1. Introduction
The Milky Way (MW) formed over billions of years through a series of accretion and merger events, whose imprints remain in the form of stellar streams and substructures (e.g., Majewski Reference Majewski2004; Belokurov et al. Reference Belokurov, Erkal and Evans2018; Helmi et al. Reference Helmi, Babusiaux and Koppelman2018; Koppelman et al. Reference Koppelman, Helmi and Massari2019; Ibata et al. Reference Ibata, Malhan and Martin2019; Naidu et al. Reference Naidu, Conroy and Bonaca2020; Malhan et al. Reference Malhan, Ibata and Sharma2022). These streams, originating from the tidal disruption of dwarf galaxies and globular clusters or from perturbations of the disk, preserve valuable orbital and chemical information. Since the orbital properties of stellar streams and substructures are effective tracers of a galaxy’s formation history and gravitational potential (Law & Majewski Reference Law and Majewski2010), their accurate measurement is essential for understanding the origin and evolution of these features, as well as the broader assembly of the MW.
Recently, Mateu (Reference Mateu2023) published the Galstreams library, a uniform compilation of the orbital parameters for nearly hundred known stellar streams. This work also assessed uncertainties in individual stream parameters, providing guidance for future improvements. Complementary efforts based on Gaia EDR3 and ground-based spectroscopy have identified additional substructures. Using a hierarchical clustering method with integrals of motion (E,
$L_z$
,
$L_\perp$
), Lövdal et al. (Reference Lövdal, Ruiz-Lara and Koppelman2022) unveiled 6 main groups or substructures, with further population properties discussed in Ruiz-Lara et al. (Reference Ruiz-Lara, Matsuno and Lövdal2022). Together, these studies reinforce the role of stellar streams and substructures as powerful tracers of the MW’s assembly history.
Within this framework, the Gaia-Sausage-Enceladus (GSE) has been identified as one of the most prominent accreted components, originally interpreted as the remnant of a single massive merger that shaped the inner halo and thick disk (Helmi et al. Reference Helmi, Babusiaux and Koppelman2018 Belokurov et al. Reference Belokurov, Erkal and Evans2018). However, recent studies (Donlon et al. Reference Donlon, Newberg and Kim2022; Donlon & Newberg Reference Donlon and Newberg2023) reveal kinematic and chemical diversity within GSE inconsistent with a single origin, suggesting instead multiple accretion events and distinct substructures such as the Virgo Radial Merger, Cronus, Nereus, and Thamnos. These findings highlight the complex assembly of the MW and emphasize the significance of accretion in shaping globular clusters and dwarf galaxies, which preserve crucial information about the Galaxy’s formative phases and continue to influence the halo’s dynamical and chemical evolution.
2. Data and Methods
For the local halo, we adopt the compilation of Lövdal et al. (Reference Lövdal, Ruiz-Lara and Koppelman2022), which combines Gaia EDR3 RVS (Gaia Collaboration et al. 2021) with complementary radial velocities from LAMOST DR6 (Liu et al. Reference Liu, Fu and Zong2019), RAVE DR6 (Steinmetz et al. Reference Steinmetz, Matijevič and Enke2020), GALAH DR3 (Buder et al. Reference Buder, Sharma and Kos2021), and APOGEE DR16 (Ahumada et al. Reference Ahumada, Allende Prieto and Almeida2020). This results in a sample of 51,671 stars. Besides, we also use the data from the DESI Early Data Release (DESI Collaboration et al. 2016), adopting the stellar catalog of Zhang et al. (Reference Zhang, Xiang and Ting2024), which provides abundances and kinematics for 136,877 stars after selection.
For the inner halo, we use the LAMOST DR5 catalog of K giant stars (Liu et al. Reference Liu, Deng and Carlin2014), cross-matched with Gaia DR3. After applying criteria, we obtain a final sample of 8,099 K giants with kinematic information.
We apply the GS
$^3$
Hunter method to identify and analyze cluster candidates. Compared to earlier approaches, it integrates Mahalanobis and Euclidean distances for more accurate clustering, employs neural networks for efficiency on high-dimensional data, and can detect both cold and hot stellar streams. Detailed procedures are referred to Wang et al. (Reference Wang, Wang and Luo2024), Wang et al. in prep. (2025).
