arxiv: 2602.19647 · v1 · submitted 2026-02-23 · 🌌 astro-ph.GA · astro-ph.SR

Searching for and characterizing halo substructures with the GALAH DR4 survey

Iryna Kushniruk , Kristopher Youakim , Karin Lind , Sven Buder , Janes Kos , Diane Feuillet , Sarah L. Martell , Richard de Grijs

show 7 more authors

Geraint F. Lewis Joss Bland-Hawthorn Gary Da Costa Michael Hayden Daniel Zucker Tomaz Zwitter Sanjib Sharma

This is my paper

Pith reviewed 2026-05-15 20:47 UTC · model grok-4.3

classification 🌌 astro-ph.GA astro-ph.SR

keywords Milky Way halostellar substructuresGaia-Sausage-EnceladusThamnoschemical taggingaction spaceGALAH surveyaccretion events

0 comments

The pith

The Milky Way stellar halo contains five distinct substructures recovered via kinematic overdensities and 15-element chemical tagging.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper uses wavelet transforms on sqrt(J_r) and L_z from GALAH DR4 and Gaia to locate overdensities in action space, then applies t-SNE to tag stars chemically with 15 abundances. This process isolates the disk, the Splash, Gaia-Sausage-Enceladus with a clean high-action component, and two Thamnos groups. The groups display different alpha-element levels, iron-peak enhancements in the Splash, and stronger r-process signatures in the halo populations. These differences indicate the halo formed through multiple separate accretion events rather than a single process.

Core claim

Wavelet analysis in sqrt(J_r)-L_z space on GALAH DR4 stars recovers five structures: the Galactic disk, the Splash, Gaia-Sausage-Enceladus (GSE) whose clean population sits above sqrt(J_r) ~40 kpc km s^-1 while a lower peak reflects disk contamination, and Thamnos1 plus Thamnos2 linked to three peaks. Chemical tagging shows Thamnos2 with elevated [alpha/Fe], iron-peak elements stronger in the Splash, and halo groups retaining a clearer r-process signature. The multiply peaked structures indicate the splashed disk reaches beyond prograde orbits, and the overall chemo-dynamical distinctions support an extragalactic origin for the halo substructures.

What carries the argument

Wavelet transforms to detect overdensities in sqrt(J_r) and L_z action space, combined with t-SNE clustering on 15 elemental abundances for origin assignment and contamination assessment.

If this is right

The splashed disk extends to retrograde orbits.
GSE's cleanest sample lies at sqrt(J_r) above 40 kpc km s^-1.
Thamnos2 exhibits higher [alpha/Fe] than the other recovered groups.
Halo substructures preserve stronger r-process signatures than disk or Splash stars.
Multiple accretion events are needed to account for the observed chemical and kinematic diversity.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Larger samples from future surveys could split these groups into still finer progenitor remnants.
The same wavelet-plus-tagging approach could test whether other spiral galaxies show analogous multi-peak halo structures if comparable data become available.
Simulations of galaxy assembly could be checked for the specific multiply-peaked action distributions reported here.

Load-bearing premise

Wavelet overdensities in action space mark physically separate stellar populations rather than selection biases or mixing artifacts, and the 15-element tagging assigns origins without large misclassification from overlapping abundances.

What would settle it

High-resolution follow-up spectra of stars assigned to the high-action GSE peak showing abundance patterns indistinguishable from the lower-action disk-contaminated peak would falsify the claimed separation of distinct populations.

Figures

Figures reproduced from arXiv: 2602.19647 by Daniel Zucker, Diane Feuillet, Gary Da Costa, Geraint F. Lewis, Iryna Kushniruk, Janes Kos, Joss Bland-Hawthorn, Karin Lind, Kristopher Youakim, Michael Hayden, Richard de Grijs, Sanjib Sharma, Sarah L. Martell, Sven Buder, Tomaz Zwitter.

