Unsupervised Chemo-Dynamical Dissection of the Inner Galactic Halo: Discovery of Five Accreted Substructures with SDSS-V and Gaia astropy ( 8 ; 6 ; 7 ) , galpy ( 14 ) , python ( 65 ) , matplotlib ( 38 ) , scikit-learn ( 57 )
Furkan Akbaba, Olcay Plevne · 2026
Read Unsupervised Chemo-Dynamical Dissection of the Inner Galactic Halo: Discovery of Five Accreted Substructures with SDSS-V and Gaia astropy ( 8 ; 6 ; 7 ) , galpy ( 14 ) , python ( 65 ) , matplotlib ( 38 ) , scikit-learn ( 57 ) by Furkan Akbaba, Olcay Plevne (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-23801.txt.
Introduction
Akbaba and Plevne present a data-driven, 12-dimensional analysis of the inner Galactic halo using SDSS-V and Gaia DR3. Using an unsupervised framework based on UMAP and HDBSCAN, they run a search for clustered populations within a chemically picked ex-situ sample of 2,185 stars without pre-selection. The pipeline recovers nine groupings matching seven known substructures, including Gaia-Enceladus/Sausage, the Helmi Streams, and Sequoia. Four candidates are confirmed as robust clumps, while FO2 shows a striking nitrogen enhancement hinting at tidal debris from a disrupted massive globular cluster.
Catalog record
| CREATOR | Furkan Akbaba, Olcay Plevne |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 68887 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.23801
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Unsupervised Chemo-Dynamical Dissection of the Inner Galactic Halo: Discovery of Five Accreted Substructures with SDSS-V and Gaia astropy ( 8 ; 6 ; 7 ) , galpy ( 14 ) , python ( 65 ) , matplotlib ( 38 ) , scikit-learn ( 57 ) Furkan Akbaba, Olcay Plevne Abstract The inner Galactic halo is a complex graveyard of the Milky Way’s earliest accretion events, where severe orbital phase-mixing challenges traditional dynamical stream-finding techniques. In this work, we present a purely data-driven, 12-dimensional chemo-dynamical analysis of the inner halo using the latest data from the SDSS-V Milky Way Mapper (DR19) and Gaia DR3. Utilizing an unsupervised machine learning framework based on UMAP and HDBSCAN, we perform a blind search for clustered populations within a chemically selected ex-situ sample of 2,185 stars without imposing kinematic pre-selection criteria. Our pipeline successfully recovers nine kinematic groupings corresponding to seven previously known canonical substructures (including Gaia -Enceladus/Sausage, the Helmi Streams, and Sequoia), validating the robustness of the high-dimensional feature space. Furthermore, we report the discovery of five new tightly bound candidate substructures, designated FO1–FO5 ( E tot ≤ − 1.8 × 10 5 km 2 s − 2 E_{\rm tot}\leq-1.8\times 10^{5}~\mathrm{km^{2}~s^{-2}} ). Statistical validation confirms four of the five candidates (FO1, FO3, FO4, and FO5) as robust chemo-dynamical overdensities, while the tentative candidate FO2 exhibits a striking nitrogen enhancement ( [ N / Fe ] = + 0.83 ± 0.16 [\mathrm{N/Fe}]=+0.83\pm 0.16 ) indicative of the tidal debris from a disrupted massive globular cluster. Finally, we demonstrate that high-dimensional chemical information is critical for resolving dynamical degeneracies in the crowded inner halo, clearly differentiating structures that share similar orbits but distinct chemical signatures (e.g., FO5 and Shiva), as well as the reverse case where chemical similarity masks dynamical differences (e.g., FO3 and the Helmi Streams). These findings emphasize that the deepest regions of the Galactic potential remain far from fully mapped, preserving a rich and diverse record of the Galaxy’s assembly history. show][email protected] I Introduction The Milky Way has grown through numerous merger and accretion events over billions of years. As a result of these processes, the remnants of disrupted dwarf galaxies, globular clusters, and stellar streams can still be observed today as fossil structures within the Galactic stellar halo. Investigating the chemical and dynamical properties of these structures constitutes one of the primary goals of Galactic archaeology, which aims to reconstruct the formation history of the Galaxy. The Galactic halo provides an environment capable of preserving the signatures of past merger events more efficiently than other Galactic components due to its long dynamical timescales and relatively low stellar density. Consequently, halo stars serve as a unique laboratory for studying the chemical and dynamical imprints of past accretion events ( 35 ; 37 ; 22 ) . In recent years, the combination of the precise astrometric measurements