Draft version May 12, 2021 Typeset using LATEX twocolumn style in AASTeX63

A Search for Wandering Black Holes with Gaia and DECaLS

1 1 2, 1, 3, 4 5, 6 2, Jenny E. Greene, Lachlan Lancaster, Yuan-Sen Ting, Sergey E. Koposov, Shany Danieli, ∗ 1 7 8, 9 Song Huang, Fangzhou Jiang, Johnny P. Greco, † and Jay Strader

1Department of Astrophysical Sciences, Princeton University, Princeton, NJ 08544, USA 2Institute for Advanced Study, 1 Einstein Drive, Princeton, NJ 08540, USA 3Observatories of the Carnegie Institution of Washington, 813 Santa Barbara Street, Pasadena, CA 91101, USA 4Research School of Astronomy & Astrophysics, Australian National University, Cotter Rd., Weston, ACT 2611, Australia 5Institute for Astronomy, University of Edinburgh, Royal Observatory, Blackford Hill, Edinburgh EH9 3HJ, UK 6Institute of Astronomy, University of Cambridge, Madingley Rd, Cambridge, CB3 0HA, UK 7TAPIR, California Institute of Technology, Pasadena, CA 91125, USA 8Center for Cosmology and AstroParticle Physics (CCAPP), The Ohio State University, Columbus, OH 43210, USA 9Center for Data Intensive and Time Domain Astronomy, Department of Physics and Astronomy, Michigan State University, East Lansing, MI 48824, USA

(Dated: Feb 2021)

ABSTRACT We present a search for “hyper-compact” star clusters in the Milky Way using a combination of Gaia and the Dark Energy Camera Legacy Survey (DECaLS). Such putative clusters, with sizes of 1 pc and containing 500-5000 stars, are expected to remain bound to intermediate-mass black holes ∼ 3 5 (MBH 10 10 M ) that may be accreted into the Milky Way halo within dwarf satellites. Using ≈ − the semi-analytic model SatGen we find an expected 100 wandering intermediate-mass black holes ∼ with if every infalling satellite hosts a black hole. We do not find any such clusters. Our upper limits rule out 100% occupancy, but do not put stringent constraints on the occupation fraction. Of course, we need stronger constraints on the properties of the putative star clusters, including their assumed sizes as well as the fraction of stars that would be compact remnants. 1. WANDERING BLACK HOLES black holes that are found outside of galaxy nuclei may Supermassive black holes are ubiquitous in the centers be particularly sensitive to the seeding mechanism (see of massive galaxies today. They may play an important arguments in Greene et al. 2020). role in regulating star formation in galaxies (e.g., Silk We expect to find black holes wandering in galaxy & Rees 1998; McConnell & Ma 2013; Kormendy & Ho halos regardless of seeding mechanism because by hi- 2013). We do not know when or how supermassive black erarchical merging, galaxies are accreted by the Milky holes are formed. Upcoming gravitational wave experi- Way throughout its history, and some will be totally ments like Laser Interferometer Space Antenna (LISA; stripped apart from a stellar nucleus (e.g., Zinnecker e.g., Amaro-Seoane et al. 2015) will potentially be sensi- et al. 1988). Some of these satellites, at least at the mas- tive to the mergers of the first “seed” black holes. How- sive end, may bring intermediate-mass black holes into ever, with limited knowledge of both the rates and the the galaxy with them (Volonteri & Perna 2005; Bellovary mass functions of these black holes, the LISA observa- et al. 2010). The central black hole can retain a small arXiv:2105.04581v1 [astro-ph.GA] 10 May 2021 tions cannot uniquely determine either. An important cluster of bound stars, which we will refer to as a hyper- complementary clue will be provided by the study of compact star cluster (e.g., Merritt et al. 2009; Lena et al. relic intermediate-mass black holes, with the mass distri- 2020). A related mechanism for depositing off-nuclear bution and environment of these black holes today car- black holes is gravitational recoil of black holes from the rying some memory of when and how they were formed centers of merger remnants at early times (e.g., Volon- (e.g., Volonteri 2010; van Wassenhove et al. 2010; Ri- teri et al. 2003a; O’Leary & Loeb 2009). Finally, if black carte & Natarajan 2018; Bellovary et al. 2019). The holes are made through gravitational runaway at star cluster centers (e.g., Miller & Hamilton 2002; Portegies Zwart & McMillan 2002), then there may be a popula- ∗ NASA Hubble Fellow tion of 1000 M black holes whose surrounding cluster ∼ † NSF Astronomy & Astrophysics Postdoctoral Fellow has mostly dissolved, with only a hyper-compact cluster 2 remaining (e.g., Fragione et al. 2018). A primary goal of rather large on the sky and easily resolvable with exist- this paper is to search for these star clusters that would, ing ground-based imaging surveys. Therefore, in §3 and in turn, point to a population of black hole wanderers. §4, we describe a joint search using the spatial resolu- We do see evidence that star clusters come into our tion of Gaia and the depth of the Dark Energy Camera galaxy with their satellite. More specifically, nuclear Legacy Survey (DECaLS; Dey et al. 2019). In §5, we star clusters are massive stellar clusters that are found at summarize the limits we derive from our non-detections the centers of galaxies. It is thought that some fraction and in §6, we consider the future prospects of this work. of the most massive known star clusters are the remnants of a stripping process that leaves nothing more than the 2. EXPECTED PROPERTIES OF THE nucleus (see, e.g., Pfeffer et al. 2016; Neumayer et al. WANDERERS 2020). The cluster M54 is the nucleus of the disrupt- Our goal is to search for the hyper-compact stellar ing Sagittarius dwarf (Ibata et al. 1995). Another likely clusters that should accompany an intermediate-mass stripped nucleus is ω Cen, which shows evidence for mul- black hole wandering through our galaxy. In this section tiple episodes of star formation (e.g., Norris et al. 1996; we summarize the relevant theoretical understanding of Pfeffer et al. 2016). We have not yet found definitive ev- the size, mass, and stellar content of these hypothetical idence of intermediate-mass black holes in either cluster objects, so that we can hone our search strategy. It is (Ibata et al. 2009; Noyola et al. 2010; L¨utzgendorfet al. worth emphasizing that we tune our search parameters 2012; Baumgardt 2017; Baumgardt et al. 2019), nor for to the specific case of “ex-situ” wanderers that formed in a similar cluster (G1), in Andromeda (Gebhardt et al. an external dwarf galaxy and were subsequently accreted 2005). So far there are no definitive detections of black by the Milky Way. In order for the galaxy to be fully holes in Milky Way globular clusters (see discussion in stripped, we expect such clusters to live relatively close Greene et al. 2020). (. 50 kpc) to the Galactic Center based on modeling There is, however, evidence in galaxies more mas- presented in §2.2. sive than the Milky Way for a population of wandering Numerous successful searches in modern wide-field black holes. There are numerous dynamical detections imaging surveys for globular clusters, dwarf galaxies, of black holes in “ultra-compact” dwarfs (Seth et al. and stellar streams have been carried out in the Milky 2014; Ahn et al. 2018; Voggel et al. 2019). A number Way over the past two decades (e.g., Willman et al. of the most massive ultra-compact dwarfs now have dy- 2005; Belokurov et al. 2006; Koposov et al. 2008; Bech- namically detected black holes at their centers, with the tol et al. 2015; Koposov et al. 2015; Drlica-Wagner et al. black hole accounting for 10% of the mass of the sys- ∼ 2015; Shipp et al. 2018; Torrealba et al. 2019). However, tem in some cases. As outlined above, the most likely no search has been tuned to the compact sizes and low explanation for these objects is that they were formed numbers of stars that we believe the hyper-compact star in a more massive galaxy that was then stripped when clusters may contain. Therefore, we thought it worth- falling into its current host halo (e.g., Pfeffer et al. 2014). while to perform a customized search matched to the Perhaps the most compelling case for a wandering small scales of our target population. black hole is the “hyper-luminous” X-ray source HLX1 (Farrell et al. 2009). HLX1 is an accreting black hole, 2.1. Predicted size and mass 4 with a likely mass of MBH 10 M (e.g., Davis et al. ≈ The most detailed calculations of the dynamical evo- 2011; Webb et al. 2012). It is found in a cluster of stars lution of bound remnant clusters have been made by (Farrell et al. 2014) sitting a few kpc from a more mas- O’Leary & Loeb(2009), followed by confirming simu- sive galaxy ESO 243-49 at z = 0.022, which is very lations in O’Leary & Loeb(2012). These authors fo- likely the remnant of a stripped dwarf galaxy that was cus on wanderers formed through very early mergers, in accreted by the more massive system. which the merged remnant is ejected from the proto- Our goal here is to search for lower-mass stellar clus- Milky Way center via gravitational wave recoil (e.g., ters that would be bound because of the presence of an Peres 1962; Campanelli et al. 2007), a model also ex- intermediate-mass black hole. O’Leary & Loeb(2012) plored by Volonteri & Perna(2005). Their modeling and Lena et al.(2020) investigate a search for hyper- of the subsequent dynamical evolution of the cluster is compact star clusters using stellar colors. In this work, very likely relevant to the final mass and size of the star we instead focus almost exclusively on spatial clustering clusters we consider as well, although the mechanism for information. As described in §2, we expect any clusters forming the wanderers is different. deposited from accreted satellites to be found relatively If we assume that the black hole is able to retain near to the Galactic Center (< 50 kpc) and thus to be roughly its mass in stars within its sphere of influence, 3 then the calculations of O’Leary & Loeb(2012) suggest Second, considerable work on high-resolution hydro- that by the present day, the cluster will lose roughly 60- dynamical simulations have yielded a new generation of 80% of that mass through relaxation (see arguments in semi-analytic models that can simultaneously capture Rashkov & Madau 2014). At the same time, the clusters the evolution of satellites in the tidal field of their host, will grow in size as t1/3, leading to present-day clusters and allow us to examine a large suite of Milky Way- of 0.5-1 pc in size, with mass 20% of the black hole like models in contrast with individual high-resolution ∼ mass. simulations that have been used for this purpose in the We highlight two major potential caveats here. First, past (e.g., Rashkov & Madau 2014). Specifically, we there is the possibility that the compact cluster is dom- use the semi-analytic model SatGen (Jiang et al. 2020). inated not by visible stars, but rather by stellar-mass This model builds upon Monte-Carlo dark matter halo black holes. Such compact remnants preferentially re- merger trees and follows the evolution of satellites using side at the cluster core and they would not be visible tidal-evolution tracks calibrated against high-resolution by electromagnetic means (e.g., Baumgardt et al. 2019; idealized N-body simulations. The model also takes Gieles et al. 2021). O’Leary & Loeb(2012) acknowl- into account the response of dark matter halos to bary- edge this possibility as well, but do not pursue it. We onic feedback, as formulated from zoom-in cosmological will take the same approach, but also note that extreme hydro-simulations, and the gravitational influence of a mass-ratio inspiral events, the detection of the merger galactic disk on satellite evolution. In this way, it is pos- of a stellar-mass black hole with an intermediate-mass sible to generate a large suite of Milky-Way-mass sys- black hole, would be one way to detect such clusters in tems emulating those from high-resolution zoom-in cos- the future with LISA (e.g., Gair et al. 2010). mological simulations regarding satellite statistics, with The second caveat is the possibility that there is no a more complete sampling of the halo-to-halo variance stellar cusp around the black hole. Bahcall & Wolf and at numerical resolutions comparable to or higher (1976) calculate the stellar distribution around a black than that of the simulations. hole that ensues when the black hole is embedded in We use SatGen models with present-day halo masses a stellar cluster with much higher mass than the black distributed uniformly in the range M = 1012 halo − hole. However, low-mass galaxies can have very low stel- 1012.3 M . These models include an evolving disk com- lar densities if they are not nucleated, and while the ponent and are made to emulate simulations of bursty nucleation fraction of 109 M galaxies is near unity, stellar feedback such as the FIRE and NIHAO simula- ∼ that fraction drops substantially at lower galaxy stel- tions. Satellites with halo masses before infall that are 6 lar mass (e.g., S´anchez-Janssen et al. 2019; Neumayer Mh > 1.3 10 M are considered. × et al. 2020). Therefore, it is possible that in practice, From the model, we are particularly interested in the wandering black holes lack even the hyper-compact satellites that dissolve, which is assumed to happen clusters that we consider here. To partially compensate in SatGen when the sub-halo is stripped to a mass of for this possibility, we consider the case that only nucle- 106 M . As shown by Jiang et al., a large fraction of ated galaxies host observable wandering black holes in the most massive satellites, particularly those accreted §2.2. early, do pass close enough to the host halo center to be dissolved. Of course, it would be very interesting to follow these systems dynamically, including the impact 2.2. Predicted number density and radial distribution of a black hole and possible stellar cluster on the sub- sequent stellar content and dynamical friction, but that There have been a number of estimates for the number is beyond the scope of the present work. For now, we of wandering black holes that we might expect to find simply count the number, infall mass, and radial distri- in a Milky Way-like galaxy (e.g., Volonteri et al. 2003a). bution of these disrupted satellites (Figure1 & Figure We update these predictions in two ways. First, we uti- 2). lize updated scaling relations that account for both the Dissolved halos are still tracked as point masses by available data and upper limits (Greene et al. 2020). SatGen. We exclude halos that are within 10 pc of the There is considerable scatter in these relations, which galaxy center, assuming these effectively have merged are of course extrapolations to the mass regime of in- with the central black hole. In practice such occur- terest here. Nevertheless, new dynamical measurements rences are very rare, although the satellite orbits are and constraining upper limits from the centers of nearby drawn from a cosmological distribution neglecting mass, low-mass galaxies (Neumayer & Walcher 2012; Nguyen so there is some chance that SatGen underestimates can- et al. 2018, 2019a) motivate us to recalculate estimates of wanderer number density. 4

