Pith. sign in

REVIEW 4 major objections 6 minor 60 references

CNN-Derived Elemental Abundances of LAMOST DR10 Giants: Implications for Galactic Substructures

T0 review · 4 major / 6 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 This paper shows that a CNN can map LAMOST low-resolution spectra of 1.1 million giants to APOGEE-grade abundances, then uses the resulting chemistry to argue that substructure PG1 mixes GSE and thick-disk stars while PG2 is an in-situ…

desk verdict A useful public catalog of 1.1M LAMOST giants with CNN abundances, but the headline precision does not transfer to the full sample and the substructure claims rest on 11 stars. read the letter →

arxiv 2506.16135 v1 pith:GMNIQSNL submitted 2025-06-19 astro-ph.GA

Haoyang Liu , Cuihua Du , Mingji Deng , Jian Zhang This is my paper
classification astro-ph.GA
keywords convolutionalneuralnetworkstellarabundancesLAMOSTDR10APOGEEDR17giantstarsgalacticsubstructureslabeltransfercatalog
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

The paper sets out to show that a convolutional neural network can transfer precise stellar labels from high-resolution APOGEE spectra to low-resolution LAMOST spectra, yielding effective temperatures, metallicities, and six element abundances for 1.1 million giant stars. The payoff is a uniform abundance catalog for a large, wide-field sample of red giants, which the authors then use to ask what two kinematically defined halo substructures, PG1 and PG2, are made of. On the test set, the network reproduces APOGEE labels with 50 K scatter in temperature, 0.06 dex in [Fe/H], 0.13 dex in log g, and 0.03 to 0.07 dex in the element ratios. On that basis the paper concludes that PG1 mixes GSE debris with thick-disk stars, while PG2 is an in-situ thin-disk feature. A public catalog accompanies the argument.

What carries the argument

The load-bearing mechanism is label transfer through a convolutional neural network: a three-convolutional-layer, three-fully-connected-layer network that maps 5,000-bin normalized LAMOST spectra to APOGEE DR17 stellar labels, trained on 62,511 common stars. A key preprocessing step is pseudo-normalization by dividing the spectrum by an error-weighted Gaussian-smoothed version, which removes continuum shape while preserving absorption-line information. Dropout is activated at inference time to give per-label uncertainties, following the dropout-as-approximate-Bayesian-inference approach.

What would settle it

Take a sample of very metal-poor giants ([Fe/H] < -2) that were not used in training, run them through the network, and compare with their APOGEE DR17 abundances. If the scatter in C, N, or Ca approaches the paper's own VMP numbers (0.29, 0.40, and 0.48 dex) rather than the overall test-set scatters (0.03 to 0.07 dex), the generalization claim for metal-poor stars collapses; the same check applies to low-S/N stars.

Watch

Extended reading notes

Core claim

The central discovery is that low-resolution LAMOST spectra carry enough information to recover APOGEE-grade abundances of C, N, O, Mg, Si, Ca, [α/M], and [M/H] for DR10 giants, and that the resulting chemistry can discriminate between accretion and in-situ origins for halo substructures identified in integral-of-motion space. In the [Mg/Fe]-[Fe/H] plane, the common stars of PG1 split between a high-α, metal-rich clump matching the thick disk and a low-α, low-[Al/Fe] tail matching GSE, while nine of eleven PG2 stars form a tight high-α group with positive [Al/Fe], pointing to an in-situ thin-disk association. The paper frames this as evidence rather than proof, given the small overlap sample, and recommends future data releases and simulations to confirm the origins.

Load-bearing premise

The claim that the catalog is accurate for all 1.1 million giants rests on the assumption that the 62,511 stars with both LAMOST and APOGEE measurements, which supply both the training set and the test set, are representative of the whole population, including the metal-poor and low signal-to-noise stars where the paper itself reports larger scatter.

Editorial extensions

If this is right

  • The catalog enables chemical tagging of halo substructures beyond PG1 and PG2, since the derived abundances cover C, N, O, Mg, Si, and Ca for 1.1 million giants.
  • [Mg/Fe], [Si/Fe], and [α/M] keep scatter below 0.10 dex even for metal-poor and very metal-poor stars, so these three abundances can be used to study the metal-poor halo where the other element scatters grow larger.
  • The high-α thick-disk and low-α thin-disk bimodality appears in all predicted [X/Fe]-[Fe/H] planes, confirming that the network preserves known disk chemistry.
  • The predicted Kiel diagram recovers the red clump and the smooth metallicity transition along the giant branch, providing an internal consistency check for the derived parameters.
  • If PG1 is a GSE/thick-disk mixture and PG2 is an in-situ thin-disk structure, then integral-of-motion clumps do not map one-to-one onto accreted versus in-situ populations; chemistry is needed to separate the two.

