Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Causal Discovery of Latent Variables in Galactic Archaeology

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Causal discovery recovers birth radius and guiding radius from five stellar observables alone.

desk verdict A genuinely new application of latent causal discovery to chemodynamics, but the paper labels L1 as birth radius without ever comparing L1 to the true birth radius in the simulation. read the letter →

arxiv 2507.00134 v1 pith:ZA7FTKC5 submitted 2025-06-30 astro-ph.GA

Zehao Jin , Yuxi Lu , Yuan-Sen Ting , Yujia Zheng , Tobias Buck This is my paper · ORCID
classification astro-ph.GA
keywords causaldiscoverylatentvariablesgalacticarchaeologystellarmigrationbirthradiusguidingstructuralmodelchemodynamics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper asks whether automated causal discovery can find hidden physical quantities that astronomers normally have to compute by hand. Applying the rank-based latent causal discovery algorithm to five observable properties of simulated Milky Way disk stars—metallicity, age, vertical action, angular momentum, and eccentricity—the authors recover two latent variables that they identify as birth radius and guiding radius. They check this identification against the simulation's ground truth and against an established birth-radius inference method, finding agreement. If the claim holds, unobservable stellar birth conditions can be inferred directly from data without prior physical assumptions.

What carries the argument

The central object is the Rank-based Latent Causal Discovery (RLCD) algorithm, a causal discovery method that finds hidden variables from observed covariance structure. Its key tool is a rank-deficiency signature: when latent variables drive the observed quantities, the covariance matrix of the observed variables becomes rank-deficient in a way that reveals how many latents exist and which observed variables they affect. The underlying model is a linear structural causal model, $V_i = \sum_{V_j \in \mathrm{Pa}(V_i)} a_{ij} V_j + \varepsilon_{V_i}$, with latent variables normalized to unit variance for identifiability. After discovering the graph, the paper estimates edge coefficients by maximum likelihood and then recovers individual latent values per star by minimizing prediction error, which turns five observable stellar properties into two candidate physical quantities, birth radius and guiding radius.

What would settle it

Re-run the identical pipeline on a synthetic disk whose true structural equations are known and deliberately nonlinear (for instance, birth radius enters metallicity through a quadratic term); if the algorithm still returns two latent variables whose estimated values do not track the true birth and guiding radii, the linear-rank interpretation carrying the claim is falsified.

Watch

Extended reading notes

Core claim

On a sample of low-$\alpha$ disk stars drawn from a high-resolution cosmological simulation of a Milky Way-mass galaxy, the authors run a causal discovery algorithm that assumes the observed variables follow a linear structural causal model with hidden common causes. The algorithm reports exactly two latent variables, $L_1$ and $L_2$: $L_1$ is connected to metallicity and vertical action (with age following the same pattern), and $L_2$ is connected to angular momentum and eccentricity. The authors interpret $L_1$ as the stellar birth radius and $L_2$ as the guiding radius, because birth metallicity gradients and vertical heating carry memory of birth conditions, while angular momentum and eccentricity together define a guiding radius. They validate the interpretation by showing that per-star estimates of $L_2$ nearly coincide with the true guiding radius in the simulation, and that $L_1$ reproduces the birth-radius patterns of an established inference method across the age-metallicity and $\alpha$-abundance planes.

Load-bearing premise

The analysis assumes the relationships among stellar properties and hidden birth conditions are linear; if the real chemodynamic relations are strongly curved, the recovered hidden variables may mix birth time and birth radius rather than cleanly matching them.

