The hidden satellites of massive galaxies and quasars at high-redshift
MMNRAS , 1–7 (2019) Preprint 5 September 2019 Compiled using MNRAS L A TEX style file v3.0
The hidden satellites of massive galaxies and quasars athigh-redshift
Tiago Costa (cid:63) , Joakim Rosdahl & Taysun Kimm Max-Planck-Institut f¨ur Astrophysik, Karl-Schwarzschild-Straße 1, D-85748 Garching b. M¨unchen, Germany CRAL, Universit´e de Lyon I, CNRS UMR 5574, ENS-Lyon, 9 Avenue Charles Andr´e, 69561, Saint-Genis-Laval, France Department of Astronomy, Yonsei University, 50 Yonsei-ro, Seodaemun-gu, Seoul 03722, Republic of Korea
Accepted 2019
ABSTRACT
Using cosmological, radiation-hydrodynamic simulations targeting a rare ≈ × M (cid:12) halo at z = , we show that the number counts and internal properties of satellitegalaxies within the massive halo are sensitively regulated by a combination of localstellar radiative feedback and strong tidal forces. Radiative feedback operates beforethe first supernova explosions erupt and results in less tightly-bound galaxies. Satellitesare therefore more vulnerable to tidal stripping when they accrete onto the mainprogenitor and are tidally disrupted on a significantly shorter timescale. Consequently,the number of satellites with M (cid:63) > M (cid:12) within the parent system’s virial radiusdrops by up to with respect to an identical simulation performed without stellarradiative feedback. Radiative feedback also impacts the central galaxy, whose effectiveradius increases by a factor (cid:46) due to the presence of a more extended and diffusestellar component. We suggest that the number of satellites in the vicinity of massivehigh-redshift galaxies is an indication of the strength of stellar radiative feedbackand and can be anomalously low in the extreme cosmic environments of high-redshiftquasars. Key words: galaxies: evolution – galaxies: high-redshift – radiative transfer
Bright quasars at z > are powered by accreting supermas-sive black holes with estimated masses of (cid:38) M (cid:12) (Fanet al. 2001; Wu et al. 2015; Ba˜nados et al. 2018). At z = ,the Hubble time ( t H ≈
930 Myr ) corresponds to (cid:46) e-folding times t BH ≈ ( η r / . ) Myr , assuming Eddington-limited black hole accretion and a fixed radiative efficiency η r . Growth to the required masses may thus proceed at theEddington rate from the ∼
100 M (cid:12) remnants of Pop IIIstars or, alternatively, through super-Eddington accretion-limited episodes or directly from massive seed black holeswith M BH ≈ − M (cid:12) (e.g. Begelman et al. 2006). Inall scenarios, gas inflow into the sphere of influence of theaccreting black holes must be efficient, a condition whichis more likely to be fulfilled if massive black holes grow atthe centre of massive dark matter haloes (Efstathiou & Rees1988).Cosmological hydrodynamic simulations following blackhole accretion and active galactic nucleus (AGN) feedbackhave successfully reproduced the rapid assembly of ∼ M (cid:12) black holes by z = (Sijacki et al. 2009; Di Matteo et al. (cid:63) E-mail: [email protected] M vir > M (cid:12) . Since they trace high- σ peaks of the cosmic densityfield distribution at z = (Volonteri & Rees 2006), suchhaloes should exhibit a statistically significant overdensityin satellite galaxy number counts when compared to lowermass systems (e.g. Costa et al. 2014), though the associatedvariance is considerable (Habouzit et al. 2019).This prediction has not been confirmed by observationsconclusively. Wide-field imaging campaigns have returnedambiguous results, reporting both under- and over-densitiesof galaxy counts in z (cid:38) quasar fields (e.g. Kim et al. 2009;Balmaverde et al. 2017; Champagne et al. 2018). Other ob-servations have started to probe the properties of the satel-lite galaxies surrounding high-redshift quasars in greater de-tail. Trakhtenbrot et al. (2017) find a number of sub-mmgalaxies within a projected distance ≈ −