Reading between the lines

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

  • An extension the authors leave implicit is that the same network architecture could be retrained on future LAMOST releases or on medium-resolution spectra, extending the catalog without new survey design.
  • The chemical classification of PG1 and PG2 rests on only 11 common stars; a testable extension is to use the catalog's uncertainties to select additional PG1 and PG2 candidates and check whether the bimodality in [Mg/Fe] and [Al/Fe] persists with a larger sample.
  • The paper's uncertainty discussion suggests that catalog values outside the reliable ranges listed in Section 3.5 should be treated as flagged rather than as measurements, and a reliability flag column would make this explicit for users.
  • The comparisons with previous neural-network and data-driven methods imply that a shared validation set of APOGEE common stars could settle which architecture transfers labels to LAMOST most accurately; the paper does not run such a benchmark.
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

4 major / 6 minor

Summary. The authors train a CNN on 62,511 stars common to LAMOST DR10 and APOGEE DR17 to predict T_eff, log g, [Fe/H], [M/H], [C/Fe], [N/Fe], [O/Fe], [Mg/Fe], [Si/Fe], [Ca/Fe], and [alpha/M] for 1,100,858 LAMOST DR10 giants with S/N_g > 10. They report test-set scatters around 50 K, 0.06 dex, and 0.13 dex for T_eff, [Fe/H], and log g, and 0.03–0.07 dex for the elemental abundances, compare with previous label-transfer methods, estimate uncertainties via Monte-Carlo dropout, and release a public catalog. They then cross-match predicted abundances with the PG1 and PG2 substructures from Liu et al. (2024) and argue that PG1 is a GSE/thick-disk mixture and PG2 is an in-situ thin-disk substructure.

Significance. If the precision claims were established on the full 1.1M-star sample, the public catalog would be a valuable resource for Galactic archaeology, complementing APOGEE and GALAH with LAMOST's wide sky coverage. The paper usefully compares with earlier label-transfer methods, quantifies S/N-dependent behavior, and provides per-star uncertainty estimates. However, the current validation is restricted to a random split of APOGEE-LAMOST common stars, and the paper's own metal-poor subsample results show substantially larger scatters, so the catalog-level precision and the substructure interpretation built on it are not yet demonstrated at the claimed level.

major comments (4)
  1. [Abstract and §3.3 vs §5] The paper reports inconsistent test-set scatter values for the same quantities. The abstract and §3.3 state scatters of 0.07 dex for [C/Fe], 0.05 dex for [N/Fe], 0.05 dex for [O/Fe], 0.04 dex for [Mg/Fe], 0.03 dex for [Si/Fe], 0.04 dex for [Ca/Fe], and 0.06 dex for [M/H], while §5 gives 0.02, 0.03, 0.06, 0.06, 0.01, 0.02, and 0.15 dex for these labels. These are not minor typographical differences: [M/H] changes by 0.09 dex and C/N by 0.03–0.05 dex. Since the precision claims are the central deliverable, the manuscript must adopt one internally consistent set of values and verify them against the test-set calculations.
  2. [§3.2 and §4.1] The validation design does not support the catalog-level precision claim. The test set is a random 80/20 split of the 62,511 stars that have both LAMOST spectra and APOGEE labels (§3.2), and APOGEE labels were restricted to S/N > 70 and abundance errors < 0.2 dex (§2.1), whereas the catalog includes all LAMOST giants with S/N_g > 10. The manuscript itself reports in §4.1 that for metal-poor stars (−2 < [Fe/H] < −1) and very metal-poor stars ([Fe/H] < −2) the scatters rise to, e.g., 0.15/0.29 dex for C, 0.20/0.40 dex for N, and 0.15/0.48 dex for Ca, versus 0.07/0.05/0.04 dex on the test set. These populations are directly relevant to the halo-substructure claims. The reported test scatters therefore do not validate the precision of the 1.1M-star catalog; please either restrict the precision claims to the common-star-like population or add a representative validation sample that covers the full S/N and metallicity range.
  3. [§4.2 and Fig. 13] The substructure conclusions rest on very small cross-matches. §4.2 reports 4 LAMOST, 1 APOGEE, and 6 GALAH matches for PG1 and 7 LAMOST, 3 APOGEE, and 1 GALAH matches for PG2, with Figure 13 plotting these together with GSE comparison stars. From these 11 PG1/PG2 common stars, the paper concludes that 'PG2 is highly likely to be an in-situ substructure' and that PG1 is a GSE/thick-disk mixture. With n = 11 and no quantitative mixture model or hypothesis test, this conclusion is not supported. Please rephrase the interpretation as a tentative suggestion and, if possible, add a statistical assessment of how consistent the PG1 and PG2 abundance distributions are with the thick-disk and GSE reference populations.
  4. [§3.5] The discussion of reliable abundance ranges contains a likely labeling error: the list of ranges appears to include two [C/Fe] intervals (−0.75 to 0.47 dex and −0.24 to 0.82 dex), while no [N/Fe] range is given. Given that §3.5 explicitly discusses the offsets for all elements, this typo affects the usability of the recommended quality cuts and should be corrected.
minor comments (6)
  1. [§2.1] The word 'acrsecond' should be 'arcsecond'.
  2. [§3.3] The sentence 'The biases for the elements N, O and Mg are both 0 dex' should read 'are all 0 dex', since three elements are listed.
  3. [Abstract] The phrase 'The spectral from LAMOST' should be 'The spectra from LAMOST'.
  4. [Data Availability] The Data Availability section mentions only LAMOST and APOGEE; the catalog URL given in the abstract should also appear here, with a version/date of access.
  5. [§3.5] In the compiled text some phrases have missing word spaces (e.g., 'comparisonofbetweenpredictionuncertainties' near the end of §3.5); please check the typesetting in the final version.
  6. [§5] The Summary uses 'uncertainties' for what are elsewhere called 'scatters' (one standard deviation of residuals); please clarify the terminology to avoid conflating random prediction uncertainty with scatter.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the CNN predictions are anchored to external APOGEE ground truth, test-set scatters are honest holdout metrics, and the PG1/PG2 substructures were defined from Gaia astrometry in prior work independent of the derived abundances.