Editorial extensions

If this is right

  • If the central claim is correct, causal discovery can serve as a prior-free route to stellar birth conditions, estimating per-star birth radii and guiding radii from survey-like measurements alone.
  • The recovered graph independently reproduces known chemodynamic relations, suggesting the method can re-derive established physics without being told it in advance.
  • The same pipeline can be transported to real spectroscopic surveys, where selection effects and measurement noise would need to be folded in.
  • The paper's linear-model assumption marks the boundary of the claim: within linear chemodynamic relations the latent recovery is demonstrated, and outside them the physical labels would need re-testing.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, if L1 truly tracks birth radius, the same method could produce empirical birth-radius maps of the Milky Way without assuming a linear metallicity gradient, giving a direct test of how much canonical inference methods rely on that assumption.
  • Beyond the paper, the rank-deficiency signature might also separate other hidden dimensions such as birth time if additional observables sensitive to stellar age are included; the present five-variable set may not fully disentangle age from birth radius.
  • Beyond the paper, applying the algorithm to a simulation with weak radial migration would be a discriminating test: the expected latent structure should simplify or the L2-guiding-radius correspondence should weaken if the latents are physically specific.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The authors apply Rank-based Latent Causal Discovery (RLCD) to five observable properties of low-α disk stars from the NIHAO-UHD simulation g2.79e12: metallicity [Fe/H], age, vertical action Jz, angular momentum Lz, and eccentricity e. RLCD returns two latent variables, L1 and L2, with causal edges to [Fe/H], age, and Jz (L1) and to Lz and e (L2). Using polynomial transformations, the authors map L1 and L2 to physical units and compare them against the Lu et al. (2024) birth-radius estimator and the simulation's true guiding radius. They report good quantitative agreement for L2 and qualitative agreement for L1 in the age-metallicity and [O/Fe]-[Fe/H] planes, and conclude that the discovered latents correspond to the real physical properties birth radius and guiding radius. The paper is a short proof-of-concept aimed at demonstrating the potential of causal discovery in astrophysics.

Significance. The paper applies a recent causal discovery method, RLCD, to a high-resolution cosmological simulation with known ground truth, which is a promising testbed for evaluating whether latent-variable causal discovery can recover astrophysically meaningful quantities. The setup is clean and the connection to an existing birth-radius estimator is a nice touch. If the L1 validation were performed against the simulation's true birth radius and the linear-SCM assumption were tested, the result would be a valuable demonstration of automated causal discovery in galactic archaeology. As written, however, the evidence does not yet support the strong claim that the discovered latents uniquely correspond to the physical birth radius and guiding radius.

major comments (3)
  1. [Section 3 (Figure 2, left panel; Figure 3)] The validation of L1 as birth radius is not supported by the comparison shown in Figure 2: L1 is compared against the Lu et al. (2024) birth-radius estimator, which is constructed from [Fe/H] and age, two of the five observables from which L1 was discovered. Because L1 is a linear latent factor that influences [Fe/H] and age, any latent explaining covariance in these variables will correlate with the Lu et al. estimate even if it does not represent birth radius. The simulation provides true birth radii (already used in the left column of Figure 3), so the authors should report a direct comparison of L1 with the true simulation birth radii, including a scatter metric such as the Pearson or Spearman correlation and the residual scatter. The polynomial mapping from L1 to physical units is also fit post hoc; the paper must state the calibration target, and if the polynomial is fit to the Lu et al. estimates, the agreement in Figure 2 is partly by construction.
  2. [Section 2.2, Eq. (1); Section 3] The rank-deficiency signature that RLCD detects (Section 2.2, Eq. (1)) is only diagnostic under the linear SCM assumed in Eq. (1). The paper provides no evidence that a linear two-latent SCM adequately fits the joint distribution of [Fe/H], age, Jz, Lz, and e for the selected sample; known chemodynamic relations such as the age-metallicity relation and the evolution of vertical action with orbital heating are substantially nonlinear. If the true relations are nonlinear, the detected rank deficiency may not indicate two linear latent causes, and L1 and L2 could be statistical mixtures of birth time, birth radius, and other variables. The sentence in Section 3 acknowledging the linear assumption as a future limitation is not a substitute for a diagnostic, such as a comparison of the linear-model likelihood with a nonlinear alternative or a simulation experiment where the true structure is nonlinear.
  3. [Section 3 (Figure 2, right panel)] The identification of L2 as guiding radius is much weaker than presented because L2 is discovered to directly cause Lz, and the guiding radius is essentially proportional to Lz (Rg = Lz/vc for circular orbits). A tight L2–Rg correlation is therefore expected for any latent factor that influences angular momentum, and the causal edge L2→Lz does not discriminate among physical interpretations such as total angular momentum, orbital energy, or birth angular momentum. The authors should test whether L2 explains variance in Rg beyond that already contained in Lz, for example through a partial-correlation or residual analysis, or otherwise temper the interpretation of L2 as the guiding radius.
minor comments (5)
  1. [Section 2.1] The construction of the low-α subsample uses the line [O/Fe] = -0.13[Fe/H] + 0.17; please state how measurement uncertainties in [O/Fe] are propagated into the sample selection, since the quoted 0.06 dex uncertainty is non-negligible.
  2. [Section 2.2] The sentence 'fix the variance of each latent variable to unity... allows unique parameter estimation' should cite or summarize the identifiability conditions of Dong et al. (2024; 2025); without such conditions, linear latent factors are generally identifiable only up to orthogonal transformation.
  3. [Figure 2] The caption and text do not specify the polynomial degree or coefficients used to map L1 and L2 to physical units; please provide this information in the Methods or the caption so the validation is reproducible.
  4. [Abstract and Section 4] The abstract's claim that the latents 'correspond to real physical properties' is stronger than the evidence presented; consider rewording to 'are consistent with' or 'may correspond to' the birth radius and guiding radius.
  5. [Section 2.1] The selection 'located between 7-10 kpc' should specify the coordinate system used (e.g., current cylindrical Galactocentric radius) and explain why this radial selection does not bias the rank-deficiency structure that RLCD detects.