102 3 MBH σ MBH σ , Nucleated 10 − ∗ − ∗ MBH M MBH M , Nucleated − ∗ − ∗ B10 RM14 VP05 ) 2 10 F18 BH OL09 101 > M ( 1 N 10

0 100 10 103 104 105 106 103 104 105 106 MBH (M ) MBH (M )

Figure 1. Left: The cumulative number of black holes predicted assuming that the only channel to make wanderers is through accreted satellites. We estimate the central MBH of each satellite based either on the stellar mass (teal) or velocity dispersion (blue) at the time of accretion; the scatter includes both the range of accreted satellite masses in the SatGen models and scatter in the scaling relations. We show the 1σ range. As these are both extrapolations, they predict very different mass functions for the same population of disrupted satellites. We assume 100% occupation fraction in this panel, which is thus an upper limit on the number of wanderers formed via accretion of satellites. Roughly speaking, the clusters should contain ∼ 20% of the mass of the black hole (O’Leary & Loeb 2012). For comparison, we plot predictions from prior work considering satellites (red symbols; VP05, B10, RM14 Volonteri & Perna 2005; Bellovary et al. 2010; Rashkov & Madau 2014) as well as predictions from formation in globular clusters that are then disrupted (grey star; Fragione et al. 2018) or through recoil (grey plus O’Leary & Loeb 2009). These are cumulative, so we plot them at the lowest black hole mass considered. Right: Same calculation as at left, but in this case we take the nucleation fractions from S´anchez-Janssen et al.(2019) and assume that only nucleated galaxies contain black holes. nibalism if more massive satellites are preferentially on each halo a stellar mass (in this case the relation from more radial orbits. Rodr´ıguez-Puebla et al. 2017). The relationship be- To convert the observed halo and stellar properties at tween stellar and halo mass is notoriously unconstrained infall into estimated black hole masses, we must decide at these dwarf masses, with large degeneracy between what fraction of satellite halos will host black holes. The the scatter and slope of the relation (e.g., Munshi et al. so-called “occupation fraction” of black holes in dwarf 2021), leading to large uncertainty in the stellar masses galaxies is not yet well measured. However, given the within the SatGen model. Second, the relationship be- upper limit on a black hole in the nearby dwarf galaxy tween stellar mass and black hole mass has considerable M33 (Gebhardt et al. 2001), it is likely that not every intrinsic scatter and is dependent on galaxy morphology dwarf galaxy hosts a massive central black hole. We first at higher mass (e.g., Reines & Volonteri 2015). We are calculate the predicted numbers of black holes under an extrapolating a high-scatter relation into an unknown assumption of 100% occupation, but additionally calcu- regime, which adds considerable systematic uncertainty late the expected number of wanderers if we were to as- to these estimates. sume that only nucleated galaxies (galaxies with nuclear As a second MBH estimate, we take the circular ve- star clusters) host black holes. We adopt the nucleation locity at infall and, following Rashkov & Madau(2014), fraction from S´anchez-Janssen et al.(2019), specifically we calculate σ = vcirc/2.2 and then use the MBH-σ ∗ ∗ their measurements in the Virgo Cluster. Virtually all relation (Greene et al. 2020). In this case, the circular galaxies with M 109 M host nuclei, falling with velocity is securely predicted from the model, but the ∗ ≈ mass such that galaxies with M < 106 M will host conversion between circular velocity and stellar veloc- ∗ black holes < 10% of the time. ity dispersion is not well known in the dwarf regime. We compute two possible black hole masses. First, As far as the scaling relations, the MBH σ relation − ∗ we use the stellar mass at infall from the model com- is also somewhat morphology dependent (e.g., Greene bined with an MBH-M relation from Greene et al. et al. 2016), but the scatter is lower than the conversion ∗ (2020). There are two major uncertainties here. First, based on total stellar mass. We will take the vmax-based we have assumed a stellar-to-halo mass relation to assign predictions presented below as the primary predictions 5

cupation fraction of black holes in the infalling satellite

4 population. σ ,MBH > 10 ∗ 3 σ ,MBH > 10 We have an estimate of the satellite stellar masses ∗ 3 from a stellar-to-halo mass relation (Rodr´ıguez-Puebla 2 M ,MBH > 10 10 ∗ 3 F18,GC et al. 2017). Assuming black holes with MBH> 10 M , VP05,satellite in the case of the MBH-M relation, nearly all hosts will VP05,slingshot ∗ have M > 107 M . In contrast, in the σ -based scal- ∗ ∗ N ing, much lower-mass satellites host black holes. Specif- 101 ically, we find that galaxies with stellar mass > 105 M

are predicted to host black holes when scaling with σ . ∗ Thus, we find many more possible black holes in the σ -based scaling. Also, we note that even if we were ∗ 100 to extend the lower-limit on halo masses considered by the model, we would not find any more black holes with 1 2 3 10 10 MBH> 10 M .

R (kpc) Depending on the scaling relation, we not only expect different stellar masses, but also a different radial dis- Figure 2. Radial distributions of clusters that host black tribution, of satellites hosting black holes, as shown in 3 4 holes with MBH> 10 , 10 M where black hole mass is es- Fig.2. The most massive satellites, those that prefer- timated following the MBH − σ relation (blue-dotted and entially host black holes in the MBH-M case, must be ∗ ∗ purple-solid respectively) or the MBH − M relation (teal accreted early and travel close to the galaxy center to be 3 ∗ long-dash, all black holes with MBH> 10 M ). We also stripped (Jiang et al. 2019). In contrast, the much wider schematically indicate the typical radius of black holes made range of halo and stellar mass that can host black holes in recoil events from early mergers or from accreted satellites when scaling with σ also translates to a wider radial from Volonteri & Perna(2005). We highlight that wanderers ∗ deposited by disrupted globular clusters would be very close range, R . 50 kpc. We will use the larger distance limit to the Galactic Center (Fragione et al. 2018), and we note in §5 when we calculate upper limits on the number of that O’Leary & Loeb(2012) do not present a radial distri- wanderers in the Milky Way. bution, but do predict that black holes will extend to 100 In Figure1, we compare with existing similar predic- kpc. tions for the number of wandering black holes in a Milky Way-mass halo. Each paper makes slightly different as- throughout the paper, because at least we will not be sumptions about seeding mechanisms, and thus predict directly dependent on an assumed stellar-to-halo mass different mass spectra for the resulting wanderer pop- relation. At the end of this section, we will briefly dis- ulation. Volonteri & Perna(2005) use a semi-analytic cuss a third model for black hole mass in which there is model as well, and directly consider multiple seeding no scaling with galaxy properties and all black holes are mechanisms, to predict between one and ten black holes 3 relatively massive at birth. with MBH> 10 M . Bellovary et al.(2010) predict