full rationale

The core derivation is a label-transfer exercise: a CNN maps LAMOST low-resolution spectra to APOGEE stellar labels, trained on 62,511 common stars whose abundances are external ground truth (Sections 2.1, 3.1). The test-set scatters (Section 3.3) come from a random 80/20 split (Section 3.2), so they are honest holdout statistics, not refits; the catalog predictions for the 1,100,858 giants are genuine extrapolation to stars never used in fitting. The extrapolation is imperfect - Section 4.1 reports 2-10x larger scatters for metal-poor and very metal-poor stars, and the training labels require S/N>70 while the full sample only requires S/N_g>10 - but this is an external-validity/representativeness risk, not a reduction of a prediction to its inputs. No equation defines a predicted quantity from the fitted labels; the dropout-based 'uncertainties' (Section 3.5) are explicitly epistemic, and the paper states they 'do not directly correlate with the offsets between the predicted labels and the APOGEE labels.' The substructure analysis (Section 4.2) uses PG1/PG2 memberships from Liu et al. (2024), a prior paper by overlapping authors, but that catalog was built from Gaia DR3 astrometry in integral-of-motion space (E, L_z, L_perp) via HDBSCAN clustering, i.e., data disjoint from the CNN abundances; the paper also notes LEG was independently identified by Dodd et al. (2024), corroborating the catalog. The abundance-based origin conclusions (GSE/thick-disk mixture for PG1, in-situ thin disk for PG2) therefore combine new chemical information with a pre-existing kinematic classification and do not reduce by construction; no uniqueness theorem or ansatz is imported from the authors' prior work. The bimodality in the [X/Fe]-[Fe/H] planes (Figure 10) is presented as recovery of the known thick/thin-disk pattern, not as a new prediction. Flagged limitations: the Summary quotes numbers inconsistent with the Abstract and Section 3.3 (e.g., [M/H] 0.15 vs 0.06 dex; [alpha/M] 0.05 vs 0.03 dex; C/N/O/Mg/Si/Ca values differ as well), and the PG2 in-situ conclusion rests on only 11 common stars total (4 and 7 LAMOST cross-matches for PG1 and PG2) - the paper itself concedes 'the current sample size is relatively limited.' These are presentation and statistical-strength concerns, not circularity.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new physical entities, forces, or conserved quantities. Its central claim depends on the transferability of APOGEE labels, the representativeness of the training split, the normalization scheme, and the chosen CNN hyperparameters, all of which are assumptions or tuned choices rather than derived results.

free parameters (2)
  • CNN hyperparameters (kernel size 10, filters 16/8/4, dropout 0.2, learning rate 1e-5, batch size 16, 60 epochs) = selected via validation MAE
    Chosen in Section 3.2 to minimize validation loss; reported scatter values depend on these choices.
  • Smoothing length L in pseudo-normalization = 50 Å
    Set in Section 2.2 as broader than typical atomic lines; affects all normalized spectra used as model input.
assumptions (3)
  • domain assumption APOGEE stellar labels are accurate and transferable to LAMOST spectra as intrinsic properties
    The entire label-transfer method (Section 1 and 2.1) assumes that labels are instrument-independent.
  • domain assumption The random 80/20 train-test split is representative of the full 1.1M giant population
    Section 3.2 splits only the 62,511 common stars; extrapolation to all LAMOST giants is assumed without external validation.
  • domain assumption Pseudo-normalization (Eq. 1-2) removes calibration differences while preserving abundance-sensitive features
    Section 2.2 adopts this normalization following Ho et al. (2017), but no proof is given that abundance information is fully retained.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CNN-Derived Elemental Abundances of LAMOST DR10 Giants: Implications for Galactic Substructures." pith.science (2026). https://pith.science/paper/GMNIQSNL