Circularity Check

2 steps flagged · score 6.0 of 10

Latent content is validated against quantities derived from the same observables: L2 is a relabeling of angular momentum, and L1 is calibrated to a [Fe/H]+age birth-radius estimator instead of the simulation's true birth radius.

  1. self definitional [Section 3 'Results', guiding-radius validation paragraph (text around Figure 2)]
    "For guiding radius (right panel), the agreement between L2 and the ground truth is notable, with minimal scatter about the one-to-one relation. This tight correlation confirms that L2 captures guiding radius—unsurprising given that guiding radius is directly encoded in the angular momentum that L2 influences."

    The discovered causal graph has L2 as a direct cause of Lz, and the paper states that Lz and eccentricity 'together define a star's guiding radius'; in practice Rg is computed from Lz. Any latent variable that is identified as influencing Lz will therefore correlate tightly with Rg. The validation is a restatement of L2's definitional relation to Lz, not an independent confirmation that L2 is the physical guiding radius.

  2. fitted input called prediction [Section 3 'Results', birth-radius validation paragraph (around Figure 2) and Methods latent-value estimation]
    "After estimating individual stellar values for L1 and L2 using the identified parameters, we map these dimensionless quantities to physical units via polynomial transformations, accounting for the centering and standardization in our analysis. For L1, we compare against birth radii inferred using the established method of Lu et al. (2024), which estimates birth radii from [Fe/H] and age based on assumptions from (Lu et al., 2022a)."

    L1 is discovered as a latent factor influencing [Fe/H] and age (Section 3: 'The latent variable L1 influences both metallicity ([Fe/H]) and vertical action (Jz), while age shows similar causal connections'). The Lu et al. (2024) estimator is itself a function of [Fe/H] and age, and the polynomial map from dimensionless L1 to physical units is fitted after the fact to match that estimator. The reported agreement therefore measures how well a fitted transformation of a latent built from [Fe/H]+age reproduces a [Fe/H]+age-based estimate; it does not test L1 against the simulation's true birth radius, which is available but used only qualitatively in Figure 3.

