo M., eds, IAU Symposium Vol. 309, Galaxiesin 3D across the Universe. pp 21–28Bois M. et al., 2011, MNRAS, 416, 1654Brodie J. P. et al., 2014, ApJ, 796, 52Bromm V., Coppi P. S., Larson R. B., 2002, ApJ, 564, 23Bromm V., Loeb A., 2003, ApJ, 596, 34Bryant J. J. et al., 2019, MNRAS, 483, 458Cappellari M. et al., 2011, MNRAS, 416, 1680Croom S. M. et al., 2012, MNRAS, 421, 872Davies R. L. et al., 2001, ApJ, 548, L33Davies R. L., Sadler E. M., Peletier R. F., 1993, MNRAS, 262,650Davis T. A. et al., 2011, MNRAS, 417, 882De Silva G. M. et al., 2015, MNRAS, 449, 2604de Zeeuw P. T. et al., 2002, MNRAS, 329, 513Dolag K., Borgani S., Murante G., Springel V., 2009, MNRAS,399, 497Dopita M. A. et al., 2015, ApJS, 217, 12Faber S. M., 1973, ApJ, 179, 731Gilmore G. et al., 2012, The Messenger, 147, 25Goddard D. et al., 2017, MNRAS, 465, 688Green A. W. et al., 2018, MNRAS, 475, 716Haardt F., Madau P., 1996, ApJ, 461, 20Hinshaw G. et al., 2013, ApJS, 208, 19 Hoffman L., Cox T. J., Dutta S., Hernquist L., 2010, ApJ, 723,818Hosokawa T., Hirano S., Kuiper R., Yorke H. W., Omukai K.,Yoshida N., 2016, ApJ, 824, 119Jesseit R., Naab T., Peletier R. F., Burkert A., 2007, MNRAS, 376,997Khochfar S. et al., 2011, MNRAS, 417, 845Kobayashi C., 2004, MNRAS, 347, 740Kobayashi C., 2005, MNRAS, 361, 1216Kobayashi C., 2016, Nat, 540, 205Kobayashi C., Karakas A. I., Umeda H., 2011, MNRAS, 414,3231Kobayashi C., Nakasato N., 2011, ApJ, 729, 16Kobayashi C., Nomoto K., 2009, ApJ, 707, 1466Kobayashi C., Springel V., White S. D. M., 2007, MNRAS, 376,1465Kobayashi C., Umeda H., Nomoto K., Tominaga N., Ohkubo T.,2006, ApJ, 653, 1145Kormendy J., Djorgovski S., 1989, ARA&A, 27, 235Koushiappas S. M., Bullock J. S., Dekel A., 2004, MNRAS, 354,292Krajnovi´c D. et al., 2008, MNRAS, 390, 93Krajnovi´c D. et al., 2011, MNRAS, 414, 2923Kroupa P., 2008, in Knapen J. H., Mahoney T. J., Vazdekis A.,eds, Astronomical Society of the Pacific Conference Series Vol.390, Pathways Through an Eclectic Universe. p. 3Kuntschner H. et al., 2010, MNRAS, 408, 97Ma C.-P., Greene J. E., McConnell N., Janish R., Blakeslee J. P.,Thomas J., Murphy J. D., 2014, ApJ, 795, 158Madau P., Rees M. J., 2001, ApJ, 551, L27Majewski S. R. et al., 2017, AJ, 154, 94Naab T. et al., 2014, MNRAS, 444, 3357Penoyre Z., Moster B. P., Sijacki D., Genel S., 2017, MNRAS,468, 3883Regan J. A., Johansson P. H., Wise J. H., 2016, MNRAS, 459,3377S´anchez S. F. et al., 2012, A&A, 538, A8Schaye J. et al., 2015, MNRAS, 446, 521Schneider R., Ferrara A., Natarajan P., Omukai K., 2002, ApJ,571, 30Schulze F., Remus R.-S., Dolag K., Burkert A., Emsellem E., vande Ven G., 2018, MNRAS, 480, 4636Schweizer F., Seitzer P., 1992, AJ, 104, 1039Schweizer F., Seitzer P., Faber S. M., Burstein D., Dalle OreC. M., Gonzalez J. J., 1990, ApJ, 364, L33Spolaor M., Kobayashi C., Forbes D. A., Couch W. J., HauG. K. T., 2010, MNRAS, 408, 272Spolaor M., Proctor R. N., Forbes D. A., Couch W. J., 2009, ApJ,691, L138Springel V., 2005, MNRAS, 364, 1105Springel V. et al., 2005, Nat, 435, 629Stott J. P. et al., 2014, MNRAS, 443, 2695Sutherland R. S., Dopita M. A., 1993, ApJS, 88, 253Taylor P., Federrath C., Kobayashi C., 2017, MNRAS, 469, 4249Taylor P., Federrath C., Kobayashi C., 2018, MNRAS, 479, 141Taylor P., Kobayashi C., 2014, MNRAS, 442, 2751Taylor P., Kobayashi C., 2015a, MNRAS, 448, 1835Taylor P., Kobayashi C., 2015b, MNRAS, 452, L59Taylor P., Kobayashi C., 2016, MNRAS, 463, 2465Taylor P., Kobayashi C., 2017, MNRAS, 471, 3856Thomas A. D. et al., 2017, ApJS, 232, 11van de Sande J. et al., 2018, MNRAS c (cid:13) RAS, MNRAS , 1–10 P. Taylor, C. Kobayashi, and C. Federrath
Vogelsberger M. et al., 2014, Nat, 509, 177Wisnioski E. et al., 2015, ApJ, 799, 209Yan R. et al., 2016, AJ, 152, 197Zheng Z. et al., 2017, MNRAS, 465, 4572This paper has been typeset from a TEX/ L A TEX file prepared by theauthor. c (cid:13) RAS, MNRAS000