@misc{pith2026250616135,
  author       = {Pith},
  title        = {Pith review of: CNN-Derived Elemental Abundances of LAMOST DR10 Giants: Implications for Galactic Substructures},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GMNIQSNL}},
  note         = {Machine review of arXiv:2506.16135}
}
abstract

Stellar parameters and abundances provide crucial insights into stellar and Galactic evolution studies. In this work, we developed a convolutional neural network (CNN) to estimate stellar parameters: effective temperature ($T_{\text{eff}}$), surface gravity (log $g$) and metallicity (both [Fe/H] and [M/H]) as well as six $\alpha$-elements (C, N, O, Mg, Si, Ca) and [$\alpha$/M]. We selected giant stars with \( 3500 \, \text{K} < T_{\text{eff}} < 5500 \, \text{K} \) and \( 0 \, \text{dex} < \log g < 3.6 \, \text{dex} \) from the LAMOST and APOGEE surveys, while requiring (S/N)$_g$ of the LAMOST low-resolution spectra $>$ 10, which leaves 1,100,858 giant stars. The spectral from LAMOST and the labels from APOGEE for 62,511 common stars were used as our training set. The corresponding test set yields scatters 50 K, 0.06 dex and 0.13 dex for $T_{\text{eff}}$, [Fe/H] and log $g$, respectively. For $\alpha$ elements O, Mg, Si and Ca, the scatters are 0.05 dex, 0.04 dex, 0.03 and 0.04 dex, respectively. For C and N elements, the scatters are 0.07 dex and 0.05 dex. For [$\alpha$/M] and [M/H], the scatters are 0.03 dex and 0.06 dex. The mean absolute error (MAE) of most elements are between 0.02 $-$ 0.04 dex. The predicted abundances were cross-matched with previously identified substructures PG1 and PG2, with their origins subsequently analyzed. Finally, the catalog is available at https://nadc.china-vo.org/res/r101529/.

Figures

Figures reproduced from arXiv: 2506.16135 by the authors.

Figure 1
Figure 1. The schematic diagram of CNN architecture used in this work. The model adopts three convolutional layers with 16, 8, and 4 filters of size 10, each followed by a max pool layer of size 2. A dropout layer follows the third convolutional layer, leading to three fully connected layers with 256, 128, and 64 neurons, respectively. Note that LeakyReLU activation function is used in this model. errors for effective tempera… view at source ↗
Figure 2
Figure 2. The loss values for training and test samples at each epoch. The model is trained for 60 epochs because the test loss value does not significantly decrease and starts to fluctuate. of 20 % consistent with their findings. The Adam optimizer is a combination of two gradient descent methods: Momentum and Root Mean Square Propagation (RMSProp) (Kingma & Ba 2014). Given its greater efficiency and lower memory requirement… view at source ↗
Figure 3
Figure 3. Distributions of training (blue) and test (orange) samples in 𝑇eff, [Fe/H] and log 𝑔. There are 50,008 stars in the training sample and 12,503 stars in the test sample [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: The upper row: the predicted labels vs. APOGEE labels in 𝑇eff, [Fe/H] and log 𝑔. The red line represents a 1:1 ratio, while the green dashed line indicates the best-fit linear regression of the labels (k is the slope and b is the interception). The bottom row: the resi…
Figure 5
Figure 5. Figure 5: The upper row: the predicted labels vs. APOGEE labels in C, N, O and Mg elements. The red line represents a 1:1 ratio, while the green dashed line indicates the best-fit linear regression of the labels (k is the slope and b is the interception). The bottom row: the res…
Figure 6
Figure 6. Figure 6: The upper row: the predicted labels vs. APOGEE labels in Si, Ca elements as well as overall metallicity and 𝛼-abundance. The red line represents a 1:1 ratio, while the green dashed line indicates the best-fit linear regression of the labels (k is the slope and b is the…
Figure 7
Figure 7. Figure 7: The dependence for scatter values (blue) and MAE values (orange) as a function of (S/N)𝑔 for each stellar label. The (S/N)𝑔 between 10 and 250 are evenly divided into 10 bins with scatter and MAE values calculated in each bin. Equal-count binning method is also applied…
Figure 8
Figure 8. Figure 8: The comparison of the mean prediction uncertainties for common stars between our work and Li et al. (2022) reveals that our model exhibits significantly lower uncertainties in most cases. However, the two models show comparable uncertainties in (overall) metallicity. N…
Figure 9
Figure 9. Figure 9: The comparison between LASP estimates and predictions in our work. Top three rows: comparisons of 𝑇eff, 𝑙𝑜𝑔g, [Fe/H] and [𝛼/M] as a function of (S/N)𝑔. These rows do not include a color bar, as they are meant only for comparative purposes. Bottom row: comparison of mea…
Figure 10
Figure 10. Figure 10: Predicted elemental labels for all LAMOST DR10 giants and corresponding elemental distributions. The bimodal structure of Galactic disks is clearly observed, with the high-𝛼 clump corresponding to the thick disk and the low-𝛼 clump representing the thin disk [PITH_FU…
Figure 11
Figure 11. Figure 11: Elemental distributions of common stars for comparison with the predicted distributions for LAMOST giants. Note that abundances here are estimated from APOGEE [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]
Figure 12
Figure 12. Figure 12: The Kiel diagrams for common stars of APOGEE label, test samples of predicted labels and LAMOST DR10 giants of predicted labels color-coded by metallicity. The predicted labels for LAMOST giants well recovered the main features in Kiel diagram. MNRAS 000, 1–13 (2025) …
Figure 13
Figure 13. Figure 13: The selected GSE, PG1 and PG2 common stars are from Liu et al. (2024). Left panel: the distributions in [Mg/Fe]-[Fe/H] plane. Gray dots represent GSE member stars with elemental abundance estimates from LAMOST DR10. Triangles denote PG1, with blue, red, and coral colo…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