In calculating the expected number of wandering black slightly higher numbers (5-15), in this case with a heavy holes, we include the halo-to-halo scatter by using the seeding model. Rashkov & Madau(2014), like us, have 85 Jiang et al. Milky Way-like models. We also vary the no seeding model, but instead assign MBH based on mapping between halo and black hole mass, by drawing properties of the halos, and predict comparable cumula- 100 black holes per satellite from within the published tive numbers as we do, using a similar MBH-σ relation ∗ scatter in the scaling relations. The resulting cumulative based on peak maximum halo velocity. Both Tremmel numbers of predicted wanderers are shown in Fig.1 for et al.(2018) and Ricarte et al.(2021) find 5 10 wan- ∼ − the σ and M scalings respectively. dering black holes in Milky Way-mass halos using the ∗ ∗ Roughly speaking, the stellar mass in the clusters will Romulus simulation. Our predictions, and those from be 20% of M . Thus, we limit our attention to the literature, span two orders of magnitude in number ∼ BH black holes > 103 M , where we might hope to detect because we are extrapolating assumptions about black the bound stellar cluster. Under an assumption of full hole scaling relations and occupation fractions into an occupation, we expect to find a few to 100 clusters with unknown regime. These are the primary systematic un- a hundreds to a few thousand stars, depending on both certainties in our predictions, rather than detailed as- the black hole scaling relation adopted and on the oc- sumptions in the cosmological models that we use. 6

The predicted radial distributions in prior work are time (O’Leary & Loeb 2012). We assume a fixed size also similar to ours (such as Bellovary et al. 2010; for our mock clusters. If in reality they are considerably Rashkov & Madau 2014), particularly those that model larger (or smaller) by more than a factor of two, then wandering black holes from disrupted satellites alone. our search is unlikely to find them. The distribution is more extended than the wandering We assume the clusters are 10 Gyr old, similar to their black holes expected from the dissolution of globular likely accretion time. To span the possible metallic- clusters should black holes form efficiently in their cen- ity range of such clusters, we take [Fe/H]= 1.5, 0.5, − − ters (e.g., Fragione et al. 2018), but is more centrally which is measured for nuclear star clusters in host galax- concentrated than models like Volonteri & Perna(2005) ies with M = 108 109 M (Neumayer et al. 2020). In ∗ − or O’Leary & Loeb(2012) that also include a component practice, the nuclei of such low-mass galaxies accreted from early recoil events. In this work we do not attempt at z > 1 are likely to have even lower metallicities, but to search for more distant (spatially unresolved) clus- we conservatively adopt these limits since the hotter and ters, which would require a different approach. brighter stars in clusters at lower metallicity are easier Finally, it is interesting to consider how our predic- to find at fixed stellar mass. tions would change if we adopted a heavy seeding pre- In Figure3, we show the color-magnitude diagrams scription with a lower-mass limit of 104 105 M , as (bottom panel) and spatial distributions (middle panel) − might be expected in some heavy seeding models (e.g., of example model clusters at the limits of our detection Bellovary et al. 2010; Inayoshi et al. 2020). If we as- threshold (§4.3), with stars detectable in Gaia and DE- sume that every satellite, regardless of mass, is seeded CaLS in red and grey, respectively. The upper panel with a heavy seed, then the number of expected black shows the artificial clusters injected into DECaLS im- 3 holes with MBH> 10 M would be very high ( 1000) ages. It is worth noting that there is a large amount of ∼ and would become very sensitive to our halo mass limit. stochasticity in the number of giant stars for a N = 1000 Such a model is very easily ruled out, as we will show in star cluster. While the median number of expected stars the following sections. If, on the other hand, we assume is five, it can range from zero to 15. We are not complete a reasonable drop in occupation fraction with mass, then beyond the distance where we lose main-sequence stars the number and mass distribution of black holes under (§4.3) for these low-N models, and so we are not very a heavy seeding model would be very similar to Figure1 sensitive to this stochasticity here. But, future searches (right), since the occupation fraction becomes a strong at larger distances would be unless they are deep enough function of stellar mass, and this factor matters more to reach below the main-sequence turn-off. than exactly how we assign black hole masses. 3. DATA 2.3. Predicted stellar content Having defined the parameters of the expected hyper- We use the ArtPop software package (Danieli et al. compact clusters surrounding the putative intermediate- 2018; Greco et al. 2021; J. Greco & S. Danieli, in prepa- mass black holes, we now turn to search for them. We ration) to generate model clusters for calibration and first describe the data sets, and then in §4 we describe completeness analyses. Using the MIST isochrones, we our search. synthesize stellar populations of a given age and metal- 3.1. The Gaia catalog licity, and simulate realistic images of stellar systems based on their physical and structural parameters. In Gaia is a European Space Agency astrometry mission. this work, we use ArtPop models to generate color- Here we use the Early Data Release Three of the Gaia magnitude-position predictions in Gaia and DECaLS for mission (Gaia EDR3; Gaia Collaboration et al. 2016, expected clusters. 2018, 2020) to make an initial search for stellar overden- To build our model clusters, we assume a standard sities on 15 00 scales by looking for stars with anoma- ∼ Kroupa(2001) initial mass function, with slope ψ lously large numbers of neighbors on this scale. To do α ∝ 1 m− with α = 1.3 for stars below 0.5 M , and α = 2.3 this, we utilize a custom-built catalog from S. Koposov at higher stellar mass. We set the lower and upper based on the EDR3 that records the number of neigh- mass limits based on the MIST isochrones of a given bors that each source has with angular separation less age and metallicity. As described in §2.1, we do not ex- plore the possibility that the present-day mass function 1 The catalog has been built based on the vanilla EDR3 Gaia source is dominated by stellar-mass black holes. We assume catalog using the Whole Sky SQL Database maintained in Cam- a Plummer(1911) profile with a scale radius of 1 pc; bridge using the Q3C spatial query software Koposov & Bartunov the clusters are expected to grow to this large size with (2006). 7

Model clusters in DECaLS images

Nstars=500, D=25 kpc Nstars=1000, D=40 kpc

30

20

10

(arcsec) 0 . Dec

δ 10 − 20 − 30 − 20 0 20 20 0 20 − δR.A.(arcsec) − δR.A.(arcsec) 17 DECaLS 18 Gaia

19

20 g

21

22

23 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 g r g r − − Figure 3. Top: Two model hyper-compact clusters injected into DECaLS images with N = 500, D = 25 kpc (left) and N = 1000, D = 40 kpc stars (right). These clusters should be comfortably detected by our Gaia search (see §3). Middle: The spatial distribution of the stars in these clusters as seen by Gaia (red) and DECaLS (grey). Bottom: Noiseless color-magnitude diagram for the clusters, colors as above. We should note that the number of giant stars for these low-mass clusters will vary quite significantly, from near zero up to fifteen for the N = 1000 star cluster, with a median of five. than x arcsec for 10 aperture choices of x. The aper- tures range from 0.25 to 128 arcsec, with each aperture 8 increasing by a factor of two. It is convenient to work we present here, we have decided to combine the spa- with the counts within “annuli” (i.e. number of neigh- tial resolution of Gaia with the depth of DECaLS to bors between y and x arcsec), since each annulus can be search efficiently for possible clusters. First, we use the modeled independently. We will refer to the number of Gaia EDR3 catalog to search for clusters of N > 4 stars neighbors between x/2 and x arcec as annuli counts Ax. within . 1500, as expected for these hyper-compact star Our fundamental data set is therefore a vector of counts clusters. The benefits of Gaia include the high spatial resolution and the potential to filter on proper motions. A (A ,A , ...A ,A ) (1) However, with a depth of G 21.5 mag (Gaia Collabo- ≡ 0.25 0.5 64 128 ∼ ration et al. 2020), we will detect only a few of the stars within these apertures for each EDR3 source. with Gaia (as we see in Fig.3). 3.1.1. Masking of known galaxies and star clusters To illustrate this point, in Figure4, we show the ex- pected number of stars that can be observed with Gaia There are two main types of contaminant that we and DECaLS in a cluster of 500-5000 stars as a func- mask before performing our cluster search. First, we tion of distance. We use 100 different realizations of the mask all known groupings of stars in the Milky Way, cluster to include stochasticity in how many stars pop- including open clusters, globular clusters, and known ulate the red giant branch for the lower-mass clusters. satellites compiled by Torrealba et al.(2018) from sev- For N = 1000, we expect a large range in the number eral different sources (McConnachie 2012; Harris 2010). of giant stars, between zero and twelve, with a median In general we mask a two degree radius around each of five. However, over the distance range where we are satellite, except in the cases of the Large and Small Mag- sensitive with this Gaia search (§4.3), we do reach be- ellanic Clouds, along with Fornax, which we mask with low the main-sequence turn-off, meaning that our overall a 10 degree radius (while the Large and Small Magel- N is dominated by main-sequence stars. lanic Clouds themselves do not fall within the DECaLS star,Gaia The grey regions show the numbers for the higher- footprint, their outskirts turn out to be a major con- metallicity clusters, showing that the range in metal- taminant). licity leads to only a small difference in the number We also find that the cores of nearby galaxies can ap- of stars, so throughout we will focus on models with pear in Gaia as a set of point sources, and so we also [Fe/H]= 1.5, which seems more likely. We see that mask the New General Catalog (NGC) galaxies2. The − overdensities of stars identified with Gaia should be ac- majority of these galaxies are given a 1-2 radius mask, 0 companied by an increase of three to five times in the while a small subset of galaxies with larger tabulated number of stars at detected at DECaLS depth, if we radii are masked over a full degree radius. In all, we are identifying star clusters. This will be true even with mask roughly 1000 deg2 of the 9000 deg2 area covered some mild crowding, as discussed in §4.2.1. by DECaLS. In this section, we first present the model that we use 3.2. DECaLS to identify possible clusters as outliers in spatial clus- tering (§4.1). We then calibrate the methods using arti- The DECam Legacy Survey (DECaLS) has been ficial clusters (§4.2), argue that crowding is unlikely to carried out with the Dark Energy Camera (DECam; greatly impact our search (§4.2.1), and determine our Flaugher et al. 2015; Dey et al. 2019) at the Mayall completeness (§4.3). Finally, with a list of cluster can- 4m telescope at the Cerro Tololo Inter-American Ob- didates in hand from Gaia, we use DECaLS to invali- servatory. DECaLS reaches 5σ point-source depths of date most candidates based their total number of stars grz = 23.95, 23.54, 22.50 AB mag over 9000 deg2. We at DECaLS depths (§4.4) along with their colors (§4.5). utilize gr photometry from Data Release 9 of DECaLS3 in our search. 4.1. Negative Binomial Model 4. SEARCH WITH GAIA+DECALS Our method relies on modeling the distribution of As described in §2, we seek hyper-compact star clus- neighbor counts around each star in annuli (as described ters comprising 500-5000 stars within 1 pc. These in 3.1) and looking for outliers, or low probability points, ∼ clusters likely fall within 50 kpc of the Galactic Cen- in this distribution. ter (see Fig.2). In the proof of concept search that For each star, we model the vector A, which is the neighbor counts in angular annuli on the sky over all scales up to 12800. We model each annulus Ax in a small 2 https://github.com/mattiaverga/OpenNGC/blob/master/NGC.csv region of sky as an independent probability distribution, 3 https://www.legacysurvey.org/dr9 with no correlations between annuli. Then we have that 9