full rationale

The central claim that the discovered latents correspond to birth radius and guiding radius is partially circular. The guiding-radius validation is definitional: L2 is identified as a cause of Lz, while Rg is derived from Lz, so the tight L2-Rg relation is built into the construction. The birth-radius validation is also partly circular: L1 is extracted from [Fe/H], age, and other observables, and it is then compared to the Lu et al. (2024) birth-radius estimate that is itself a function of [Fe/H] and age, with a polynomial mapping from L1 to physical units fitted to that same estimate. The paper does not quantitatively validate L1 against the simulation's true birth radius even though it claims ground truth is available. The RLCD method itself (Dong et al., 2024; 2025) is external and tested on synthetic data, and the self-citation for variance normalization is not load-bearing for the physical interpretation. The linear-SCM assumption in Eq. (1) is acknowledged as a limitation rather than smuggled in as a validated fact, so it affects correctness risk more than circularity. The remaining independent content is the unsupervised recovery of a two-latent structure from covariance patterns and the qualitative consistency in chemical-abundance planes, but the specific physical labeling of L1 and the quantitative 'validation' of both latents reduce substantially to the inputs used to define them. Score 6 reflects one or more predictions that reduce by construction while leaving some independent methodological content.

Assumptions & free parameters 3 free parameters · 3 assumptions · 2 invented entities

The central claim rests on a linear latent causal model, a comparison to a model-based birth radius estimator that uses the same observables, and a fitted polynomial mapping to physical units. The L1 validation is circular; L2 is near-tautological. No direct ground-truth comparison for L1 is reported.

free parameters (3)
  • Polynomial mapping coefficients for L1 and L2 to physical units = not reported
    Section 3: latent values are mapped to physical units via polynomial transformations; this is a post-hoc calibration and no coefficients or holdout are given.
  • SCM edge coefficients aij = estimated by MLE, values in Figure 1
    Section 2.2: coefficients are estimated from data and used to reconstruct latent values; these are fitted model parameters.
  • Sample selection thresholds = [Fe/H] > -1, R = 7-10 kpc, low-alpha split -0.13[Fe/H]+0.17
    Section 2.1: chosen by hand to ensure adequate sampling; could affect the discovered rank structure.
assumptions (3)
  • domain assumption The causal relationships among the five observables and latent variables are linear with additive noise (Eq. 1).
    Section 2.2 states the linear SCM; the authors later note this is a proof-of-concept assumption.
  • domain assumption Latent variables create detectable rank deficiencies in the observed covariance matrix, and RLCD's identifiability conditions hold.
    Section 2.2: this is the key algorithmic premise; no local checks or sensitivity are reported.
  • domain assumption The Lu et al. (2024) birth radius estimates are a valid proxy for true birth radius.
    Section 3: L1 is validated against Lu et al. estimates rather than against the simulation's true birth radius.
invented entities (2)
  • Latent variable L1
    purpose: Explains covariation among [Fe/H], age, and Jz; interpreted as birth radius.
    No direct quantitative comparison to true birth radius is reported; validation is against a model-based proxy that uses the same observables.
  • Latent variable L2
    purpose: Explains covariation between Lz and eccentricity; interpreted as guiding radius.
    Validated against simulation guiding radius, but guiding radius is essentially derived from Lz, which L2 directly influences, so the confirmation is weak; also the mapping is fitted.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Causal Discovery of Latent Variables in Galactic Archaeology." pith.science (2026). https://pith.science/paper/ZA7FTKC5

@misc{pith2026250700134,
  author       = {Pith},
  title        = {Pith review of: Causal Discovery of Latent Variables in Galactic Archaeology},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZA7FTKC5}},
  note         = {Machine review of arXiv:2507.00134}
}
read the original abstract

Galactic archaeology--the study of stellar migration histories--provides insights into galaxy formation and evolution. However, establishing causal relationships between observable stellar properties and their birth conditions remains challenging, as key properties like birth radius are not directly observable. We employ Rank-based Latent Causal Discovery (RLCD) to uncover the causal structure governing the chemodynamics of a simulated Milky Way galaxy. Using only five observable properties (metallicity, age, and orbital parameters), we recover in a purely data-driven manner a causal graph containing two latent nodes that correspond to real physical properties: the birth radius and guiding radius of stars. Our study demonstrates the potential of causal discovery models in astrophysics.

Figures

Figures reproduced from arXiv: 2507.00134 by the authors.