60 extracted references · 8 canonical work pages

  1. [1]

    Abdurro'uf et al., 2022, @doi [ ] 10.3847/1538-4365/ac4414 , https://ui.adsabs.harvard.edu/abs/2022ApJS..259...35A 259, 35

    doi:10.3847/1538-4365/ac4414 2022

  2. [2]

    F., 2018, @doi [arXiv e-prints] 10.48550/arXiv.1803.08375 , https://ui.adsabs.harvard.edu/abs/2018arXiv180308375A p

    Agarap A. F., 2018, @doi [arXiv e-prints] 10.48550/arXiv.1803.08375 , https://ui.adsabs.harvard.edu/abs/2018arXiv180308375A p. arXiv:1803.08375

    arXiv doi:10.48550/arxiv.1803.08375 2018

  3. [3]

    Amarante J. A. S., Debattista V. P., Beraldo e Silva L., Laporte C. F. P., Deg N., 2022, @doi [ ] 10.3847/1538-4357/ac8b0d , https://ui.adsabs.harvard.edu/abs/2022ApJ...937...12A 937, 12

    doi:10.3847/1538-4357/ac8b0d 2022

  4. [4]

    Bailer-Jones C. A. L., Rybizki J., Fouesneau M., Demleitner M., Andrae R., 2021, @doi [ ] 10.3847/1538-3881/abd806 , https://ui.adsabs.harvard.edu/abs/2021AJ....161..147B 161, 147

    doi:10.3847/1538-3881/abd806 2021

  5. [5]

    W., Koposov S

    Belokurov V., Erkal D., Evans N. W., Koposov S. E., Deason A. J., 2018, @doi [ ] 10.1093/mnras/sty982 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.478..611B 478, 611

    doi:10.1093/mnras/sty982 2018

  6. [6]

    L., Fattahi A., Smith M

    Belokurov V., Sanders J. L., Fattahi A., Smith M. C., Deason A. J., Evans N. W., Grand R. J. J., 2020, @doi [ ] 10.1093/mnras/staa876 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.494.3880B 494, 3880

    doi:10.1093/mnras/staa876 2020

  7. [7]

    S., 2014, @doi [ ] 10.1051/0004-6361/201322631 , https://ui.adsabs.harvard.edu/abs/2014A&A...562A..71B 562, A71

    Bensby T., Feltzing S., Oey M. S., 2014, @doi [ ] 10.1051/0004-6361/201322631 , https://ui.adsabs.harvard.edu/abs/2014A&A...562A..71B 562, A71

    doi:10.1051/0004-6361/201322631 2014

  8. [8]

    arXiv:2409.19858

    Buder S., et al., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2409.19858 , https://ui.adsabs.harvard.edu/abs/2024arXiv240919858B p. arXiv:2409.19858

    arXiv doi:10.48550/arxiv.2409.19858 2024

Show all 60 references
  1. [9]

    Cui X.-Q., et al., 2012, @doi [Research in Astronomy and Astrophysics] 10.1088/1674-4527/12/9/003 , https://ui.adsabs.harvard.edu/abs/2012RAA....12.1197C 12, 1197

    2012 doi

  2. [10]

    arXiv:1611.00036

    DESI Collaboration et al., 2016, @doi [arXiv e-prints] 10.48550/arXiv.1611.00036 , https://ui.adsabs.harvard.edu/abs/2016arXiv161100036D p. arXiv:1611.00036

    2016 doi arXiv

  3. [11]

    M., et al., 2015, @doi [ ] 10.1093/mnras/stv327 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.449.2604D 449, 2604

