The origin of metal-poor stars on prograde disk orbits in FIRE simulations of Milky Way-mass galaxies
Isaiah B. Santistevan, Andrew Wetzel, Robyn E. Sanderson, Kareem El-Badry, Jenna Samuel, Claude-André Faucher-Giguère
MMNRAS , 1–16 (2020) Preprint 9 February 2021 Compiled using MNRAS L A TEX style file v3.0
The origin of metal-poor stars on prograde disk orbits in FIREsimulations of Milky Way-mass galaxies
Isaiah B. Santistevan ★ , Andrew Wetzel , Robyn E. Sanderson , , Kareem El-Badry ,Jenna Samuel , Claude-André Faucher-Giguère Department of Physics & Astronomy, University of California, Davis, CA 95616, USA Department fo Physics & Astronomy, University of Pennsylvania, Philadelphia, PA 19104, USA Center for Computational Astrophysics, Flatiron Institute, New York, NY 10010, USA Department of Astronomy, Theoretical Astrophysics Center, University of California Berkeley, Berkeley, CA 94720, USA Department of Physics and Astronomy and CIERA, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA
Accepted XXX. Received YYY; in original form ZZZ
ABSTRACT
In hierarchical structure formation, metal-poor stars in and around the Milky Way (MW)originate primarily from mergers of lower-mass galaxies. A common expectation is thereforethat metal-poor stars should have isotropic, dispersion-dominated orbits that do not correlatestrongly with the MW disk. However, recent observations of stars in the MW show thatmetal-poor ([Fe/H] (cid:46) −
2) stars are preferentially on prograde orbits with respect to the disk.Using the FIRE-2 suite of cosmological zoom-in simulations of MW/M31-mass galaxies,we investigate the prevalence and origin of prograde metal-poor stars. Almost all (11 of12) of our simulations have metal-poor stars preferentially on prograde orbits today andthroughout most of their history: we thus predict that this is a generic feature of MW/M31-mass galaxies. The typical prograde-to-retrograde ratio is ∼ (cid:46) −
1. These trends predicted by our simulations agree well withMW observations. Prograde metal-poor stars originate largely from a single LMC/SMC-massgas-rich galaxy merger, typically 7 − . 𝑧 = Key words: stars – metal-poor, galaxies – formation
The populations of stars in different regions of a galaxy, such asthe disk, the bulge, and the halo, can reveal its formation history. Inthe standard picture of galaxy formation, unenriched gas collapsesin dark matter (DM) halos to form stars and galaxies (e.g. Ostriker& Tremaine 1975; Rees & Ostriker 1977; White & Rees 1978;Fall & Efstathiou 1980). Many works have studied the hierarchicalnature of galaxy formation, such as determining the ages of globularclusters throughout the MW’s stellar halo, and the formation of diskgalaxies within their DM halos (e.g. Searle & Zinn 1978; White &Frenk 1991; Mo et al. 1998). Eggen et al. (1962) first suggested thatold, metal-poor stars orbit the MW in elliptical orbits compared to ★ E-mail: [email protected] metal-rich stars, which move in more circular orbits, a picture thatbroadly holds today (e.g. Chiba & Beers 2000). Many authors haveexplored how metal-poor stars were deposited into the stellar halo ofthe MW by mergers of low-mass dwarf galaxies (e.g. Bullock et al.2001; Bullock & Johnston 2005; Ibata et al. 1994; Newberg et al.2003; Nissen & Schuster 2010). We now understand that most ofthe mass in the stellar halo came from a few massive mergers, whichis a natural consequence of the steepness of the stellar mass halomass relation. Deason et al. (2016) showed that only 1 − 𝑀 star ∼ − M (cid:12) deposited a majority of thestars in the halo, and dwarf galaxies with 𝑀 star ∼ − M (cid:12) contributed the bulk of metal-poor ([Fe/H] (cid:46) −
2) stars. A recentobservational study by Naidu et al. (2020) suggests that at distancesgreater than ∼
15 kpc from the galactic center, ∼
80 per cent of halostars came from 2 mergers: Gaia-Enceladus (Helmi et al. 2018; © a r X i v : . [ a s t r o - ph . GA ] F e b Santistevan et al.
Belokurov et al. 2018) and Sagittarius (Ibata et al. 1994; Newberget al. 2003; Majewski et al. 2003).Many studies focused on the properties and origins of theseold and/or metal-poor stars, such as their spatial and elemental dis-tributions within the MW, the masses of the accreted dwarf galaxiesthey came from, and the formation of the oldest Population IIIstars (e.g. Scannapieco et al. 2006; Brook et al. 2007; Deason et al.2016; Griffen et al. 2018). Spectroscopic surveys that have observedstars with [Fe/H] (cid:46) −
2, such as RAVE, GALAH, LAMOST, andAPOGEE, provide the elemental abundances and ages of stars indifferent regions of the MW and now allow for an investigation intothe early formation period (Steinmetz et al. 2006; De Silva et al.2015; Li et al. 2015; Majewski et al. 2017). The Pristine survey alsohas pushed metallicity measurements down to [Fe/H] < −
3, withthe potential of reaching the ‘ultra metal-poor’ ([Fe/H] < −
4) popu-lation (Starkenburg et al. 2017b). Finally, the recent Hectochelle inthe Halo at High Resolution (H3) Survey is targeting stars outside ofthe disk and bulge regions of the MW, and is reaching metallicitiesdown to [Fe/H] ∼ − 𝑧 form > < −
2) stars in 3 of the FIRE-2simulations that we examine here, El-Badry et al. (2018b) proposedthat this spatially extended distribution of metal-poor stars in a MW-mass galaxy results from a combination of kinematic heating of starsto larger orbits in the host galaxy from a time-varying galactic po-tential and from early mergers that deposited these stars on radialorbits. However, they did not examine any preference for metal-poor stars to be on prograde orbits aligned with the disk. Similarly,using the APOSTLE simulations, Starkenburg et al. (2017a) foundthat the fraction of stars with [Fe/H] < − .
5, older than (cid:38)
13 Gyr,increases with increasing distance from the center of the galaxy.Although these studies agree that the number density of old stars islargest near the central bulge region, the overabundance of youngerstars can make finding older stars difficult.The origin of the MW’s stellar halo and its disk are likelyconnected, particularly for the inner stellar halo and the thick disk.Gallart et al. (2019) showed that stellar populations in the inner haloand thick disk partially overlap in kinematic space. The thick disk ex-tends (cid:38) − (cid:46) [Fe/H] (cid:46) − . ∼ > − < −
2) were more retrograde on av-erage and could have been part of the early-forming halo. They alsonoted that many globular clusters in the MW halo orbit in the pro-grade direction. Another study by Bonaca et al. (2017) concludedthat metal-rich stars ([Fe/H] > −
1) near the MW have progradeorbits, while metal-poor stars have no average orbit direction. Thissuggests the metal-rich halo stars formed in-situ in the MW galaxy,while the metal-poor stars likely were accreted. The authors foundsimilar features in one of the simulated galaxies in our sample, m12i.Finally, recent work focused on kinematically, or elementally, co-herent structures throughout the MW suggest that many previousmergers currently comprise most of the retrograde population in theMW halo (Myeong et al. 2019; Naidu et al. 2020).Recently, Sestito et al. (2019) analyzed all known MW starswith [Fe/H] < − ± 𝐽 𝜙 > 𝐽 𝜙 < (cid:46) − ∼ ∼
31 per centof the sample have prograde orbits that are confined to ± MNRAS000
31 per centof the sample have prograde orbits that are confined to ± MNRAS000 , 1–16 (2020) rograde metal-poor stars ii) How does this prograde bias depend on the metallicity and/orage of the stars?iii) What process(es) cause this prograde bias among metal-poorstars?iv) Do any properties of the MW/M31-mass galaxy or theirgalaxy mergers correlate with this prograde bias? We use cosmological zoom-in baryonic simulations of MW/M31-mass galaxies from the Feedback In Realistic Environments (FIRE)project (Hopkins et al. 2018). We ran all of the simulations usingthe same 𝑁 -body gravitational plus hydrodynamics code Gizmo(Hopkins 2015), with the mesh-free finite-mass (MFM) hydrody-namics method and the FIRE-2 physics model (Hopkins et al. 2018).Across a temperature range of 10 − K, the FIRE-2 modelincludes several radiative cooling and heating processes for gassuch as free-free emission, photoionization/recombination, Comp-ton scattering, photoelectric, metal-line, molecular, fine-structure,dust-collisional, and cosmic-ray heating, including the spatially uni-form, redshift-dependent cosmic UV background from Faucher-Giguère et al. (2009), where HI reionization occurs at 𝑧 reion ∼ = −
4. We choosethis metallicity floor as a rough model for the level of enrichmentwe expect from Population III stars, which also corresponds to thelowest metallicity in one of the observational studies we compareour results to, Sestito et al. (2019). As we will show, the strength ofthe prograde bias does not depend on metallicity at low metallici-ties, so the choice of the metallicity floor is unimportant. Stars formin gas that is self-gravitating, Jeans unstable, molecular (followingKrumholz & Gnedin 2011), and dense ( 𝑛 H > 1000 cm − ). A starparticle inherits the mass and elemental abundances from its progen-itor gas particle, and represents a single stellar population, assuminga Kroupa (2001) initial mass function. It then evolves along stel-lar population models from STARBURST99 v7.0 (Leitherer et al.1999). Furthermore, the FIRE-2 simulations include the followingfeedback processes: core-collapse and Ia supernovae, stellar winds,radiation pressure, photoionization, and photo-electric heating.We generated the cosmological zoom-in initial conditions foreach simulation embedded within periodic cosmological boxes oflengths 70 . −
172 Mpc using the code MUSIC (Hahn & Abel 2011),at 𝑧 ≈
99. Each simulation saves of 600 snapshots down to 𝑧 = ≈
25 Myr, and all assume flat Λ CDM cosmologywith the following range in cosmological parameters, consistentwith Planck Collaboration et al. (2018): Ω m = . − . Ω b = . − . 𝜎 = . − . 𝑛 s = . − .