delivered by the Gaia mission ( 26 ; 27 ) with large-scale spectroscopic surveys such as APOGEE ( 46 ) , GALAH ( 21 ) , LAMOST ( 72 ) , H3 ( 18 ) , RAVE ( 63 ) , and SEGUE ( 69 ) has revolutionized the field of Galactic archaeology. This wealth of data has enabled the discovery of numerous halo substructures, including Gaia -Enceladus/Sausage (GES; 11 ; 34 ; 33 ; 45 ), Sequoia ( 10 ; 48 ; 53 ) , Thamnos ( 43 ) , LMS-1 ( 71 ) , Arjuna and I’itoi ( 54 ) , Heracles ( 36 ) , and the recently identified Shiva and Shakti structures ( 47 ) . Detailed investigations of their chemical abundance patterns, age distributions, and dynamical properties ( 20 ; 31 ; 16 ; 37 ) have significantly improved our understanding of the Milky Way’s merger history. Despite these advances, the inner Galactic halo still remains highly complex and difficult to disentangle. Due to strong phase mixing, stars originating from ancient merger events have become spatially dispersed over time. Furthermore, remnants of different accretion events substantially overlap with the Milky Way’s in-situ populations within integrals-of-motion space ( 12 ; 59 ) . This significantly increases the risk that selection methods relying solely on dynamical parameters may group physically unrelated structures into the same population. Recent discoveries such as the Shiva and Shakti structures ( 47 ) have demonstrated that systems exhibiting similar orbital properties may nevertheless possess fundamentally different chemical origins. Consequently, modern Galactic archaeology increasingly requires data-driven approaches capable of simultaneously analysing high-dimensional chemical and dynamical information. In this context, unsupervised machine learning techniques have become increasingly common in Galactic archaeology studies ( 70 ; 44 ; 56 ; 58 ) . In particular, high-dimensional approaches combining chemical and dynamical information help reduce the limitations of classical selection methods based on only a few parameters or two-dimensional projections, enabling the identification of more complex substructures. Methods such as Principal Component Analysis (PCA), Uniform Manifold Approximation and Projection (UMAP), and HDBSCAN provide powerful tools for identifying compact and chemically coherent stellar groups in high-dimensional chemo-dynamical space. In this work, we combine SDSS-V Milky Way Mapper DR19 and Gaia DR3 data to investigate hidden accreted substructures within the inner halo of the Milky Way. After applying a PCA-based pre-separation in chemical space to isolate the ex-situ population, we search for new substructure candidates using UMAP and HDBSCAN clustering within a 12-dimensional chemo-dynamical feature space. This paper is structured as follows. In Section II , we describe the target selection criteria and the parent sample derived from the SDSS-V MWM DR19 and Gaia DR3 datasets. Section III details our unsupervised machine learning pipeline, including the chemical isolation via PCA and the subsequent clustering in the 12-dimensional chemo-dynamical parameter space using UMAP and HDBSCAN. In Section IV , we present our results, detailing the recovery of canonical literature structures and the properties of the five new candidate substructures. Section V provides a comprehensive physical interpretation of these groups, discussing their potential origins as either disrupted star clusters or remnants of ancient accretion events. Finally, we summarize our core findings and conclude in Section VI . II DATA AND SAMPLE SELECTION II.1 Spectroscopic and Astrometric Data This study relies on combining astrometric data from the Gaia Data Release 3 ( 27 , DR3;) with spectroscopic observations from the nineteenth data release (DR19) of the SDSS-V Milky Way Mapper ( 41 , MWM ;) . As a core component of the fifth-generation Sloan Digital Sky Survey ( 42 , SDSS-V :) , the MWM provides high-resolution ( R\approx 22,500 ) near-infrared spectra across both hemispheres. This full-sky coverage is achieved through twin spectrographs ( 68 ) installed on the 2.5m Apache Point Observatory telescope in New Mexico ( 30 ) and the Du Pont telescope at Las Campanas Observatory in Chile ( 15 ) . For the DR19 dataset, stellar parameters and elemental abundances for \approx 1,400,000 stars were derived using the newly developed astra framework (A. R. Casey et al., in prep). This framework implements a modified ASPCAP pipeline ( 28 ) , powered by the FEREE code ( 3 ; 4 ) alongside atomic and molecular line lists from 32 , 19 , and 62 . Comprehensive details regarding the MWM DR19 target selection and