3. Results
3.1. Local Halo Results
After applying our method to the local halo data sets, we detect 38 clusters near the Sun, of which 21 satisfy our candidate criterion (data fraction
$>$
95%). Thirteen of these are associated with known substructures, including GSE (Helmi et al. Reference Helmi, Babusiaux and Koppelman2018; Belokurov et al. Reference Belokurov, Erkal and Evans2018), the hot thick disk (Di Matteo et al. Reference Di Matteo, Haywood and Lehnert2019; Helmi et al. Reference Helmi, Babusiaux and Koppelman2018), L-RL3 (Dodd et al. Reference Dodd, Callingham and Helmi2023), Thamnos (Koppelman et al. Reference Koppelman, Helmi and Veljanoski2018), the Helmi stream (Ruiz-Lara et al. Reference Ruiz-Lara, Matsuno and Lövdal2022), and ED-1 (Dodd et al. Reference Dodd, Callingham and Helmi2023), with cluster #11 matching cluster #38 from Ruiz-Lara et al. (Reference Ruiz-Lara, Matsuno and Lövdal2022). The remaining eight clusters are presented as new discoveries.
As shown in Figure 1, we recover clustering features in regions consistent with those reported by Lövdal et al. (Reference Lövdal, Ruiz-Lara and Koppelman2022), with different colors marking the corresponding substructures. Notably, GSE, Thamnos, and the Helmi stream appear more extended in energy-angular momentum space, likely due to methodological differences. The newly identified clusters and streams provide valuable targets for future studies of their stellar populations and chemical properties.
The left panel shows the six main groups in the E-L
$_z$
plane from Lövdal et al. (Reference Lövdal, Ruiz-Lara and Koppelman2022). The right panel presents our results, with colored labels: solid lines mark our
$3\sigma$
substructure regions, while dashed lines show those from Lövdal et al. (Reference Lövdal, Ruiz-Lara and Koppelman2022). Our regions are generally more extended, especially for the Hot Thick Disk. Differences reflect sample coverage and methodology, though several structures overlap.
Recent studies suggest that the Gaia-Sausage-Enceladus (GSE) may comprise multiple components with distinct origins (e.g., Donlon & Newberg Reference Donlon and Newberg2023). By this view, we applied GS
$^3$
Hunter to the GALAH DR3 sample (using the same selection as Donlon & Newberg Reference Donlon and Newberg2023) and identified 30 clusters. Our results show good agreement with Donlon & Newberg (Reference Donlon and Newberg2023). Given the complexity of the local halo, different GSE selection strategies inevitably capture substructure mixtures (Donlon et al. Reference Donlon, Newberg and Kim2022).
To further constrain and explore the origin of the Gaia-Sausage/Enceladus (GSE) structure, we extend the analysis with DESI EDR data, GS
$^3$
Hunter identifies 17 structures. Within the GSE region, we resolve four distinct components, labeled GSE-GSH1 through GSE-GSH4. Kernel density maps in
$\alpha$
-[Fe/H] chemical space (Figure 2) reveal multiple overdensities within each component, likely tracing separate star-forming episodes or chemically distinct ISM regions.
This figure presents the KDE distributions of the four GSE-related structures in chemical abundance space. A colorbar on the right indicates the density levels. In many of the chemical abundance planes, multiple clumps are visible, which may correspond to distinct chemical evolution pathways.
3.2. Inner Halo Results and Simulation results
In the inner halo region, we detect 45 clusters, of which 29 are candidate groups, including 4 linked to Sagittarius (Ibata et al. Reference Ibata, Irwin and Lewis2001), 4 to the Virgo Overdensity (Newberg et al. Reference Newberg, Yanny and Rockosi2002), and 2 to the Hercules-Aquila Cloud (Belokurov et al. Reference Belokurov, Evans and Bell2007), with six more reported in recent literature. The remaining 13 groups (1891 K-giants) show no clear association with previously known structures, to be explored in future study. We further tested GS
$^{3}$
Hunter on the FIRE-2 “Latte” and “ELVIS on FIRE” simulations (Wetzel et al. Reference Wetzel, Hopkins and Kim2016; Panithanpaisal et al. Reference Panithanpaisal, Sanderson and Wetzel2021), which provide rich datasets of Milky Way-like galaxies. Applying our method, we identified 33 groups, including eight true progenitors. GS
$^{3}$
Hunter successfully recovers most substructures, demonstrating its robustness.
4. Summary
We anticipate that GS
$^3$
Hunter will become a useful tool for the community dedicated to the search for stellar streams and structures in the Milky Way (MW) and the Local Group, thus helping advance our understanding of the stellar inner and outer halos and the assembly and tidal stripping history in and around the MW.
z
3σ
Open access