    De Silva G. M., et al., 2015, @doi [ ] 10.1093/mnras/stv327 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.449.2604D 449, 2604

    2015 doi

  4. [12]

    M., Cassisi S., Fern \'a ndez-Alvar E., Surot F., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2408.13763 , https://ui.adsabs.harvard.edu/abs/2024arXiv240813763D p

    Dodd E., Ruiz-Lara T., Helmi A., Gallart C., Callingham T. M., Cassisi S., Fern \'a ndez-Alvar E., Surot F., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2408.13763 , https://ui.adsabs.harvard.edu/abs/2024arXiv240813763D p. arXiv:2408.13763

    2024 doi arXiv

  5. [13]

    A., O'Briain T., Bialek S., Kielty C

    Fabbro S., Venn K. A., O'Briain T., Bialek S., Kielty C. L., Jahandar F., Monty S., 2018, @doi [ ] 10.1093/mnras/stx3298 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.475.2978F 475, 2978

    2018 doi

  6. [14]

    Fernandes L., et al., 2023, @doi [ ] 10.1093/mnras/stac3543 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.519.3611F 519, 3611

    2023 doi

  7. [15]

    Fuhrmann K., 1998, , https://ui.adsabs.harvard.edu/abs/1998A&A...338..161F 338, 161

    1998

  8. [16]

    arXiv:1506.02142

    Gal Y., Ghahramani Z., 2015, @doi [arXiv e-prints] 10.48550/arXiv.1506.02142 , https://ui.adsabs.harvard.edu/abs/2015arXiv150602142G p. arXiv:1506.02142

    2015 doi arXiv

  9. [17]

    J., Brook C

    Gallart C., Bernard E. J., Brook C. B., Ruiz-Lara T., Cassisi S., Hill V., Monelli M., 2019, @doi [Nature Astronomy] 10.1038/s41550-019-0829-5 , https://ui.adsabs.harvard.edu/abs/2019NatAs...3..932G 3, 932

    2019 doi

  10. [18]

    E., et al., 2016, @doi [ ] 10.3847/0004-6256/151/6/144 , https://ui.adsabs.harvard.edu/abs/2016AJ....151..144G 151, 144

    Garc \' a P \'e rez A. E., et al., 2016, @doi [ ] 10.3847/0004-6256/151/6/144 , https://ui.adsabs.harvard.edu/abs/2016AJ....151..144G 151, 144

    2016 doi

  11. [19]

    Gilmore G., Reid N., 1983, @doi [ ] 10.1093/mnras/202.4.1025 , https://ui.adsabs.harvard.edu/abs/1983MNRAS.202.1025G 202, 1025

    1983 doi

  12. [20]

    Gilmore G., et al., 2012, The Messenger, https://ui.adsabs.harvard.edu/abs/2012Msngr.147...25G 147, 25

    2012

  13. [21]

    Guiglion G., et al., 2020, @doi [ ] 10.1051/0004-6361/202038271 , https://ui.adsabs.harvard.edu/abs/2020A&A...644A.168G 644, A168

    2020 doi

  14. [22]

    E., et al., 2006, @doi [ ] 10.1086/500975 , https://ui.adsabs.harvard.edu/abs/2006AJ....131.2332G 131, 2332

    Gunn J. E., et al., 2006, @doi [ ] 10.1086/500975 , https://ui.adsabs.harvard.edu/abs/2006AJ....131.2332G 131, 2332

    2006 doi

  15. [23]

    J., et al., 2022, @doi [ ] 10.3847/1538-3881/ac97e9 , https://ui.adsabs.harvard.edu/abs/2022AJ....164..249H 164, 249

    Han J. J., et al., 2022, @doi [ ] 10.3847/1538-3881/ac97e9 , https://ui.adsabs.harvard.edu/abs/2022AJ....164..249H 164, 249

    2022 doi

  16. [24]

    Hasselquist S., et al., 2021, @doi [ ] 10.3847/1538-4357/ac25f9 , https://ui.adsabs.harvard.edu/abs/2021ApJ...923..172H 923, 172

    2021 doi

  17. [25]

    H., Massari D., Veljanoski J., Brown A

    Helmi A., Babusiaux C., Koppelman H. H., Massari D., Veljanoski J., Brown A. G. A., 2018, @doi [ ] 10.1038/s41586-018-0625-x , https://ui.adsabs.harvard.edu/abs/2018Natur.563...85H 563, 85

    2018 doi

  18. [26]

    Ho A. Y. Q., et al., 2017, @doi [ ] 10.3847/1538-4357/836/1/5 , https://ui.adsabs.harvard.edu/abs/2017ApJ...836....5H 836, 5

    2017 doi

  19. [27]

    Horta D., et al., 2021, @doi [ ] 10.1093/mnras/staa2987 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.500.1385H 500, 1385

    2021 doi

  20. [28]

