Bloodhound Unleashed: Particle-based Substructure Tracking for Cosmological Simulations

Abstract

Modern studies of galaxy formation rely heavily on numerical simulations, which in turn require tools to identify and track self-bound structures in stars and dark matter. In this paper, we present Bloodhound, a new halo tracking algorithm optimized to track and characterize substructure in cosmological simulations, a regime that is crucial for studies of the nature of dark matter but where standard methods often have difficulties. Using simulations of Milky Way-mass haloes, we demonstrate that Bloodhound extends subhalo tracking by $3-4\, \mathrm{Gyr}$ on average, and significantly longer for subhaloes with small pericentres, relative to the widely used ROCKSTAR $+$ consistent-trees halo tracking pipeline. We also show that Bloodhound provides continuous tracking, mitigating an issue for the standard technique where subhaloes can be lost and then found again -- but assigned to a new merger tree -- after several snapshots. This improved tracking leads to a substantially larger number of surviving subhaloes in the inner regions of dark matter haloes, which has several implications for studies of the Milky Way's satellite galaxy system and its use for constraining properties of dark matter. For example, within the radius where current surveys are complete to ultra-faint galaxies ($D_{\rm MW} \lesssim 50$ kpc), Bloodhound finds more than twice as many subhaloes above the atomic cooling scale relative to the standard tracking method. Our results underscore the importance of robust subhalo tracking techniques in advancing our understanding of galaxy formation and cosmological models.

Figures (20)

• Figure 1: Evolution of the physical distance (black) between an example subhalo and the centre of its host halo, along with the subhalo's ${V}_\mathrm{max}$ (purple), as a function of lookback time, as traced by consistent-trees. The two panels show the progression for the same subhalo with ${V}_\mathrm{peak} \simeq 35km\,s^{-1}$ infalling at ${t}_\mathrm{lookback} = 10.9Gyr$ (first vertical dashed line) in the DMO (left) and Disc (right) runs. The black dotted line gives ${R}_\mathrm{vir}$($t$) of the host halo. In the DMO run, the subhalo completes one pericentric passage and its ${V}_\mathrm{max}$ value is reduced to $\simeq 25km\,s^{-1}$ before disrupting completely at ${t}_\mathrm{lookback} = 9.7Gyr$ (second vertical dashed line). In the Disc run, the same subhalo completes two pericentric passages before disrupting at ${t}_\mathrm{lookback} = 8.6Gyr$ with ${V}_\mathrm{max} \simeq 22km\,s^{-1}$. In both cases, it is puzzling to have such a massive subhalo (as measured at $t_{\rm disrupt}$) disrupt after at most 2 fairly mild pericentric passages (${D}_\mathrm{peri} \simeq 20kpc$). The evolution of this subhalo's particles is explored in more detail in Fig. \ref{['fig:orbit with particles']}.
• Figure 2: Illustration of the orbit and particles of the example subhalo shown in Fig. \ref{['fig:distance and vmax over time ct']} for the DMO (left) and Disc (right) runs in the frame of the host halo. For $40$ snapshots starting from ${z}_\mathrm{infall}=2.2$ (${t}_\mathrm{lookback}=10.9Gyr$, marked as location 1 here and in Fig. \ref{['fig:distance and vmax over time ct']}) to $z=0.8$ (${t}_\mathrm{lookback}=7Gyr$, marked as location 3), particles belonging to the subhalo at infall and residing within $8kpc$ of its centre at each time are plotted for a slice of depth 3 kpc. Particles shown in cyan are the $2$ per cent most bound at infall. Curved black arrows show the subhalo's trajectory while the red arrow points to the subhalo at the time of disruption according to consistent-trees, ${z}_\mathrm{disrupt,CT}$, highlighted with a red circle (location 2 in the left panel here and Fig. \ref{['fig:distance and vmax over time ct']}). For snapshots after ${z}_\mathrm{disrupt, CT}$, particles are shown in brown. The dashed arrow traces the direction of the orbit immediately after $z=0.8$. Particles after this point are not shown except for those at $z=0$, highlighted with a grey square. In the left panel, the $z=0$ particles have been offset to an empty region of the figure for visibility. In both runs, the subhalo shows no signs of disruption at ${z}_\mathrm{disrupt, CT}$ or for many subsequent snapshots. In fact, the $z=0$ particle distributions show that there still is a tightly bound core with a significant number of bound particles remaining and that the subhalo survives in both simulations.
• Figure 3: Illustration of "broken-link" tree subhaloes found in the standard method. Left: Evolution of a subhalo identified by consistent-trees to form deep within the host halo at ${t}_\mathrm{lookback}=1.7Gyr$, counter to general expectation. A plausible explanation is the standard tracking method's inability to follow the object at earlier epochs. Right: Median cumulative distributions of ${V}_\mathrm{infall}$ for all surviving subhaloes identified by consistent-trees across all of the DMO (black) and Disc (blue) runs are plotted as solid lines. Dashed lines depict the ${V}_\mathrm{infall}$ distribution of subhaloes with merger tree starting inside ${R}_\mathrm{vir}$ of the host halo, constituting up to 15 per cent of the total population of subhaloes identified by consistent-trees at $z=0$.
• Figure 4: Illustration of Bloodhound's halo disruption criteria for a subhalo infalling at $t_{\rm lookback} = 11.5Gyr$ with ${V}_\mathrm{infall} \simeq 17.5km\,s^{-1}$. Top panel: The black line shows the physical separation between the centre computed for tracked subhalo particles and the host halo, as a function of the lookback time. The purple line traces ${R}_\mathrm{max}$ normalised by its value at infall, over the same time range. Bottom panel: Evolution of the physical concentration parameter $c_V$ is shown in black. Over-plotted in purple is the ratio ${V}_\mathrm{max}\, / {V}_\mathrm{infall}$. Two vertical lines mark the disruption time, ${t}_\mathrm{disrupt}$, according to Bloodhound (brown) and consistent-trees (black).
• Figure 5: A reanalysis of the evolution of the example subhalo shown in Fig. \ref{['fig:distance and vmax over time ct']} with Bloodhound's tracking result shown as thin solid lines. In both runs, the subhalo survives to $z=0$ in Bloodhound, while consistent-trees deems it disrupted prior to $z=1$. Left: In the DMO version, the subhalo completes 8 additional pericentric passages after ${t}_\mathrm{disrupt, CT} = 9.7Gyr$ (vertical dashed line) for a total of 9 close encounters with the host, ending as a ${V}_\mathrm{max}(z=0) \approx 16km\,s^{-1}$ subhalo. Right: In the Disc version, the subhalo completes 6 additional pericentric passages after ${t}_\mathrm{disrupt, CT} = 8.6Gyr$ and ends up as a ${V}_\mathrm{max}(z=0) \approx 6km\,s^{-1}$ subhalo. This lower final ${V}_\mathrm{max}$ value reflects the tidal effects of the disc potential. Filled (open) circles in each panel show the snapshots where the subhalo is (is not) identified in the Rockstar halo catalogue. Both the DMO and Disc runs exhibit multiple prolonged stretches where the subhalo is lost from the catalogue followed by periods where it is found again. The discontinued identification over several contiguous snapshots in Rockstar leads to consistent-trees tracking failure.