arxiv: 2604.26491 · v1 · submitted 2026-04-29 · 🌌 astro-ph.GA

Recognition: unknown

TwinSpecNet: Extending APOGEE's chemical reach to low-S/N spectra via empirical paired learning

Cristina Chiappini, Samir Nepal, Weijia Sun

Authors on Pith no claims yet

Pith reviewed 2026-05-07 13:20 UTC · model grok-4.3

classification 🌌 astro-ph.GA

keywords APOGEEstellar spectroscopymachine learningstellar parameterschemical abundancessignal-to-noise ratiopaired learning

0 comments

The pith

TwinSpecNet recovers precise stellar parameters and abundances from low signal-to-noise APOGEE spectra by training on paired observations of the same stars.

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

Large surveys like APOGEE collect many spectra at low signal-to-noise, which limits precise measurements of stellar properties for faint and distant stars. TwinSpecNet addresses this by using pairs of low- and high-quality spectra from repeated observations of identical stars to train a model that removes noise while matching the labels from standard high-quality processing. This approach allows extending reliable chemical information to populations in the bulge, outer halo, and satellites that would otherwise be excluded. It demonstrates that survey-specific noise patterns can be learned empirically from the data itself rather than modeled theoretically.

Core claim

TwinSpecNet reduces label scatter relative to visit-level processing for spectra with signal-to-noise below 60, reproducing the ASPCAP scale with residual scatters below 19 K in effective temperature, 0.06 dex in surface gravity, and 0.03 dex in iron abundance. It tightens abundance dispersions within star clusters, yields cleaner chemical sequences in disk, bulge and satellite samples, and improves the precision of ages derived from carbon-to-nitrogen ratios in giant stars from 1.70 to 1.59 Gyr.

What carries the argument

Vision Transformer encoder trained on empirical low- and high-signal-to-noise spectral pairs of the same stars, with objectives to reconstruct clean flux and predict parameters with uncertainties.

Load-bearing premise

High signal-to-noise spectra provide an unbiased ground truth for the star's true parameters and abundances, with visit-to-visit differences arising only from random noise.

What would settle it

Compare TSN labels on low-S/N spectra against independent measurements from a different high-resolution spectrograph for the same stars and check for systematic offsets.

Figures

Figures reproduced from arXiv: 2604.26491 by Cristina Chiappini, Samir Nepal, Weijia Sun.

**Figure 1.** Figure 1: Signal-to-noise distributions for the label-training sam view at source ↗

**Figure 2.** Figure 2: Schematic overview of the TwinSpecNet architecture. view at source ↗

**Figure 3.** Figure 3: Label recovery and uncertainty estimates on the test set. For each label we show the bias (green triangles), standard deviation view at source ↗

**Figure 4.** Figure 4: Comparison of TSN and ASPCAP parameters at low S view at source ↗

**Figure 6.** Figure 6: Intra-cluster abundance dispersion ratios for APOGEE view at source ↗

**Figure 5.** Figure 5: Comparison of TSN and ASPCAP abundances against view at source ↗

**Figure 7.** Figure 7: Inner-disk abundance-plane comparison at low S view at source ↗

**Figure 8.** Figure 8: Bulge abundance-plane comparison at low S view at source ↗

**Figure 9.** Figure 9: Satellite-system abundance patterns. Shown are [Mg view at source ↗

**Figure 10.** Figure 10: C/N-based ages versus asteroseismic ages. The horizontal axis shows APOKASC-3 asteroseismic ages (Pinsonneault et al. 2025), while the vertical axis shows C/N-based ages inferred from TSN (red) and ASPCAP (green) abundances for low-S/N stars. Ages are estimated from the empirical [C/N]–age calibration of Roberts et al. (2025), which provides a metallicitydependent mapping from post-dredge-up C/N ratios… view at source ↗