97, and ℎ = . − . 𝑧 =
0. Halfof our sample comes from the Latte suite of isolated MW/M31-massgalaxies first introduced in Wetzel et al. (2016). The Latte suite has FIRE project web site: http://fire.northwestern.edu gas and initial star particle masses of 7100 M (cid:12) , but because of stel-lar mass loss, the average star particle mass is ≈ (cid:12) . Themass of DM particles in the zoom-in region is 3 . × M (cid:12) . Thegravitational softening lengths for star and DM particles are fixed at4 and 40 pc (Plummer equivalent), comoving at 𝑧 > 𝑀 = − × M (cid:12) and have no other similar-mass halos within 5 × 𝑅 , but weimposed no additional selection criteria beyond this.The other half of our sample comes from the ‘ELVIS on FIRE’suite of LG-like MW+M31 pairs, introduced in Garrison-Kimmelet al. (2019a,b), which have approximately 2 × better mass resolu-tion, with gas and initial star particle masses of 3500 − (cid:12) .Each pair of halos was chosen based on their masses (eachwith 𝑀 = − × M (cid:12) and total LG mass between2 − × M (cid:12) ), current separation (600 − 𝑧 = 𝜐 rad < . 𝑧 ∼ We generate (sub)halo catalogs using only DM particles, at all 600snapshots, using the ROCKSTAR 6-D halo finder (Behroozi et al.2013a). To generate merger trees, we use CONSISTENT-TREES(Behroozi et al. 2013b). All (sub)halos that we examine have zerocontamination by low-resolution DM particles, because of the largezoom-in region that we generate for each host.We assign star particles to (sub)halos in post-processing, whichwe outline below, but refer the reader to Samuel et al. (2020b) formore details. We identify star particles with positions within 0 . × 𝑅 halo (out to a maximum distance of 30 kpc) and velocities within 2 × 𝑉 circ , max of a (sub)halo’s center-of-mass velocity. Following theseinitial criteria, we keep star particles if they are within 1 . × 𝑅 star , (the radius enclosing 90 per cent of the stellar mass) of the currentmember star particle’s center-of-mass and halo center position. Thisguarantees that the centers-of-mass of both the galaxy and (sub)haloare consistent with one another. We then keep star particles within2 × 𝜎 vel , star (the velocity dispersion of current member star particles)of the center-of-mass velocity of member star particles, and weiterate on both criteria until the (sub)halo’s stellar mass convergesto within 1 per cent.We use two publicly available analysis packages: HaloAnaly- MNRAS , 1–16 (2020)
Santistevan et al.
Table 1.
Properties of the 12 MW/M31-mass galaxies in the FIRE-2 simulation suite that we analyze, ordered by decreasing prograde-to-retrograderatio for metal-poor stars. Simulations with ‘m12’ names are isolated hosts from the Latte suite, while the others are from the ELVIS on FIRE suite ofLG-like galaxies. We measure all masses and mass fractions at 𝑧 =
0, using only metal-poor ([Fe/H] < − .
5) stars (except for the total stellar mass ofthe galaxy, which uses all stars). Columns: name; 𝑀 star , is the host’s stellar mass within 𝑅 star , , the disk radius enclosing 90 per cent of the stellarmass within 20 kpc; 𝑀 star , pro / 𝑀 star , ret is the prograde bias, defined as the mass ratio of prograde to retrograde stars; 𝑡 lb , merger , is the lookback time ofthe merger that contributed the most prograde stars; 𝑓 merger , is the fraction of all prograde stars that came from the primary (most contributing) merger; 𝑓 merger , top 3 is the fraction of prograde stars from the 3 most contributing mergers; 𝑓 in − situ is the fraction of prograde stars that formed in-situ (within 15kpc); 𝑓 merger , + 𝑓 in − situ is the sum; 𝑓 merger , top 3 + 𝑓 in − situ is the sum of fractions from the 3 most contributing mergers and the fraction that formed in-situ.Name 𝑀 star , 𝑀 star , pro / 𝑀 star , ret 𝑡 lb , merger , 𝑓 merger , 𝑓 merger , top 3 𝑓 in − situ 𝑓 in − situ + 𝑓 merger , 𝑓 merger , top 3 + 𝑓 in − situ [10 M (cid:12) ] [Gyr]m12m a b c c c c c d e d c f Note:
Simulation introduced in: a: Hopkins et al. (2018), b: Samuel et al. (2020b), c: Garrison-Kimmel et al. (2019b), d: Garrison-Kimmel et al.(2019a), e: Garrison-Kimmel et al. (2017), f: Wetzel et al. (2016). sis (Wetzel & Garrison-Kimmel 2020a) for assigning star particlesto halos and for reading and analyzing halo catalogs/trees, and Giz-moAnalysis (Wetzel & Garrison-Kimmel 2020b) for reading andanalyzing particles from Gizmo snapshots. For each MW/M31-mass host galaxy, we define the ‘prograde bias’as the ratio of the total stellar mass of prograde stars to that ofretrograde stars, 𝑀 star , pro / 𝑀 star , ret . To measure the prograde bias,we first select star particles in the simulations at 𝑧 = ± −
12 kpcradially from the center of the galaxy, which ensures that the sampleis not contaminated significantly from bulge stars; (2) with ironabundance [Fe/H] < − .
5. These conditions define our fiducialsample, however, we explore how our results vary with [Fe/H],and with how we spatially and kinematically select star particles inSection 3.1. We then define prograde or retrograde motion basedon the star particle’s action variables, 𝐽 𝜙 and 𝐽 𝑧 . 𝐽 𝜙 is equal to thez-component of angular momentum, 𝐿 𝑧 . We use the approximation 𝐽 𝑧 ≈ | 𝑧 𝜐 𝑧 | , where 𝑧 is a star particle’s vertical position and 𝜐 𝑧 is itsvertical velocity, relative to the host’s disk. Our choice is motivatedby the definition for specific action (per unit mass), 𝐽 𝑧 ≡ ∮ 𝜐 𝑧 𝑑𝑧 (1)where the integral is over one orbit in z. In the epicyclic approxi-mation ( 𝐽 𝑧 (cid:28) | 𝐽 𝜙 | ), 𝐽 𝑧 ≈ | 𝜐 𝑧 | 𝑧 max . (2) https://bitbucket.org/awetzel/halo_analysis https://bitbucket.org/awetzel/gizmo_analysis Because 𝑧 follows simple harmonic motion, (cid:104)| 𝑧 |(cid:105) = 𝑧 max /
2, where (cid:104)| 𝑧 |(cid:105) is the time-averaged height. Therefore, for a population oforbits described by the epicyclic approximation, trading time-averaging for spatial-averaging, we get 𝐽 𝑧 ≈ | 𝑣 𝑧 𝑧 | for the popu-lation. Even though this approximation is good only for stars nearthe plane of the disk on near-circular orbits, our geometric selec-tion of stars avoids stars that orbit more than 3 kpc above or belowthe disk at 𝑅 = . <𝐽 𝜙 / 𝐽 𝜙 (cid:12) < . 𝐽 𝑧 / 𝐽 𝑧 (cid:12) < − . < 𝐽 𝜙 / 𝐽 𝜙 (cid:12) < − . 𝐽 𝑧 / 𝐽 𝑧 (cid:12) < 𝐽 𝜙 (cid:12) = − kpc and 𝐽 𝑧 (cid:12) = .
35 km s − kpc.Our simulated MW/M31-mass galaxies have a range of rotationalvelocities ( 𝜐 rot ∼ −
230 km s − ), so we normalize 𝐽 𝜙 by8 × 𝜐 rot km s − kpc, where 𝜐 rot is the rotational velocity of the hostgalaxy, for stars at a similar galactocentric distance as the Sun.Given that we approximate the 𝐽 𝑧 action variable, we use the sameMW values to normalize 𝐽 𝑧 as in Sestito et al. (2020), and keepstars with | 𝐽 𝑧 | <
438 km s − kpc. As we will show below, our qual-itative results do not depend on the details of our selection window,if for example, we select all prograde ( 𝐽 𝜙 >
0) and all retrograde( 𝐽 𝜙 <
0) stars.Figure 1 shows the action space coordinates of all metal-poor([Fe/H] < − .
5) star particles within our selection region, for oneof our simulated galaxies, m12w. The black rectangles show ourfiducial selection windows to define prograde and retrograde stars.An overabundance of metal-poor stars lie in the prograde region(right box) compared to the retrograde region (left box), similar tothe MW results in Sestito et al. (2020).We track these stars, which we select at 𝑧 =
0, back in timeto track their origin, specifically, to determine whether they formed(a) in-situ, within 15 kpc physical of the most massive progenitorof the host or (b) in another galaxy that merged into the host. At
MNRAS000
MNRAS000 , 1–16 (2020) rograde metal-poor stars J /(8 kpc × rot [km s ]) J z [ k m s k p c ] ProgradeRetrograde m12w
Figure 1.
Metal-poor ([Fe/H] < − .
5) stars in the disk (4 ≤ R ≤
12 kpc , | Z | ≤ < − .