Figure 2
Figure 2. Validation of discovered latent variables. Left: Com￾parison between renormalized L1 and birth radius inferred using the Lu et al. (2024) method. Right: True guiding radius from simulation versus renormalized L2. The discovered L1 achieves comparable performance to (Lu et al., 2024) for birth radius infer￾ence, while L2 directly recovers the true guiding radius. orbits through gravitational scattering with giant mol… view at source ↗
Figure 3
Figure 3. Distribution of birth radii in chemical abundance space. Left: True birth radius from simulation. Middle: Birth radius inferred using the Lu et al. (2024) method. Right: Renormalized L1 from causal discovery. Top row shows the age-metallicity plane, bottom row shows the alpha-abundance plane. All three methods exhibit consistent patterns: younger, metal-rich stars originate from smaller galactic radii (yellow) while… view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. A Supermassive Black Hole Mass Measurement in NGC 5102 with Schwarzschild Orbit-superposition Modeling

    astro-ph.GA 2026-07 conditional novelty 5.0 of 10

    The central black hole in NGC 5102 has a mass of (1.30+0.19−0.18)×10^6 solar masses, measured with Schwarzschild orbit-superposition models of STIS and MUSE stellar kinematics.

Reference graph

Works this paper leans on

44 extracted references · 14 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Bensby , T., Feltzing , S., and Oey , M. S. Exploring the Milky Way stellar disk. A detailed elemental abundance study of 714 F and G dwarf stars in the solar neighbourhood . , 562: 0 A71, February 2014. doi:10.1051/0004-6361/201322631

    doi:10.1051/0004-6361/201322631 2014

  3. [3]

    On the origin of the chemical bimodality of disc stars: a tale of merger and migration

    Buck, T. On the origin of the chemical bimodality of disc stars: a tale of merger and migration . Monthly Notices of the Royal Astronomical Society, 491 0 (4): 0 5435--5446, 11 2019. ISSN 0035-8711. doi:10.1093/mnras/stz3289. URL https://doi.org/10.1093/mnras/stz3289

    doi:10.1093/mnras/stz3289 2019

  4. [4]

    K., Macci \`o , A

    Buck , T., Ness , M. K., Macci \`o , A. V., Obreja , A., and Dutton , A. A. Stars Behind Bars. I. The Milky Way's Central Stellar Populations . , 861 0 (2): 0 88, July 2018. doi:10.3847/1538-4357/aac890

    work page doi:10.3847/1538-4357/aac890 2018

  5. [5]

    V., and Dutton , A

    Buck , T., Ness , M., Obreja , A., Macci \`o , A. V., and Dutton , A. A. Stars behind Bars II: A Cosmological Formation Scenario for the Milky Way s Central Stellar Structure . , 874 0 (1): 0 67, March 2019. doi:10.3847/1538-4357/aaffd0

    work page doi:10.3847/1538-4357/aaffd0 2019

  6. [6]

    V., Minchev , I., Dutton , A

    Buck , T., Obreja , A., Macci \`o , A. V., Minchev , I., Dutton , A. A., and Ostriker , J. P. NIHAO-UHD: the properties of MW-like stellar discs in high-resolution cosmological simulations . , 491 0 (3): 0 3461--3478, January 2020. doi:10.1093/mnras/stz3241

    doi:10.1093/mnras/stz3241 2020

  7. [7]

    Optimal structure identification with greedy search

    Chickering, D. Optimal structure identification with greedy search. Journal of Machine Learning Research, 3: 0 507--554, 01 2002. doi:10.1162/153244303321897717

    doi:10.1162/153244303321897717 2002

  8. [8]

    H., Naidu , R

    Conroy , C., Weinberg , D. H., Naidu , R. P., Buck , T., Johnson , J. W., Cargile , P., Bonaca , A., Caldwell , N., Chandra , V., Han , J. J., Johnson , B. D., Speagle , J. S., Ting , Y.-S., Woody , T., and Zaritsky , D. Birth of the Galactic Disk Revealed by the H3 Survey . arXiv e-prints, art. arXiv:2204.02989, April 2022. doi:10.48550/arXiv.2204.02989

    arXiv doi:10.48550/arxiv.2204.02989 2022

Show all 44 references
  1. [9]

