Pith. sign in

REVIEW 3 major objections 6 minor 86 references

Efficient Search for Extremely Metal Poor Galaxies in the Local Universe using Convolutional Neural Networks

T0 review · 3 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a three-stage CNN pipeline can find extremely metal-poor galaxies directly from SDSS images, with all 45 spectroscopically followed-up candidates confirmed as metal-poor.

desk verdict Genuinely new CNN pipeline for XMP discovery with strong extreme-tail validation, but the precision claims for the full candidate list go beyond the evidence. read the letter →

arxiv 2502.01638 v1 pith:JTJHHB5Y submitted 2025-02-03 astro-ph.GA

Ting-Yun Cheng , Ryan J. Cooke This is my paper · ORCID
classification astro-ph.GA
keywords extremelymetal-poorgalaxiesconvolutionalneuralnetworksN2indexSDSSimaginggalaxymetallicitydwarfmachinelearninginastronomyemission-linespectroscopy
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

Extremely metal-poor galaxies are rare local laboratories for studying primitive galaxy formation, but finding them usually requires spectroscopy or human colour selection. This paper proposes a convolutional neural network pipeline that works directly from five-band SDSS images and predicts each galaxy's $N_2$ index, a spectroscopic proxy for oxygen abundance. The pipeline first filters for metal-poor galaxies, then for extremely metal-poor galaxies, then predicts $N_2$; applied to 7.7 million SDSS galaxies without spectra, it produced 232,954 candidates. The team observed the 45 most extreme candidates with two telescopes, and all 45 were confirmed metal-poor, 28 of them for the first time. If the confirmation rate holds beyond the observed tail, the method turns broadband survey images into a fast discovery engine for the most primitive galaxies.

What carries the argument

The load-bearing object is the $N_2$ index, $N_2 \equiv \log([\mathrm{N\,II}]\,\lambda6585/\mathrm{H}\alpha)$, used as a metallicity proxy, together with a three-stage CNN pipeline that predicts it from 32$\times$32-pixel, five-band SDSS cutouts normalized relative to the $r$-band. Stage (i) classifies metal-poor galaxies ($N_2 \le -1.0$); stage (ii) classifies extremely metal-poor galaxies ($N_2 \le -1.5$) among predicted MPs; stage (iii) predicts the $N_2$ value for those candidates, with extra loss weight on the most metal-poor systems ($N_2 \le -2.1$). Each stage is an ensemble of nine CNNs (three training splits times three initializations), combined by the median prediction, and the final $N_2$ prediction is used to rank candidates for follow-up.

What would settle it

Take a random subsample of candidates from the full 232,954 list with classification probabilities between 0.5 and 0.99 and predicted $N_2$ between $-1.8$ and $-1.5$, obtain spectra, and compare the metallic confirmation fraction with the roughly 96--99 percent precision reported for the training set; a large drop would show the pipeline's search precision holds only for its brightest, most extreme predictions.

Watch

Extended reading notes

Core claim

The central claim is that a sequential CNN pipeline can select extremely metal-poor galaxies from broadband imaging with a high confirmation rate at the most extreme prediction tail. From over seven million SDSS DR17 galaxies without spectroscopy, the pipeline selected 232,954 XMP candidates; restricting to the highest-probability, lowest-$N_2$ subset left 390 targets, of which 45 were observed spectroscopically with INT and SOAR. Spectroscopy confirmed all 45 as metal-poor---28 are new discoveries---and the predicted $N_2$ values agree with observed values to a mean absolute error of about 0.16 dex for secure detections. Derived oxygen abundances span $7.1 \le 12+\log(\mathrm{O/H}) \le 8.7$ (2$\sigma$ upper limits), with 21 systems below roughly $0.1\,Z_\odot$, and four objects may host low-metallicity AGNs.

Load-bearing premise

The confirmation rate measured on the 45 brightest, most extreme candidates is assumed to carry over to the full 232,954-candidate list, including galaxies fainter than the training data and outside the known colour-morphology locus; the observed sample does not test that transfer.