500 500 1000 1000 5000 5000 102 102 Gaia , DECaLS ,

star 1 star N 101 10 N

100 100 101 102 101 102 D(kpc) D(kpc)

Figure 4. Left: The number of detectable stars in clusters with N = 500, 1000, 5000 total stars (in blue, green, red respectively) as a function of distance from the Sun. The grey shaded regions are the [Fe/H]= −0.5 models, showing the small difference introduced over this modest metallicity range. We will show (§4.3) that when there are more than ∼ 4 stars, we can reliably identify the cluster in the Gaia data. Right: Same as left, but for DECaLS imaging with a conservative magnitude limit of g = 23.5 mag and only considering stars within 1500 of the cluster center.

In order to account for this possibility, we specify P (Ax)

Starfield with the Negative Binomial distribution, which is a gen- 4 500, 30kpc eralization of a Poisson distribution, and can be thought 10 1000, 50kpc 5000, 60kpc as a Poisson distribution whose mean is sampled from a gamma distribution. 3 10 The Negative Binomial is a discrete probability distri- bution describing the number of ‘unsuccessful’ trials k 102 that occur in some series of repeated, independent tri- Number als, each with success probability p, before r successful 101 trials occur. This distribution can be written as

k + r 1 100 p(k r, p) = − pr(1 p)k . (3) | r 1 − 60 40 20 0 − − − − Log(Probability) This distribution approaches the Poisson distribution if one sends p 1 and r while keeping the mean → → ∞ of the distribution [λ = (1 p)r/p] constant. Unlike in Figure 5. Log of the probability that the model stars (col- − the Poisson distribution, the variance [σ2 = (1 p)r/p2] ored lines) are drawn from the parent distribution that de- − and mean are two independent quantities. scribes the probability of the annuli Ax for a random field (black). The probabilities are calculated assuming that the We determine the parameters of the Negative Bino- counts within annuli are randomly distributed according to mial φ = rx, px by modeling the Ax of stars in neighbor- a Negative Binomial distribution. ing regions set to be a few times the expected scale of the hyper-compact clusters themselves. We divide the sky into healpix pixels with NSIDE= 256 ( 2 0) pixels, ∼ and fit the Negative Binomial coefficients to stars in the P (A φ) = ΠxP (Ax φx), (2) | | surrounding nearest eight healpix pixels. We then use where φx are the model parameters. If all of the counts that parameter vector φˆ, comprising each fitted rx, px were random, we could model each probability distribu- for each annulus, to estimate the likelihood of each star tion as a Poisson distribution. In practice, however, the P (A φˆ). We find p to range from 0.7-0.9 at large radii, | dispersion may well be larger than the mean due to den- with significant improvements to χ2 over a Poisson fit. sity variations over the fields of view that we consider. In the smallest annuli < 8 00, p approaches 1, and we 10 therefore adopt the Poisson distribution fits at these nal motions of the cluster stars should be smaller than smallest annuli. their bulk motions. However, in practice we are finding The final probability is calculated as the product of typically only 3-4 stars that may be associated with each the probabilities from the fit to each annulus (Equation other (§4.5), and some of these are too faint for reliable 2 above). Stars that live in anomalously dense regions of proper motions. In practice we therefore do not use the sky on 15 arcsec scales will have very low probability proper motion measurements in our search. ∼ in this model, which assumes that stars are distributed randomly. In the next section we present mock tests 4.2.1. Possible impact of crowding that we use to pick an outlier threshold. Here we investigate what fraction of stars in the core of our clusters may be lost in the DECaLS catalog due 4.2. Selecting outliers to crowding (Fig.3). While the clusters we are seeking will be a few arcsec across on the sky, as Figure3 shows, To hone our outlier selection and test our complete- we could expect some blending to affect the number of ness, we use the ArtPop models described in §2.3 to in- stars in the DECaLS catalogs. While this blending will ject artificial clusters into our data, and calculate their not impact our completeness corrections, that are based Negative Binomial probabilities, as a function of num- entirely on Gaia, they could impact our ability to detect ber of stars in the cluster, distance to the cluster, and the larger population of cluster stars that we expect at the background stellar density in each field. DECaLS depths. We have generated 100 realizations for each of three 2 We expect there to be 0.07 stars arcsec− within the cluster masses (Nstar = 500, 1000, 5000 M ) and two half-light radius of a 500-star cluster at our detection metallicities ([Fe/H]= 1.5, 0.5). These 100 models 2 − − limit of 30 kpc, 0.17 stars arcsec− in a 1000-star are each inserted into 100 random locations that span ∼ 2 cluster at our limit of 40 kpc, and 0.9 stars arcsec− 4 ∼ a range of background stellar density, for a total of 10 in a 5000-star cluster at 50 kpc, beyond which we do models for each Nstar. For book-keeping purposes we not expect many clusters. We investigate how the DE- track only the star closest to the center, creating annu- CaLS catalog behaves using the star clusters Reticulum lar counts that include both artificial stars in the cluster and Whiting 1. Denser and more luminous clusters are and real stars in the random location we have chosen. modeled differently within the DECaLS catalog, which We then calculate the probability of each star in the uses Gaia data to identify stars and does photometry Negative Binomial model. While these 100 models cap- only on these stars4. ture the variation in stars on the main sequence well, To get a handle on crowding using these clusters, we they will not capture the full stochasticity on the giant make two assumptions. First, we assume that all the branch. Therefore we ran a second set of 1000 real- objects at the centers of these two clusters are actually izations for the N = 500, 1000 low-metallicity models stars (probably nearly true). Second, we consider each to verify that our completeness and thresholds do not object of any type to represent a single star. This is not change. always true. For instance, a single exponential object The distributions of probabilities from the Negative can be a blend of two or three stars. However, it will Binomial fits are shown in Figure5 for the lower- at least give us an idea of the magnitude of crowding. metallicity [Fe/H]= 1.5 model. For display purposes, 2 − Whiting 1 has 0.09 stars arcsec− , with 70% of those we select models at a distance and number of stars that identified as point sources. Reticulum is denser, with is roughly at our 80% completeness limit, as shown in 2 1.7 stars arcsec− and 50% of those identified as stars. Figure6. We then select a probability threshold for can- Thus, at the crowding levels that we expect for the ma- didate clusters that retains a high fraction of the model jority of our cases, it is safe to assume that 30% of ∼ cluster stars without swamping us with “normal” back- stars may be lost to crowding. Note that things will ground stars. There are very few stars with probabilities be worse in the center of a 5000-star cluster, but there log P < 10, but we can still maintain high complete- − would be many more stars outside the core that we ness in the mock clusters (§4.3). Generally, with this would find. This is of course approximate, but it demon- cut we select a fraction . 0.01% of the stars, for a total strates that at the level of crowding that we expect, it is of 22,200 candidates. still safe to use the DECaLS photometry. We also note In principle, we also have access to proper motion data that we do not expect any saturated stars. for the Gaia stars. We would expect the hyper-compact clusters to have velocities and positions consistent with halo objects, and we would expect the stars to have 4 https://www.legacysurvey.org/dr9/external/#globular-clusters- proper motions consistent with each other, as the inter- planetary-nebulae 11

be used to determine whether or not we have found hints 1.0 of a real cluster. By cross-matching with DECaLS, we are able to ap- ply the following additional cuts. We only consider can- didates that have at least Nstar,DECaLS = 9 stars in the DECaLS catalog within 15 arcsec of the star in question. 0.5 The aperture is chosen to be roughly twice the effective radius of a cluster around our completeness limit. We Completeness 500 also apply a color criterion that the stars must fall in the 1000 range 0.2 < g r < 1.5, inspired by the color range of 5000 − 0.0 ClusterBoundary the color-magnitude diagram of an old and metal-poor stellar population (see Figure3). Selecting nine stars as 10 20 30 40 50 a limit is conservatively lower than the number we ex- D(kpc) pect within < 15 arcsec (25 5 at our detection limit for ± Figure 6. The fraction of clusters that we recover as outliers a 500-star cluster, Fig.4). Thus we allow for some loss from our Negative Binomial fit for three cluster masses (500 of stars from confusion. However, with nine stars we still stars in blue, 1000 stars green, 5000 stars red) as a function have sufficient numbers to crudely fit a color-magnitude of distance. We define our survey volume by the distance relation (§4.5). where we reach 80% completeness (dashed vertical blue for As an additional sanity check on the impact of crowd- 500 stars, dashed vertical green for 1000 stars). Based on ing, we also include in our “star” count objects that the SatGen models, we expect most hyper-compact clusters DECaLS has classified as “compact” exponential sources would lie within 50 kpc of the Galactic Center, excluding the (REX). We only consider REX sources with sizes smaller grey shaded region. than the PSF, but in this way we can crudely evaluate whether we are missing stars that have been classified 4.3. Completeness as extended due to blending. In practice, we find that We use the ArtPop simulations to quantify our com- adding these makes little difference to our final list of pleteness. For each cluster mass, and each distance, targets, since most REX-dominated candidates tend to we first verify that there is no difference in complete- be distant galaxy clusters that are eliminated by our ness as a function of background stellar density in each color cut. field. This independence is not surprising, given that We find 176 stars from 86 candidate clusters with at the search for very compact objects is limited by Pois- least nine DECaLS stars falling in a reasonable color son noise in the number of stars at the cluster scale range, once we remove some contamination from inade- rather than the Poisson noise in the background. We quately masked nearby galaxies. then measure the fraction of clusters that are identified We expect 25 5 stars in a 1500 aperture for the ∼ ± by our outlier threshold based on the Negative Binomial 500-star clusters at our detection limit (Fig.4). There fits, as a function of distance in Figure6. Our probed is only one candidate cluster in our outlier sample with volume is defined within the distance where we fall to Nstar,DECaLS > 12, and 6 with Nstar,DECaLS > 11. This 80% completeness (30, 40 kpc for 500, 1000-star clus- number is already 2σ lower than the low end of the ∼ ters, and beyond 50 kpc for 5000-star clusters). We will expected range at our distance detection limit, assuming use these numbers to quantify the limits on the number Poisson errors. However, in order to determine whether of such clusters that may still be lurking in the Milky we have identified any candidates, it is useful to also Way in §5. examine the colors and magnitudes of the candidates. With color and magnitude we can isolate associated 4.4. Eliminating candidates with DECaLS matching stars from foreground/background objects. As shown in Figure4, we expect to detect many tens of stars from a cluster at our detection limit in DECaLS, 4.5. Eliminating candidates with a color-magnitude fit where we might only expect four to six stars in Gaia. For all cluster candidates, we next determine whether Therefore, a straightforward way to winnow down can- the DECaLS gr color-magnitude distribution is consis- didates identified with Gaia is simply to ask how many tent with an old coeval stellar population. We limit stars are detected at that position in DECaLS. From the our attention to those candidates that have at least DECaLS photometry, we count the number of stars and N = 9 stars with 0.2 < g r < 1.5 within star,DECaLS − extract color-magnitude distributions, both of which can 1500. 12