    C., Rankawat, M., Lacoste-Julien, S., Bauer, S., and Bengio, Y

    Deleu, T., G \'o is, A., Emezue, C. C., Rankawat, M., Lacoste-Julien, S., Bauer, S., and Bengio, Y. Bayesian structure learning with generative flow networks. In The 38th Conference on Uncertainty in Artificial Intelligence, 2022. URL https://openreview.net/forum?id=HElfed8j9g9

    2022

  2. [10]

    A versatile causal discovery framework to allow causally-related hidden variables

    Dong, X., Huang, B., Ng, I., Song, X., Zheng, Y., Jin, S., Legaspi, R., Spirtes, P., and Zhang, K. A versatile causal discovery framework to allow causally-related hidden variables. In The Twelfth International Conference on Learning Representations, 2024

    2024

  3. [11]

    On the parameter identifiability of partially observed linear causal models

    Dong, X., Ng, I., Huang, B., Sun, Y., Jin, S., Legaspi, R., Spirtes, P., and Zhang, K. On the parameter identifiability of partially observed linear causal models. Advances in Neural Information Processing Systems, 37: 0 30740--30771, 2025

    2025

  4. [12]

    and Heckerman, D

    Geiger, D. and Heckerman, D. Learning gaussian networks. In Proceedings of the Tenth International Conference on Uncertainty in Artificial Intelligence, UAI'94, pp.\ 235--243, San Francisco, CA, USA, 1994. Morgan Kaufmann Publishers Inc. ISBN 1558603328

    1994

  5. [13]

    R., Bovy , J., Holtzman , J

    Hayden , M. R., Bovy , J., Holtzman , J. A., Nidever , D. L., Bird , J. C., Weinberg , D. H., Andrews , B. H., Majewski , S. R., Allende Prieto , C., Anders , F., Beers , T. C., Bizyaev , D., Chiappini , C., Cunha , K., Frinchaboy , P., Garc \' a-Her \'n andez , D. A., Garc \'...

    2015

  6. [14]

    C., Monari , G., Famaey , B., de Jong , R

    Hilmi , T., Minchev , I., Buck , T., Martig , M., Quillen , A. C., Monari , G., Famaey , B., de Jong , R. S., Laporte , C. F. P., Read , J., Sanders , J. L., Steinmetz , M., and Wegg , C. Fluctuations in galactic bar parameters due to bar-spiral interaction . , 497 0 (1): 0 93...

    2020 doi

  7. [15]

    Generalized score functions for causal discovery

    Huang, B., Zhang, K., Lin, Y., Sch\" o lkopf, B., and Glymour, C. Generalized score functions for causal discovery. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD '18, pp.\ 1551--1560, New York, NY, USA, 2018. Associati...

    2018

  8. [16]

    L., Deleu, T., Luo, Y., Cho, C., Lemos, P., Perreault-Levasseur, L., Bengio, Y., Kang, X., Macciò, A

    Jin, Z., Pasquato, M., Davis, B. L., Deleu, T., Luo, Y., Cho, C., Lemos, P., Perreault-Levasseur, L., Bengio, Y., Kang, X., Macciò, A. V., and Hezaveh, Y. Causal discovery in astrophysics: Unraveling supermassive black hole and galaxy coevolution. The Astrophysical Journal, 97...

    2025 doi

  9. [17]

    Causal discovery in physical systems from videos

    Li, Y., Torralba, A., Anandkumar, A., Fox, D., and Garg, A. Causal discovery in physical systems from videos. Advances in Neural Information Processing Systems, 33: 0 9180--9192, 2020

    2020

  10. [18]

    Lu , Y., Buck , T., Minchev , I., and Ness , M. K. Reliability and limitations of inferring birth radii in the Milky Way disc . , 515 0 (1): 0 L34--L38, September 2022 a . doi:10.1093/mnrasl/slac065

    2022 doi

  11. [19]

    L., Ness , M

    Lu , Y. L., Ness , M. K., Buck , T., and Carr , C. Turning points in the age-metallicity relations - created by late satellite infall and enhanced by radial migration . , 512 0 (4): 0 4697--4714, June 2022 b . doi:10.1093/mnras/stac780

    2022 doi

  12. [20]