5) in the MW. Wenormalize 𝐽 𝜙 by what ‘solar’ values would be in the simulations, where 𝜐 rot is the rotational velocity of the galaxy at 8 kpc. Metal-poor stars in m12whave a strong preference for prograde orbits, with 𝑀 star , pro / 𝑀 star , ret ≈ early times, this distance may encompass other nearby low-massprogenitor galaxies, so we tested how the in-situ fractions changedusing stars within 10 and 20 kpc, and we found that they variedby approximately ± We first examine how the present prograde bias varies with ironabundance, [Fe/H]. Figure 2 shows the cumulative prograde biasfor all stars below a given [Fe/H], and the prograde bias for starswithin 0.5 dex bins of [Fe/H], for all of our hosts, as a functionof [Fe/H] at 𝑧 =
0. The thick black lines show the median acrossall 12 hosts, and we overlay Pristine observations of the MW fromSestito et al. (2019) in black crosses labeled ‘MW (S19)’. Solidlines show our 6 isolated hosts, while dashed line show our 6 hostsin LG-like pairs. We show stars only up to [Fe/H] = − . = − < − < − .
75 for the binned(middle panel) population. A few hosts (such as m12b) show a slightrise at [Fe/H] (cid:46) −
3. The prograde bias rapidly rises at higher ironabundance, as the stellar population quickly transitions to the moremetal-rich, younger, thin disk (and thus highly prograde) componentof each host. The observed values from Sestito et al. (2019) areclose to our median lines, so the MW is both consistent and typicalcompared with our simulation suite.
Figure 2.
The prograde bias, defined as the mass ratio of prograde toretrograde stars, 𝑀 star , pro / 𝑀 star , ret , versus stellar iron abundance for our12 MW/M31-mass galaxy simulations at 𝑧 = Top and middle panels :the prograde bias as a function of cumulative [Fe/H] and binned [Fe/H],respectively. Solid lines show the 6 isolated hosts, while dotted lines showthe 6 LG-like hosts. The thick black lines show the medians across all 12hosts, which are consistent with MW observations from the Pristine survey(Sestito et al. 2019, S19), which we show with black crosses. 11 of the 12galaxies in our sample show a significant prograde bias ( >
1) in metal-poor([Fe/H] < − .
5) stars. For most hosts, the prograde bias is nearly constantwith iron abundance at [Fe/H] (cid:46) − .
5, and it rises rapidly at higher [Fe/H].
Bottom : The binned prograde bias, split into in-situ (red) and ex-situ (blue)stars, where we select ‘in-situ’ stars as formed within 15 kpc of the host.The lines show the medians, and the shaded regions show the 68 per centscatter across 12 hosts. Ex-situ and in-situ stars have similar prograde biasat [Fe/H] (cid:46) − .
7, while ex-situ stars show a slightly higher prograde bias at − . (cid:46) [Fe/H] (cid:46) − .
7. As we will show, a single SMC/LMC-mass mergertypically drives this trend. At higher [Fe/H], the in-situ component quicklyrises in prograde bias, as stars transition to the thin-disk component.MNRAS , 1–16 (2020)
Santistevan et al.
Figure 2 shows that nearly all (11 of 12) of our hosts havesignificant prograde bias at all [Fe/H]. The key exception is m12i,which does not have any significant prograde bias at [Fe/H] (cid:46) − . (cid:38) − = − .
5, respectively.There are differences in the in-situ formation times between LG-likeand isolated galaxies, with LG-like galaxies forming around 𝑧 ∼ . 𝑡 lb ∼ . 𝑧 ∼ . 𝑡 lb ∼ . < − .
5. At [Fe/H] (cid:46) − .
75, in-situ stars tend to bea little more prograde, but at higher [Fe/H] up to ∼ − .
75, ex-situstars in fact show a slightly stronger preference for prograde orbits.At [Fe/H] > − .
75, in-situ stars quickly become the more progradepopulation, because we now transition to stars that primarily formedin the host’s thin disk.Figure 2 shows that the prograde bias depends only weakly oniron abundance at [Fe/H] (cid:46) −
2. Thus, to maintain consistency withthe selection in Sestito et al. (2020), we present results in Section 3.2and after (and Table 1) using [Fe/H] < − . Recently, the H3 Spectroscopic Survey (Conroy et al. 2019) alsoexamined the orbits of metal-poor ([Fe/H] < −
2) stars in the MW’shalo (out of the disk; Carter et al. 2020). They also found a strongprograde bias, with nearly 70 per cent of their stars on progradeorbits (defined as 𝐽 𝜙 > < | 𝑍 | < | 𝑏 | > ◦ .Following Carter et al. (2020), we calculate the prograde bias assimply the ratio of the mass of star particles with 𝐽 𝜙 > 𝐽 𝜙 <
0. As in Figure 2, Figure 3 shows each of our hostsseparately, and the thick black line shows the median.The results in Figure 3 are similar to those in Figure 2 (middle).The median prograde bias is flat across a wider range of iron abun-dance, − . < [Fe/H] < − .
75, before rising at higher [Fe/H]. Themedian prograde bias is consistently lower, with a value of ∼ . [Fe/H] M s t a r , p r o / M s t a r , r e t m12bm12cm12fm12im12mm12w RomeoJulietThelmaLouiseRomulusRemus medianMW (C20) Figure 3.
The prograde bias versus stellar iron abundance, similar to Figure 2(middle panel), but using using a ‘stellar halo’ selection function to mimicthe H3 survey (1 < | 𝑍 | < | 𝑏 | > ◦ ) and computing theprograde bias as the mass ratio of stars with 𝐽 𝜙 > 𝐽 𝜙 < < − . (cid:38) − at [Fe/H] = − . ∼ 𝐽 𝑧 , we now select stars with orbits farther from thedisk, which leads to a somewhat smaller prograde bias.Again, our simulation suite broadly agrees with MW obser-vations from Carter et al. (2020), although less well than with theresults of Sestito et al. (2020). At − . < [Fe/H] < −
1, the obser-vations are consistent with being constant, but they show an upwardtrend at lower [Fe/H]. The values below [Fe/H] < − . > − 𝐽 𝜙 > 𝐽 𝜙 <
0) using the Pristinespatial selection window in Section 3.1.1 and saw nearly identicalresults to Figure 3. The medians are essentially the same up to[Fe/H] ∼ − .
5, where the H3 selection has a slightly smaller valueof ∼
10 compared to ∼
15 when using the Pristine window. Thissuggests that implementing a more stringent spatial selection andignoring the disk midplane, as in the H3 survey, does not change theprograde bias results when compared to our entire disk selection.As a final check, we examined how the prograde bias varies usingthe stellar populations solely above versus below the disk: bothpopulations have nearly identical behavior.
In general, a star’s metallicity correlates with its age, though withsignificant scatter. Thus we investigate how the prograde bias at 𝑧 = MNRAS000
In general, a star’s metallicity correlates with its age, though withsignificant scatter. Thus we investigate how the prograde bias at 𝑧 = MNRAS000 , 1–16 (2020) rograde metal-poor stars
10 11 12 13
Stellar Age [Gyr] M s t a r , p r o / M s t a r , r e t m12bm12cm12fm12im12mm12w RomeoJulietThelmaLouiseRomulusRemus median Figure 4.
The prograde bias versus stellar age (regardless of iron abundance)at 𝑧 =
0. Solid lines show the 6 isolated hosts, dotted lines show the 6 LG-like hosts, and the thick black line shows the median across all 12 hosts.The prograde bias decreases with stellar age, such that younger stars tendto be on more prograde orbits, consistent with the expectation that youngerstars typically form on diskier orbits. However, even stars in our oldest agebin ( (cid:38) . ∼ .
2. All but one host(m12i) shows significant prograde bias to arbitrarily old stellar ages, and allhosts show prograde bias > (cid:46) . broadly similar results following the methodology of Carter et al.(2020) from the H3 Survey.Figure 4 shows the the prograde bias at 𝑧 = ∼ . ∼
100 for stars of age ∼ ∼ . (cid:38) . ∼ . 𝑧 = We next investigate the origin of the prograde bias in metal-poorstars, where we define ‘metal-poor’ as [Fe/H] < − . 𝑓 in − situ , primary M star, pro /M star, ret f r a c t i o n o f p r o g r a d e s t a r s f in situ f merger,1 f merger,1 + f in situ Figure 5.
For each host, the fraction of all of its prograde metal-poor([Fe/H] < − .
5) stars that formed in different components as a function ofthe host’s prograde bias at 𝑧 =
0. One host, m12m, has a particularly highprograde bias of ≈
9, as the break in the x-axis indicates. Blue open squaresshow the fraction from in-situ ( 𝑑 form <
15 kpc) stars that formed in the(most massive) progenitor of the host, while orange open triangles show thefraction from the ‘primary’ merger that contributed the most prograde stars.In nearly all (10 of 12) hosts, the primary merger contributed more progradestars than in-situ formation. Red squares show the sum of these two, whichaccounts for 30 −
70 per cent of all prograde metal-poor stars. merger fractions, 𝑓 merger , , and fraction of stars contributed fromthe primary, secondary, and tertiary mergers combined, 𝑓 merger , top3 .Figure 5 shows the fractions of all prograde metal-poor starsthat came from the primary merger (empty orange triangles) andthat formed in-situ (empty blue squares) versus the prograde bias at 𝑧 = 𝑓 merger , ≈ .
24 and 𝑓 in − situ ≈ . Thus, the primary merger gen-erally was responsible for more prograde metal-poor stars than the(most massive progenitor of the) host galaxy.
The primary mergerfractions range from ≈ −
55 per cent: the merger in Romulus con-tributed the most, and the mergers in Juliet and Louise contributedthe least. The in-situ fractions range from ≈ −
30 per cent. Further-more, on average, the primary merger contributed more progradestars ( ∼
24 per cent) than retrograde stars ( ∼
16 per cent) to the hostgalaxy.Figure 6 shows the ratio of the stellar masses contributed bythe primary (red), secondary (green), and tertiary (blue) mergers,relative to in-situ formation, of prograde metal-poor stars at 𝑧 = ≈
10 in the primarymerger to in-situ ratio, though we truncate the plot for clarity. Theprimary merger contributed twice as many stars as in-situ formation,so the primary merger is the single most important contributorto prograde metal-poor stars . Furthermore, the secondary mergercontributed comparable stellar mass as in-situ formation. Finally,the tertiary merger contributed typically only ∼ / (cid:38)
12 Gyr ago; this includes both stars that form in-situ and ex-situ.Almost all metal-poor stars formed prior to these three mergers
MNRAS , 1–16 (2020)
Santistevan et al.