50 kpc from threeout of the six z ≈ quasar fields probed by their ALMA ob-servations, suggesting that quasar host galaxies often expe-rience major mergers. Similarly, based on an ALMA surveyof 25 z > . quasars, Decarli et al. (2017) serendipitouslydiscovered [CII] bright galaxies around four of the quasars.While detected satellites tend to be nearly as massive as the © a r X i v : . [ a s t r o - ph . GA ] S e p Costa, Rosdahl & Kimm quasar hosts, the upcoming James Webb Space Telescope(JWST) will enable the discovery of fainter satellites andplace tighter constraints on their environments.In this paper, we present cosmological, radiation-hydrodynamic, ‘zoom-in’ simulations of a ≈ . × M (cid:12) halo at z = as a likely host to a bright quasar (see e.g.Costa et al. 2014). In particular, we compare simulationsthat include or neglect stellar radiative feedback and ex-plore the impact of stellar radiation on the demographicsand properties of satellites. We describe our simulations inSection 2, present our results in Section 3, summarise theimplications of our findings in Section 4 and our conclusionsin Section 5. A flat Λ CDM cosmology with Ω m = . , Ω Λ = . , Ω b = . and h = . is adopted through-out this paper. We perform cosmological, radiation-hydrodynamic, ‘zoom-in’ simulations targeting a massive halo with virial mass M vir ≈ . × M (cid:12) and virial radius R vir ≈ . , cor-responding to an angular scale of ≈ . (cid:48)(cid:48) , at z = . Thisis the second most massive halo found at z = within acosmological volume of comoving side length h − Mpc ,as presented in Costa et al. (2018). The massive galaxiesevolving in such haloes are the best candidates for hostingrare supermassive black holes with masses M BH ∼ M (cid:12) at z = , as shown in Costa et al. (e.g. 2014).We employ Ramses-RT (Teyssier 2002; Rosdahl et al.2013; Rosdahl & Teyssier 2015) to evolve the coupled evolu-tion of gas hydrodynamics, radiative transfer of stellar radi-ation and N-body dynamics of stellar populations and darkmatter, modelled with particles of mass m (cid:63) = . × M (cid:12) and m DM = × M (cid:12) , respectively. In order to increasethe numerical resolution, we refine a cell if its total enclosedmass satisfies M DM + Ω m Ω b M bar > × m DM , where M DM and M bar are the total dark matter and baryonic masses in thecell, respectively. The minimum cell size is ∆ x min ≈
40 pc .All spatial coordinates are given in physical units.We follow non-equilibrium cooling of hydrogen and he-lium (coupled to the radiative fluxes present in the simu-lation), metal-line cooling down to T =
10 K and star for-mation using a Schmidt law with a variable star formationefficiency, as described in Kimm et al. (2017). In order tomodel the ionising flux of external sources, we adopt thespatially homogeneous and time-evolving UV background ofFaucher-Gigu`ere et al. (2009). Supernova feedback is mod-elled through injection of thermal energy, if the Sedov-Taylorphase is resolved, and momentum otherwise, using the for-mulation of Kimm et al. (2015). Supernova events occur
10 Myr after their parent stellar particle forms.We focus on two simulations which differ only inwhether they include or neglect stellar radiative feedback.All other parameters are kept unchanged and all other feed-back processes are modelled identically. We name our simu-lation with radiative transfer
SN+RT and our simulation with-out stellar radiation SN . The virial radius is defined as the radius enclosing a mean den-sity 200 times the critical density of the Universe. In SN+RT , we follow photo-ionisation, photo-heating andradiation pressure from stellar radiation. Time-integrationin
Ramses-RT is performed explicitly, such that the time-step is limited by the speed-of-light. In order to preventour simulations from becoming computationally prohibitive,we adopt a reduced speed of light of . c , which has beenshown in AppendixD of Rosdahl et al. (2015) to result in wellconverged stellar masses, morphologies, outflow rates andISM properties. The emission spectra are discretised intofive radiation bins as in Rosdahl et al. (2015), i.e. infraredand optical radiation, which couple to gas solely throughradiation pressure on dust, and three UV radiation groupswith lower energy limits corresponding to the ionising po-tentials of H + , He + and He ++ (as in Rosdahl et al. 2015).UV radiation couples to gas through radiation pressure ondust, ionising radiation pressure, photo-heating and photo-ionisation. Radiation pressure on dust is treated both in thesingle- and multi-scattering regimes, for which we select spe-cific opacities of κ ss = ( Z / Z (cid:12) ) cm g − , where Z is thegas-phase metallicity, and κ ms = ( Z / Z (cid:12) ) cm g − , respec-tively.Our aim is to quantify the differential effect of stel-lar radiation on the structure of a massive, high-z galaxy.AGN-driven outflows are typically far more powerful thanthose driven by supernovae (e.g Costa et al. 2014). In ad-dition, supernovae- and AGN-driven outflows can interactnon-linearly (Costa et al. 2015; Biernacki & Teyssier 2018),which would prevent us from isolating the impact stellar ra-diative feedback cleanly. We therefore exclude radiation andmechanical feedback from AGN and black hole growth fromour simulations.Haloes and galaxies are identified using AdaptaHop (Aubert et al. 2004; Tweed et al. 2009) in the most massivesubmaxima (MSM) mode. We require a minimum of 20 par-ticles per halo. Haloes are selected from matter overdensitieshigher than ρ TH = ρ mean , where ρ mean is the mean densityof the Universe. We choose parameters N SPH = , whichgives the number of nearest-neighbours used to smooth thedensity field around each dark matter particle, N HOP = ,as the number of particle neighbours used to determine thedensity gradient around each particle when assigning it to itslocal patch, and f Poisson = , which ensures that only clumpsidentified at σ significance are retained in our catalogues(see AppendixB in Aubert et al. 2004, for a description of allparameters). We follow two approaches in identifying galax-ies: (i) we assign galaxies to haloes by adding up the massesof all stellar particles found within of a given halo’svirial radius (as in Rosdahl et al. 2018) and, as an alterna-tive, (ii) we explore identifying galaxies from the stellar par-ticle field directly using AdaptaHop with ρ TH = ρ mean , N SPH = , N HOP = and f Poisson = . In Fig. 1, we show the entropy distribution of gas within acube of side length
500 kpc centred on the target galaxy at z = . The top panel, which shows results for SN , shows aprominent bubble of high entropy gas that extends out to ≈ − R vir ( ≈ −
300 kpc ), where R vir is indicated witha circle. The bubble, which is composed of hot gas heatedby supernova-driven blasts and accretion shocks, encircles MNRAS , 1–7 (2019) he hidden satellites of quasars at z =
100 kpc
SNSN+RT z = 6100 kpc
A B C R e ↵ = 235 pc
Mass-weighted entropy within a cubic volume of sidelength centred on the most massive galaxy at z = in thesimulation without stellar radiative feedback (top, left panel) andwith stellar radiation (bottom, left panel). The black circle marksthe halo’s virial radius. By suppressing star formation, stellar ra-diation leads to weaker supernova-driven outflows, explaining thelower entropies seen in the simulation with radiative feedback. Inthe right-hand panels, we show the stellar surface density in thecentral few kpc of the massive galaxy. In the simulation with-out stellar radiation, the targeted galaxy is more compact, spinsfaster and is significantly clumpier. In the simulation with radia-tive feedback, the massive galaxy is more spatially extended andhas a significantly more diffuse and smoother stellar component.The clumps highlighted with black circles are example remnantsof accreted satellites, which are more thoroughly destroyed in thesimulation with stellar radiation (see text). The red circles showexamples of satellite galaxies that exist in both simulations. the targeted galaxy as well as other massive satellites in itsvicinity.If radiative feedback is included, both the spatial scaleof the high entropy region and the typical entropy within thebubble lessen substantially, as shown in the bottom panel ofFig. 1; the importance of shocks diminishes in the presenceof stellar radiation. High entropy gas becomes rarer, becausesupernova-driven outflows become weaker in the presence ofstellar radiation. On the one hand, this is because the totalstellar mass in SN+RT is at most times lower by − , suchthat there are fewer supernova events overall (Rosdahl et al.2015; Agertz et al. 2019) and, on the other hand, becausesupernova explosions become less spatially and temporallycorrelated (e.g. Kimm et al. 2018).The stronger supernova-driven winds generated in theabsence of radiative feedback occur in response to the moreefficient gas collapse that takes place in SN , where the ab-sence of processes regulating gas accretion onto star-formingsites before supernova feedback operates (see e.g. Peters et al. 2017) results in generally denser stellar structures. Atfixed halo mass, we find moderate but systematic enhance-ments in stellar mass already at z ≈ ; the mean stellar massof all galaxies in the high-resolution volume is higher by . - − . in SN than in SN+RT across the full halo mass range.More strikingly, galaxies are typically more tightly-bound in SN than in SN+RT at these early times; we find a mean peakstellar circular velocity of v pk ≈
65 kms − in SN , compared to v pk ≈
