Identifying substructure associations in the Milky Way halo using chemo-kinematic tagging

TL;DR

This study advances Milky Way halo archaeology by applying chemo-kinematic tagging to a large, multi-source catalog of halo substructures, using 6D phase-space data plus metallicity and a suite of conserved orbital parameters. By applying t-SNE across multiple scales, the authors identify nine main groups, 16 subgroups, and 20 stream-progenitor associations, recovering major mergers such as Gaia-Sausage-Enceladus, Thamnos, Sequoia, LMS-1/Wukong, and Sagittarius, while unveiling internal structure within GSE and refining progenitor memberships for several globular clusters and streams. The work yields a 44% accreted globular cluster fraction, links Omega Centauri to Thamnos, and confirms a notable Orphan-Chenab–Grus II association as sharing a common progenitor. These findings refine the Milky Way’s accretion history, constrain the properties of past mergers, and motivate targeted high-precision abundance studies to resolve remaining ambiguities in progenitor classifications.

Abstract

The Milky Way halo has been built-up over cosmic time through the accretion and dissolution of star clusters and dwarf galaxies as well as through their complex interactions with the Galactic disc. Traces of these accreted structures persist to the present day in the chemical and kinematic properties of stars and their orbits and allow for the disentangling of the accretion history of the Galaxy through observations of Milky Way stars. We utilised 6D phase-space information in combination with [Fe/H] measurements to facilitate a clustering analysis of stars using their kinematics and chemistry simultaneously, a technique known as chemo-kinematic tagging. Using t-distributed stochastic neighbour embedding (t-SNE), we performed dimensionality reduction and identify stars from clusters and streams that are co-localised in the kinematic and chemical parameter space. We included E, Jr, Jz, Lz, r_apo, r_peri, and eccentricity as well as [Fe/H] as input into the algorithm, and used a sample of 5347 stars from 229 individual Milky Way substructures compiled from various sources in the literature. Most notably, we recovered several large-scale structures that have been reported in the literature, including GSE, Thamnos, Sequoia, I'itoi, LMS-1/Wukong, Sagittarius, Kraken/Koala, the splashed disc, and a candidate structure recently found in another work. We find that 44\% of Milky Way globular clusters are consistent with having an accreted origin. We also find that the chemo-dynamic properties of omega cen are consistent with a common accretion with the Thamnos structure. In addition, we identified many stream-progenitor associations, most notably a connection between the Orphan-Chenab stream and the Grus II ultra-faint dwarf galaxy, which supports previous findings that these two objects were brought into the Galaxy in the same accretion event.

Figures (7)

• Figure 1: Top: t-SNE latent space at a perplexity of p = 50. Stream stars from STREAMFINDER are shown as filled coloured points, and stream stars from S5 as open coloured points, GCs are open coloured squares, satellite galaxies are open coloured circles, halo substructures are filled coloured circles, and extra stream stars from the literature are star symbols, coloured according to the legend. Bottom: t-SNE latent space coloured by the selected groups, with the left panel showing the selection of the main groups, the middle panel showing the selection of the subgroups, and the right panel showing the selection of the stream-cluster progenitors. Each group, subgroup and stream-progenitor pair is labelled by its group number which corresponds to the groups listed in Tables \ref{['tab:small_associations']}, \ref{['tab:selected_stream_progenitor']}, and \ref{['tab:large_associations']}.
• Figure 2: Validation plots for the nine identified groups. Top: Kinematic parameter spaces of E-L$_z$ (left panel) and J$_r$-J$_z$ (right panel). Bottom: Differential [Fe/H] of each structure vs the mean [Fe/H] for all structures in the group (see Section \ref{['subsec:results_validation']} for a more detailed explanation.) The grey dashed line shows the mean value for each group, and the faint dashed lines show a 0.2 dex dispersion to guide the eye. A horizontal offset of 0.01 dex has been applied between streams and clusters within the groups in order to better visualise the distributions. Structures are coloured consistently with the selection shown in the bottom left panel of Figure \ref{['fig:tsne_plot']} and described in Table \ref{['tab:small_associations']}. Filled and open points are stream stars from the STREAMFINDER sample and the S5 sample, respectively. Large open square symbols are GCs and large open circle symbols are Local Group dwarf galaxies.
• Figure 3: Validation plots for the 16 identified subgroups. Structures are coloured consistently with the selection shown in the bottom-left middle panel of Figure \ref{['fig:tsne_plot']} and described in Table \ref{['tab:large_associations']}. Filled and open points are stream stars from the STREAMFINDER sample and the S5 sample, respectively. Large open square symbols are GCs, and large open circle symbols are Local Group dwarf galaxies.
• Figure 4: Top: Streams and associated progenitors plotted in Galactic coordinates. Middle: Kinematic parameter spaces E-L$_z$ and J$_r$-J$_z$ are shown in the left and right panels, respectively. Bottom: Differential [Fe/H] of each structure vs the mean [Fe/H] for all streams and progenitors in the group. Filled and open points are stream stars from the STREAMFINDER sample and the S5 sample, respectively. Large open square symbols are for GCs, and large open circle symbols are for Local Group dwarf galaxies.
• Figure 5: t-SNE latent space projections for different values of perplexity showing the different scales of clustering used in the selection of groups. Markers are the same as described in Figure 1.