Editorial extensions

If this is right

  • The pipeline identifies 232,954 XMP candidates among 7.7 million SDSS DR17 galaxies that lack spectroscopy, demonstrating that imaging alone can rank galaxies by likely metallicity.
  • The 45 observed candidates, all drawn from the most extreme predicted tail, are all confirmed metal-poor, including 28 new discoveries and 36 first $N_2$ measurements.
  • Agreement between predicted and observed $N_2$ at the 0.16 dex level means the network's metallicity ranking can be trusted well enough to prioritize telescope time.
  • 18 of 45 galaxies lack detectable [N ii], so their metallicities are only upper limits; deeper follow-up is needed to know how many of these approach record-low abundances.
  • The observed sample's $g$-band brightness and blueberry-like colours suggest the population overlaps low-redshift analogues of green pea and Ly$\alpha$-emitting galaxies.

Reading between the lines

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

  • Because the 45 confirmations all lie at the extreme tail ($P>0.99$, $N_2<-1.8$), the pipeline's precision for the full 232,954-candidate working list remains unmeasured; a representative follow-up sample at lower thresholds would test whether the 'efficient search' claim extends beyond the tail.
  • The observed outliers whose true $N_2$ is lower than predicted have ordinary colours and morphologies; if real, they point to galaxies whose nitrogen content is decoupled from broadband appearance, possibly through outflows, gas stripping, or a different nitrogen production channel---a physical discovery that the CNN made accessible.
  • The same architecture could be retrained on other wide-area surveys or on direct oxygen-abundance labels, and the sequential classifier-plus-regressor design could be reused for other rare astrophysical classes where labelled examples are sparse.
  • A concrete extension: apply the pipeline to the fainter objects (up to $r\approx22$) that were included but not observed, and to galaxies just outside the colour selection box, to map how precision degrades with signal-to-noise and colour.
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 / 6 minor

Summary. The paper presents a machine-learning pipeline for identifying extremely metal-poor (XMP) galaxies from SDSS multi-band imaging. The pipeline consists of three CNNs: a metal-poor classifier, an XMP classifier, and an N2-index predictor. It is applied to 7.7 million SDSS galaxies without spectroscopy, yielding 232,954 XMP candidates with PMP>0.5 and PXMP>0.5, from which a visually cleaned subset of 390 with PMP>0.99, PXMP>0.99, and N2<-1.8 was selected. New INT and SOAR spectroscopy of 45 of these candidates confirms all 45 to be metal-poor (N2<-1.0), including 28 new discoveries; the predicted and observed N2 values agree with an MAE of 0.16 dex. The paper also reports oxygen abundances, SFRs, stellar masses, and a first look at possible low-metallicity AGN candidates.

Significance. If the reported confirmation rate holds over the entire candidate list, the pipeline would be a significant tool for discovering local extremely metal-poor galaxies, which are important for studying primitive galaxy formation and primordial nucleosynthesis. The 45/45 confirmation rate on the most extreme tail is a strong, independently obtained result, and the paper demonstrates a novel architecture (sequential classifier-predictor CNNs) that goes beyond earlier photometric-magnitude-based machine learning searches such as Kojima et al. (2020). The new spectroscopic sample of 45 XMPs, including 28 new discoveries, is itself a valuable contribution. However, the paper's central claim of an 'efficient search' at scale rests on precision estimates for a candidate list that has not been externally validated outside the extreme tail, and the N2 predictor accuracy is first reported on training data. These issues need to be addressed before the pipeline can be adopted as a calibrated discovery engine.