35 kms − in SN+RT in galaxies hosted by haloes with M vir > M (cid:12) .Such differences in galactic internal structure persist atlower redshift and are pronounced also for the most massivegalaxy. At z ≈ , we find that the star formation rate in SN+RT starts exceeding that of SN , as the gas which fails toform stars in progenitor galaxies due to radiative feedbackat higher redshift undergoes star formation in the massivesystem instead. Accordingly, the stellar mass of the mas-sive galaxy at z = , estimated by adding up the massesof all stellar particles within of the host’s virial radius,is ≈ . × M (cid:12) in SN and ≈ . × M (cid:12) in SN+RT . Thedifference in the target galaxy’s stellar mass is more sig-nificant at higher redshift, e.g. ≈ . × M (cid:12) in SN and ≈ . × M (cid:12) in SN+RT at z = . . At no point in thesimulation, however, does the difference in the host stel-lar mass exceed between the two simulations. Accord-ingly, the stellar-to-halo mass ratio for the targeted galaxyis M (cid:63) / M vir ≈ . and M (cid:63) / M vir ≈ . in SN and SN+RT , re-spectively. Both values are in reasonable agreement with theabundance matching expectations of Moster et al. (2018)who predict M (cid:63) / M vir ≈ . at z = , somewhat higherthan those of Behroozi et al. (2013) ( M (cid:63) / M vir ≈ . ) but invery close agreement with the recent simulations of a sim-ilarly massive galaxy at z = of Lupi et al. (2019), where M (cid:63) / M vir ≈ . .The right-hand panels of Fig. 1 display the stellar sur-face density fields around the targeted system. With a stellarhalf-mass radius of R eff =
235 pc , the stellar component ismore concentrated in SN than in SN+RT , where R eff =
650 pc ,a size difference of a factor (cid:46) . In addition, the mean stel-lar rotational velocity around the galaxy’s angular momen-tum vector, averaged over the redshift range < z < . and evaluated within R eff equals v rot ≈
960 km s − in SN and v rot ≈
720 km s − in SN+RT . That other systems are also moretightly-bound in SN can be seen directly from the stellar sur-face density maps shown in Fig. 1, where the central surfacedensities of the satellites are typically higher in SN than in SN+RT . There is also a clear excess of compact “clumps” inthe simulation without radiative feedback, which makes thegalaxy of SN appear to be richer in structure than in SN+RT ,where the stellar component looks decidedly smoother. Stel-lar radiation results in less tightly-bound, somewhat lessmassive galaxies (in agreement with Rosdahl et al. 2015;Kimm et al. 2018; Hopkins et al. 2018) as well as in morespatially extended, “puffed-up” systems with less structure.The clumps appearing in the stellar surface density mapon the top, right-hand panel of Fig. 1 have typical massesof − M (cid:12) ; for instance clumps A, B and C marked inFig. 1 have approximate masses of × M (cid:12) , × M (cid:12) and × M (cid:12) , respectively. While the additional stellar systems,including clumps A, B and C may, at first glance, resemblethe stellar clumps that form in-situ within simulated high- MNRAS000
720 km s − in SN+RT . That other systems are also moretightly-bound in SN can be seen directly from the stellar sur-face density maps shown in Fig. 1, where the central surfacedensities of the satellites are typically higher in SN than in SN+RT . There is also a clear excess of compact “clumps” inthe simulation without radiative feedback, which makes thegalaxy of SN appear to be richer in structure than in SN+RT ,where the stellar component looks decidedly smoother. Stel-lar radiation results in less tightly-bound, somewhat lessmassive galaxies (in agreement with Rosdahl et al. 2015;Kimm et al. 2018; Hopkins et al. 2018) as well as in morespatially extended, “puffed-up” systems with less structure.The clumps appearing in the stellar surface density mapon the top, right-hand panel of Fig. 1 have typical massesof − M (cid:12) ; for instance clumps A, B and C marked inFig. 1 have approximate masses of × M (cid:12) , × M (cid:12) and × M (cid:12) , respectively. While the additional stellar systems,including clumps A, B and C may, at first glance, resemblethe stellar clumps that form in-situ within simulated high- MNRAS000 , 1–7 (2019)
Costa, Rosdahl & Kimm N g a l z < 6 SN+RT (< R vir ) SN (< R vir ) SN+RT ( R vir < R <2 R vir ) SN ( R vir < R <2 R vir ) log ( M /M ) N g a l ( > M ) SN+RT (< R vir ) SN (< R vir ) SN+RT ( R vir /2< R < R vir ) SN ( R vir /2< R < R vir ) Figure 2.