**Figure 1.** Figure 1: Galactocentric distance, R, as a function of distance from the Galactic plane, Z, in the direction of the North Galactic Pole, Z = 90◦ , for 124 618 stars selected from GALAH DR4. Dashed lines show the Solar values R = 8 kpc, and Z = 20.8 pc and the yellow star shows the location of the Sun. The bin size is 0.05 × 0.05 kpc. To avoid selection effects that might arise while mixing different types of stars a… view at source ↗

**Figure 2.** Figure 2: Panel (a): Density map in Lz − √ Jr space of 124 618 stars that were selected from GALAH DR4. The bin size is 11.7 × 0.2 kpc km s−1 . Panel (b): The wavelet transform map of 500 Monte-Carlo-generated samples in the Lz − √ Jr space for scale J = 5. Centres of the detected structures are shown with white crosses. The structures are the disk, Splash, GSE, and Thamnos. Panel (c): The wavelet transform map of 5… view at source ↗

**Figure 3.** Figure 3: The binned distributions of 116 047 stars selected from the GALAH DR4 in the [Mg/Cu] – [Na/Fe] plane. The bin size is 0.01 × 0.01. The dashed line shows the division of stars into the disk (below the line) and halo (above the line) stars. 2020). Similarly to [Mg/Fe], [Na/Fe] allows tracking differences between the in situ and accreted stars (e.g. Nissen & Schuster 2010). We also selected stars for which th… view at source ↗

**Figure 4.** Figure 4: A scatter plot of all 116 047 stars selected from GALAH DR4 in the [Mg/Cu] – [Na/Fe] plane is shown in gray. The markers of different colors show stars in the following kinematic structures: Splash (Panel (a), orange circles), GSE (Panel (b), blue circles), Thamnos1 (Panel (c), green circles), and Thamnos2 (Panel (d), red circles). The dashed line is the same as on panel (a) and divides the stars into the … view at source ↗

**Figure 5.** Figure 5: Panel (a): The plot shows the orbital energy as a function of the angular momentum. Gray dots show a scatter plot of the total sample, and other colors and symbols correspond to the kinematic structures detected in the legend. Chemically-defined halo stars were excluded from the Splash, and chemically-defined disk stars were excluded from the rest of the groups. Panel (b): Similar to the plot on panel (a),… view at source ↗

**Figure 6.** Figure 6: Normalized probability density metallicity [Fe/H] with uncertainty bands for kinematic groups listed in the legend. The chemicallydefined halo stars were excluded from the Splash, and chemicallydefined disk stars were excluded from the rest of the groups. The plot is generated using kernel density estimation (KDE) with a bandwidth of 0.075 and Gaussian kernel. The shaded regions around each curve repres… view at source ↗

**Figure 7.** Figure 7: Panels (a) – (d) show individual distributions of elemental abundances for the Splash (orange box), GSE (blue box), Thamnos1 (green box), and Thamnos2 (red box) are depicted using box plots. The analysis covers 32 chemical elements, X, related to iron, [X/Fe], and iron abundance, which is related to hydrogen, [Fe/H]. Each box plot adheres to standard conventions, with the first and third quartiles defining… view at source ↗

**Figure 8.** Figure 8: Distribution of stars selected from GALAH DR4 in the following elemental abundance planes: (a) [Mg/Fe] – [Fe/H], (b) [Ba/Fe] – [Fe/H], (c) [Ba/Mg] – [Fe/H]. The gray dots represent the total sample, while the colored lines show running means with shaded error regions for stars from kinematic structures as labeled in the legend. Stars in Thamnos2 are shown as dots due to the low number of stars in the group… view at source ↗

**Figure 9.** Figure 9: t-SNE latent space projection of all stars in the GALAH sample, shown in gray. Halo stars from GSE and Thamnos1 and 2 are shown in blue, green and red, respectively, while disk stars from the Splash are shown in orange. Filled points show stars identified as belonging to the halo based on the cuts applied in [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗

read the original abstract

Recent studies show that the Milky Way stellar halo is composed of populations of different origins, shaped by multiple accretion events. To better understand the formation of the Milky Way and other spiral galaxies, we characterize the chemical and kinematic properties of halo substructures using GALAH DR4 and Gaia data. We apply wavelet transforms in the space of sqrt(J_r) and azimuthal action (L_z) to identify kinematic overdensities. Stars in the detected structures are analyzed in elemental abundance space to determine their origin. We further assess contamination using the unsupervised machine-learning algorithm t-distributed stochastic neighbor embedding (t-SNE), performing chemical tagging with 15 elemental abundances. We recover five structures: the Galactic disk, the Splash, Gaia-Sausage-Enceladus (GSE), Thamnos1, and Thamnos2. GSE shows two peaks; one at sqrt(J_r) ~ 25 kpc km s^-1 is due to disk contamination, while the other above sqrt(J_r) ~ 40 kpc km s^-1 represents the cleanest GSE population. Thamnos exhibits three peaks linked to Thamnos1 and Thamnos2. Thamnos2 shows higher [alpha/Fe], iron-peak elements are enhanced in the Splash, and halo groups retain a stronger r-process signature. The multiply peaked structures suggest that the splashed disk extends beyond prograde orbits. The distinct chemo-dynamical properties of the halo groups support their extragalactic origin.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This is a solid data update on known Milky Way halo substructures using GALAH DR4, recovering the usual components with some added peak details, but the claims rest on standard methods without clear robustness checks.