major comments (3)
  1. [§2.3, §4.1, §6] The 232,954-candidate list is defined by PMP>0.5 and PXMP>0.5 (Section 6) and includes galaxies up to 2 mag fainter than the training set (Section 2.3), but the only external validation is the 45 observed candidates, all selected with PMP>0.99, PXMP>0.99, N2<-1.8 (Section 4.1). The 45/45 confirmation therefore validates the extreme tail at high probability thresholds, not the full candidate list. The internal precision values (96% XMP precision, 99% MP precision in the sequential process, Section 3.2) are computed on the training/testing distribution and do not account for the magnitude and selection shift of the working sample. The summary statement that 'There are 232,954 XMP candidates' is consequently not supported by the evidence presented. Either additional validation spanning the full threshold and magnitude range should be provided, or the paper should explicitly state that the candidate list is an uncalibrated product and restrict the claimed confirmation rate to the observed tail.
  2. [§3.1, Fig. 2, §5.2] The N2 predictor is reported to have RMSD 0.031 dex and MAD 0.015 dex, but these are computed on the full training+testing set, as stated in Section 3.1 ('...their whole assigned samples (including training and testing sets)'). This is not an independent measure of predictive accuracy; it reflects memorization of training data. The independent estimate from the new observations is the MAE 0.16 dex in Section 5.2, which should be the headline scatter for the N2 predictor. The paper should also state how many of the 45 observed targets lie outside the 0.16 dex band in Fig. 5, and investigate whether the outliers are preferentially at faint magnitudes or extreme N2 values, since this determines whether the predictor is reliable in the extrapolated regime that motivates the survey.
  3. [§5.2] The interpretation of the lower outliers (observed N2 lower than predicted) as evidence for gas outflows or different nitrogen production routes is not uniquely supported by the data. The same pattern could be produced by systematic errors in the continuum subtraction or flux calibration for the faintest targets, or by the N2 predictor's extrapolation to fainter galaxies. The paper should at least test for correlations between the residuals and galaxy brightness, Hα S/N, or redshift, and should soften the physical interpretation until the proposed [O ii] follow-up is available.
minor comments (6)
  1. [Table 3] The caption states the order is 'sorted based on the predicted N2 values', but the text before the table says the order is by right ascension; the entries appear to be in RA order. Please make the captions consistent.
  2. [§2.2] The bin notation '[−2.1,)' is incomplete; use 'N2 ≤ −2.1' or '[−2.6, −2.1]'.
  3. [§5.3.3] The phrase 'divided the value by a factor of correct the "diet" Salpeter IMF' is ungrammatical; rephrase to 'divided the value by a factor to correct the "diet" Salpeter IMF to a Chabrier (2003) IMF'.
  4. [Table 3] The reference list in Table 3 contains 'W ang et al. (2018)' with an unintended space; it should read 'Wang et al. (2018)'.
  5. [§6] The star formation rate range '10−3.9−10−0.035 M⊙/yr' should be rendered with proper superscripts and a clear separator, e.g., '10^-3.9 to 10^-0.035 M⊙/yr'.
  6. [Data Availability] The statement that catalogues 'will be available upon the publication of this manuscript' is vague; please specify a permanent archive or repository, and state whether the trained CNN models will be released to facilitate reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the CNN validation rests on new, independent spectra rather than on training labels.

full rationale

The derivation chain is not circular. The CNN models are trained on SDSS MPA-JHU N2 labels and literature metallicities (Section 2.1), and the central validation is a set of 45 targets whose [N ii] λ6585/Hα ratios were measured from new INT and SOAR spectra (Sections 4.2–5.2). These observed N2 values are not derived from the CNN predictions, so the '45/45 confirmed metal-poor' result is an external check rather than an echo of training labels. The oxygen abundances use the literature Y07 calibration (Eq. 2), which is independent of the pipeline. The extreme-tail selection (P>0.99, N2<−1.8) means the high confirmation rate applies only to the most extreme predicted candidates; that is a scope limitation, not circularity. The in-sample RMSD of 0.031 dex reported in Section 3.1 is computed including training data and is not a valid precision estimate, but it is not a by-construction reduction: the learned mapping is not identical to its labels. Self-citations (Cheng et al. 2020, 2021, 2023; Cooke et al. 2014) support data augmentation, extrapolation claims, and spectral fitting, but none of these carries the load of the central observational validation.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central claim rests on standard assumptions about the N2 metallicity proxy, the fidelity of SDSS/MPA-JHU training labels, and the generalization of CNN predictions to fainter, non-spectroscopic galaxies. No new physical entities are introduced.