Top: Average number of satellites per logarithmic stel-lar mass within R vir (filled histograms) and in the radius range R vir < R < R vir (open and hatched histograms), as obtained byaveraging over all simulations snapshots between z = . and z = . Violet histograms show the satellite number count in thesimulation without stellar radiative feedback, while the orangeand hatched histograms give the result for the simulation follow-ing stellar radiation self-consistently. Bottom: Mean cumulativenumber of galaxies above any given stellar mass at z = within R vir (solid curves) and within R vir / < R < R vir (dashed curves).Stellar radiation results in a significantly smaller satellite galaxypopulation within the virial radius. The mean number of systemswith M (cid:63) > ( ) M (cid:12) is ± ( ± ) in SN and ± ( ± )in SN+RT ; the number of galaxies within the halo drops by up to with the addition of stellar radiation. Error bars and shadedregions denote σ intervals. redshift galaxies (e.g. Mandelker et al. 2017), we find, bytracing their constituent stellar particles back in time, thatmany of them form at z ≈ − in separate dark matter haloes −
100 kpc away from the main progenitor. For instance,clumps A, B and C in Fig. 1 form half of their stellar mass at z ≈ within dark matter haloes with M vir ∼ M (cid:12) . Thereare no evident counterparts for these clumps for SN+RT , eventhough this simulation also contains some (but markedlyfewer) clumps and some satellites that also exist in SN (redcircles).We are thus led to investigate why there may be hiddensatellites in SN+RT . We first investigate the satellite popu- lation, identified through the method (i) described in thelast paragraph of Section 2, at z = in both simulations.Remarkably, we find 6 satellites with M (cid:63) > M (cid:12) within R vir in SN , compared to in SN+RT . The discrepancy be-tween the number counts of massive satellites is significantlyweaker if we instead select systems outside R vir ; there are 12galaxies with M (cid:63) > M (cid:12) between R vir and × R vir in SN and 11 in SN+RT . If we select systems from within the wholehigh-resolution region, excluding those within R vir , the dis-crepancy in number counts is (cid:46) . The difference in thenumber count of massive systems is amplified close to thequasar host galaxy.Accreted satellites, however, often lose their dark mat-ter haloes due to strong tidal interactions with the massivecentral, such that method (i) does not pick out many of theclumps seen in in Fig. 1. In the top panel of Fig. 2, we plotthe number of galaxy satellites, as identified directly using AdaptaHop , i.e. the method (ii) described in Section 2,as a function of stellar mass. The values shown in the toppanel of Fig. 2 correspond to the mean number, as obtainedby averaging over all the snapshots in the redshift range . < z < . Excluding stellar radiation leads to a significantexcess in the number of galaxies within R vir in almost everystellar mass bin, but in particular for M (cid:63) < M (cid:12) . The de-crease in galaxy number counts is also clear in the bottompanel of Fig. 2, where cumulative galaxy counts are shownfor z = within two different spatial scales. The excessis of ≈ massive clumps with M (cid:63) > M (cid:12) or ≈ with M (cid:63) > M (cid:12) , within the virial radius. Both for M (cid:63) > M (cid:12) or M (cid:63) > M (cid:12) , the inclusion of stellar radiation thus leadsto a ≈ − reduction in the number of satellites andstellar clumps within the virial radius. Notably, the distribu-tions of galaxy counts as a function of stellar mass are similarbetween SN and SN+RT (open and hatched histograms in thetop panel and dashed curves in the bottom panel of Fig. 2)in the outskirts of the halo and beyond the virial radius.In order to gain insight into why there is a loss of struc-ture when we include stellar radiation into our simulations,we select three dark matter haloes with M vir ∼ M (cid:12) at z = , that are known to end up within the most massivehalo at z = , and extract all the stellar particles residingwithin R vir . By matching the IDs of dark matter parti-cles in SN and SN+RT , we ensure we identify the same halo inboth simulations. Fig. 3 shows the spatial distribution of theselected stellar particles at three different redshifts (shownin different columns) in SN (top row) and in SN+RT (bottomrow). While the spatial configuration of the stellar compo-nent is similar between both simulations at z (cid:38) , theirevolution differs significantly at lower redshift. Prominenttails of stripped material are visible both in SN and SN+RT ,indicating that the selected systems are strongly tidally dis-rupted as they approach the central galaxy. Crucially, vari-ous tightly-bound stellar cores survive intact in SN , while theselected galaxies dissipate almost entirely in SN+RT , form-ing a diffuse stellar envelope around the central galaxy. Theclearest example is the galaxy shown in green in Fig. 3. Asit passes through the central galaxy, this system survivesrelatively undisturbed in SN (clump E), but is completelyshattered in SN+RT . Similarly, the galaxy shown in grey iscompletely shredded in
SN+RT , while it evolves into a sys-tem of multiple tightly-bound cores in SN (clumps B, C alsoshown in Fig. 1, and D). Finally, even if the system shown in MNRAS , 1–7 (2019) he hidden satellites of quasars at z = SN z = 7 z = 6.5 z = 6 SN+RT
AB C
15 km s -1
15 km s -1
47 km s -1
11 km s -1 -1
20 kpc20 kpc
39 km s -1 -1
45 km s -1 -1
106 km s -1
29 km s -1
53 km s -1 D E
103 km s -1
45 km s -1
25 km s -1
96 km s -1
22 km s -1
38 km s -1 Figure 3.