Primary Secondary Tertiary
Merger01234 P r o g r a d e M m e r g e r / M i n s i t u Figure 6.
The ratio of the mass of prograde metal-poor stars contributedfrom a merger to that from in-situ formation, for the primary (red), secondary(green), and tertiary (blue) galaxy mergers, defined as the 3 progenitorgalaxies that deposited the most mass to the prograde metal-poor population.Squares show the median values across all 12 hosts, and the dark and lightvertical bars indicate the 68 per cent scatter and full range, respectively. Thered arrow indicates that the full range of values for the primary-to-in-situratios extend beyond the axis (to ∼ ≈ × more stars, the secondary merger wascomparable to in-situ, and the tertiary merger contributed only ∼ / 𝑧 = ( (cid:38)
93 per cent), thus, these mergers did not induce in-situ formationthat contributed to the prograde bias.Figure 5 and Table 1 show that in-situ formation together withthe primary merger account for 30 −
70 per cent of all progrademetal-poor stars. Considering the in-situ fraction along with the top3 mergers, this accounts for 43 −
96 per cent. Juliet has the small-est summed fraction, so an unusually large number of additionalmergers contributed significantly to its prograde metal-poor stars.In m12m, m12c, and Romulus, the combined fractions reach ≈ ≈ 𝑧 =
0, andthat the fractions of such stars from in-situ formation or the primarymerger do not predict the strength of this kinematic feature.We thus conclude that mergers dominate the origin of pro-grade metal-poor stars, and the primary merger dominates over anyother merger. We thus focus most of our subsequent analysis on theproperties of the primary merger.
We next investigate how the prograde bias evolved over time. Wecalculate the prograde bias at each snapshot spaced ≈
500 Myr,selecting stars based on the criteria in Section 2.3. We measured theorientation of the host’s disk separately at each snapshot, defined according to the moment of inertia tensor of all star particles in thehost. We then calculated the prograde bias at each snapshot basedon the host’s rotational velocity, 𝜐 rot , for normalizing 𝐽 𝜙 . Finally,we do not limit this analysis across time to star particles that areconfined to the plane of the disk throughout their lifetimes, onlythose that are within the action selection windows discussed above.Figure 7 (top sub-panels) shows the evolution of the progradebias for 6 of our hosts as a function of lookback time. These hostsshow a wide range of prograde biases and histories, and they arerepresentative of the other 6 hosts that we do not show. The verticalarrows show when each primary, secondary, and tertiary mergertook place (darkest to lightest, respectively).Figure 7 (top left) shows m12m, the host with the strongest pro-grade bias, ≈
9. As with many hosts, the prograde bias at early timeswas highly variable, caused by rapid merger/accretion activity andno well-defined long-lived disk. The prograde bias increased dra-matically after the secondary and primary mergers, with continuedgradual growth after the primary merger ∼ . ∼ . ∼ . ∼ ≈
2. Here, the (early) secondaryand tertiary mergers did not have much of an effect. m12b shows thata galaxy can have no significant long-term prograde bias throughoutmost of its history, until a late merger induces one. Both Thelma andm12c (not shown) show similar behavior, with a negligible progradebias until ∼ − ∼ ∼ . ∼ .
5, however, thesecondary merger caused the prograde bias to completely vanisharound 0 . ∼ . ∼ . ∼ .
5, where it remainedto 𝑧 =
0. This primary merger in Romulus contributed the highestfraction of prograde metal-poor stars compared to all other hosts(53 per cent). Both m12b and Romulus show how the timing ofthese mergers affects the prograde bias.We tested several variations in how to measure the prograde
MNRAS000
MNRAS000 , 1–16 (2020) rograde metal-poor stars Figure 7.
The formation histories for 6 galaxies in our sample. We show half of our simulated galaxies, which represent the diverse origins of the prograde biasacross our suite; the other galaxies have similar behavior. Arrows indicate the times of the top three galaxy mergers, ranked by the mass that they contribute toprograde metal-poor ([Fe/H] < − .
5) stars at 𝑧 =
0, with arrow darkness indicating the rank, such that the primary merger is the darkest. For each galaxy, theupper sub-panel shows the evolution of the prograde bias: the dotted horizontal line at 1 indicates no bias (equal mass in prograde and retrograde stars). Atearly times, the prograde bias was typically near 1, though this reflects the fact that the disk had not formed yet, so prograde and retrograde are not well defined;see discussion below. For all galaxies except m12i (top right), the prograde bias increased sharply following either the primary merger (as in m12m, m12w,m12b, Romulus) or the secondary merger (as in Romeo). m12w also shows an additional increase after the tertiary merger. For each galaxy, the lower sub-panelshows the evolution of the offset angle between the angular momentum vector of the galaxy’s disk and the angular momentum vector of the primary merger’sorbit, where we freeze the latter at its final orientation vector after the merger coalesces with the host (see Section 3.4). At early times, the offset angle of thehost’s disk was un-aligned with the (future) merger, and it rapidly shifted given the high merger/accretion rates, and that the disk itself was only marginallydefined. However, the host’s disk typically aligned itself with the orbital plane of the primary merger (as in m12m, m12w, m12b, Romulus) during/after themerger event. These mergers were typically gas-rich and drove (via torquing and gas deposition) the formation and/or prograde orientation of the host’s disk,seeding the origin of prograde metal-poor stars.MNRAS , 1–16 (2020) Santistevan et al. bias for each host across time, and we found similar overall trendsusing all of them. First, we tested using the same 𝐽 𝜙 selection regionfor each host at all snapshots, instead of scaling to each host’s 𝜐 rot at each snapshot. Second, while Figure 7 shows the prograde biasusing all metal-poor stars within the host’s selection region at eachsnapshot, we tested following back only the star particles that are inthe selection region at 𝑧 =
0. Finally, we examined how the progradebias evolves selecting all metal-poor stars within a radius of 15 kpcfrom the host, as opposed to our fiducial ‘disk’ spatial selection.For all of these variations, we found the same general evolutionarytrends, which reinforces that the prograde bias is a global feature ofthese MW/M31-mass galaxies.
We next investigate the relation between the primary merger’s orbitand the orientation of the stellar disk of the host galaxy. In particular,we examine whether the merger was on an orbit that was alignedwith the host’s pre-existing disk, whether the merger torqued thedirection of a pre-existing disk, and/or whether the merger helpedto seed the formation and orientation of the disk.The bottom sub-panels in Figure 7 show the evolution of theoffset angle, θ offset , for each host, defined as the angle between theangular momentum of the primary merger’s orbit and the angularmomentum of the host galaxy’s disk. We store the final angularmomentum vector of the primary merger just it merged into thehost, and we compare it with the evolving angular momentum ofthe disk after the merger.At early times, θ offset changed rapidly, caused by rapid accre-tion and mergers with many galaxies that built the host galaxy (e.g.Santistevan et al. 2020). As a result, the angular momentum of thehost’s (often poorly defined) disk quickly changed. All hosts in Fig-ure 7 except m12b show a rapid transition from a rapidly varying θ offset to a settled long-lived disk orientation. This transition oftencoincides with one of the major mergers.In both m12m and m12w (top left and middle left panels,respectively), just prior to the primary merger θ offset dipped sig-nificantly, and after the merger it went to near 0 ◦ . This means themerging galaxy was aligned with the resultant host’s disk afterthe merger , that is, the galaxy merged into the host, deposited themetal-poor stars that remain until 𝑧 =
0, and defined the progradedirection of the host’s disk. These events occurred early and helpedseed the formation of a stable long-lived stellar disk in the host. Af-ter these merger events, the disk continued to gradually change itsorientation, presumably from interaction with other galaxies and/oraccretion events. In m12m, the disk ends up tilted by ≈ ◦ withrespect to the primary merger orbit. However, in m12w, the disk didnot rotate much, remaining at ≈ ◦ with respect to the primarymerger orbit.Romulus (bottom right panel) shows similar behavior as m12mand m12w; starting ∼
500 Myr before the merger, the host’s diskflipped from ≈ ◦ down to ≈ ◦ and stayed there for ∼ ≈ ◦ ,similar to both Romeo and m12i (middle right and top right panels,respectively). Both Juliet and m12c are similar to m12m, m12w,and Romulus, because there were temporary merging events thatdrove θ offset down, while subsequent merger events caused θ offset to rise/rotate again.Romeo and m12i are different from the rest of the hosts shownhere, because θ offset remained rapidly variable both before and afterthe primary merger, stabilizing only ∼ − θ offset atearly times, including before, during, and after the primary merger,but eventually the disk settleed and θ offset remained fairly constantat 30 − ◦ .Finally, m12b (bottom left panel) is unique among our suite,because the primary merger occurred late, ∼ . <
20 per cent of the stellar massof the host, the merger only moderately torqued the orientation ofthe host’s disk. m12b provides our single example that a late-time merger can drive significant prograde bias as well.As we discuss in further detail in Section 3.5, the strength ofthe prograde bias does not correlate with the gas or stellar masses ofthe primary mergers, nor the gas or stellar mass ratios of the primarymergers to their host galaxies. At first glance, it may seem surprisingthat there is no correlation of these merger/host properties with theprograde bias because these mergers largely source the progrademetal-poor population, however, as we mentioned above, they alsodeposit gas that contributes to the formation of the host’s disk andset the prograde direction. These lacks of correlations hold trueacross different times (when the primary merger occurred, whenthe primary merger was at its peak stellar mass, 300 Myr beforethe primary merger) and across different ways we spatially andkinematically select gas particles in the galaxies.