free parameters (5)
  • Observation selection thresholds = P_MP > 0.99, P_XMP > 0.99, predicted N2 < -1.8
    Hand-chosen thresholds in Section 4.1 used to select 390 candidates from 232,954; the 45 observed targets are drawn from this extreme tail, so the validation result depends on these cuts.
  • Classifier decision thresholds = P > 0.5 for both classifiers
    Used in Section 3.2 to define candidate lists in the sequential process; affects the number of candidates and the reported precision.
  • Color query cuts = mag_r 16-22, (u-g)<=1.7, (g-r)<=0.6, etc.
    Table 1 query criteria applied to the working sample; cover >99% of MP training samples but may exclude potential XMPs outside these colors.
  • N2 binning and loss weight = 11 bins of 0.1 dex plus last bin N2<=-2.1; loss weight 3x for N2<=-2.1
    Data augmentation and loss weighting choices in Section 3.1 that affect the N2 predictor's training and its behavior on extreme values.
  • CNN hyperparameters = Learning rates 1e-4/4e-4, channels 16-256, kernels 3-7 (Table 2)
    Optimized per model via Bayesian optimization; these fitted parameters determine the model outputs but are not the scientific claim itself.
assumptions (6)
  • domain assumption The N2 index is a reliable proxy for gas-phase oxygen abundance.
    Used throughout (Sections 2, 5.3.1) to define MP/XMP and derive 12+log(O/H) via the Y07 calibration (Eq. 2).
  • domain assumption MPA-JHU N2 measurements and literature N2 values are accurate and unbiased.
    Training labels in Section 2.1 come from these catalogs; any systematic error propagates to the CNN and the abundance estimates.
  • ad hoc to paper The r-band flux normalization in Eq. 1 preserves per-pixel color information needed for N2 prediction.
    The preprocessing choice in Section 2 is not derived from physical principles and could suppress faint features.
  • domain assumption The CNN trained on SDSS spectroscopy generalizes to the 7.7 million galaxies without spectroscopy, including objects up to 2 mag fainter than the training set.
    The working sample in Section 2.3 extends beyond the training distribution; the paper cites Cheng et al. (2021, 2023) for extrapolation ability.
  • domain assumption The color query criteria in Table 1 do not bias the XMP candidate sample.
    The paper states the criteria cover >99% of MP samples and only use upper limits, but this is not independently verified for the working sample.
  • domain assumption The Y07 calibration (Eq. 2) remains valid for N2 values below -1.8.
    Used in Section 5.3.1 to report oxygen abundances; the calibration was not derived for the most extreme metallicities and has 0.159 dex scatter.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Search for Extremely Metal Poor Galaxies in the Local Universe using Convolutional Neural Networks." pith.science (2026). https://pith.science/paper/JTJHHB5Y

@misc{pith2026250201638,
  author       = {Pith},
  title        = {Pith review of: Efficient Search for Extremely Metal Poor Galaxies in the Local Universe using Convolutional Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JTJHHB5Y}},
  note         = {Machine review of arXiv:2502.01638}
}
abstract

Nearby extremely metal-poor galaxies (XMPs) allow us to study primitive galaxy formation and evolution in greater detail than is possible at high redshift. This work, for the first time, promotes the use of convolutional neural networks (CNNs) to efficiently search for XMPs in multi-band imaging data based on their predicted N2 index (N2\,$\equiv\log$\{\rNii/\Ha\}). We developed a sequential characterisation pipeline, composed of three CNN procedures: (i) a classifier for metal-poor galaxies, (ii) a classifier for XMPs, and (iii) an N2 predictor. The pipeline is applied to over 7.7 million SDSS DR17 imaging data without SDSS spectroscopy. The predicted N2 values are used to select promising candidates for observations. This approach was validated by new observations of 45 candidates with redshifts less than 0.065 using the 2.54~m Isaac Newton Telescope (INT) and the 4.1~m Southern Astrophysical Research (SOAR) Telescope between 2023 and 2024. All 45 candidates are confirmed to be metal-poor, including 28 new discoveries. There are 18/45 galaxies lacking detectable \rNii\ lines ($S/N<2$); for these, we report $2\sigma$ upper limits on their oxygen abundance. Our XMPs have estimated oxygen abundances of $7.1\leq$\OH$\leq8.7$ ($2\sigma$ upper limit), based on the N2 index, and 21 of them with estimated metallicity $<0.1~Z_{\odot}$. Additionally, we identified 4 potential candidates of low-metallicity AGNs at $\lesssim0.1Z_{\odot}$. Finally, we found that our observed samples are mostly brighter in the $g-$band compared to other filters, similar to blueberry (BB) galaxies, resembling green pea galaxies and high-redshift Ly$\alpha$ emitters.

Figures

Figures reproduced from arXiv: 2502.01638 by the authors.