read the letter

This paper applies wavelet transforms in sqrt(J_r)-L_z space to GALAH DR4 plus Gaia data and follows up with t-SNE on 15 abundances to tag stars in the detected overdensities. It recovers the disk, Splash, GSE, Thamnos1, and Thamnos2, noting two GSE peaks (one likely disk contamination) and three in Thamnos, plus abundance trends like higher alpha in Thamnos2 and stronger r-process in the halo groups. The new data release gives cleaner abundance measurements than earlier surveys, which is the real incremental value here. The multiple-peak findings suggest the splashed disk reaches beyond prograde orbits, and the chemo-dynamical separation supports extragalactic origins for the halo pieces. That part is straightforward and matches what prior work has found with other datasets. The soft spots sit in the peak identification step. The abstract does not describe forward modeling of the survey selection function or injection-recovery tests on mocks, so it is hard to rule out that the reported overdensities partly reflect magnitude, color, or SNR cuts rather than true dynamical groups. Thresholds for wavelet detection and how overlapping abundances are handled in t-SNE also need explicit checks to confirm the clean GSE population above sqrt(J_r) ~40 is not partly an artifact. These are fixable with more detail in the methods, but they matter for the strength of the multiple-peak claims. This is for galactic archaeologists who track merger debris and want the latest GALAH numbers on GSE and Thamnos properties. It will not rewrite the formation picture, but the refined maps could feed into simulations of accretion timelines. A reader already citing the original GSE or Thamnos papers would find the updated characterizations worth a look. I would send it to peer review so referees can verify the data cuts and contamination handling against the full distributions.

Referee Report

2 major / 2 minor

Summary. The paper applies wavelet transforms to sqrt(J_r) and L_z action space using GALAH DR4 and Gaia data to detect kinematic overdensities in the Milky Way halo, then uses t-SNE chemical tagging on 15 elemental abundances to characterize five recovered structures (Galactic disk, Splash, GSE with two peaks, Thamnos1, Thamnos2). It argues that distinct chemo-dynamical properties, including enhanced alpha/Fe in Thamnos2, iron-peak elements in the Splash, and r-process signatures in halo groups, support extragalactic origins for the accreted components, with multiply-peaked structures implying an extended splashed disk.

Significance. If the wavelet peaks and chemical assignments prove robust, the work adds to the growing catalog of Milky Way accretion events by providing a large-sample chemo-dynamical dissection of known structures (GSE, Thamnos) and the Splash, with potential to constrain the timing and mass of progenitor mergers in galaxy-formation simulations.

major comments (2)

[Methods / Kinematic detection] Wavelet transform analysis: the identification of distinct overdensities (e.g., GSE peak above sqrt(J_r) ~40 kpc km s^-1 and Thamnos peaks) is presented as evidence of physically distinct populations, yet no injection-recovery tests on mocks or forward-modeling of the GALAH DR4 selection function (magnitude, color, SNR cuts) are described; these cuts are non-uniform in action space and could generate artificial density contrasts when combined with Gaia astrometry.
[Chemical analysis / Results] t-SNE chemical tagging: the claim that 15-abundance clustering cleanly assigns origins and supports extragalactic provenance for GSE and Thamnos1/2 rests on separability, but the abstract notes overlapping iron-peak and r-process signatures; quantitative misclassification rates or contamination fractions from the clustering are not reported, leaving the origin inferences vulnerable to dynamical mixing or abundance overlaps.

minor comments (2)

[Results] The distinction between the two GSE peaks (one attributed to disk contamination) would benefit from an explicit table or figure panel showing the abundance distributions of each peak for direct comparison.
[Methods] Notation for action-space coordinates (sqrt(J_r), L_z) is introduced without a brief reminder of the adopted potential or action-calculation method; a short methods paragraph or reference would improve accessibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive report and the opportunity to clarify our methods. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