Finally, we investigate correlations between properties of theMW/M31-mass host galaxies, or their galaxy mergers, and theirprograde biases. Figure 8 shows distributions of key properties ofthe primary mergers and lists the Spearman correlation coefficientsand p-values with respect to the prograde bias. Table 2 lists all ofthe correlations that we tested.Given that both in-situ stars and the primary merger contributesignificantly to the prograde stars, we explore whether hosts whoseprimary merger’s orbit was more aligned with its disk orientation before the merger show a stronger prograde bias. Specifically, weexamine the median θ offset over the 500 Myr before the primarymerger, and Figure 8 (top left) shows its relation to the progradebias at 𝑧 = θ offset , median ranges from ∼ − ◦ ,with Romulus having the smallest and m12m having the largestoffset. In particular, m12m has both the largest pre-merger offsetangle and the largest prograde bias. m12w and m12i have values inthe middle of the sample, even though they have the second-mostand least prograde biases, respectively. We thus conclude that theorientation of the primary merger’s orbit with the host’s pre-existingdisk has no significant role in driving prograde bias for metal-poorstars.
Figure 8 (top right) shows the prograde bias versus the look-back time of the primary merger. Almost all primary mergers oc-curred > ≈ MNRAS000
Figure 8 (top right) shows the prograde bias versus the look-back time of the primary merger. Almost all primary mergers oc-curred > ≈ MNRAS000 , 1–16 (2020) rograde metal-poor stars r s = 0.01 p s -value = 0.97 M s t a r , p r o / M s t a r , r e t offset, median m12bm12cm12fm12im12mm12w RomeoJulietThelmaLouiseRomulusRemus r s = 0.25 p s -value = 0.43 M s t a r , p r o / M s t a r , r e t t lb, merger [Gyr] r s = 0.36 p s -value = 0.26 M s t a r , p r o / M s t a r , r e t M star, merger [ M ] r s = 0.28 p s -value = 0.38 M s t a r , p r o / M s t a r , r e t f gas Figure 8.
The prograde bias of metal-poor ([Fe/H] < − .
5) stars in each MW/M31-mass galaxy at 𝑧 = Top left :the median offset angle between the angular momentum vector of the host’s disk and the angular momentum vector of the primary merger’s orbit, θ offset , median ,over 500 Myr prior to the merger. θ offset , median indicates how prograde the merger’s orbit was with respect to the host’s existing disk, with 0 ◦ indicating completealignment. The prograde bias does not correlate with θ offset , median ; rather, the primary merger drove the orientation of the host’s disk after the merger. Top right :the lookback time when the primary merger occurred, 𝑡 lb , merger . Most mergers occurred 7 . − . ≈ 𝑡 lb , merger , such that earlier primary mergers cause a lower prograde biases, likely because the hostgalaxies have had more time to merge with other galaxies, which can phase-mix the stars. Bottom left : the stellar mass of the primary merger, 𝑀 star , merger ,which ranges from 10 − M (cid:12) , and which ranges from 0 . − × the stellar mass of the host at the time of the merger. The prograde bias increases weaklywith 𝑀 star , merger , as expected. Bottom right : the gas fraction of the primary merger, 𝑓 gas , merger = 𝑀 gas , merger / (cid:0) 𝑀 gas , merger + 𝑀 star , merger (cid:1) . This ranges from ≈ . − .
95, though most (10 of 12) primary mergers have 𝑓 gas , merger > .
5. The prograde bias decreases weakly with 𝑓 gas , merger . Each panel lists theSpearman correlation coefficient and p-value: we find no strong correlation between the prograde bias and these or almost any other property that we tested(see Table 2). Thus, while metal-poor stars nearly ubiquitously prefer prograde orbits in our simulations, the strength of this prograde bias has a complexdependence on formation/merger history. spans 10 − M (cid:12) , roughly evenly in log mass. Given the stellarmasses of the host galaxies at these merger times (Santistevan et al.2020), the primary merger galaxy was ∼ −
95 per cent as massive asthe host during the merger. As expected, the prograde bias increaseswith the stellar mass of the primary merger, but the strength of thiscorrelation is weak, as is the correlation with the the ratio of stellarmasses between the primary merger and the host (see Table 2).We also examine whether the gas content of the primary mergercorrelates with the prograde bias. The prograde metal-poor starsformed (cid:38)
12 Gyr ago, before the primary merger occurred, so thegas content during the merger event did not contribute to this popula-tion. Rather, the (more metal-rich) gas deposited by the merger couldcontribute to the formation/stabilization of the host’s disk, becausethe merger deposited significant gas mass on a single angular mo-mentum vector. Thus, one might expect that higher gas content in the primary merger would drive stronger prograde-ness of the depositedmetal-poor stars with respect to the resultant disk. Because, gas canbe stripped as the merger orbits around and coalesces with the host,we measure the gas mass of the merging galaxy 300 Myr beforemerging. We include all gas that is within 𝑅 star , , the radius that en-closes 90 per cent of the stellar mass of the merging galaxy, and thathas relative velocity < × 𝜎 star , the stellar velocity dispersion of themerging galaxy. Most of these primary mergers were gas-rich, bring-ing in between ≈ − M (cid:12) of gas. We then calculate their gasfractions as 𝑓 gas , merger = 𝑀 gas , merger /( 𝑀 gas , merger + 𝑀 star , merger ) .Figure 8 (bottom right) shows the prograde bias versus the pri-mary merger’s 𝑓 gas , merger . Most (10 of 12) primary mergers weregas-rich, with 𝑓 gas , merger > .
5, and more than half had values > .
75. The two exceptions are m12b, which merged most recently,and Thelma, whose primary merger occurred ∼ MNRAS , 1–16 (2020) Santistevan et al.
Table 2.
Properties that we tested for a correlation withthe prograde bias of metal-poor stars at 𝑧 =
0. Columnlist: property name; Spearman correlation coefficient, r s ;Spearman p-value. We calculate the mass properties/ratiosat the following different times: present-day ( 𝑡 ); at the timeof the primary merger ( 𝑡 merger , ); at the time the primarygalaxy merger had its peak stellar mass ( 𝑡 merger , peak ); and300 Myr before the primary merger ( 𝑡 ).Property r s p s value 𝑓 in − situ -0.11 0.73 𝑓 merger , 𝑀 star , host ( 𝑡 ) 𝑀 halo , host ( 𝑡 ) 𝑀 gas , host ( 𝑡 ) 𝑀 star , host / 𝑀 halo , host ( 𝑡 ) 𝑀 star , merger ( 𝑡 merger , peak ) 𝑀 gas , merger ( 𝑡 ) -0.09 0.78 𝑀 star , merger / 𝑀 star , host ( 𝑡 merger ) 𝑀 star , merger / 𝑀 star , host ( 𝑡 merger , peak ) 𝑀 gas , merger / 𝑀 gas , host ( 𝑡 ) -0.57 0.05 𝑀 gas , total ( 𝑡 ) 𝑓 gas , merger -0.28 0.38 𝑡 lb , merger -0.25 0.43 𝑡 lb , -0.43 0.16 𝑡 lb , -0.28 0.38 𝑡 lb , form -0.34 0.27 𝑁 sat 𝜐 rot / 𝜎 θ offset , median -0.01 0.97 was gas poor. Perhaps surprisingly, we find that more gas-rich pri-mary mergers correspond to slightly weaker prograde bias, thoughagain this correlation is weak. For example, m12i has the small-est prograde bias but also the second-highest 𝑓 gas , merger ( ≈ . 𝑓 gas , merger .In total, we examined correlations of the prograde bias at 𝑧 = 𝑟 𝑠 ) and p-value at present-day ( 𝑡 ), at the time of the primary merger( 𝑡 merger , ), at the time the primary galaxy merger had its peak stellarmass ( 𝑡 merger , peak ), and 300 Myr before the primary merger ( 𝑡 ).We briefly address each below: • 𝑓 in − situ and 𝑓 merger , : Figure 5 also shows these correlations.While 𝑓 merger , does positively correlate with the prograde bias, thestrength of its correlation is weak, indicating additional dependence,such as time and gas-richness of the merger. Importantly, 𝑓 in − situ shows no meaningful correlation with the prograde bias. • The ratio of stellar mass of the primary merger and the hostdoes not correlate significantly with prograde bias. We testedmeasuring this ratio both at the time when the primary mergerreached its peak stellar mass ( 𝑀 star , merger / 𝑀 star , host ( 𝑡 merger , peak ) ,and at the time immediately before merging into the host galaxy( 𝑀 star , merger / 𝑀 star , host ( 𝑡 merger ) . • 𝑀 gas , merger / 𝑀 gas , host ( 𝑡 ) : the ratio of the gas mass in theprimary merger to that in the host galaxy. We explored four ways of measuring gas masses, including 300 Myr or 600 Myr before theprimary merger, using only cold gas ( 𝑇 < × K), or measur-ing gas out to 2 × 𝑅 . Interestingly, this ratio 300 Myr before themerger is the most significant correlation that we find (p-value =0.05). However, we caution the reader to over-interpreted this result,because the strength of this correlation varies greatly with how wemeasure gas masses (p-values = 0.23-0.78). In particular, the gasmass of the primary merger is relatively robust across our methods,but the gas mass of the host galaxy varies more significantly, whichmay be because of ambiguity in mass association during or justbefore a merger. • Host mass: 𝑀 star , host ( 𝑡 ) , 𝑀 halo , host ( 𝑡 ) , 𝑀 gas , host ( 𝑡 ) , and 𝑀 star , host / 𝑀 halo , host ( 𝑡 ) , correlate at most weakly with progradebias. The strongest correlation is the current ratio of stellar mass todark-matter halo mass. • 𝑀 gas , total ( 𝑡 ) : the total gas mass, summing that of the pri-mary merger and the host galaxy, does not correlate significantlywith prograde bias. • 𝑡 lb , , 𝑡 lb , , and 𝑡 lb , form : we investigated three metrics of hostgalaxy formation: the time when the galaxy formed 50 per cent ofits final mass, the time when it reached 90 per cent of its final mass,and the time when it transitioned from having primarily ex-situto primarily in-situ stellar mass. All of these correlate negativelywith prograde bias, however, the significance of these correlationsremains weak. 𝑡 lb , correlates most strongly, perhaps because itcorrelates best with typical times of the most important mergers.Despite differences in the formation histories between galaxies inLG-like pairs, and galaxies in isolated environments mentioned inSantistevan et al. (2020), and the small positive correlation betweenthe primary merger times ( 𝑡 lb , merger ) and host galaxy formationtimes ( 𝑡 lb , form ; p-value=0.05), we found no significant differencesin the metal-poor, prograde biases between LG-like and isolatedhost galaxies, which suggests prograde bias does not correlate withhost galaxy formation time. • 𝑁 sat : the number of satellite dwarf galaxies ( 𝑀 star > M (cid:12) )within the host halo at 𝑧 = • Primary merger ordering: we compare the relative timing ofthe primary merger to the secondary or tertiary mergers, that is,whether the primary merger occurred first, second, or third, whichwe define as the primary merger ordering. We find only a weakcorrelation, such that hosts whose primary merger happened themost recently out of the 3 have a somewhat stronger prograde bias. • 𝜐 rot / 𝜎 : the ratio of the rotational velocity to the velocity dis-persion of stars in the host at 𝑧 = 𝜐 rot / 𝜎 is a kinematic metric of‘diskiness’. We calculated both 𝜐 rot and 𝜎 using all stars in the disk,however, we also calculated the ratio using only stars that are metal-poor (threshold from − . < [Fe/H] < − . 𝜐 rot / 𝜎 , but it is weak. We investigated the prevalence and origin of the preference ofmetal-poor stars to be on prograde disk orbits using 12 MW/M31-mass galaxies from the FIRE-2 simulation suite. We reiterate thequestions that we articulated in the introduction, and we summarize