Figure 1
Figure 1. Schematic diagram of the CNN architecture used in this work. The input is a galaxy image of 5 different filters (u, g, r, i, z). The ‘Conv 1’ and ‘Conv 2’ represent convolutional layers, and each layer is followed by a pooling layer (Pool 1 and Pool 2), respectively. Finally, two dense layers (Dense 1 and Dense 2) are used before the output layer. is the same (see [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Each panel presents the result applying the 9 trained CNN models to the specific dataset (including training and testing samples) for each procedure (Section 2.2). The left and middle panels are the confusion matrices of the MP and XMP classifiers. The classification probability thresholds for assigning classes are > 0.5. The value in each quadrant indicates the fraction (number) of the samples predicted by CNN in e… view at source ↗
Figure 3
Figure 3. Unlike [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Examples of the reduced spectra with (left; > 10σ) and without (right; < 2σ) clear detection of the [N ii] λ6585 lines. measured the [O iii] λ5008/Hβ flux ratio (i.e. the O3 index; O3 ≡ log{[O iii] λ5008/Hβ}), to assess the existence of AGN activity in our samples. The…
Figure 5
Figure 5. Figure 5: Comparison of the N2 index between CNN predictions and the observed values. The gray dots show the values of our training samples. The dashed lines indicate the MAE measured using the samples marked as squares. The blue and orange squares represent the values measured …
Figure 6
Figure 6. Figure 6: The distributions of oxygen abundance, star formation rate, and stellar mass of our XMP galaxy sample. The gray histograms show the values of all samples, while the blue histograms in the first panel represent the values of the samples with significantly detected [N ii…
Figure 7
Figure 7. Figure 7: Diagnostic diagram of [N ii] λ6585/Hα versus [O iii] λ5008/Hβ. The gray dashed line shows Equation 1 from Kauffmann et al. (2003b), which delineates star-forming galaxies and AGN. The black solid line represents the mean of local star-forming sequences for SDSS galaxie…
Figure 8
Figure 8. Figure 8: Colour distributions of training MP samples (gray shadings), training XMP samples (blue shadings), and the observed XMPs (unfilled blue histogram). The red dashed lines indicate the upper limit applied to each colour for querying the working samples (see [PITH_FULL_IM…
Figure 9
Figure 9. Figure 9: The comparison of the colour-colour diagram and his￾tograms between the observed XMP (blue contours and his￾tograms), samples of BB galaxies from Y17 (black dots and his￾togram), and training XMP samples (TrainXMP) from this work (grey dots and histogram). The red dash…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

86 extracted references · 15 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTION or pop #1...

  2. [2]

    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

  3. [3]

    B., Seo M., Ha D

    Ann H. B., Seo M., Ha D. K., 2015, @doi [ ] 10.1088/0067-0049/217/2/27 , https://ui.adsabs.harvard.edu/abs/2015ApJS..217...27A 217, 27

    work page doi:10.1088/0067-0049/217/2/27 2015

  4. [4]

    Astropy Collaboration et al., 2022, @doi [ ] 10.3847/1538-4357/ac7c74 , https://ui.adsabs.harvard.edu/abs/2022ApJ...935..167A 935, 167

    doi:10.3847/1538-4357/ac7c74 2022

  5. [5]

    A., Olive K

    Aver E., Berg D. A., Olive K. A., Pogge R. W., Salzer J. J., Skillman E. D., 2021, @doi [ ] 10.1088/1475-7516/2021/03/027 , https://ui.adsabs.harvard.edu/abs/2021JCAP...03..027A 2021, 027

    doi:10.1088/1475-7516/2021/03/027 2021

  6. [6]

    A., Phillips M

    Baldwin J. A., Phillips M. M., Terlevich R., 1981, @doi [ ] 10.1086/130766 , https://ui.adsabs.harvard.edu/abs/1981PASP...93....5B 93, 5

    doi:10.1086/130766 1981

  7. [7]

    F., McIntosh D

    Bell E. F., McIntosh D. H., Katz N., Weinberg M. D., 2003, @doi [ ] 10.1086/378847 , https://ui.adsabs.harvard.edu/abs/2003ApJS..149..289B 149, 289

    doi:10.1086/378847 2003

  8. [8]

    R., Roweis S., 2007, @doi [ ] 10.1086/510127 , https://ui.adsabs.harvard.edu/abs/2007AJ....133..734B 133, 734