read the original abstract

Large spectroscopic surveys rely on automated pipelines to deliver homogeneous stellar labels, but a substantial fraction of observations are at low signal-to-noise ratio (S/N), where label estimates become imprecise or are omitted. In APOGEE, these low-S/N spectra visits sample faint and distant populations -- the bulge, outer halo, and satellite systems -- yet still encode recoverable chemical information. We present TwinSpecNet (TSN), a paired-learning framework that exploits APOGEE's multi-visit observing strategy: by training on empirical low-/high-S/N spectral twins of the same stars, TSN learns to suppress stochastic noise while preserving the ASPCAP label scale. TSN employs a Vision Transformer encoder with dual objectives: reconstructing high-S/N flux from low-S/N visits and predicting stellar parameters and abundances with calibrated uncertainties. TSN reduces label scatter relative to visit-level ASPCAP for S/N<60 visits. TSN reproduces the ASPCAP scale with residual scatters of $\sigma$< 19 K in $T_{\mathrm{eff}}$, $\sigma\sim$0.06 dex in $\log g$, and $\sigma\sim$0.03 dex in Fe/H. TSN tightens intra-cluster abundance dispersions, recovers cleaner chemical sequences in inner-disk and bulge and satellite samples, and improves C/N-based age precision for APOKASC giants from 1.70 to 1.59 Gyr. By learning survey-specific noise patterns from repeated observations, TSN demonstrates how empirical paired learning can extend the chemical reach of existing spectroscopic data, providing a template applicable to other multi-visit surveys.

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

TwinSpecNet uses real paired APOGEE visits to train a ViT for denoising and label recovery on low-S/N spectra, with reported gains that look usable but rest on the untested claim that high-S/N labels are always the clean ground truth.

read the letter

The paper's core contribution is an empirical paired-learning setup that trains on actual low-S/N and high-S/N visits of the same stars instead of simulations. A Vision Transformer handles both flux reconstruction and label prediction in one model, and the authors report concrete reductions in scatter for S/N below 60: under 19 K in Teff, 0.06 dex in log g, and 0.03 dex in Fe/H, plus tighter cluster dispersions and a modest improvement in C/N age precision from 1.70 to 1.59 Gyr. That is the practical advance for people who want to keep more faint bulge or halo stars in their samples without throwing them out at the pipeline stage. The approach is straightforward and makes good use of the survey's repeat observations, which is a real strength over purely synthetic training routes. The numbers are given against the existing ASPCAP scale, so the method at least stays consistent with what the community already uses. The main limitation is that performance is measured against the high-S/N labels themselves, so the test does not independently verify whether visit-to-visit differences are only noise or include small systematics from calibration, fiber response, or mild variability. The intra-cluster dispersion tightening is helpful supporting evidence but still indirect. If those systematics exist at the level of the quoted residuals, the network could suppress real signals or learn spurious corrections. The paper would benefit from explicit checks on paired-visit consistency beyond the main results. This work is aimed at galactic archaeology groups already using APOGEE or similar multi-visit surveys who need to push fainter. It is incremental rather than transformative, but the problem it tackles is common enough that a careful referee could improve the validation and make the claims more robust. I would send it to peer review.

Referee Report

2 major / 2 minor

Summary. The manuscript presents TwinSpecNet (TSN), a Vision Transformer encoder trained via paired empirical learning on APOGEE multi-visit spectra. It maps low-S/N visits to high-S/N flux reconstructions and ASPCAP-derived stellar parameters/abundances, claiming reduced label scatter for S/N<60 visits relative to standard ASPCAP processing, with residual scatters of σ<19 K in Teff, σ∼0.06 dex in log g, and σ∼0.03 dex in Fe/H. Additional claims include tightened intra-cluster abundance dispersions, cleaner chemical sequences in bulge/satellite samples, and improved C/N-based age precision (1.70 to 1.59 Gyr) for APOKASC giants.

Significance. If the central results are robust, the work offers a practical template for recovering chemical information from the substantial fraction of low-S/N observations in large surveys like APOGEE, directly benefiting studies of faint bulge, halo, and satellite populations. The empirical paired-learning strategy is a notable strength, as it learns survey-specific noise patterns without synthetic spectra and supplies calibrated uncertainties; this could be broadly applicable to other multi-visit spectroscopic datasets.

major comments (2)

[Abstract and validation/results sections] The central claim that TSN reduces label scatter while preserving the ASPCAP scale rests on the assumption that high-S/N ASPCAP labels are unbiased ground truth and that differences between paired visits of the same star are purely stochastic noise (no systematic offsets from tellurics, fiber response, or variability). However, the reported residual scatters are measured directly against these high-S/N labels, providing no independent validation of the assumption. Intra-cluster dispersion tightening is supporting but indirect evidence only. Explicit tests for visit-to-visit systematics are required in the validation section to substantiate the claims.
[Methods section] The manuscript provides insufficient detail on data selection criteria for twin pairs, cross-validation strategy, handling of potential label biases in ASPCAP, and architecture/training hyperparameters. Without these, it is not possible to confirm that the numerical improvements (e.g., the quoted residual scatters and age precision gain) are free of selection effects or overfitting.