    L., Minchev , I., Buck , T., Khoperskov , S., Steinmetz , M., Libeskind , N., Cescutti , G., Freeman , K

    Lu , Y. L., Minchev , I., Buck , T., Khoperskov , S., Steinmetz , M., Libeskind , N., Cescutti , G., Freeman , K. C., and Ratcliffe , B. There is no place like home - finding birth radii of stars in the Milky Way . , 535 0 (1): 0 392--405, November 2024. doi:10.1093/mnras/stae2364

    2024 doi

  13. [21]

    Chemodynamical evolution of the Milky Way disk

    Minchev , I., Chiappini , C., and Martig , M. Chemodynamical evolution of the Milky Way disk. I. The solar vicinity . , 558: 0 A9, October 2013. doi:10.1051/0004-6361/201220189

    2013 doi

  14. [22]

    S., Kordopatis , G., Majewski , S

    Minchev , I., Anders , F., Recio-Blanco , A., Chiappini , C., de Laverny , P., Queiroz , A., Steinmetz , M., Adibekyan , V., Carrillo , I., Cescutti , G., Guiglion , G., Hayden , M., de Jong , R. S., Kordopatis , G., Majewski , S. R., Martig , M., and Santiago , B. X. Estimati...

    2018 doi

  15. [23]

    A., Macci \`o , A

    Obreja , A., Dutton , A. A., Macci \`o , A. V., Moster , B., Buck , T., van de Ven , G., Wang , L., Stinson , G. S., and Zhu , L. NIHAO XVI: the properties and evolution of kinematically selected discs, bulges, and stellar haloes . , 487 0 (3): 0 4424--4456, August 2019. doi:1...

    2019 doi

  16. [24]

    Obreja , A., Buck , T., and Macci \`o , A. V. A first estimate of the Milky Way dark matter halo spin . , 657: 0 A15, January 2022. doi:10.1051/0004-6361/202140983

    2022 doi

  17. [25]

    L., and Macci \`o , A

    Pasquato , M., Jin , Z., Lemos , P., Davis , B. L., and Macci \`o , A. V. Causa prima: cosmology meets causal discovery for the first time . arXiv e-prints, art. arXiv:2311.15160, November 2023. doi:10.48550/arXiv.2311.15160

    2023 doi arXiv

  18. [26]

    Causality

    Pearl, J. Causality. Cambridge university press, 2009

    2009

  19. [27]

    Pearl, J., Glymour, M., and Jewell, N. P. Causal inference in statistics : a primer. Wiley, Chichester, West Sussex, 2016. ISBN 9781119186854

    2016

  20. [28]

    D., Mu \ n oz-Mar \' , J., et al

    Runge, J., Bathiany, S., Bollt, E., Camps-Valls, G., Coumou, D., Deyle, E., Glymour, C., Kretschmer, M., Mahecha, M. D., Mu \ n oz-Mar \' , J., et al. Inferring causation from time series in earth system sciences. Nature communications, 10 0 (1): 0 2553, 2019

    2019

  21. [29]

    R., Kalchbrenner, N., Goyal, A., and Bengio, Y

    Sch \"o lkopf, B., Locatello, F., Bauer, S., Ke, N. R., Kalchbrenner, N., Goyal, A., and Bengio, Y. Toward causal representation learning. Proceedings of the IEEE, 109 0 (5): 0 612--634, 2021

    2021

  22. [30]

    Sellwood , J. A. and Binney , J. J. Radial mixing in galactic discs . , 336 0 (3): 0 785--796, November 2002. doi:10.1046/j.1365-8711.2002.05806.x

    2002

  23. [31]

    F., Navarro , J

    Sestito , F., Buck , T., Starkenburg , E., Martin , N. F., Navarro , J. F., Venn , K. A., Obreja , A., Jablonka , P., and Macci \`o , A. V. Exploring the origin of low-metallicity stars in Milky-Way-like galaxies with the NIHAO-UHD simulations . , 500 0 (3): 0 3750--3762, Janu...