    Blanton M. R., Roweis S., 2007, @doi [ ] 10.1086/510127 , https://ui.adsabs.harvard.edu/abs/2007AJ....133..734B 133, 734

    doi:10.1086/510127 2007

Show all 86 references
  1. [10]

    Bromm V., Yoshida N., 2011, @doi [ ] 10.1146/annurev-astro-081710-102608 , https://ui.adsabs.harvard.edu/abs/2011ARA&A..49..373B 49, 373

    2011 doi

  2. [11]

    F., 2009, @doi [ ] 10.1038/nature07990 , https://ui.adsabs.harvard.edu/abs/2009Natur.459...49B 459, 49

    Bromm V., Yoshida N., Hernquist L., McKee C. F., 2009, @doi [ ] 10.1038/nature07990 , https://ui.adsabs.harvard.edu/abs/2009Natur.459...49B 459, 49

    2009 doi

  3. [13]

    J., Lavaux G., Hudson M

    Carrick J., Turnbull S. J., Lavaux G., Hudson M. J., 2015, @doi [ ] 10.1093/mnras/stv547 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.450..317C 450, 317

    2015 doi

  4. [14]

    Chabrier G., 2003, @doi [ ] 10.1086/376392 , https://ui.adsabs.harvard.edu/abs/2003PASP..115..763C 115, 763

    2003 doi

  5. [15]

    Cheng T.-Y., et al., 2020, @doi [ ] 10.1093/mnras/staa501 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.493.4209C 493, 4209

    2020 doi

  6. [16]

    Cheng T.-Y., et al., 2021, @doi [ ] 10.1093/mnras/stab2142 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.507.4425C 507, 4425

    2021 doi

  7. [17]

    Y., et al., 2023, @doi [ ] 10.1093/mnras/stac3228 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.518.2794C 518, 2794

    Cheng T. Y., et al., 2023, @doi [ ] 10.1093/mnras/stac3228 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.518.2794C 518, 2794

    2023 doi

  8. [19]

    C., Crain J

    Clemens J. C., Crain J. A., Anderson R., 2004, in Moorwood A. F. M., Iye M., eds, Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series Vol. 5492, Ground-based Instrumentation for Astronomy. pp 331--340, @doi 10.1117/12.550069

    2004 doi

  9. [20]

    J., Pettini M., Jorgenson R

    Cooke R. J., Pettini M., Jorgenson R. A., Murphy M. T., Steidel C. C., 2014, @doi [ ] 10.1088/0004-637X/781/1/31 , https://ui.adsabs.harvard.edu/abs/2014ApJ...781...31C 781, 31

    2014 doi

  10. [21]

    Denicol \'o G., Terlevich R., Terlevich E., 2002, @doi [ ] 10.1046/j.1365-8711.2002.05041.x , https://ui.adsabs.harvard.edu/abs/2002MNRAS.330...69D 330, 69

    2002

  11. [22]

    I., Terlevich R., 2019, @doi [ ] 10.1093/mnras/stz1433 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.487.3221F 487, 3221

    Fern \'a ndez V., Terlevich E., D \' az A. I., Terlevich R., 2019, @doi [ ] 10.1093/mnras/stz1433 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.487.3221F 487, 3221

    2019 doi

  12. [23]

    I., 2018, @doi [arXiv e-prints] 10.48550/arXiv.1807.02811 , https://ui.adsabs.harvard.edu/abs/2018arXiv180702811F p

    Frazier P. I., 2018, @doi [arXiv e-prints] 10.48550/arXiv.1807.02811 , https://ui.adsabs.harvard.edu/abs/2018arXiv180702811F p. arXiv:1807.02811

    2018 doi arXiv

  13. [24]

    Fukugita M., Kawasaki M., 2006, @doi [ ] 10.1086/505109 , https://ui.adsabs.harvard.edu/abs/2006ApJ...646..691F 646, 691

    2006 doi

  14. [25]

    arXiv:2401.06450

    Fukushima K., Nagamine K., Matsumoto A., Isobe Y., Ouchi M., Saitoh T., Hirai Y., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2401.06450 , https://ui.adsabs.harvard.edu/abs/2024arXiv240106450F p. arXiv:2401.06450

    2024 doi arXiv

  15. [26]

    J., Shlens J., Szegedy C., 2014, @doi [arXiv e-prints] 10.48550/arXiv.1412.6572 , https://ui.adsabs.harvard.edu/abs/2014arXiv1412.6572G p