We select three systems with M vir ∼ M (cid:12) at z = , matching them between the simulations without- and with radiativefeedback. We track the stellar particles within R vir forward in time and show the resulting spatial distribution at z = (first column), z = . (second column) and z = (third column) in SN (top row) and SN+RT (bottom row). We provide the peak stellar circular velocitynext to each system, taking the location of the system’s centre as the position of the most tightly-bound stellar particle. The galaxiesundergo significant tidal interactions with the massive central, exhibiting prominent tidal tails. Crucially, the less tightly-bound galaxiesin
SN+RT are more easily tidally disrupted, often dispersing entirely. In SN , all selected galaxies survive in the form of compact cores,including clumps A, B and C identified in Fig. 1. In the rightmost panels, we show the spatial distribution of all tracked stellar particlesat z = . Many of the stellar clumps orbiting around the massive galaxy are the tightly-bound cores of accreted satellites which, byvirtue of their lower binding energy, are less long-lived in the presence of stellar radiation. orange survives as a core in both simulations, this has a massof M (cid:63) ≈ × M (cid:12) in SN+RT , while it has M (cid:63) ≈ × M (cid:12) in SN . Many of the “stellar clumps” seen in Fig. 1 are there-fore the remnant cores of accreted satellite galaxies whichare destroyed in SN+RT , but survive in SN . It is clear that thedifference in satellite counts highlighted in Fig. 2 is causedby the more efficient tidal disruption in SN+RT . Stellar radiation reduces the stellar masses of the galaxypopulation that emerges in our simulations by a factor (cid:46) (in agreement with Rosdahl et al. 2015; Kimm et al. 2018;Hopkins et al. 2018; Kannan et al. 2018). In addition, byacting on surrounding gas immediately after the formationof stellar populations, stellar radiation counters gas collapsein galaxies residing in haloes with M vir (cid:46) M (cid:12) , such thatthey become more diffuse and less tightly-bound.At z > , systems with stellar masses of M (cid:63) ∼ - − M (cid:12) typically represent central galaxies. In our simu-lations, they instead consist of the progenitors of a massive galaxy capable of harbouring a quasar, with many such sys-tems orbiting within the massive halo as satellites. The deepgravitational potential well of the massive galaxy, with highpeak stellar circular velocities of v pk >
700 km s − rapidlydisrupts all but the most tightly-bound of the accreted sys-tems. We have shown that this process is more efficient ifearly feedback operates.Our findings have potentially far-reaching implicationsfor future observational missions aiming at probing the prop-erties and number counts of satellites within close proximityof high-redshift massive galaxies and quasars. These may befewer in number than na¨ıvely expected, a result which couldbe misinterpreted as an indication that the mass of the darkmatter haloes hosting z > quasars must be lower than typ-ically assumed. The preprocessing of low-mass galaxies athigh-redshift by stellar radiation may also reduce the num-ber of satellites around low-redshift galaxies, though we ex-pect the shredding of these satellites to become less efficientthan seen for the massive z = haloes presented here, dueto the weaker tidal fields. It will be important to test themechanism outlined in this paper in lower mass haloes andat lower redshift. MNRAS000
700 km s − rapidlydisrupts all but the most tightly-bound of the accreted sys-tems. We have shown that this process is more efficient ifearly feedback operates.Our findings have potentially far-reaching implicationsfor future observational missions aiming at probing the prop-erties and number counts of satellites within close proximityof high-redshift massive galaxies and quasars. These may befewer in number than na¨ıvely expected, a result which couldbe misinterpreted as an indication that the mass of the darkmatter haloes hosting z > quasars must be lower than typ-ically assumed. The preprocessing of low-mass galaxies athigh-redshift by stellar radiation may also reduce the num-ber of satellites around low-redshift galaxies, though we ex-pect the shredding of these satellites to become less efficientthan seen for the massive z = haloes presented here, dueto the weaker tidal fields. It will be important to test themechanism outlined in this paper in lower mass haloes andat lower redshift. MNRAS000 , 1–7 (2019)
Costa, Rosdahl & Kimm
Due to efficient tidal stripping, a significant numberof the neighbouring systems of z > massive galaxies andquasars should also appear irregular and, in the likely casethat only the tightly-bound cores are detected, unusuallycompact. Based on our results, we also expect quasar hostgalaxies to be surrounded by a rich web of stellar streamscomposed of the tidally stripped stellar components of ac-creted dwarf galaxies and to contain a diffuse stellar compo-nent out to radii of ≈