Referee: [Methods / Kinematic detection] Wavelet transform analysis: the identification of distinct overdensities (e.g., GSE peak above sqrt(J_r) ~40 kpc km s^-1 and Thamnos peaks) is presented as evidence of physically distinct populations, yet no injection-recovery tests on mocks or forward-modeling of the GALAH DR4 selection function (magnitude, color, SNR cuts) are described; these cuts are non-uniform in action space and could generate artificial density contrasts when combined with Gaia astrometry.

Authors: We agree that injection-recovery tests and explicit forward-modeling of the GALAH DR4 selection function would strengthen the robustness claims. The GALAH selection function is complex and not fully public in a form that allows straightforward forward modeling in action space. We validated the peaks by cross-matching with literature identifications of GSE and Thamnos and by confirming their distinct chemical signatures. In the revised manuscript we will add a dedicated subsection discussing possible selection biases and include a limited set of mock tests using a simplified selection model to quantify the risk of artificial contrasts. revision: partial
Referee: [Chemical analysis / Results] t-SNE chemical tagging: the claim that 15-abundance clustering cleanly assigns origins and supports extragalactic provenance for GSE and Thamnos1/2 rests on separability, but the abstract notes overlapping iron-peak and r-process signatures; quantitative misclassification rates or contamination fractions from the clustering are not reported, leaving the origin inferences vulnerable to dynamical mixing or abundance overlaps.

Authors: We acknowledge that quantitative misclassification rates were not provided. While the t-SNE projection and subsequent chemical tagging demonstrate clear separation for the main groups, overlaps in iron-peak and r-process elements do exist. In the revised version we will report silhouette scores for the t-SNE clusters, estimate contamination fractions by cross-validation against literature labels for GSE and disk stars, and explicitly discuss the impact of possible dynamical mixing on the origin assignments. revision: yes

Circularity Check

0 steps flagged

No significant circularity: purely empirical data analysis

full rationale

The paper applies standard wavelet transforms to identify overdensities in sqrt(J_r)-L_z space and t-SNE clustering on 15 elemental abundances from GALAH DR4 and Gaia observations. No mathematical derivation chain exists that reduces any claimed result to fitted inputs by construction, self-definitional equations, or load-bearing self-citations. Structure recovery (disk, Splash, GSE, Thamnos1/2) is presented as direct output from the data processing pipeline without renaming known results or smuggling ansatzes via prior work. The analysis remains self-contained against external benchmarks and does not invoke uniqueness theorems or predictions that collapse to the input selection.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard assumptions from galactic dynamics and stellar nucleosynthesis without new free parameters or invented entities.

axioms (2)

domain assumption Stellar actions (J_r, L_z) are approximately conserved in the Galactic potential and can be used to identify kinematic substructures
Invoked when applying wavelet transforms to detect overdensities.
domain assumption Elemental abundance patterns reflect the birth environment and can be used to tag stellar origins
Basis for the chemical tagging step with 15 elements and t-SNE.

pith-pipeline@v0.9.0 · 5630 in / 1349 out tokens · 51107 ms · 2026-05-15T20:47:05.944839+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

54 extracted references · 54 canonical work pages

[1]

2022, ApJS, 259, 35

Abdurro’uf, Accetta, K., Aerts, C., et al. 2022, ApJS, 259, 35

work page 2022
[2]

2012, MNRAS, 426, L1

Antoja, T., Helmi, A., Bienayme, O., et al. 2012, MNRAS, 426, L1

work page 2012
[3]

Bailer-Jones, C. A. L. 2023, AJ, 166, 269

work page 2023
[4]

2023, A&A, 674, A194

Bellazzini, M., Massari, D., De Angeli, F., et al. 2023, A&A, 674, A194

work page 2023
[5]

W., Koposov, S

Belokurov, V ., Erkal, D., Evans, N. W., Koposov, S. E., & Deason, A. J. 2018, MNRAS, 478, 611

work page 2018
[6]

L., Fattahi, A., et al

Belokurov, V ., Sanders, J. L., Fattahi, A., et al. 2020, MNRAS, 494, 3880

work page 2020
[7]