minor comments (2)

[Abstract] Notation for uncertainties (e.g., σ∼0.03 dex) should be made consistent with the more precise bounds given for Teff; clarify whether these are 1σ or other quantiles.
[Figures] Figure captions and axis labels should explicitly state the S/N range and sample size for each comparison panel to aid immediate interpretation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed comments, which have helped us strengthen the validation and methodological transparency of the manuscript. We address each major comment below and have revised the paper accordingly.

read point-by-point responses

Referee: [Abstract and validation/results sections] The central claim that TSN reduces label scatter while preserving the ASPCAP scale rests on the assumption that high-S/N ASPCAP labels are unbiased ground truth and that differences between paired visits of the same star are purely stochastic noise (no systematic offsets from tellurics, fiber response, or variability). However, the reported residual scatters are measured directly against these high-S/N labels, providing no independent validation of the assumption. Intra-cluster dispersion tightening is supporting but indirect evidence only. Explicit tests for visit-to-visit systematics are required in the validation section to substantiate the claims.

Authors: We agree that the primary validation relies on high-S/N ASPCAP labels and that explicit checks for visit-to-visit systematics would strengthen the claims. While intra-cluster dispersion tightening offers supporting evidence, we acknowledge it is indirect. In the revised manuscript, we have added a new subsection to the validation section that performs explicit tests: residuals between TSN predictions and high-S/N labels are examined as functions of fiber ID, airmass, telluric absorption strength, and available stellar variability indicators. No significant correlations are found, consistent with predominantly stochastic differences. We also added a brief discussion of ASPCAP limitations and how the empirical paired approach mitigates certain biases. These changes directly address the concern. revision: yes
Referee: [Methods section] The manuscript provides insufficient detail on data selection criteria for twin pairs, cross-validation strategy, handling of potential label biases in ASPCAP, and architecture/training hyperparameters. Without these, it is not possible to confirm that the numerical improvements (e.g., the quoted residual scatters and age precision gain) are free of selection effects or overfitting.

Authors: We appreciate the referee highlighting the need for greater detail. The revised Methods section now includes: (i) explicit twin-pair selection criteria (minimum S/N contrast, shared APOGEE ID, exclusion of flagged visits); (ii) star-level k-fold cross-validation to avoid leakage; (iii) a subsection on ASPCAP label biases, noting known pipeline systematics and how empirical learning from real pairs reduces sensitivity to them; (iv) full ViT architecture specifications (layers, heads, dimensions) and all training hyperparameters (optimizer, rates, batch size, epochs, loss weights). An appendix with ablation studies on key hyperparameters has also been added to confirm robustness against overfitting and selection effects. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical supervised learning on independent high-S/N labels

full rationale

The paper describes a supervised Vision Transformer trained to map low-S/N spectra to high-S/N flux reconstructions and to ASPCAP stellar labels obtained from paired high-S/N visits of the same stars. Reported residual scatters and improvements in intra-cluster dispersions are measured on held-out data against those independent high-S/N ASPCAP labels, which are not constructed from the low-S/N inputs or from the network itself. No equation or claim reduces by definition to a fitted parameter that is then relabeled as a prediction, and no load-bearing step relies on a self-citation chain or uniqueness theorem imported from the authors' prior work. The method is therefore self-contained as an empirical denoising procedure whose performance claims rest on generalization to unseen visits rather than on any internal redefinition of its targets.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Based solely on the abstract, the central claim rests on the domain assumption that paired visits differ only by stochastic noise and on the empirical training data itself. No new physical entities are postulated. The neural network contains many free parameters typical of deep learning models, but none are enumerated in the abstract.

free parameters (1)

neural network hyperparameters and architecture choices
Standard for Vision Transformer models; optimized during training but not specified in the abstract.

axioms (1)

domain assumption High-S/N spectra of the same star serve as unbiased reference for low-S/N visits
Invoked by the paired-learning setup that treats differences as pure noise.

pith-pipeline@v0.9.0 · 5600 in / 1330 out tokens · 218388 ms · 2026-05-07T13:20:56.743557+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