10 kpc , just under (cid:48)(cid:48) .The different satellite destruction timescales also carryimplications for the morphology of the massive galaxy. Earlyradiative feedback leads to satellites which are destroyed ona timescale shorter than the dynamical friction timescaleand therefore add to the diffuse stellar halo that surroundsthe massive galaxy already at z = . The more tightly-bound satellites that would form otherwise are more likelyto sink into the centre of the galaxy by dynamical friction,contributing to the growth of the central bulge instead.Future work should therefore explore the evolution ofmassive galaxies such as that explored here down to z = .According to the few cosmological, radiation-hydrodynamicsimulations performed down to z = , stellar radiation op-erates efficiently in dwarfs with M vir (cid:46) M (cid:12) during re-ionisation (Agertz et al. 2019; Katz et al. 2019), but in-troduces only modest corrections to bulk quantities such asstellar mass in more massive systems with M vir ∼ M (cid:12) aswell as moderately flatter stellar density profiles (e.g. Hop-kins et al. 2018) by z = . Rosdahl et al. (2015), Kimm et al.(2018) and Hopkins et al. (2018) highlight how stellar radi-ation, however, results in a smaller number of tightly-boundstar clusters, such that the clearest signature of stellar ra-diation may be in the morphology of the galaxy and theproperties of its star forming regions. If we extrapolate thesefindings to the galaxy studied in this paper, we may expectstellar radiation to contribute negligibly to the final stellarmass of the galaxy, which is more likely set by AGN feed-back, but to significantly reduce the central stellar densityand the number of star clusters and satellites in its vicinity.While we have accounted for stellar radiation in our sim-ulations, we have neglected the radiation field of the quasaritself, which may have a strong impact on the ability of gasto accrete onto dwarf galaxies, leading to potentially lowermass and less tightly-bound progenitors. In previous simu-lations (see Costa et al. 2018) we followed quasar radiationin the same halo at z < . and found no difference inthe satellite demographics. However, if quasar radiation is‘switched-on’ at an earlier time, it remains possible for theprocess outlined here to be amplified.In addition, since stellar radiation regulates star forma-tion to some extent, we may na¨ıvely expect that it could, inturn, suppress black hole accretion, particularly in the lowermass progenitors at very high-z. However, since stellar radi-ation (i) stabilises larger masses of gas against collapse and(ii) results in weaker large-scale outflows, we find that themassive galaxy at z = , as well as its progenitors contain larger central gas reservoirs in SN+RT out to z (cid:38) (see Ta-ble 1), a time before most black hole growth is likely to occur(see Costa et al. 2014). This counter-intuitive results echoesone of the findings of Costa et al. (2014), who showed thatmore highly suppressed star formation due to stronger stel-lar feedback leads, counter-intuitively, to a higher black holemass. The higher abundance of cold gas within the central Simulation z = z = z = SN . × M (cid:12) . × M (cid:12) . × M (cid:12) SN+RT . × M (cid:12) . × M (cid:12) . × M (cid:12) Table 1.
The total gas mass in the central kpc of the most mas-sive galaxy at different redshifts in SN (top) and SN+RT (bottom).There is a systematic excess of gas in the simulation with stellarradiative feedback, where both star formation and outflows aresuppressed simultaneously. We speculate that the excess gas inthe galactic nucleus can fuel more intense black hole growth. regions of the targeted, massive galaxy and its progenitors,in fact, suggests that stellar radiation may aid in growingsupermassive black holes by z = , an exciting possibilitythat deserves to be explored in a future study.Besides investigating the impact of quasar radiation onthe satellite population, it will be important to simulate alarger number of massive z = haloes to quantify the impactof cosmic variance on our findings. Increasing resolution will,in turn, allow for yet more detailed studies of the internalstructure and kinematics of satellite galaxies. Using cosmological, radiation-hydrodynamic simulationstargeting a rare M vir ≈ × M (cid:12) halo at z = , wehave shown that stellar radiative feedback results in lesstightly-bound central and satellite galaxies. As opposed tosupernovae, which take ∼
10 Myr to erupt, stellar radiationcounters gas accretion immediately after stars form. Stel-lar radiative feedback is gentle, but has surprising effects: itresults in weaker supernova feedback and in less structure.While it does not strongly alter the demographics of galaxiesresiding far from the massive targeted galaxy, stellar radia-tive feedback is ultimately responsible for a smaller satellitepopulation within the latter’s virial radius. Before they sinkinto the galactic nucleus on a dynamical friction timescale,the less dense satellites are efficiently tidally disrupted andincorporated into the diffuse stellar halo that envelopes thequasar host galaxy. We suggest that this anomaly is a uniqueimprint of the extreme environments of high-redshift galax-ies and bright quasars at z = and should be sensitive to thehost halo mass. Its magnitude should, in turn, be a measureof the importance of early feedback in galaxy formation.Crucially, our findings suggest that future missions, per-formed with instruments such as JWST, may unveil a sur-prisingly low number of galactic satellites with M (cid:63) > M (cid:12) around z > quasars, which could be erroneously attributedto a lower parent halo mass. ACKNOWLEDGEMENTS
The authors thank the anonymous referee for a useful andconstructive report. TC is grateful to Benny Trakhtenbrotfor thoroughly reading the manuscript and for providingmany insightful comments that greatly improved its clar-ity. TC further acknowledges Martin Haehnelt, ThorstenNaab, R¨udiger Pakmor, Debora Sijacki, Volker Springel andFreeke van de Voort for helpful comments and discussions.