MNRAS000
MNRAS000 , 1–16 (2020) rograde metal-poor stars our answers to them:i) How commonly do MW/M31-mass galaxies show a preferencefor prograde motions in their metal-poor stars, as observed in theMW, and what is the range in strength of this prograde bias? •
11 of 12 of our MW/M31-mass galaxies show a preferencefor prograde orbits among metal-poor stars (Figure 1). • Our simulations predict that this prograde bias is a generalfeature of MW/M31-mass galaxies throughout most of their his-tory, and its presence does not depend significantly on the waythat we spatially or kinematically select metal-poor stars (Fig-ures 2, 3, 4). • The present-day prograde biases for our sample range from0 . − .
14, with a median value of 2 . • We find no significant differences in prograde bias betweenLG-like and isolated host galaxies (Figure 2).ii)
How does this prograde bias depend on the metallicity and/orage of the stars? • We find little-to-no dependence of the prograde bias on theiron abundance of stars at [Fe/H] (cid:46) − • Our results broadly agree with recent observations of theMW from the Pristine and H3 surveys (Figures 2 & 3). •
11 of 12 of our galaxies have a prograde bias for arbitrarilyold stars ( (cid:38) . (cid:46) . What process(es) cause this prograde bias among metal-poorstars? • We find no large difference in the prograde bias for stars thatformed in-situ versus ex-situ (Figure 2). The sample of progrademetal-poor stars (both in-situ and ex-situ) almost entirely formed (cid:38)
12 Gyr ago, prior to the primary mergers. • A single galaxy merger is typically the most important causeof the prograde bias, with additional but sub-dominant contribu-tions from in-situ star formation and lesser mergers (Figures 5 &6). This primary merger typically contributed ∼
24 per cent of allprograde metal-poor stars, while the fraction that formed in-situis ∼
16 per cent. Combining in-situ stars with those from the top3 mergers accounts for ∼
70 per cent, on average (Table 1). • We do not find any significant correlation in orientationbetween the primary merger’s orbit and any pre-existing disk inthe host (Figure 7). • The primary merger occurred typically 7 − . 𝑓 gas , merger (cid:38) . • In 3 of our simulated hosts, the host stellar disk orientationbecame settled/long-lived within ∼
500 Myr of 1 of the 3 mergerswe tracked, and in two cases, the orientation became settled, but itsubsequently changed via other mergers (Figure 7). This suggeststhat mergers can be a dominant factor in the formation of stellardisks.iv)
Do any properties of the MW/M31-mass galaxy or theirgalaxy mergers correlate with this prograde bias? • We explored correlations with 21 properties of the hostgalaxy or primary merger, but we find few clear correlations with the strength of the prograde bias (Table 2). This indicates thatthe strength of the prograde bias has a complicated dependenceon galaxy formation/merger history, which limits our immediateinterpretation of how to translate recent observations of the pro-grade bias for the MW into a robust constraint on its formationhistory.
A few galaxy mergers predominantly drive the growth of the metal-poor component of MW/M31-mass galaxies/halos. For example,using a suite of dark-matter-only simulations of MW-mass ha-los and applying abundance matching to assign stellar mass to(sub)halos, Deason et al. (2016) found that most of the accretedstellar mass in the stellar halo comes from only 1-2 dwarf galaxieswith 𝑀 star = − M (cid:12) . Similarly, in Santistevan et al. (2020)we used the same 12 simulated galaxies in this paper and found thatstars that form ex-situ (in galaxies other than the most massive pro-genitor) and accrete into the host galaxy largely come from the mostmassive progenitor galaxies, and that this is true for progenitors atall redshifts. Because just a few galaxies are responsible for mostof the accreted (metal-poor) population, they leave a kinematic im-print, as we have shown. Accreted substructures populate our owngalaxy, such as Gaia-Enceladus (Helmi et al. 2018; Belokurov et al.2018), the Sagittarius stream (Ibata et al. 1994; Newberg et al. 2003;Majewski et al. 2003), Sequoia (Barbá et al. 2019; Myeong et al.2019; Matsuno et al. 2019), and a few others (e.g., see Naidu et al.2020). Therefore, we consider it likely that one (or more) of thesemergers corresponds the origin of the MW’s prograde bias, thoughwe defer a more detailed connection to future work.In their analysis of the prograde bias in the H3 survey, Carteret al. (2020) suggest that these metal-poor stars likely came fromeither (1) a combination of in-situ and ex-situ star formation duringthe early assembly and formation of the MW, which is broadlysimilar to what we find, or (2) late accretion of dwarf galaxieson prograde orbits with respect to the disk. However, because thestars in the MW have not been kinematically heated enough tochange their prograde orbits, the authors claim that the angularmomentum of the MW has been in place for at least 12 Gyr. Fromthe bottom sub-panels of Figure 7, we see find changes in θ offset at early times, with a sharp transition, after which the angle slowlyprecesses over time. This transition marks when the stellar diskorientation stabilized in the host, and although the disk can precessafterwards, θ offset remains relatively stable. We find a range of timeswhen this transition occurs, across ≈ − . ∼ . 𝑧 orientation ofthe disk at 𝑧 = MNRAS , 1–16 (2020) Santistevan et al. the Illustris simulation (Vogelsberger et al. 2014) and showed thateven though 38 of them experienced a (cid:38) 𝑧 ∼ . 𝑧 =
0. The most important factorin determining the final outcome of such a merger is the amountof gas, given that stars can form from this gas after the merger tobuild a disk, while the older pre-existing stars can be heated duringthe merger to form a thick-disk/halo/ellipsoidal component in thegalaxy. In our simulated sample, 300 Myr prior to the primarymergers, the gas mass ratios of the primary merging galaxy to thehost span values from ∼ .
002 in Thelma, to as high as ∼ . ∼ × − × M (cid:12) . In 3 of our hosts (m12m, m12w, and Romulus),the orientation of the disk settles from the rapid variations caused byearly accretion/mergers within ≈
500 Myr of the three mergers weinvestigated here, suggesting that these mergers drove the formationand prograde orientation of the disk. In 2 other hosts (m12c andJuliet), 1 of the 3 mergers caused the orientation of the disk tosettle, but other subsequent mergers caused the disk to then changeorientation again. Our results qualitatively agree with Peschken et al.(2020) and suggest that mergers were a dominant mode of promotingthe transition to thin-disk morphology, with much of the gas massmentioned above likely contributing to subsequent star formation inthe disk region.Recently, Sestito et al. (2021) performed a similar study as oursusing 5 of the NIHAO-UHD cosmological zoom-in simulations ofMW-mass galaxies (Buck et al. 2020). Building on their previouswork, they found that the simulated galaxies also show a prevalencefor a prograde bias. The authors also showed that a rotating spheroidof metal-poor stars, akin to a halo-like or pressure supported dis-tribution of stars from the early accretion phase, cannot explain theorigin of the prograde bias in a galaxy. Rather, the early accretionand mergers of progenitor galaxies is the dominant source of metal-poor planar stars, with a smaller additional contribution from lateraccretion of satellite galaxies on prograde orbits that get draggedand disrupted into the proto-disk. Furthermore, they suggest that ∼ −
93 per cent of metal-poor stars ([Fe/H] < −
2) in thosesimulations are older than ∼
12 Gyr, where the retrograde planarpopulation traces the early accretion of the main galaxy and theprograde planar population better traces the full history. Our resultsqualitatively agree with this picture, and we discuss the similaritiesand differences below.Both our analysis and Sestito et al. (2021) find that this popu-lation of metal-poor stars within the disks of MW-mass galaxies hasa preference of prograde motion as opposed to retrograde. Sestitoet al. (2021) find a prograde bias in all 5 NIHAO-UHD simulatedgalaxies, and we find a prograde bias in 11 of our 12 galaxies. Thisreinforces that the prograde bias for metal-poor stars is a nearly(but not completely) ubiquitous feature of MW/M31-mass galax-ies. Our sample of 12 host galaxies is larger than that in Sestitoet al. (2021), which is likely why we find one host without a pro-grade bias. The strengths of the prograde biases differ significantlybetween our analysis and theirs. Sestito et al. (2021) find progradebiases of ∼ . − .