92 extracted references · 10 canonical work pages · 3 internal anchors

[1]

F., Argudo-Fern \'a ndez, M., et al.\ 2023, , 267, 2, 44

Almeida, A., Anderson, S. F., Argudo-Fern \'a ndez, M., et al.\ 2023, , 267, 2, 44

2023
[2]

Anders, F., Chiappini, C., Minchev, I., et al.\ 2017, , 600, A70

2017
[3]

Anders, F., Gispert, P., Ratcliffe, B., et al.\ 2023, , 678, A158

2023
[4]

H., Weinberg, D

Andrews, B. H., Weinberg, D. H., Sch \"o nrich, R., et al.\ 2017, , 835, 2, 224

2017
[5]

Baba, J.\ 2025, , 77, 4, 916

2025
[6]

Bensby, T., Feltzing, S., Gould, A., et al.\ 2017, , 605, A89

2017
[7]

& Vasiliev, E.\ 2024, , 527, 2, 1915

Binney, J. & Vasiliev, E.\ 2024, , 527, 2, 1915

2024
[8]

D., et al.\ 2025, , 994, 1, 126

Borbolato, L., Rossi, S., Perottoni, H. D., et al.\ 2025, , 994, 1, 126

2025
[9]

C., Aguirre B rsen-Koch, V., Helmi, A., et al.\ 2022, , 514, 2, 2527

Borre, C. C., Aguirre B rsen-Koch, V., Helmi, A., et al.\ 2022, , 514, 2, 2527

2022
[10]

J., et al.\ 2016, , 455, 1, 987

Casagrande, L., Silva Aguirre, V., Schlesinger, K. J., et al.\ 2016, , 455, 1, 987

2016
[11]

Casali, G., Montalb \'a n, J., Miglio, A., et al.\ 2025, , 541, 3, 2631

2025
[12]

J., Serenelli, A

Chaplin, W. J., Serenelli, A. M., Miglio, A., et al.\ 2020, Nature Astronomy, 4, 382

2020
[13]

Ciuc a , I., Kawata, D., Miglio, A., et al.\ 2021, , 503, 2, 2814

2021
[14]

C., et al.\ 2014, , 9147, 91470L

Dalton, G., Trager, S., Abrams, D. C., et al.\ 2014, , 9147, 91470L

2014
[15]

Das, P., Hawkins, K., & Jofr \'e , P.\ 2020, , 493, 4, 5195

2020
[16]

S., Agertz, O., Berbel, A

de Jong, R. S., Agertz, O., Berbel, A. A., et al.\ 2019, The Messenger, 175, 3

2019
[17]

M., Freeman, K

De Silva, G. M., Freeman, K. C., Bland-Hawthorn, J., et al.\ 2015, , 449, 3, 2604

2015
[18]

M., Sneden, C., Paulson, D

De Silva, G. M., Sneden, C., Paulson, D. B., et al.\ 2006, , 131, 1, 455

2006
[19]

Dosovitskiy, A., Beyer, L., Kolesnikov, A., et al.\ 2020, , arXiv:2010.11929

work page internal anchor Pith review arXiv 2020
[20]

Ernandes, H., Barbuy, B., Chiappini, C., et al.\ 2026, arXiv:2602.12415

work page arXiv 2026
[21]

A., O'Briain, T., et al.\ 2018, , 475, 3, 2978

Fabbro, S., Venn, K. A., O'Briain, T., et al.\ 2018, , 475, 3, 2978

2018
[22]

& Chiba, M.\ 2013, , 57, 3-4, 80

Feltzing, S. & Chiba, M.\ 2013, , 57, 3-4, 80

2013
[23]

C., Horta, D., et al.\ 2023, , 519, 3, 3611

Fernandes, L., Mason, A. C., Horta, D., et al.\ 2023, , 519, 3, 3611

2023
[24]

Gaia Collaboration, Vallenari, A., Brown, A. G. A., et al.\ 2023, , 674, A1

2023
[25]

E., Allende Prieto, C., Holtzman, J

Garc \' a P \'e rez, A. E., Allende Prieto, C., Holtzman, J. A., et al.\ 2016, , 151, 6, 144

2016
[26]

H., Johnson, J

Griffith, E., Weinberg, D. H., Johnson, J. A., et al.\ 2021, , 909, 1, 77

2021
[27]