    Goodfellow I. J., Shlens J., Szegedy C., 2014, @doi [arXiv e-prints] 10.48550/arXiv.1412.6572 , https://ui.adsabs.harvard.edu/abs/2014arXiv1412.6572G p. arXiv:1412.6572

    2014 doi arXiv

  16. [27]

    L., et al., 2011, @doi [ ] 10.1088/2041-8205/736/1/L22 , https://ui.adsabs.harvard.edu/abs/2011ApJ...736L..22G 736, L22

    Griffith R. L., et al., 2011, @doi [ ] 10.1088/2041-8205/736/1/L22 , https://ui.adsabs.harvard.edu/abs/2011ApJ...736L..22G 736, L22

    2011 doi

  17. [28]

    arXiv:2501.18498

    Grossi M., et al., 2025, arXiv e-prints, https://ui.adsabs.harvard.edu/abs/2025arXiv250118498G p. arXiv:2501.18498

    2025

  18. [30]

    G., Izotov Y

    Guseva N. G., Izotov Y. I., Papaderos P., Fricke K. J., 2007, @doi [ ] 10.1051/0004-6361:20066067 , https://ui.adsabs.harvard.edu/abs/2007A&A...464..885G 464, 885

    2007 doi

  19. [31]

    G., Izotov Y

    Guseva N. G., Izotov Y. I., Fricke K. J., Henkel C., 2017, @doi [ ] 10.1051/0004-6361/201629181 , https://ui.adsabs.harvard.edu/abs/2017A&A...599A..65G 599, A65

    2017 doi

  20. [32]

    S., et al., 2016, @doi [ ] 10.3847/0004-637X/822/2/108 , https://ui.adsabs.harvard.edu/abs/2016ApJ...822..108H 822, 108

    Hirschauer A. S., et al., 2016, @doi [ ] 10.3847/0004-637X/822/2/108 , https://ui.adsabs.harvard.edu/abs/2016ApJ...822..108H 822, 108

    2016 doi

  21. [33]

    J., Prochaska J

    Hsyu T., Cooke R. J., Prochaska J. X., Bolte M., 2017, @doi [ ] 10.3847/2041-8213/aa821f , https://ui.adsabs.harvard.edu/abs/2017ApJ...845L..22H 845, L22

    2017 doi

  22. [34]

    J., Prochaska J

    Hsyu T., Cooke R. J., Prochaska J. X., Bolte M., 2018, @doi [ ] 10.3847/1538-4357/aad18a , https://ui.adsabs.harvard.edu/abs/2018ApJ...863..134H 863, 134

    2018 doi

  23. [35]

    J., Prochaska J

    Hsyu T., Cooke R. J., Prochaska J. X., Bolte M., 2020, @doi [ ] 10.3847/1538-4357/ab91af , https://ui.adsabs.harvard.edu/abs/2020ApJ...896...77H 896, 77

    2020 doi

  24. [36]

    Isobe Y., et al., 2022, @doi [ ] 10.3847/1538-4357/ac3509 , https://ui.adsabs.harvard.edu/abs/2022ApJ...925..111I 925, 111

    2022 doi

  25. [37]

    I., Thuan T

    Izotov Y. I., Thuan T. X., 2007, @doi [ ] 10.1086/519922 , https://ui.adsabs.harvard.edu/abs/2007ApJ...665.1115I 665, 1115

    2007 doi

  26. [38]

    I., Thuan T

    Izotov Y. I., Thuan T. X., 2009, @doi [ ] 10.1088/0004-637X/690/2/1797 , https://ui.adsabs.harvard.edu/abs/2009ApJ...690.1797I 690, 1797

    2009 doi

  27. [39]

    I., Lipovetsky V

    Izotov Y. I., Lipovetsky V. A., Chaffee F. H., Foltz C. B., Guseva N. G., Kniazev A. Y., 1997, @doi [ ] 10.1086/303664 , https://ui.adsabs.harvard.edu/abs/1997ApJ...476..698I 476, 698

    1997 doi

  28. [40]

    I., Papaderos P., Guseva N

    Izotov Y. I., Papaderos P., Guseva N. G., Fricke K. J., Thuan T. X., 2006, @doi [ ] 10.1051/0004-6361:20065100 , https://ui.adsabs.harvard.edu/abs/2006A&A...454..137I 454, 137