& Bovy, J

Bennett, M. & Bovy, J. 2019, MNRAS, 482, 1417

work page 2019
[8]

& Tepper-García, T

Bland-Hawthorn, J. & Tepper-García, T. 2021, MNRAS, 504, 3168

work page 2021
[9]

E., et al

Buder, S., Kos, J., Wang, X. E., et al. 2025, PASA, 42, e051

work page 2025
[10]

K., et al

Buder, S., Lind, K., Ness, M. K., et al. 2022, MNRAS, 510, 2407

work page 2022
[11]

2021, MNRAS, 506, 150

Buder, S., Sharma, S., Kos, J., et al. 2021, MNRAS, 506, 150

work page 2021
[12]

2019, ApJ, 883, 107

Conroy, C., Bonaca, A., Cargile, P., et al. 2019, ApJ, 883, 107

work page 2019
[13]

2023, A&A, 670, A106

Contursi, G., de Laverny, P., Recio-Blanco, A., et al. 2023, A&A, 670, A106

work page 2023
[14]

C., et al

Dalton, G., Trager, S., Abrams, D. C., et al. 2020, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, V ol. 11447, Ground- based and Airborne Instrumentation for Astronomy VIII, ed. C. J. Evans, J. J. Bryant, & K. Motohara, 1144714

work page 2020
[15]

C., et al

Dalton, G., Trager, S., Abrams, D. C., et al. 2018, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, V ol. 10702, Ground- based and Airborne Instrumentation for Astronomy VII, ed. C. J. Evans, L. Simard, & H. Takami, 107021B de Jong, R. S., Agertz, O., Berbel, A. A., et al. 2019, The Messenger, 175, 3 De Silva, G. M., Freema...

work page 2018
[16]

Deason, A. J. & Belokurov, V . 2024, New A Rev., 99, 101706

work page 2024
[17]

& Feuillet, D

Feltzing, S. & Feuillet, D. 2023, ApJ, 953, 143

work page 2023
[18]

K., Feltzing, S., Sahlholdt, C

Feuillet, D. K., Feltzing, S., Sahlholdt, C. L., & Casagrande, L. 2020, MNRAS, 497, 109

work page 2020
[19]

K., Sahlholdt, C

Feuillet, D. K., Sahlholdt, C. L., Feltzing, S., & Casagrande, L. 2021, MNRAS, 508, 1489

work page 2021
[20]

& Bland-Hawthorn, J

Freeman, K. & Bland-Hawthorn, J. 2002, ARA&A, 40, 487 Gaia Collaboration, Vallenari, A., Brown, A. G. A., et al. 2023, A&A, 674, A1

work page 2002
[21]

D., et al

Haywood, M., Di Matteo, P., Lehnert, M. D., et al. 2018, ApJ, 863, 113

work page 2018
[22]

2020, arXiv e-prints, arXiv:2002.04340

Helmi, A. 2020, arXiv e-prints, arXiv:2002.04340

work page arXiv 2020
[23]

H., et al

Helmi, A., Babusiaux, C., Koppelman, H. H., et al. 2018, Nature, 563, 85

work page 2018
[24]

Helmi, A., White, S. D. M., de Zeeuw, P. T., & Zhao, H. 1999, Nature, 402, 53

work page 1999
[25]

& Roweis, S

Hinton, G. & Roweis, S. 2002, Advances in Neural Processing Systems, 15, 833

work page 2002
[26]

A., Hasselquist, S., Shetrone, M., et al

Holtzman, J. A., Hasselquist, S., Shetrone, M., et al. 2018, AJ, 156, 125

work page 2018
[27]

P., Mackereth, J

Horta, D., Schiavon, R. P., Mackereth, J. T., et al. 2021, MNRAS, 500, 1385

work page 2021
[28]

P., Mackereth, J

Horta, D., Schiavon, R. P., Mackereth, J. T., et al. 2023, MNRAS, 520, 5671

work page 2023
[29]

P., Li, T

Ji, A. P., Li, T. S., Hansen, T. T., et al. 2020, AJ, 160, 181

work page 2020
[30]

I., & Lugaro, M

Kobayashi, C., Karakas, A. I., & Lugaro, M. 2020, ApJ, 900, 179

work page 2020
[31]

2018, ApJ, 860, L11

Koppelman, H., Helmi, A., & Veljanoski, J. 2018, ApJ, 860, L11

work page 2018
[32]