Grisoni, V., Chiappini, C., Miglio, A., et al.\ 2024, , 683, A111

2024
[28]

K., Zinn, J

Grunblatt, S. K., Zinn, J. C., Price-Whelan, A. M., et al.\ 2021, , 916, 2, 88

2021
[29]

Guiglion, G., Nepal, S., Chiappini, C., et al.\ 2024, , 682, A9

2024
[30]

J., Davies, G

Hall, O. J., Davies, G. R., van Saders, J., et al.\ 2021, Nature Astronomy, 5, 707

2021
[31]

R., Lian, J., et al.\ 2021, , 923, 2, 172

Hasselquist, S., Hayes, C. R., Lian, J., et al.\ 2021, , 923, 2, 172

2021
[32]

Hasselquist, S., Shetrone, M., Smith, V., et al.\ 2017, , 845, 2, 162

2017
[33]

A., Shetrone, M., Johnson, J

Holtzman, J. A., Shetrone, M., Johnson, J. A., et al.\ 2015, , 150, 5, 148

2015
[34]

P., Mackereth, J

Horta, D., Schiavon, R. P., Mackereth, J. T., et al.\ 2023, , 520, 4, 5671

2023
[35]

Hunt, E. L. & Reffert, S.\ 2024, , 686, A42

2024
[36]

A., et al.\ 2023, , 954, 2, 124

Imig, J., Price, C., Holtzman, J. A., et al.\ 2023, , 954, 2, 124

2023
[37]

Jofr \'e , P., Heiter, U., & Soubiran, C.\ 2019, , 57, 571

2019
[38]

A., Allende Prieto, C., et al.\ 2020, , 160, 3, 120

J \"o nsson, H., Holtzman, J. A., Allende Prieto, C., et al.\ 2020, , 160, 3, 120

2020
[39]

What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?

Kendall, A. & Gal, Y.\ 2017, , arXiv:1703.04977

work page Pith review arXiv 2017
[40]

SDSS-V: Pioneering Panoptic Spectroscopy

Kollmeier, J. A., Zasowski, G., Rix, H.-W., et al.\ 2017, , arXiv:1711.03234

work page internal anchor Pith review arXiv 2017
[41]

Lagarde, N., Reyl \'e , C., Chiappini, C., et al.\ 2021, , 654, A13

2021
[42]

Lehtinen, J., Munkberg, J., Hasselgren, J., et al.\ 2018, , arXiv:1803.04189

work page Pith review arXiv 2018
[43]

Leung, H. W. & Bovy, J.\ 2024, , 527, 1, 1494

2024
[44]

W., Bovy, J., Mackereth, J

Leung, H. W., Bovy, J., Mackereth, J. T., et al.\ 2023, , 522, 3, 4577

2023
[45]

Li, J., Wong, K. W. K., Hogg, D. W., et al.\ 2024, , 272, 1, 2

2024
[46]

T., Bovy, J., Leung, H

Mackereth, J. T., Bovy, J., Leung, H. W., et al.\ 2019, , 489, 1, 176

2019
[47]

R., Schiavon, R

Majewski, S. R., Schiavon, R. P., Frinchaboy, P. M., et al.\ 2017, , 154, 3, 94

2017
[48]

Martig, M., Fouesneau, M., Rix, H.-W., et al.\ 2016, , 456, 4, 3655

2016
[49]

& Gilmore, G.\ 2015, , 453, 2, 1855

Masseron, T. & Gilmore, G.\ 2015, , 453, 2, 1855

2015
[50]

Matsuno, T., Aoki, W., Casagrande, L., et al.\ 2021, , 912, 1, 72

2021
[51]

T., et al.\ 2021, , 645, A85

Miglio, A., Chiappini, C., Mackereth, J. T., et al.\ 2021, , 645, A85

2021
[52]

Miglio, A., Chiappini, C., Morel, T., et al.\ 2013, European Physical Journal Web of Conferences, 43, 03004

2013
[53]

T., Miglio, A., et al.\ 2021, Nature Astronomy, 5, 640

Montalb \'a n, J., Mackereth, J. T., Miglio, A., et al.\ 2021, Nature Astronomy, 5, 640

2021
[54]

C., et al.\ 2025, , 704, A44

Montelius, M., Starkenburg, E., Woudenberg, H. C., et al.\ 2025, , 704, A44

2025
[55]