1, higher than the MW, while our simulationshave 0 . − .
1, with a median value nearly identical to the MW.In particular, 11 of our hosts have prograde biases (cid:46)
3, below thesmallest prograde bias in Sestito et al. (2021), which may indicatethe importance on these results from differences in physical and/ornumerical implementations between our FIRE-2 and the NIHAO-UHD simulations.We qualitatively agree about the origins of prograde metal-poor stars. These star particles primarily came from a number of mergers, both from the early accretion and late growth phases. Eventhough a non-negligible fraction of these stars form in-situ (5 − 𝑀 star (cid:38) M (cid:12) ) merged intothe host on orbits aligned with the angular momentum of the stellardisk around the time the host galaxy formed 25 per cent of its finalmass ( 𝑡 ). At these times, they report that ∼ −
90 per cent of theprograde planar stars are already in the host galaxy from the earlyassembly phase. However, from the bottom sub-panels in Figure 7,we show that the orientation of the proto-disk is not well defineduntil the primary merger (or another lesser merger) occurs, implyingthat the primary merger helped to seed the formation and set theorientation of the host’s disk by depositing a significant supply ofgas on a coherent angular momentum vector. Thus, this means thatthe primary merger determined the prograde direction in the host,and that these progenitor galaxies do not merge into the host onorbits aligned with the disk because the disk was not well-definedbefore this. Our answer to the origin of prograde metal-poor stars inthe disks of MW-mass galaxies is then most similar to explanation(iii) from Sestito et al. (2019): these stars formed inside one or moreprogenitor galaxies that merged into the MW-mass host as it wasforming.As a final comparison, we note important differences betweenour analysis and Sestito et al. (2021). The authors do not explorethe stability of the stellar disk orientations in their Figure 5 at 𝑡 .Therefore, it is unclear what effect, if any, the satellite progenitorgalaxies had on the orientation of the host. Although they quantifythe growth of the accreted stellar mass of the host galaxy, they donot examine the gas mass of the merging galaxies, which plays alarge role in shaping the disk. In this paper, we only focus on theevolution of the ratio of the stellar mass of prograde to retrogradestars, but Sestito et al. (2021) suggests that 74 −
90 per cent of theretrograde planar population are primarily accreted before 𝑡 , whilethe prograde population is accreted throughout most of cosmic time.This paper also investigated several properties of the host galaxy tolook for possible correlations with the strength of its prograde bias.Finally, we do not analyze other regions in kinematic space, such asthe high eccentricity feature in Sestito et al. (2021).Evidence for the prograde, metal-poor population in the MWdisk is further supported by a recent study involving stars located inthe stellar halo and thick disk from the SkyMapper Survey for ex-tremely metal-poor stars (Da Costa et al. 2019). In a sample of 475stars with iron abundances in the range of − . (cid:46) [Fe/H] (cid:46) − ∼
21 per cent of their sam-ple have disk-like orbits confined to 3 kpc of the disk’s midplane.The authors suggest that this sub-sample of metal-poor stars withdisk-like orbits consists of a high-eccentricity and low-eccentricitypopulation, where the low-eccentricity population are on primarilyprograde orbits (defined as 𝐽 𝜙 >
0) with a prograde bias of ∼ . | 𝑍 | < and were accreted early in the MW’s formationhistory.Similar studies focused on old, metal-poor stars in the MWalso halo have investigated their origin and kinematics. By select-ing stars within 3 kpc of the Sun with halo-like kinematics, Bonacaet al. (2017) found that there are an excess of metal-rich halo stars MNRAS000
0) with a prograde bias of ∼ . | 𝑍 | < and were accreted early in the MW’s formationhistory.Similar studies focused on old, metal-poor stars in the MWalso halo have investigated their origin and kinematics. By select-ing stars within 3 kpc of the Sun with halo-like kinematics, Bonacaet al. (2017) found that there are an excess of metal-rich halo stars MNRAS000 , 1–16 (2020) rograde metal-poor stars ([Fe/H] > −
1) with prograde orbits compared to the metal-poorhalo stars ([Fe/H] < −
1) which have no orbital preference. Theauthors compared this with one of the simulations in our sam-ple, m12i, and found similar results. They found that the two halopopulations also have different origins, with the metal-rich starsprimarily forming in-situ, and the metal-poor population accretingfrom mergers (ex-situ). A similar study using the APOSTLE sim-ulations by Starkenburg et al. (2017a) suggests that the fractionsof metal-poor ([Fe/H] < − .
5) and old ( 𝑡 form < . ∼
60 per cent of thestars orbiting outside of the solar circle. We did not examine theprograde bias of the halo region in our simulations, but rather, wecompared how the prograde bias evolution changed when selecting all metal-poor ([Fe/H] < − .
5) stars within 15 kpc, radially. Wesaw the same qualitative evolution of the prograde biases as in thetop sub-panels of Figure 7. This is not a strict ‘halo selection’ ofstars, but given that we showed that the prograde bias is a generalfeature, independent of the kinematic and spatial selection windowswe examined, we do not expect our results to change significantlyby making a more strict selection of halo stars. Similarly, El-Badryet al. (2018b) investigated the spatial distribution and dynamics ofthe oldest stars (z form (cid:38)
5) in MW-mass galaxies and found thatthe majority formed ex-situ. The authors found that metal-rich stars([Fe/H] > −
1) typically have disk-like orbits, while the more metal-poor stars reflect an isotropic distribution. Furthermore, when onlyselecting the oldest stars ( 𝑧 form (cid:38) 𝜐 𝜙 > − .Instead of selecting stars based on their distributions in velocityspace, as in a Toomre diagram, one recent study tested a Gaussianmixture model to deconstruct the different components of the MWusing the star’s velocities and iron abundances (Nikakhtar et al.,in prep). The authors used data from APOGEE (Majewski et al.2017) and Gaia DR2 (Gaia Collaboration et al. 2018), and createdmock catalogs using 3 of the simulations in our study (m12f, m12i,m12m), and showed that the MW is best described by five differentcomponents: the stellar halo, the thin disk, the metal-rich thick disk,and a 2-component metal-poor thick disk. Nikakhtar et al. (in prep)suggest that the 3-component thick disk originates from a combina-tion of kinematic heating (via gravitational interactions) and radialmigration of early forming stars: we show that a significant frac-tion of the most metal-poor stars in the disk come from multiplemergers during the galaxy’s early assembly. In a different approach,using the same 12 simulated galaxies as in our sample, Yu et al.(in prep) differentiated thin disk and thick disk stars based on theircircularity (angular momenta relative to the angular momentum ofa circular orbit) and suggest that the transition from thick disk tothin disk formation correlates with the transition from early, burstystar formation to steady, near constant star formation. This transitionoccurred some time between ∼ . − . ACKNOWLEDGEMENTS
We greatly appreciate interesting and fruitful discussions with bothFederico Sestito and Nicolas Martin, as well as Courtney Carter andCharlie Conroy. We also express our gratitude for their sharing ofobservational data. IBS, AW, and JS received support from NASAthrough ATP grants 80NSSC18K1097 and 80NSSC20K0513; HSTgrants GO-14734, AR-15057, AR-15809, and GO-15902 fromSTScI; a Scialog Award from the Heising-Simons Foundation; and aHellman Fellowship. AW performed this work in part at KITP, sup-ported by NSF grant PHY-1748958. CAFG was supported by NSFthrough grants AST-1715216 and CAREER award AST-1652522;by NASA through grant 17-ATP17-0067; by STScI through grantHST-AR-16124.001-A; and by a Cottrell Scholar Award and a Scia-log Award from the Research Corporation for Science Advance-ment. RES acknowledges support from NASA grant 19-ATP19-0068, NSF grant AST-2009828, and HST-AR-15809 from the SpaceTelescope Science Institute (STScI), which is operated by AURA,Inc., under NASA contract NAS5-26555. We ran simulations using:XSEDE, supported by NSF grant ACI-1548562; Blue Waters, sup-ported by the NSF; Pleiades, via the NASA HEC program throughthe NAS Division at Ames Research Center.This paper used various python packages including NumPy(Harris et al. 2020), SciPy (Virtanen et al. 2020), and Matplotlib(Hunter 2007), as well as NASA’s Astrophysics Data System.