19

20

g 21

22

23

0.0 0.5 1.0 1.5 g r −

Figure 7. Left: A cutout from DECaLS around the position of the cluster candidate corresponding to the CMD at right. Right: An example of the isochrone fit to the candidate cluster shown at left. The red points are drawn from a 3000 radius around the candidate, while the grey-scale represents our best-fit isochrone. Even with this wider aperture, we still only find a maximum of six stars that may be associated with the same isochrone.

To fit the color-magnitude diagrams (CMDs), we in- reliable distance (roughly > 40 kpc), but we can achieve clude two components, a background model built from reasonable fits when the number of stars in the cluster the data, and a simple stellar population model at a is Nstar,DECaLS > 9. range of distances. The model of the background is con- Our most likely CMD fits yield a refined Nstar,DECaLS structed from all of the stars in 10 degree patches. The for each candidate cluster. Although we have iden- possible additional cluster component is made from the tified overdensities in space, the colors of the stars MESA Isochrones and Stellar Tracks isochrones (MIST; suggests that the majority are foreground/background Dotter 2016; Choi et al. 2016) built from the Mod- stars. According to our fits, the largest number of ules for Experiments in Stellar Astrophysics (MESA stars that may be associated with a common isochrone Paxton et al. 2011, 2013, 2015, 2018) models sampled is N = 4 5 stars (two candidates), or star,DECaLS − with a Kroupa IMF. For simplicity, and because we do Nstar,DECaLS = 3 stars, (14), while all the other poten- not expect young stars in our hyper-compact star clus- tial cluster stars are drawn from the background. Since ters, we fix the age to 10 Gyr and the metallicity to the low end of our predicted number of stars is 25 5 at ± [Fe/H]= 1.5. There are two free parameters in our our detection limit, we conclude that with high signifi- − fit, the relative fraction of background and cluster stars, cance, we have not detected a candidate hyper-compact and the distance for the cluster component, allowed to star cluster as defined in §2. range from 10 to 90 kpc. As an example of an apparent overdensity, we high- We build a two-dimensional histogram of the theoret- light one candidate cluster from the list of 86 (Figure ical color-magnitude diagram thus constructed, includ- 7). Within 1500, this candidate has three stars associ- ing a scatter in the magnitude and color derived from ated with the same isochrone, which grows to 5-6 when the DECaLS data. We then select the distance modulus we open the aperture to 3000. This candidate comprises and background weight that maximizes the likelihood a clear overdensity spatially. As a sanity check, we fit (log = Σ logCMD ), where CMD is the normalized Poisson distributions to randomly selected 1500 apertures L i i probability density at the position of each of i stars in within 30 of the candidate, and find that it is a 3 σ out- our candidate cluster. An example fit is shown in Figure lier in density. However, as judged from the isochrone 7. fit, only a few of those stars may be associated with each We test this method using our mock clusters generated other. They do have consistent proper motions, but even by ArtPop, described above in §2.3. Specifically, we if they are physically associated, there are not enough of embed the model clusters in random DECaLS fields, add them to represent the clusters that we are looking for. photometric noise, and then run our fit on this collection of stars. Once main sequence turnoff stars are too faint 5. LIMITS ON WANDERING BLACK HOLES IN to be detected by DECaLS, we can no longer estimate a THE MILKY WAY 13

We searched 8000 deg2 of sky for hyper-compact where the halo velocity function is given by N (v ) ∼ h max star clusters of 500-5000 stars within 50 kpc of the and the occupation fraction is λocc (vmax). Galactic Center, that would be the signature of wan- The halo mass function in our predictions is derived dering black holes accreted with infalling satellites. We from the SatGen simulation. The MBH scaling relation did not find any plausible candidates, so in this section is extrapolated from the observed MBH σ relation, − ∗ we translate our non-detection into upper limits on this assuming that the maximum halo circular velocity can population. be converted directly to a stellar velocity dispersion with To calculate the upper limits, we assume that we are a constant scale factor. Both the extrapolation and the sensitive within the volume defined by our 80% com- conversion are uncertain. As described in §2.2, we do not pleteness, as indicated in Figure6. We further assume consider the M MBH scaling relation here, given the ∗− that the black holes lie within 50 kpc, as calculated us- additional uncertainty added by the unknown stellar-to- ing the MBH σ relation to populate halos with black halo mass relation. − ∗ holes (see Figure2). Currently, the occupation fraction is not constrained The resulting limits are shown in Figure8 and Table1. in the regime of interest to us, below galaxy stellar 9 1 The predicted number of black holes (or corresponding masses of M 10 M , or σ 20 km s− . There ∗ ≈ ∗≈ star clusters on the top axis) under the assumption of are observational constraints on occupation fraction for 100% occupation fraction are shown by the dotted line, more massive dwarf galaxies with stellar masses of M & ∗ while taking the nuclear cluster occupation fraction as 109 M (see summary in Greene et al. 2020), where the the black hole occupation fraction is shown in solid. In occupation fraction is consistent with at least 50% of what follows we discuss what we can and cannot con- galaxies hosting a central black hole, based both on stel- clude from these limits. lar dynamical results (e.g., Nguyen et al. 2018, 2019b) and X-ray studies (e.g., Miller et al. 2015; She et al. 5.1. Limits on the occupation fraction 2017). At stellar masses M < 109 M , there are ∗ In this paper we have searched for and place limits on precious few constraints on the occupation fraction, al- the number density of star clusters with < 5000 stars. though some AGN have been found in galaxies of this We are interested in the corresponding number of black mass (e.g., Baldassare et al. 2019; Reines et al. 2019). holes. We translate between the cluster mass limit and Our measured limits are inconsistent with 100% oc- cupation for sub-halos of the Milky Way. They are well MBH by simply assuming that the cluster mass is 20% below predictions for the number of black holes we would of MBH. The number of black holes that we expect is basically expect if every satellite contained a black hole with mass the product of the halo mass function, the occupation as predicted by the MBH σ scaling. As mentioned in − ∗ fraction, and a convolution with the halo to black hole §2.2, a heavy-seeding model in which all halos down to 6 scaling relation. If we assume that σ can be related 10 M are seeded is ruled out even more conclusively. ∼ ∗ To take a concrete example of a non-unity occupation to the maximum halo circular velocity as σ = vmax/2.2, ∗ then we can take the black hole scaling relation with fraction, we adopt the nucleation fraction (the fraction of galaxies containing nuclear star clusters) as a proxy vmax to have the form: log MBH= α + βlog vmax. For for the black hole occupation fraction, shown as a solid intrinsic scatter in this relation of σint, we then have (see, e.g., Marconi et al. 2004; Gallo & Sesana 2019): line in Fig.8). All black holes discovered dynamically in galaxies with M 109 1010 M are found in nu- ∗ ≈ − clear star clusters (e.g., Seth et al. 2008; Nguyen et al. P (MBH vmax) = 2018, 2019b), potentially suggesting a relationship be- |  2 tween the two (see details in Neumayer et al. 2020). 1 1 log MBH α β log vmax exp − − Nucleation fractions are near unity at 109.5 M , and √2πσint −2 2σint ∼ then fall to < 10% by M 106 M (e.g., S´anchez- (4) ∗ ∼ Janssen et al. 2019; Neumayer et al. 2020). We then write the number of black holes in terms of the Our observed limits are consistent with an occupa- halo maximum velocity function, the occupation frac- tion fraction that falls with mass in a similar way to tion λocc(vmax), and P (log MBH vmax): the nucleation fraction. In fact, the details of how we | seed the satellites become relatively unimportant in this case, because the number of black holes is set by the Z 1 N(MBH) = Nh(vmax)λocc(vmax)P (MBH vmax) dvmax small number of available dissolved satellites with ini- MBH | tial M > 107 M , where most of the black holes would (5) ∗ 14

Table 1. Number Density Limits 5.2. Important limitations to our assumptions Dynamical modeling that resolves the scale of the Nstar MBH(M ) N(< MBH) sphere of influence of the black hole is needed to be 2 3 5 × 10 10 24 secure in the properties of the hyper-compact star clus- 3 3 10 5 × 10 11 ters. There are at least three effects that we have ignored 3 4 5 × 10 10 5 or oversimplified that could strongly impact our conclu- 104 5 × 104 3 sions. First, we assume that once a satellite reaches a 6 Note—Limits on number of hyper- stellar mass of 10 M in SatGen, it will fully dissolve.

compact star clusters, and corre- This assumption should be tested in the context of a sponding limits on the number of central intermediate-mass black hole. wandering massive black holes in- Second, we have ignored the possibility that hyper- ferred in this work. compact star clusters may be composed mostly of com- pact remnants. Given that neutron stars and black holes will settle to the cluster center through mass- segregation, there is some chance that these clusters ex- ist but have much higher mass-to-light ratios than as- sumed here. In fact, a mass fraction in stellar-mass rem- Nstar nants of 20%, which sounds high compared with some 3 4 5 ∼ 10 10 10 models (e.g., Zocchi et al. 2019; Baumgardt et al. 2019),

focc = 1 has recently been suggested to explain the low-density focc=NSC of Palomar 5 and potentially other fluffy globular clus- Limits σ ∗ ters (Gieles et al. 2021). It would be very interesting 102 to know how the black hole pathways investigated by ) Gieles et al. play out with an intermediate-mass black BH hole at the cluster center.