    2021 doi

  24. [32]

    R., and Bland-Hawthorn , J

    Sharma , S., Hayden , M. R., and Bland-Hawthorn , J. Chemical enrichment and radial migration in the Galactic disc - the origin of the [ Fe] double sequence . , 507 0 (4): 0 5882--5901, November 2021. doi:10.1093/mnras/stab2015

    2021 doi

  25. [33]

    O., Hyvrinen, A., and Kerminen, A

    Shimizu, S., Hoyer, P. O., Hyvrinen, A., and Kerminen, A. J. A linear non-gaussian acyclic model for causal discovery. Journal of Machine Learning Research, 7 0 (4): 0 2003--2030, 2006

    2003

  26. [34]

    Causal inference in the presence of latent variables and selection bias

    Spirtes, P., Meek, C., and Richardson, T. Causal inference in the presence of latent variables and selection bias. In Proceedings of the Eleventh Conference on Uncertainty in Artificial Intelligence, pp.\ 499--506, August 1995

    1995

  27. [35]

    Causation, Prediction, and Search

    Spirtes, P., Glymour, C., and Scheines, R. Causation, Prediction, and Search. The MIT Press, 01 2001. ISBN 9780262284158. doi:10.7551/mitpress/1754.001.0001. URL https://doi.org/10.7551/mitpress/1754.001.0001

    2001 doi

  28. [36]

    Star formation and feedback in smoothed particle hydrodynamic simulations - I

    Stinson , G., Seth , A., Katz , N., Wadsley , J., Governato , F., and Quinn , T. Star formation and feedback in smoothed particle hydrodynamic simulations - I. Isolated galaxies . , 373 0 (3): 0 1074--1090, December 2006. doi:10.1111/j.1365-2966.2006.11097.x

    2006

  29. [37]

    S., Bovy , J., Rix , H

    Stinson , G. S., Bovy , J., Rix , H. W., Brook , C., Ro s kar , R., Dalcanton , J. J., Macci \`o , A. V., Wadsley , J., Couchman , H. M. P., and Quinn , T. R. MaGICC thick disc - I. Comparing a simulated disc formed with stellar feedback to the Milky Way . , 436 0 (1): 0 625--...

    2013 doi

  30. [38]

    W., Keller , B

    Wadsley , J. W., Keller , B. W., and Quinn , T. R. Gasoline2: a modern smoothed particle hydrodynamics code . , 471 0 (2): 0 2357--2369, October 2017. doi:10.1093/mnras/stx1643

    2017 doi

  31. [39]

    K., and Buck , T

    Wang , K., Carrillo , A., Ness , M. K., and Buck , T. The individual abundance distributions of disc stars across birth radii in GALAH . arXiv e-prints, art. arXiv:2307.04724, July 2023. doi:10.48550/arXiv.2307.04724

    2023 doi arXiv

  32. [40]

    A., Stinson , G

    Wang , L., Dutton , A. A., Stinson , G. S., Macci \`o , A. V., Penzo , C., Kang , X., Keller , B. W., and Wadsley , J. NIHAO project - I. Reproducing the inefficiency of galaxy formation across cosmic time with a large sample of cosmological hydrodynamical simulations . , 454 ...

    2015 doi

  33. [41]

    and Rix , H.-W

    Xiang , M. and Rix , H.-W. A time-resolved picture of our Milky Way's early formation history . , 603 0 (7902): 0 599--603, March 2022. doi:10.1038/s41586-022-04496-5

    2022 doi

  34. [42]

    Marrying causal representation learning with dynamical systems for science

    Yao, D., Muller, C., and Locatello, F. Marrying causal representation learning with dynamical systems for science. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

    2024

  35. [43]

    On the completeness of orientation rules for causal discovery in the presence of latent confounders and selection bias

    Zhang, J. On the completeness of orientation rules for causal discovery in the presence of latent confounders and selection bias. Artificial Intelligence, 172 0 (16-17): 0 1873--1896, 2008

    2008

  36. [44]

    Causal representation learning from multiple distributions: A general setting

    Zhang, K., Xie, S., Ng, I., and Zheng, Y. Causal representation learning from multiple distributions: A general setting. In International Conference on Machine Learning, pp.\ 60057--60075. PMLR, 2024

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.