DATA AVAILABILITY [Fill out later]
REFERENCES
Athanassoula E., Rodionov S. A., Peschken N., Lambert J. C., 2016, ApJ,821, 90Barbá R. H., Minniti D., Geisler D., Alonso-García J., Hempel M., Monach-esi A., Arias J. I., Gómez F. A., 2019, ApJ, 870, L24Behroozi P. S., Wechsler R. H., Wu H.-Y., 2013a, ApJ, 762, 109Behroozi P. S., Wechsler R. H., Wu H.-Y., Busha M. T., Klypin A. A.,Primack J. R., 2013b, ApJ, 763, 18Belokurov V., Erkal D., Evans N. W., Koposov S. E., Deason A. J., 2018,Monthly Notices of the Royal Astronomical Society, 478, 611Bird J. C., Kazantzidis S., Weinberg D. H., Guedes J., Callegari S., MayerL., Madau P., 2013, ApJ, 773, 43Bird J. C., Loebman S. R., Weinberg D. H., Brooks A., Quinn T. R., Chris-tensen C. R., 2020, arXiv e-prints, p. arXiv:2005.12948Bonaca A., Conroy C., Wetzel A., Hopkins P. F., Kereš D., 2017, ApJ, 845,101Bournaud F., Combes F., Jog C. J., 2004, A&A, 418, L27Bournaud F., Jog C. J., Combes F., 2005, A&A, 437, 69Brook C. B., Kawata D., Scannapieco E., Martel H., Gibson B. K., 2007,ApJ, 661, 10Buck T., Obreja A., Macciò A. V., Minchev I., Dutton A. A., Ostriker J. P.,2020, MNRAS, 491, 3461Bullock J. S., Johnston K. V., 2005, ApJ, 635, 931Bullock J. S., Kravtsov A. V., Weinberg D. H., 2001, ApJ, 548, 33Carney B. W., Laird J. B., Latham D. W., Aguilar L. A., 1996, AJ, 112, 668Carter C., et al., 2020, arXiv e-prints, p. arXiv:2012.00036Chiba M., Beers T. C., 2000, AJ, 119, 2843Conroy C., et al., 2019, ApJ, 883, 107Cordoni G., et al., 2020, MNRAS,Da Costa G. S., et al., 2019, MNRAS, 489, 5900De Silva G. M., et al., 2015, MNRAS, 449, 2604Deason A. J., Belokurov V., Evans N. W., 2011, MNRAS, 411, 1480Deason A. J., Mao Y.-Y., Wechsler R. H., 2016, ApJ, 821, 5MNRAS , 1–16 (2020) Santistevan et al.
Di Matteo P., Spite M., Haywood M., Bonifacio P., Gómez A., Spite F.,Caffau E., 2020, A&A, 636, A115Eggen O. J., Lynden-Bell D., Sandage A. R., 1962, ApJ, 136, 748El-Badry K., et al., 2018a, MNRAS, 473, 1930El-Badry K., et al., 2018b, Monthly Notices of the Royal AstronomicalSociety, 480, 652Escala I., et al., 2018, MNRAS, 474, 2194Fall S. M., Efstathiou G., 1980, MNRAS, 193, 189Faucher-Giguère C.-A., Lidz A., Zaldarriaga M., Hernquist L., 2009, ApJ,703, 1416Freeman K., Bland-Hawthorn J., 2002, ARA&A, 40, 487Gaia Collaboration et al., 2018, A&A, 616, A1Gallart C., Bernard E. J., Brook C. B., Ruiz-Lara T., Cassisi S., Hill V.,Monelli M., 2019, Nature Astronomy, 3, 932Garrison-Kimmel S., et al., 2017, MNRAS, 471, 1709Garrison-Kimmel S., et al., 2018, MNRAS, 481, 4133Garrison-Kimmel S., et al., 2019a, MNRAS, 487, 1380Garrison-Kimmel S., et al., 2019b, MNRAS, 489, 4574Gilmore G., Reid N., 1983, MNRAS, 202, 1025Grand R. J. J., Springel V., Gómez F. A., Marinacci F., Pakmor R., CampbellD. J. R., Jenkins A., 2016, MNRAS, 459, 199Griffen B. F., Dooley G. A., Ji A. P., O’Shea B. W., Gómez F. A., Frebel A.,2018, MNRAS, 474, 443Guedes J., Callegari S., Madau P., Mayer L., 2011, ApJ, 742, 76Hahn O., Abel T., 2011, MNRAS, 415, 2101Harris C. R., et al., 2020, Nature, 585, 357Haywood M., Di Matteo P., Lehnert M. D., Snaith O., Khoperskov S., GómezA., 2018, ApJ, 863, 113Helmi A., Babusiaux C., Koppelman H. H., Massari D., Veljanoski J., BrownA. G. A., 2018, Nature, 563, 85Hopkins P. F., 2015, MNRAS, 450, 53Hopkins P. F., 2016, MNRAS, 455, 89Hopkins P. F., Cox T. J., Younger J. D., Hernquist L., 2009, ApJ, 691, 1168Hopkins P. F., et al., 2018, MNRAS, 480, 800Hunter J. D., 2007, Computing in Science Engineering, 9, 90Ibata R. A., Gilmore G., Irwin M. J., 1994, Nature, 370, 194Kroupa P., 2001, MNRAS, 322, 231Krumholz M. R., Gnedin N. Y., 2011, ApJ, 729, 36Leitherer C., et al., 1999, ApJS, 123, 3Li H.-N., Zhao G., Christlieb N., Wang L., Wang W., Zhang Y., Hou Y.,Yuan H., 2015, ApJ, 798, 110Li H., Tan K., Zhao G., 2018, ApJS, 238, 16Loebman S. R., Roškar R., Debattista V. P., Ivezić Ž., Quinn T. R., WadsleyJ., 2011, ApJ, 737, 8Ma X., Hopkins P. F., Wetzel A. R., Kirby E. N., Anglés-Alcázar D., Faucher-Giguère C.-A., Kereš D., Quataert E., 2017, MNRAS, 467, 2430Majewski S. R., Skrutskie M. F., Weinberg M. D., Ostheimer J. C., 2003,ApJ, 599, 1082Majewski S. R., et al., 2017, AJ, 154, 94Matsuno T., Aoki W., Suda T., 2019, ApJ, 874, L35Mo H. J., Mao S., White S. D. M., 1998, MNRAS, 295, 319Myeong G. C., Vasiliev E., Iorio G., Evans N. W., Belokurov V., 2019,MNRAS, 488, 1235Naidu R. P., Conroy C., Bonaca A., Johnson B. D., Ting Y.-S., Caldwell N.,Zaritsky D., Cargile P. A., 2020, arXiv e-prints, p. arXiv:2006.08625Newberg H. J., et al., 2003, ApJ, 596, L191Nissen P. E., Schuster W. J., 2010, A&A, 511, L10Norris J. E., Ryan S. G., 1989, ApJ, 336, L17Ostriker J. P., Tremaine S. D., 1975, ApJ, 202, L113Peschken N., Athanassoula E., Rodionov S. A., 2017, MNRAS, 468, 994Peschken N., Łokas E. L., Athanassoula E., 2020, MNRAS, 493, 1375Planck Collaboration et al., 2018, arXiv e-prints, p. arXiv:1807.06209Rees M. J., Ostriker J. P., 1977, MNRAS, 179, 541Robertson B., Bullock J. S., Cox T. J., Di Matteo T., Hernquist L., SpringelV., Yoshida N., 2006, ApJ, 645, 986Rodionov S. A., Athanassoula E., Peschken N., 2017, A&A, 600, A25Samuel J., Wetzel A., Chapman S., Tollerud E., Hopkins P. F., Boylan- Kolchin M., Bailin J., Faucher-Giguère C.-A., 2020a, arXiv e-prints, p.arXiv:2010.08571Samuel J., et al., 2020b, MNRAS, 491, 1471Sanderson R. E., et al., 2018, ApJ, 869, 12Sanderson R. E., et al., 2020, ApJS, 246, 6Santistevan I. B., Wetzel A., El-Badry K., Bland-Hawthorn J., Boylan-Kolchin M., Bailin J., Faucher-Giguère C.-A., Benincasa S., 2020, MN-RAS, 497, 747Scannapieco E., Kawata D., Brook C. B., Schneider R., Ferrara A., GibsonB. K., 2006, ApJ, 653, 285Searle L., Zinn R., 1978, ApJ, 225, 357Sestito F., et al., 2019, MNRAS, 484, 2166Sestito F., et al., 2020, MNRAS, 497, L7Sestito F., et al., 2021, MNRAS, 500, 3750Sparre M., Springel V., 2017, MNRAS, 470, 3946Springel V., Hernquist L., 2005, ApJ, 622, L9Starkenburg E., Oman K. A., Navarro J. F., Crain R. A., Fattahi A., FrenkC. S., Sawala T., Schaye J., 2017a, MNRAS, 465, 2212Starkenburg E., et al., 2017b, MNRAS, 471, 2587Steinmetz M., et al., 2006, AJ, 132, 1645Su K.-Y., Hopkins P. F., Hayward C. C., Faucher-Giguère C.-A., Kereš D.,Ma X., Robles V. H., 2017, MNRAS, 471, 144Tapia T., Eliche-Moral M. C., Aceves H., Rodríguez-Pérez C., Borlaff A.,Querejeta M., 2017, A&A, 604, A105Toomre A., Toomre J., 1972, ApJ, 178, 623Venn K. A., et al., 2020, MNRAS, 492, 3241Villalobos Á., Helmi A., 2008, MNRAS, 391, 1806Virtanen P., et al., 2020, Nature Methods, 17, 261Vogelsberger M., et al., 2014, MNRAS, 444, 1518Walker I. R., Mihos J. C., Hernquist L., 1996, ApJ, 460, 121Wetzel A., Garrison-Kimmel S., 2020a, HaloAnalysis: Read and analyzehalo catalogs and merger trees (ascl:2002.014)Wetzel A., Garrison-Kimmel S., 2020b, GizmoAnalysis: Read and analyzeGizmo simulations (ascl:2002.015)Wetzel A. R., Hopkins P. F., Kim J.-h., Faucher-Giguère C.-A., Kereš D.,Quataert E., 2016, ApJ, 827, L23White S. D. M., Frenk C. S., 1991, ApJ, 379, 52White S. D. M., Rees M. J., 1978, MNRAS, 183, 341Zinn R., 1993, in Smith G. H., Brodie J. P., eds, Astronomical Society ofthe Pacific Conference Series Vol. 48, The Globular Cluster-GalaxyConnection. p. 38This paper has been typeset from a TEX/L A TEX file prepared by the author.MNRAS000