> M Third, we have assumed that all of the clusters have

( 1 10 the same size of 1 pc, which is clearly an oversimpli- N fication and is also related to the detailed dynamical evolution of the cluster. If the clusters are significantly larger than the 1 pc that we assume here, then their 100 ∼ clustered signal would drop, as would our sensitivity. As 103 104 105 106 an example, if the Nstar = 1000 cluster had a size of 5 pc, MBH (M ) then our completeness would drop roughly by a factor of two compared to what is shown in Figure6.

Figure 8. Upper limits on the number of hyper-compact star clusters in the Milky Way halo within 50 kpc. Theo- 5.3. Other ways to search for wandering black holes retical expectations from SatGen, assuming either that every While we have been focused on low-mass hyper- satellite hosts a 103 −105 M black hole are shown in dotted, compact clusters, one promising avenue for continued or in solid the perhaps more realistical case that the fraction of galaxies hosting black holes is the same as the fraction of study is certainly to identify all of the Milky Way star galaxies that host nuclear star clusters. Black hole masses clusters that may be stripped nuclei, using both stellar are estimated from the halo maximum velocity (blue). Our population and orbital information (e.g., Massari et al. limits assume that all clusters would be found within 50 kpc 2019; Pfeffer et al. 2021). Future observations with ex- of the Galactic Center. tremely large telescopes will provide far more stringent limits on their possible central black hole content. be found. If we improve the probed area by a factor Another angle is to think about the other channels of a few (e.g., with the Rubin Observatory; Ivezi´cet al. that may generate hyper-compact clusters surround- 2019), we might be able to constrain the occupation frac- ing black holes at larger galactocentric radius, includ- tion mass dependence further. In the meantime, we still ing early ejection from the gravitational slingshot (e.g., expect that there must be a wandering black hole popu- Volonteri et al. 2003b; O’Leary & Loeb 2009; Lena et al. lation in Milky-Way like galaxies, and we explore other 2020). It is therefore interesting to consider searches ways to search for them in §5.3. that will reach larger galactocentric radii in the Milky 15

Way, as well as complementary techniques for searching though these remain very challenging measurements. In for extragalactic analogs. the era of extremely large telescopes, we can hope to Euclid and the Vera Rubin Observatories Legacy Sur- detect 104 M black holes dynamically out to 5 10 ∼ − vey of Space and Time (LSST; Ivezi´cet al. 2019) will Mpc, which will reach a much larger number of massive provide a more sensitive search of these potentially more globular cluster and UCDs (Greene et al. 2019). distant hyper-compact star clusters. Here we consider There are promising search avenues via accretion sig- what these clusters would look like in next-generation natures as well. One is through tidal capture of individ- imaging surveys. From the ground, the LSST will ual stars in the cluster (e.g., MacLeod et al. 2016), which produce single-epoch images with 5 σ depths of r = may be what we are observing in the hyper-luminous X- 24.7 mag. At this depth, we could expect to uncover ray source HLX1 (e.g., Farrell et al. 2009; Lasota et al. 500-star hyper-compact clusters out to 70 kpc, where 2011; Soria et al. 2017). Tidal disruption events should ∼ we still expect 25 5 stars to be detected within a 300 probe wandering black hole populations, and there are ∼ ± radius, although crowding will likely be significant. A some potential tidal events in off-nuclear stellar systems 1000-star cluster will have r = 17 18 mag within 200 (e.g., Lin et al. 2018). − ∼ at 70-90 kpc, making them easily detectable in prin- It is possible that wandering black holes in more mas- ciple. However, these clusters will be very crowded at sive halos (e.g., the Virgo cluster) may encounter suf- ground-based spatial resolution, and so techniques using ficient gas to be detected via accretion of inter-cluster colors may be required for these more massive clusters medium (e.g., Guo et al. 2020). This detection channel beyond 50 kpc. Such clusters may be selected as out- is highly unlikely in Milky Way-mass systems. It may, ∼ liers in the color-magnitude diagram, as they will be however, be possible to detect very low Eddington-ratio far more luminous than expected for their color (e.g., accretion onto black holes in the centers of normal extra- Lena et al. 2020). One might imagine combining such galactic globular clusters in the radio, which is relatively a ground-based search with higher-resolution data from boosted at low accretion rates (e.g., Maccarone et al. a mission like Euclid (Racca et al. 2016) to see if the 2005). On the other hand, searches with deep radio sur- putative clusters are resolved into a couple of stars at veys have not been successful to date, either in the Milky higher resolution. Way (Tremou et al. 2018) or external galaxies (Wrobel Finally, an additional prospect is opened by the Ro- et al. 2016). Next-generation radio facilities will be very man Space Telescope (Spergel et al. 2015) to search for powerful to further the reach of these searches (Wrobel the 5000 star hyper-compact clusters in halos of ex- et al. 2019). ∼ ternal galaxies. While these clusters would be point Finally, wandering black holes may eventually be de- sources, they would be anomalously bright for their tected as extreme mass-ratio inspiral events through color. Since such a cluster would have an integrated the merger of a stellar mass black hole with the cen- magnitude of z 22 23 mag at 10 Mpc, one could tral intermediate-mass black hole (e.g., Gair et al. 2010; ≈ − imagine performing a search leveraging the proposed Ro- Amaro-Seoane et al. 2015; Eracleous et al. 2019). Using man Infrared Nearby Galaxy Survey (RINGS) (Williams the tools developed here, it would be useful to calculate 2015) to search for hyper-compact star clusters at the the number densities of extragalactic wandering black very low-luminosity end of the star cluster luminosity holes to estimate a detection rate for space-based grav- function, using the high spatial resolution to remove itational wave experiments. background galaxies and the colors to distinguish from halo stars, as extragalactic globular clusters are cur- 6. SUMMARY rently found. However, ultimately identifying these as We have presented a search for hyper-compact star real cluster candidates will be very challenging, as will clusters that would be the hosts of wandering black certainly require spectroscopy. holes. If such black holes are deposited by infalling satel- Another tool, and the most robust one, is to search lites, we expect them to be 1 pc in size, have 500-5000 for the dynamical signatures of black holes in the mo- ∼ stars, and fall within 50 kpc of the Galactic Center. tions of stars. At somewhat higher mass, stripped ∼ We use Gaia+DECaLS data over 8000 deg2. Using remnant clusters surrounding massive black holes have ∼ a Negative Binomial model to describe the distributions been found from dynamical signatures in ultra-compact of stellar counts around each target star in Gaia EDR3, dwarf galaxies (Seth et al. 2014; Ahn et al. 2017, 2018). we identify large outliers in count space. Real stellar Closer to home in Andromeda, lower-mass stellar clus- clusters with normal mass functions would have three- ters, likely to be ultra-compact dwarfs, also show some five times as many stars in DECaLS, allowing us to effi- signs of nuclear black holes (Gebhardt et al. 2005) al- ciently eliminate candidates through a cross-match with 16

DECaLS. We do not find any hyper-compact star clus- ACKNOWLEDGMENTS ters within 30 50 kpc (for clusters of 500-5000 M ). ∼ − We translate these limits into upper limits on the num- JEG acknowledges support from NSF grants AST- ber of intermediate-mass black holes wandering in the 1815417 and AST-1907723. YST is grateful to be sup- inner Milky Way halo. ported by the NASA Hubble Fellowship grant HST- We also use modern semi-analytic models to bracket HF2-51425.001 awarded by the Space Telescope Science the number of 103 105 M black holes that might Institute. S.D. is supported by NASA through Hub- − wander in the Milky Way halo using the SatGen code ble Fellowship grant HST-HF2-51454.001- A awarded (Jiang et al. 2019). SatGen calculates the number of by the Space Telescope Science Institute, which is oper- expected dissolved satellites as a function of galactocen- ated by the Association of Universities for Research in tric radius and halo/stellar mass at infall. We then ex- Astronomy, Incorporated, under NASA contract NAS5- trapolate black hole-galaxy scaling relations to predict 26555. J.P.G. is supported by an NSF Astronomy and the range of black hole masses that might inhabit these Astrophysics Postdoctoral Fellowship under award AST- halos. Based on our calculations, we expect as many 1801921. J.S. acknowledges support from NSF grant 3 AST-1812856 and the Packard Foundation. This paper as 100 black holes with MBH> 10 M based on scal- made use of the Whole Sky Database (wsdb) created ing from the MBH σ relation, should every satellite − ∗ carry an intermediate-mass black hole. In the context by Sergey Koposov and maintained at the Institute of of this model, we can rule out that all satellite galaxies Astronomy, Cambridge by Sergey Koposov, Vasily Be- host central black holes from our non-detection. Our lokurov and Wyn Evans with financial support from the measurements are consistent with models in which most Science & Technology Facilities Council (STFC) and the black hole hosts have stellar masses > 107 M . European Research Council (ERC).