L., Montalb \'a n, J., et al.\ 2021, , 646, A78

Morel, T., Creevey, O. L., Montalb \'a n, J., et al.\ 2021, , 646, A78

2021
[56]

M., Delgado-Mena, E., et al.\ 2022, , 660, A15

Moya, A., Sarro, L. M., Delgado-Mena, E., et al.\ 2022, , 660, A15

2022
[57]

Mucciarelli, A., Minelli, A., Bellazzini, M., et al.\ 2023, , 671, A124

2023
[58]

Nepal, S., Chiappini, C., P \'e rez-Villegas, A., et al.\ 2025, , arXiv:2507.06863

work page arXiv 2025
[59]

B., et al.\ 2024, , 688, A167

Nepal, S., Chiappini, C., Queiroz, A. B., et al.\ 2024, , 688, A167

2024
[60]

W., Rix, H.-W., et al.\ 2015, , 808, 1, 16

Ness, M., Hogg, D. W., Rix, H.-W., et al.\ 2015, , 808, 1, 16

2015
[61]

W., Rix, H.-W., et al.\ 2016, , 823, 2, 114

Ness, M., Hogg, D. W., Rix, H.-W., et al.\ 2016, , 823, 2, 114

2016
[62]

L., Hasselquist, S., Hayes, C

Nidever, D. L., Hasselquist, S., Hayes, C. R., et al.\ 2020, , 895, 2, 88

2020
[63]

H., Zinn, J

Pinsonneault, M. H., Zinn, J. C., Tayar, J., et al.\ 2025, , 276, 2, 69

2025
[64]

Queiroz, A. B. A., Chiappini, C., Perez-Villegas, A., et al.\ 2021, , 656, A156

2021
[65]

C., Casey, A

Reggiani, H., Schlaufman, K. C., Casey, A. R., et al.\ 2021, , 162, 6, 229

2021
[66]

M., Miglio, A., Chiappini, C., et al.\ 2019, , 490, 4, 4465

Rendle, B. M., Miglio, A., Chiappini, C., et al.\ 2019, , 490, 4, 4465

2019
[67]

D., Pinsonneault, M

Roberts, J. D., Pinsonneault, M. H., Johnson, J. A., et al.\ 2025, , arXiv:2509.25321

work page arXiv 2025
[68]

Salaris, M., Cassisi, S., & Weiss, A.\ 2002, , 114, 794, 375

2002
[69]

C., Stassun, K

Schonhut-Stasik, J., Zinn, J. C., Stassun, K. G., et al.\ 2024, , 167, 2, 50

2024
[70]

Sharma, S., Stello, D., Bland-Hawthorn, J., et al.\ 2019, , 490, 4, 5335

2019
[71]

Silva Aguirre, V., Bojsen-Hansen, M., Slumstrup, D., et al.\ 2018, , 475, 4, 5487

2018
[72]

V., Bizyaev, D., Cunha, K., et al.\ 2021, , 161, 6, 254

Smith, V. V., Bizyaev, D., Cunha, K., et al.\ 2021, , 161, 6, 254

2021
[73]

Stokholm, A., Aguirre B rsen-Koch, V., Stello, D., et al.\ 2023, , 524, 2, 1634

2023
[74]

Suda, T., Katsuta, Y., Yamada, S., et al.\ 2008, , 60, 1159

2008
[75]

Sundararajan, M., Taly, A., & Yan, Q.\ 2017, , arXiv:1703.01365

work page arXiv 2017
[76]

e , G., Drazdauskas, A., Mikolaitis, S ., et al.\ 2025, , 703, A4

Tautvai s ien \. e , G., Drazdauskas, A., Mikolaitis, S ., et al.\ 2025, , 703, A4

2025
[77]

Ting, Y.-S.\ 2025, The Open Journal of Astrophysics, 8, 95

2025
[78]

Ting, Y.-S., Conroy, C., Rix, H.-W., et al.\ 2019, , 879, 2, 69

2019
[79]

Valentini, M., Chiappini, C., Bossini, D., et al.\ 2019, , 627, A173

2019
[80]

G., et al.\ 2024, , 685, A150

Valle, G., Dell'Omodarme, M., Prada Moroni, P. G., et al.\ 2024, , 685, A150

2024

Showing first 80 references.