H., Helmi, A., Massari, D., Price-Whelan, A

Koppelman, H. H., Helmi, A., Massari, D., Price-Whelan, A. M., & Starkenburg, T. K. 2019, A&A, 631, L9

work page 2019
[33]

2018, MNRAS, 473, 4612

Kos, J., Bland-Hawthorn, J., Freeman, K., et al. 2018, MNRAS, 473, 4612

work page 2018
[34]

& Bensby, T

Kushniruk, I. & Bensby, T. 2019, A&A, 631, A47

work page 2019
[35]

2017, A&A, 608, A73

Kushniruk, I., Schirmer, T., & Bensby, T. 2017, A&A, 608, A73

work page 2017
[36]

Lane, J. M. M., Bovy, J., & Mackereth, J. T. 2023, MNRAS, 526, 1209

work page 2023
[37]

2019, The Journal of Open Source Software, 4, 1237

Lee, G., Gommers, R., Waselewski, F., Wohlfahrt, K., & O’Leary, A. 2019, The Journal of Open Source Software, 4, 1237

work page 2019
[38]

2016, A&A, 595, A4

Lindegren, L., Lammers, U., Bastian, U., et al. 2016, A&A, 595, A4

work page 2016
[39]

2021, A&A, 650, A110

Matsuno, T., Hirai, Y ., Tarumi, Y ., et al. 2021, A&A, 650, A110

work page 2021
[40]

2020, MNRAS, 496, 2902

Molaro, P., Cescutti, G., & Fu, X. 2020, MNRAS, 496, 2902

work page 2020
[41]

C., Vasiliev, E., Iorio, G., Evans, N

Myeong, G. C., Vasiliev, E., Iorio, G., Evans, N. W., & Belokurov, V . 2019, MNRAS, 488, 1235

work page 2019
[42]

P., Conroy, C., Bonaca, A., et al

Naidu, R. P., Conroy, C., Bonaca, A., et al. 2020, ApJ, 901, 48

work page 2020
[43]

E., Amarsi, A

Nissen, P. E., Amarsi, A. M., Skúladóttir, Á., & Schuster, W. J. 2024, A&A, 682, A116

work page 2024
[44]

Nissen, P. E. & Schuster, W. J. 2010, A&A, 511, L10

work page 2010
[45]

G., et al

Ortigoza-Urdaneta, M., Vieira, K., Fernández-Trincado, J. G., et al. 2023, A&A, 676, A140

work page 2023
[46]

2011, Journal of Machine Learning Research, 12, 2825

Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011, Journal of Machine Learning Research, 12, 2825

work page 2011
[47]

2018, A&A, 619, A72

Ramos, P., Antoja, T., & Figueras, F. 2018, A&A, 619, A72

work page 2018
[48]

L., Belokurov, V ., & Man, K

Sanders, J. L., Belokurov, V ., & Man, K. T. F. 2021, MNRAS, 506, 4321

work page 2021
[49]

D., Martell, S

Simpson, J. D., Martell, S. L., Buder, S., et al. 2021, MNRAS, 507, 43

work page 2021
[50]

& Murtagh, F

Starck, J.-L. & Murtagh, F. 2006, Astronomical Image and Data Analysis

work page 2006
[51]

2009, ARA&A, 47, 371

Tolstoy, E., Hill, V ., & Tosi, M. 2009, ARA&A, 47, 371

work page 2009
[52]

H., Coronado, J., & Rix, H.-W

Trick, W. H., Coronado, J., & Rix, H.-W. 2019, MNRAS, 484, 3291 van der Maaten, L. & Hinton, G. 2008, Journal of Machine Learning Research, 1, 1 van der Walt, S., Schönberger, J. L., Nunez-Iglesias, J., et al. 2014, PeerJ, 2, e453

work page 2019
[53]

2023, MNRAS, 524, 2630

Youakim, K., Lind, K., & Kushniruk, I. 2023, MNRAS, 524, 2630

work page 2023
[54]

2012, Research in Astronomy and Astrophysics, 12, 723 Article number, page 12 of 12

Zhao, G., Zhao, Y .-H., Chu, Y .-Q., Jing, Y .-P., & Deng, L.-C. 2012, Research in Astronomy and Astrophysics, 12, 723 Article number, page 12 of 12

work page 2012