MNRAS , 1–7 (2019) he hidden satellites of quasars at z = JR acknowledges support from the ORAGE project fromthe Agence Nationale de la Recherche under grant ANR-14-CE33-0016-03. TK was supported in part by the NationalResearch Foundation of Korea (No. 2017R1A5A1070354and No. 2018036146) and in part by the Yonsei Uni-versity Future-leading Research Initiative (RMS2-2018-22-0183). This work was partially carried out on the Dutchnational e-infrastructure with the support of SURF cooper-ative. We acknowledge that the results of this research havebeen achieved using the DECI resource Eagle, at PSNC inPoland with support from the PRACE aisbl.
REFERENCES
Agertz O., et al., 2019, MNRAS (submitted), p. arXiv:1904.02723Aubert D., Pichon C., Colombi S., 2004, MNRAS, 352, 376Ba˜nados E., et al., 2018, Nature, 553, 473Balmaverde B., et al., 2017, A&A, 606, A23Begelman M. C., Volonteri M., Rees M. J., 2006, MNRAS, 370,289Behroozi P. S., Wechsler R. H., Conroy C., 2013, ApJ, 770, 57Biernacki P., Teyssier R., 2018, MNRAS, 475, 5688Champagne J. B., et al., 2018, ApJ, 867, 153Costa T., Sijacki D., Trenti M., Haehnelt M. G., 2014, MNRAS,439, 2146Costa T., Sijacki D., Haehnelt M. G., 2015, MNRAS, 448, L30Costa T., Rosdahl J., Sijacki D., Haehnelt M. G., 2018, MNRAS,479, 2079Decarli R., et al., 2017, Nature, 545, 457Di Matteo T., Khandai N., DeGraf C., Feng Y., Croft R. A. C.,Lopez J., Springel V., 2012, ApJ, 745, L29Efstathiou G., Rees M. J., 1988, MNRAS, 230, 5pFan X., et al., 2001, AJ, 122, 2833Faucher-Gigu`ere C.-A., Lidz A., Zaldarriaga M., Hernquist L.,2009, ApJ, 703, 1416Habouzit M., Volonteri M., Somerville R. S., Dubois Y., PeiraniS., Pichon C., Devriendt J., 2019, MNRAS, p. 2120Hopkins P. F., Grudic M. Y., Wetzel A. R., Keres D., Gaucher-Giguere C.-A., Ma X., Murray N., Butcher N., 2018, MNRAS(submitted),Kannan R., Marinacci F., Simpson C. M., Glover S. C. O., Hern-quist L., 2018, MNRAS (submitted),Katz H., et al., 2019, MNRAS (submitted), p. arXiv:1905.11414Kim S., et al., 2009, ApJ, 695, 809Kimm T., Cen R., Devriendt J., Dubois Y., Slyz A., 2015, MN-RAS, 451, 2900Kimm T., Katz H., Haehnelt M., Rosdahl J., Devriendt J., SlyzA., 2017, MNRAS, 466, 4826Kimm T., Haehnelt M., Blaizot J., Katz H., Michel-Dansac L.,Garel T., Rosdahl J., Teyssier R., 2018, MNRAS, 475, 4617Lupi A., Volonteri M., Decarli R., Bovino S., Silk J., Bergeron J.,2019, MNRAS, 488, 4004Mandelker N., Dekel A., Ceverino D., DeGraf C., Guo Y., PrimackJ., 2017, MNRAS, 464, 635Moster B. P., Naab T., White S. D. M., 2018, MNRAS, 477, 1822Peters T., et al., 2017, MNRAS, 466, 3293Rosdahl J., Teyssier R., 2015, MNRAS, 449, 4380Rosdahl J., Blaizot J., Aubert D., Stranex T., Teyssier R., 2013,MNRAS, 436, 2188Rosdahl J., Schaye J., Teyssier R., Agertz O., 2015, MNRAS, 451,34Rosdahl J., et al., 2018, MNRAS, 479, 994Sijacki D., Springel V., Haehnelt M. G., 2009, MNRAS, 400, 100Teyssier R., 2002, A&A, 385, 337Trakhtenbrot B., Lira P., Netzer H., Cicone C., Maiolino R.,Shemmer O., 2017, ApJ, 836, 8 Tweed D., Devriendt J., Blaizot J., Colombi S., Slyz A., 2009,A&A, 506, 647Volonteri M., Rees M. J., 2006, ApJ, 650, 669Wu X.-B., et al., 2015, Nature, 518, 512This paper has been typeset from a TEX/L A TEX file prepared bythe author.MNRAS000