    Huang Y., et al., 2020, @doi [ ] 10.3847/1538-4365/ab994f , https://ui.adsabs.harvard.edu/abs/2020ApJS..249...29H 249, 29

    2020 doi

  21. [29]

    Hunt J. A. S., Vasiliev E., 2025, @doi [arXiv e-prints] 10.48550/arXiv.2501.04075 , https://ui.adsabs.harvard.edu/abs/2025arXiv250104075H p. arXiv:2501.04075

    2025 doi arXiv

  22. [30]

    P., Ba J., 2014, @doi [arXiv e-prints] 10.48550/arXiv.1412.6980 , https://ui.adsabs.harvard.edu/abs/2014arXiv1412.6980K p

    Kingma D. P., Ba J., 2014, @doi [arXiv e-prints] 10.48550/arXiv.1412.6980 , https://ui.adsabs.harvard.edu/abs/2014arXiv1412.6980K p. arXiv:1412.6980

    2014 doi arXiv

  23. [31]

    W., Bovy J., 2019, @doi [ ] 10.1093/mnras/sty3217 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.483.3255L 483, 3255

    Leung H. W., Bovy J., 2019, @doi [ ] 10.1093/mnras/sty3217 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.483.3255L 483, 3255

    2019 doi

  24. [32]

    Li X., Lin B., 2023, @doi [ ] 10.1093/mnras/stad831 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.521.6354L 521, 6354

    2023 doi

  25. [33]

    Li Z., Zhao G., Chen Y., Liang X., Zhao J., 2022, @doi [ ] 10.1093/mnras/stac1959 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.517.4875L 517, 4875

    2022 doi

  26. [34]

    Liu H., Du C., Ye D., Zhang J., Deng M., 2024, @doi [ ] 10.3847/1538-4357/ad8352 , https://ui.adsabs.harvard.edu/abs/2024ApJ...976..161L 976, 161

    2024 doi

  27. [35]

    L., et al., 2015, @doi [Research in Astronomy and Astrophysics] 10.1088/1674-4527/15/8/002 , https://ui.adsabs.harvard.edu/abs/2015RAA....15.1095L 15, 1095

    Luo A. L., et al., 2015, @doi [Research in Astronomy and Astrophysics] 10.1088/1674-4527/15/8/002 , https://ui.adsabs.harvard.edu/abs/2015RAA....15.1095L 15, 1095

    2015 doi

  28. [36]

    R., et al., 2017, @doi [ ] 10.3847/1538-3881/aa784d , https://ui.adsabs.harvard.edu/abs/2017AJ....154...94M 154, 94

    Majewski S. R., et al., 2017, @doi [ ] 10.3847/1538-3881/aa784d , https://ui.adsabs.harvard.edu/abs/2017AJ....154...94M 154, 94

    2017 doi

  29. [37]

    Martig M., et al., 2016, @doi [ ] 10.1093/mnras/stv2830 , https://ui.adsabs.harvard.edu/abs/2016MNRAS.456.3655M 456, 3655

    2016 doi

  30. [38]

    McWilliam A., Wallerstein G., Mottini M., 2013, @doi [ ] 10.1088/0004-637X/778/2/149 , https://ui.adsabs.harvard.edu/abs/2013ApJ...778..149M 778, 149

    2013 doi

  31. [39]

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

    Myeong G. C., Vasiliev E., Iorio G., Evans N. W., Belokurov V., 2019, @doi [ ] 10.1093/mnras/stz1770 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.488.1235M 488, 1235

    2019 doi

  32. [40]

    P., Conroy C., Bonaca A., Johnson B

    Naidu R. P., Conroy C., Bonaca A., Johnson B. D., Ting Y.-S., Caldwell N., Zaritsky D., Cargile P. A., 2020, @doi [ ] 10.3847/1538-4357/abaef4 , https://ui.adsabs.harvard.edu/abs/2020ApJ...901...48N 901, 48

    2020 doi

  33. [41]

    W., Rix H

    Ness M., Hogg D. W., Rix H. W., Ho A. Y. Q., Zasowski G., 2015, @doi [ ] 10.1088/0004-637X/808/1/16 , https://ui.adsabs.harvard.edu/abs/2015ApJ...808...16N 808, 16

    2015 doi

  34. [42]

    Pedregosa F., et al., 2011, Journal of Machine Learning Research, 12, 2825

    2011

  35. [43]

    Steinmetz M., et al., 2006, @doi [ ] 10.1086/506564 , https://ui.adsabs.harvard.edu/abs/2006AJ....132.1645S 132, 1645

    2006 doi

  36. [44]

    Sun T., et al., 2025, @doi [Nature Communications] 10.1038/s41467-025-56550-1 , https://ui.adsabs.harvard.edu/abs/2025NatCo..16.1581S 16, 1581

    2025 doi

  37. [45]