In the near future, wide-area imaging surveys from The Legacy Surveys consist of three individual and the ground and space will open up new discovery space complementary projects: the Dark Energy Camera for hyper-compact star clusters in the Milky Way and Legacy Survey (DECaLS; Proposal ID #2014B-0404; nearby galaxies. Extremely large telescopes will enable PIs: David Schlegel and Arjun Dey), the Beijing- dynamical detection of 103 M black holes in the Milky Arizona Sky Survey (BASS; NOAO Prop. ID #2015A-

Way and 104 M black holes within the Local Volume. 0801; PIs: Zhou Xu and Xiaohui Fan), and the Mayall

The Rubin Observatory LSST will increase the rate of z-band Legacy Survey (MzLS; Prop. ID #2016A-0453; detection of tidal disruption events, some of which may PI: Arjun Dey). DECaLS, BASS and MzLS together be in off-nuclear systems. Eventually gravitational wave include data obtained, respectively, at the Blanco tele- detectors in space will be sensitive to the mergers of scope, Cerro Tololo Inter-American Observatory, NSF’s stellar-mass and intermediate-mass black holes in these NOIRLab; the Bok telescope, Steward Observatory, clusters (e.g., Gair et al. 2010; Gallo & Sesana 2019). University of Arizona; and the Mayall telescope, Kitt Thus, if there are black holes wandering in galaxy halos, Peak National Observatory, NOIRLab. The Legacy Sur- we will begin to uncover them relatively soon. veys project is honored to be permitted to conduct astro- nomical research on Iolkam Du’ag (Kitt Peak), a moun- tain with particular significance to the Tohono O’odham Nation. NOIRLab is operated by the Association of Universi- ties for Research in Astronomy (AURA) under a cooper- ative agreement with the National Science Foundation. This project used data obtained with the Dark En- ergy Camera (DECam), which was constructed by the Dark Energy Survey (DES) collaboration. Funding for the DES Projects has been provided by the U.S. De- partment of Energy, the U.S. National Science Founda- tion, the Ministry of Science and Education of Spain, the Science and Technology Facilities Council of the United Kingdom, the Higher Education Funding Coun- cil for England, the National Center for Supercomput- ing Applications at the University of Illinois at Urbana- Champaign, the Kavli Institute of Cosmological Physics 17 at the University of Chicago, Center for Cosmology of Illinois at Urbana-Champaign, the Institut de Cien- and Astro-Particle Physics at the Ohio State Univer- cies de l’Espai (IEEC/CSIC), the Institut de Fisica sity, the Mitchell Institute for Fundamental Physics and d’Altes Energies, Lawrence Berkeley National Labora- Astronomy at Texas A&M University, Financiadora de tory, the Ludwig Maximilians Universitat Munchen and Estudos e Projetos, Fundacao Carlos Chagas Filho de the associated Excellence Cluster Universe, the Univer- Amparo, Financiadora de Estudos e Projetos, Funda- sity of Michigan, NSF’s NOIRLab, the University of cao Carlos Chagas Filho de Amparo a Pesquisa do Es- Nottingham, the Ohio State University, the University tado do Rio de Janeiro, Conselho Nacional de Desen- of Pennsylvania, the University of Portsmouth, SLAC volvimento Cientifico e Tecnologico and the Ministe- National Accelerator Laboratory, Stanford University, rio da Ciencia, Tecnologia e Inovacao, the Deutsche the University of Sussex, and Texas A&M University. Forschungsgemeinschaft and the Collaborating Institu- The Legacy Surveys imaging of the DESI footprint tions in the Dark Energy Survey. The Collaborat- is supported by the Director, Office of Science, Office ing Institutions are Argonne National Laboratory, the of High Energy Physics of the U.S. Department of En- University of California at Santa Cruz, the University ergy under Contract No. DE-AC02-05CH1123, by the of Cambridge, Centro de Investigaciones Energeticas, National Energy Research Scientific Computing Center, Medioambientales y Tecnologicas-Madrid, the Univer- a DOE Office of Science User Facility under the same sity of Chicago, University College London, the DES- contract; and by the U.S. National Science Foundation, Brazil Consortium, the University of Edinburgh, the Division of Astronomical Sciences under Contract No. Eidgenossische Technische Hochschule (ETH) Zurich, AST-0950945 to NOAO. Fermi National Accelerator Laboratory, the University

REFERENCES Ahn, C. P., Seth, A. C., den Brok, M., et al. 2017, ApJ, Choi, J., Dotter, A., Conroy, C., et al. 2016, ApJ, 823, 102, 839, 72, doi: 10.3847/1538-4357/aa6972 doi: 10.3847/0004-637X/823/2/102 Ahn, C. P., Seth, A. C., Cappellari, M., et al. 2018, ApJ, Danieli, S., van Dokkum, P., & Conroy, C. 2018, ApJ, 856, 858, 102, doi: 10.3847/1538-4357/aabc57 69, doi: 10.3847/1538-4357/aaadfb Amaro-Seoane, P., Gair, J. R., Pound, A., Hughes, S. A., & Davis, S. W., Narayan, R., Zhu, Y., et al. 2011, ApJ, 734, Sopuerta, C. F. 2015, in Journal of Physics Conference 111, doi: 10.1088/0004-637X/734/2/111 Series, Vol. 610, Journal of Physics Conference Series, Dey, A., Schlegel, D. J., Lang, D., et al. 2019, AJ, 157, 168, 012002, doi: 10.1088/1742-6596/610/1/012002 doi: 10.3847/1538-3881/ab089d Bahcall, J. N., & Wolf, R. A. 1976, ApJ, 209, 214, Dotter, A. 2016, ApJS, 222, 8, doi: 10.1086/154711 doi: 10.3847/0067-0049/222/1/8 Drlica-Wagner, A., Bechtol, K., Rykoff, E. S., et al. 2015, Baldassare, V. F., Geha, M., & Greene, J. 2019, arXiv ApJ, 813, 109, doi: 10.1088/0004-637X/813/2/109 e-prints, arXiv:1910.06342. Eracleous, M., Gezari, S., Sesana, A., et al. 2019, in BAAS, https://arxiv.org/abs/1910.06342 Vol. 51, 10 Baumgardt, H. 2017, MNRAS, 464, 2174, Farrell, S. A., Webb, N. A., Barret, D., Godet, O., & doi: 10.1093/mnras/stw2488 Rodrigues, J. M. 2009, Nature, 460, 73, Baumgardt, H., He, C., Sweet, S. M., et al. 2019, MNRAS, doi: 10.1038/nature08083 488, 5340, doi: 10.1093/mnras/stz2060 Farrell, S. A., Servillat, M., Gladstone, J. C., et al. 2014, Bechtol, K., Drlica-Wagner, A., Balbinot, E., et al. 2015, MNRAS, 437, 1208, doi: 10.1093/mnras/stt1924 ApJ, 807, 50, doi: 10.1088/0004-637X/807/1/50 Flaugher, B., Diehl, H. T., Honscheid, K., et al. 2015, AJ, Bellovary, J. M., Cleary, C. E., Munshi, F., et al. 2019, 150, 150, doi: 10.1088/0004-6256/150/5/150 MNRAS, 482, 2913, doi: 10.1093/mnras/sty2842 Fragione, G., Ginsburg, I., & Kocsis, B. 2018, ApJ, 856, 92, Bellovary, J. M., Governato, F., Quinn, T. R., et al. 2010, doi: 10.3847/1538-4357/aab368 ApJ, 721, L148, doi: 10.1088/2041-8205/721/2/L148 Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al. Belokurov, V., Zucker, D. B., Evans, N. W., et al. 2006, 2020, arXiv e-prints, arXiv:2012.01533. ApJL, 647, L111, doi: 10.1086/507324 https://arxiv.org/abs/2012.01533 Campanelli, M., Lousto, C., Zlochower, Y., & Merritt, D. Gaia Collaboration, Prusti, T., de Bruijne, J. H. J., et al. 2007, ApJL, 659, L5, doi: 10.1086/516712 2016, A&A, 595, A1, doi: 10.1051/0004-6361/201629272 18

Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al. Kormendy, J., & Ho, L. C. 2013, ARA&A, 51, 511, 2018, A&A, 616, A1, doi: 10.1051/0004-6361/201833051 doi: 10.1146/annurev-astro-082708-101811 Gair, J. R., Tang, C., & Volonteri, M. 2010, PhRvD, 81, Kroupa, P. 2001, MNRAS, 322, 231, 104014, doi: 10.1103/PhysRevD.81.104014 doi: 10.1046/j.1365-8711.2001.04022.x Gallo, E., & Sesana, A. 2019, ApJ, accepted, Lasota, J. P., Alexander, T., Dubus, G., et al. 2011, ApJ, arXiv:1909.02585. https://arxiv.org/abs/1909.02585 735, 89, doi: 10.1088/0004-637X/735/2/89 Gebhardt, K., Rich, R. M., & Ho, L. C. 2005, ApJ, 634, Lena, D., Jonker, P. G., Rauer, J. P., Hernandez, S., & 1093, doi: 10.1086/497023 Kostrzewa-Rutkowska, Z. 2020, MNRAS, 495, 1771, Gebhardt, K., Lauer, T. R., Kormendy, J., et al. 2001, AJ, doi: 10.1093/mnras/staa1174 122, 2469, doi: 10.1086/323481 Lin, D., Strader, J., Carrasco, E. R., et al. 2018, Nature Gieles, M., Erkal, D., Antonini, F., Balbinot, E., & Astronomy, 2, 656, doi: 10.1038/s41550-018-0493-1 Pe˜narrubia,J. 2021, Nature Astronomy, submitted L¨utzgendorf,N., Gualandris, A., Kissler-Patig, M., et al. (arXiv:2102.11348), arXiv:2102.11348. 2012, A&A, 543, A82, doi: 10.1051/0004-6361/201219062 https://arxiv.org/abs/2102.11348 Maccarone, T. J., Fender, R. P., & Tzioumis, A. K. 2005, Greco, J. P., van Dokkum, P., Danieli, S., Carlsten, S. G., MNRAS, 356, L17, doi: 10.1111/j.1745-3933.2005.08584.x & Conroy, C. 2021, ApJ, 908, 24, MacLeod, M., Trenti, M., & Ramirez-Ruiz, E. 2016, ApJ, doi: 10.3847/1538-4357/abd030 819, 70, doi: 10.3847/0004-637X/819/1/70 Greene, J., Barth, A., Bellini, A., et al. 2019, in BAAS, Marconi, A., Risaliti, G., Gilli, R., et al. 2004, MNRAS, Vol. 51, 83 351, 169, doi: 10.1111/j.1365-2966.2004.07765.x Greene, J. E., Strader, J., & Ho, L. C. 2020, ARA&A, 58, Massari, D., Koppelman, H. H., & Helmi, A. 2019, A&A, 257, doi: 10.1146/annurev-astro-032620-021835 630, L4, doi: 10.1051/0004-6361/201936135 Greene, J. E., Seth, A., Kim, M., et al. 2016, ApJL, 826, McConnachie, A. W. 2012, AJ, 144, 4, L32, doi: 10.3847/2041-8205/826/2/L32 doi: 10.1088/0004-6256/144/1/4 Guo, M., Inayoshi, K., Michiyama, T., & Ho, L. C. 2020, McConnell, N. J., & Ma, C.-P. 2013, ApJ, 764, 184, ApJ, 901, 39, doi: 10.3847/1538-4357/abacc1 doi: 10.1088/0004-637X/764/2/184 Harris, W. E. 2010, arXiv e-prints, arXiv:1012.3224. Merritt, D., Schnittman, J. D., & Komossa, S. 2009, ApJ, https://arxiv.org/abs/1012.3224 699, 1690, doi: 10.1088/0004-637X/699/2/1690 Ibata, R., Bellazzini, M., Chapman, S. C., et al. 2009, Miller, B. P., Gallo, E., Greene, J. E., et al. 2015, ApJ, 799, ApJL, 699, L169, doi: 10.1088/0004-637X/699/2/L169 98, doi: 10.1088/0004-637X/799/1/98 Ibata, R. A., Gilmore, G., & Irwin, M. J. 1995, MNRAS, Miller, M. C., & Hamilton, D. P. 2002, MNRAS, 330, 232, 277, 781, doi: 10.1093/mnras/277.3.781 doi: 10.1046/j.1365-8711.2002.05112.x Inayoshi, K., Visbal, E., & Haiman, Z. 2020, ARA&A, 58, Munshi, F., Brooks, A., Applebaum, E., et al. 2021, arXiv 27, doi: 10.1146/annurev-astro-120419-014455 e-prints, arXiv:2101.05822. Ivezi´c, Z.,ˇ Kahn, S. M., Tyson, J. A., et al. 2019, ApJ, 873, https://arxiv.org/abs/2101.05822 111, doi: 10.3847/1538-4357/ab042c Neumayer, N., Seth, A., & B¨oker, T. 2020, A&A Rv, 28, 4, Jiang, F., Dekel, A., Freundlich, J., et al. 2020, arXiv doi: 10.1007/s00159-020-00125-0 e-prints, arXiv:2005.05974. Neumayer, N., & Walcher, C. J. 2012, Advances in https://arxiv.org/abs/2005.05974 Astronomy, 2012, 709038, doi: 10.1155/2012/709038 Jiang, Y.-F., Stone, J. M., & Davis, S. W. 2019, ApJ, 880, Nguyen, D. D., Seth, A. C., Neumayer, N., et al. 2018, ApJ, 67, doi: 10.3847/1538-4357/ab29ff 858, 118, doi: 10.3847/1538-4357/aabe28 Koposov, S., & Bartunov, O. 2006, in Astronomical Society Nguyen, D. D., den Brok, M., Seth, A. C., et al. 2019a, of the Pacific Conference Series, Vol. 351, Astronomical ApJ, submitted, arXiv:1902.03813. Data Analysis Software and Systems XV, ed. C. Gabriel, https://arxiv.org/abs/1902.03813 C. Arviset, D. Ponz, & S. Enrique, 735 Nguyen, D. D., Seth, A. C., Neumayer, N., et al. 2019b, Koposov, S., Belokurov, V., Evans, N. W., et al. 2008, ApJ, ApJ, 872, 104, doi: 10.3847/1538-4357/aafe7a 686, 279, doi: 10.1086/589911 Norris, J. E., Freeman, K. C., & Mighell, K. J. 1996, ApJ, Koposov, S. E., Belokurov, V., Torrealba, G., & Evans, 462, 241, doi: 10.1086/177145 N. W. 2015, ApJ, 805, 130, Noyola, E., Gebhardt, K., Kissler-Patig, M., et al. 2010, doi: 10.1088/0004-637X/805/2/130 ApJL, 719, L60, doi: 10.1088/2041-8205/719/1/L60 19

O’Leary, R. M., & Loeb, A. 2009, MNRAS, 395, 781, Seth, A. C., Blum, R. D., Bastian, N., Caldwell, N., & doi: 10.1111/j.1365-2966.2009.14611.x Debattista, V. P. 2008, ApJ, 687, 997, —. 2012, MNRAS, 421, 2737, doi: 10.1086/591935 doi: 10.1111/j.1365-2966.2011.20078.x Seth, A. C., van den Bosch, R., Mieske, S., et al. 2014, Paxton, B., Bildsten, L., Dotter, A., et al. 2011, ApJS, 192, Nature, 513, 398, doi: 10.1038/nature13762 3, doi: 10.1088/0067-0049/192/1/3 She, R., Ho, L. C., & Feng, H. 2017, ApJ, 842, 131, Paxton, B., Cantiello, M., Arras, P., et al. 2013, ApJS, 208, doi: 10.3847/1538-4357/aa7634 4, doi: 10.1088/0067-0049/208/1/4 Shipp, N., Drlica-Wagner, A., Balbinot, E., et al. 2018, Paxton, B., Marchant, P., Schwab, J., et al. 2015, ApJS, ApJ, 862, 114, doi: 10.3847/1538-4357/aacdab 220, 15, doi: 10.1088/0067-0049/220/1/15 Silk, J., & Rees, M. J. 1998, A&A, 331, L1. Paxton, B., Schwab, J., Bauer, E. B., et al. 2018, ApJS, https://arxiv.org/abs/astro-ph/9801013 234, 34, doi: 10.3847/1538-4365/aaa5a8 Soria, R., Musaeva, A., Wu, K., et al. 2017, MNRAS, 469, Peres, A. 1962, Physical Review, 128, 2471, 886, doi: 10.1093/mnras/stx888 doi: 10.1103/PhysRev.128.2471 Spergel, D., Gehrels, N., Baltay, C., et al. 2015, arXiv Pfeffer, J., Griffen, B. F., Baumgardt, H., & Hilker, M. e-prints, arXiv:1503.03757. 2014, MNRAS, 444, 3670, doi: 10.1093/mnras/stu1705 https://arxiv.org/abs/1503.03757 Torrealba, G., Belokurov, V., Koposov, S. E., et al. 2018, Pfeffer, J., Hilker, M., Baumgardt, H., & Griffen, B. F. MNRAS, 475, 5085, doi: 10.1093/mnras/sty170 2016, MNRAS, 458, 2492, doi: 10.1093/mnras/stw498 —. 2019, MNRAS, 488, 2743, doi: 10.1093/mnras/stz1624 Pfeffer, J., Lardo, C., Bastian, N., Saracino, S., & Kamann, Tremmel, M., Governato, F., Volonteri, M., Pontzen, A., & S. 2021, MNRAS, 500, 2514, Quinn, T. R. 2018, ApJL, 857, L22, doi: 10.1093/mnras/staa3407 doi: 10.3847/2041-8213/aabc0a Plummer, H. C. 1911, MNRAS, 71, 460, Tremou, E., Strader, J., Chomiuk, L., et al. 2018, ApJ, 862, doi: 10.1093/mnras/71.5.460 16, doi: 10.3847/1538-4357/aac9b9 Portegies Zwart, S. F., & McMillan, S. L. W. 2002, ApJ, van Wassenhove, S., Volonteri, M., Walker, M. G., & Gair, 576, 899, doi: 10.1086/341798 J. R. 2010, MNRAS, 408, 1139, Racca, G. D., Laureijs, R., Stagnaro, L., et al. 2016, in doi: 10.1111/j.1365-2966.2010.17189.x Society of Photo-Optical Instrumentation Engineers Voggel, K. T., Seth, A. C., Baumgardt, H., et al. 2019, (SPIE) Conference Series, Vol. 9904, Space Telescopes ApJ, 871, 159, doi: 10.3847/1538-4357/aaf735 and Instrumentation 2016: Optical, Infrared, and Volonteri, M. 2010, A&A Rv, 18, 279, Millimeter Wave, ed. H. A. MacEwen, G. G. Fazio, doi: 10.1007/s00159-010-0029-x M. Lystrup, N. Batalha, N. Siegler, & E. C. Tong, Volonteri, M., Haardt, F., & Madau, P. 2003a, ApJ, 582, 99040O, doi: 10.1117/12.2230762 559, doi: 10.1086/344675 Rashkov, V., & Madau, P. 2014, ApJ, 780, 187, —. 2003b, ApJ, 582, 559, doi: 10.1086/344675 doi: 10.1088/0004-637X/780/2/187 Volonteri, M., & Perna, R. 2005, MNRAS, 358, 913, Reines, A., Condon, J., Darling, J., & Greene, J. 2019, doi: 10.1111/j.1365-2966.2005.08832.x ApJ, accepted, arXiv:1909.04670. Webb, N., Cseh, D., Lenc, E., et al. 2012, Science, 337, 554, https://arxiv.org/abs/1909.04670 doi: 10.1126/science.1222779 Reines, A. E., & Volonteri, M. 2015, ApJ, 813, 82, Williams, B. 2015, WINGS: WFIRST Infrared Nearby doi: 10.1088/0004-637X/813/2/82 Galaxy Survey, NASA WFIRST Proposal Ricarte, A., & Natarajan, P. 2018, MNRAS, 481, 3278, Willman, B., Dalcanton, J. J., Martinez-Delgado, D., et al. doi: 10.1093/mnras/sty2448 2005, ApJL, 626, L85, doi: 10.1086/431760 Ricarte, A., Tremmel, M., Natarajan, P., Zimmer, C., & Wrobel, J., Haiman, Z., Holley-Bockelmann, K., et al. 2019, Quinn, T. 2021, MNRAS, accepted (arXiv:2103.12124, BAAS, 51, 29. https://arxiv.org/abs/1903.04902 arXiv:2103.12124. https://arxiv.org/abs/2103.12124 Wrobel, J. M., Miller-Jones, J. C. A., & Middleton, M. J. Rodr´ıguez-Puebla, A., Primack, J. R., Avila-Reese, V., & 2016, AJ, 152, 22, doi: 10.3847/0004-6256/152/1/22 Faber, S. M. 2017, MNRAS, 470, 651, Zinnecker, H., Keable, C. J., Dunlop, J. S., Cannon, R. D., doi: 10.1093/mnras/stx1172 & Griffiths, W. K. 1988, in The Harlow-Shapley S´anchez-Janssen, R., Cˆot´e,P., Ferrarese, L., et al. 2019, Symposium on Globular Cluster Systems in Galaxies, ed. ApJ, 878, 18, doi: 10.3847/1538-4357/aaf4fd J. E. Grindlay & A. G. D. Philip, Vol. 126, 603 20

Zocchi, A., Gieles, M., & H´enault-Brunet,V. 2019, MNRAS, 482, 4713, doi: 10.1093/mnras/sty1508