    X., Woosley S

    Timmes F. X., Woosley S. E., Weaver T. A., 1995, @doi [ ] 10.1086/192172 , https://ui.adsabs.harvard.edu/abs/1995ApJS...98..617T 98, 617

    1995 doi

  38. [46]

    Ting Y.-S., Conroy C., Rix H.-W., Cargile P., 2019, @doi [ ] 10.3847/1538-4357/ab2331 , https://ui.adsabs.harvard.edu/abs/2019ApJ...879...69T 879, 69

    2019 doi

  39. [47]

    arXiv:2501.08285

    Valdenegro-Toro M., Zullich M., 2025, @doi [arXiv e-prints] 10.48550/arXiv.2501.08285 , https://ui.adsabs.harvard.edu/abs/2025arXiv250108285V p. arXiv:2501.08285

    2025 doi arXiv

  40. [48]

    Vasini A., Spitoni E., Matteucci F., 2024, @doi [ ] 10.1051/0004-6361/202347603 , https://ui.adsabs.harvard.edu/abs/2024A&A...683A.121V 683, A121

    2024 doi

  41. [49]

    Wang R., et al., 2020, @doi [ ] 10.3847/1538-4357/ab6dea , https://ui.adsabs.harvard.edu/abs/2020ApJ...891...23W 891, 23

    2020 doi

  42. [50]

    Wang C., Huang Y., Yuan H., Zhang H., Xiang M., Liu X., 2022, @doi [ ] 10.3847/1538-4365/ac4df7 , https://ui.adsabs.harvard.edu/abs/2022ApJS..259...51W 259, 51

    2022 doi

  43. [51]

    L., Zhang S., Ting Y.-S., O'Briain T., Lamost Mrs Collaboration 2023, @doi [ ] 10.3847/1538-4365/acce36 , https://ui.adsabs.harvard.edu/abs/2023ApJS..266...40W 266, 40

    Wang R., Luo A. L., Zhang S., Ting Y.-S., O'Briain T., Lamost Mrs Collaboration 2023, @doi [ ] 10.3847/1538-4365/acce36 , https://ui.adsabs.harvard.edu/abs/2023ApJS..266...40W 266, 40

    2023 doi

  44. [52]

    306, Statistical Challenges in 21st Century Cosmology

    Wu Y., Du B., Luo A., Zhao Y., Yuan H., 2014, in Heavens A., Starck J.-L., Krone-Martins A., eds, IAU Symposium Vol. 306, Statistical Challenges in 21st Century Cosmology. pp 340--342 ( @eprint arXiv 1407.1980 ), @doi 10.1017/S1743921314010825

    2014 arXiv

  45. [53]

    Xiang M., et al., 2019, @doi [ ] 10.3847/1538-4365/ab5364 , https://ui.adsabs.harvard.edu/abs/2019ApJS..245...34X 245, 34

    2019 doi

  46. [54]

    Xiang G., Chen J., Qiu B., Lu Y., 2021, @doi [ ] 10.1088/1538-3873/abd997 , https://ui.adsabs.harvard.edu/abs/2021PASP..133b4504X 133, 024504

    2021 doi

  47. [55]

    Xie J., Bu Y., Liang J., Li H., Wang X., Pan J., 2021, @doi [ ] 10.3847/1538-3881/ac1c7c , https://ui.adsabs.harvard.edu/abs/2021AJ....162..155X 162, 155

    2021 doi

  48. [56]

    Yang C., et al., 2019, @doi [ ] 10.3847/1538-4357/ab2462 , https://ui.adsabs.harvard.edu/abs/2019ApJ...880...65Y 880, 65

    2019 doi

  49. [57]

    Yanny B., et al., 2009, @doi [ ] 10.1088/0004-6256/137/5/4377 , https://ui.adsabs.harvard.edu/abs/2009AJ....137.4377Y 137, 4377

    2009 doi

  50. [58]

    Q., Wang Q

    Zhang X., Zhao G., Yang C. Q., Wang Q. X., Zuo W. B., 2019, @doi [ ] 10.1088/1538-3873/ab2687 , https://ui.adsabs.harvard.edu/abs/2019PASP..131i4202Z 131, 094202

    2019 doi

  51. [59]

    Zhang B., Liu C., Deng L.-C., 2020, @doi [ ] 10.3847/1538-4365/ab55ef , https://ui.adsabs.harvard.edu/abs/2020ApJS..246....9Z 246, 9

    2020 doi

  52. [60]

    M., Mao S., 2001, @doi [ ] 10.1086/319832 , https://ui.adsabs.harvard.edu/abs/2001ApJ...551L..85Z 551, L85

    Zhao G., Qiu H. M., Mao S., 2001, @doi [ ] 10.1086/319832 , https://ui.adsabs.harvard.edu/abs/2001ApJ...551L..85Z 551, L85

    2001 doi

Pith tools

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