    2006 doi

  29. [41]

    I., Guseva N

    Izotov Y. I., Guseva N. G., Fricke K. J., Papaderos P., 2009, @doi [ ] 10.1051/0004-6361/200911965 , https://ui.adsabs.harvard.edu/abs/2009A&A...503...61I 503, 61

    2009 doi

  30. [42]

    I., Thuan T

    Izotov Y. I., Thuan T. X., Guseva N. G., 2012, @doi [ ] 10.1051/0004-6361/201219733 , https://ui.adsabs.harvard.edu/abs/2012A&A...546A.122I 546, A122

    2012 doi

  31. [43]

    I., Thuan T

    Izotov Y. I., Thuan T. X., Guseva N. G., 2014, @doi [ ] 10.1093/mnras/stu1771 , https://ui.adsabs.harvard.edu/abs/2014MNRAS.445..778I 445, 778

    2014 doi

  32. [44]

    L., Koposov S

    James B. L., Koposov S. E., Stark D. P., Belokurov V., Pettini M., Olszewski E. W., McQuinn K. B. W., 2017, @doi [ ] 10.1093/mnras/stw2962 , https://ui.adsabs.harvard.edu/abs/2017MNRAS.465.3977J 465, 3977

    2017 doi

  33. [45]

    D., Makarova L

    Karachentsev I. D., Makarova L. N., Koribalski B. S., Anand G. S., Tully R. B., Kniazev A. Y., 2023, @doi [ ] 10.1093/mnras/stac3284 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.518.5893K 518, 5893

    2023 doi

  34. [46]

    Kauffmann G., et al., 2003a, @doi [ ] 10.1046/j.1365-8711.2003.06291.x , https://ui.adsabs.harvard.edu/abs/2003MNRAS.341...33K 341, 33

    2003

  35. [47]

    Kauffmann G., et al., 2003b, @doi [ ] 10.1111/j.1365-2966.2003.07154.x , https://ui.adsabs.harvard.edu/abs/2003MNRAS.346.1055K 346, 1055

    2003

  36. [48]

    J., 1998, @doi [ ] 10.1146/annurev.astro.36.1.189 , https://ui.adsabs.harvard.edu/abs/1998ARA&A..36..189K 36, 189

    Kennicutt Robert C. J., 1998, @doi [ ] 10.1146/annurev.astro.36.1.189 , https://ui.adsabs.harvard.edu/abs/1998ARA&A..36..189K 36, 189

    1998 doi

  37. [49]

    J., Tamblyn P., Congdon C

    Kennicutt Robert C. J., Tamblyn P., Congdon C. E., 1994, @doi [ ] 10.1086/174790 , https://ui.adsabs.harvard.edu/abs/1994ApJ...435...22K 435, 22

    1994 doi

  38. [50]

    J., Groves B., Kauffmann G., Heckman T., 2006, @doi [ ] 10.1111/j.1365-2966.2006.10859.x , https://ui.adsabs.harvard.edu/abs/2006MNRAS.372..961K 372, 961

    Kewley L. J., Groves B., Kauffmann G., Heckman T., 2006, @doi [ ] 10.1111/j.1365-2966.2006.10859.x , https://ui.adsabs.harvard.edu/abs/2006MNRAS.372..961K 372, 961

    2006

  39. [51]

    J., Dopita M

    Kewley L. J., Dopita M. A., Leitherer C., Dav \'e R., Yuan T., Allen M., Groves B., Sutherland R., 2013, @doi [ ] 10.1088/0004-637X/774/2/100 , https://ui.adsabs.harvard.edu/abs/2013ApJ...774..100K 774, 100

    2013 doi

  40. [52]

    P., Ba J., 2015, in Bengio Y., LeCun Y., eds, 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings

    Kingma D. P., Ba J., 2015, in Bengio Y., LeCun Y., eds, 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings. http://arxiv.org/abs/1412.6980

    2015 arXiv

  41. [53]

    Kojima T., et al., 2020, @doi [ ] 10.3847/1538-4357/aba047 , https://ui.adsabs.harvard.edu/abs/2020ApJ...898..142K 898, 142

    2020 doi

  42. [54]

    Kunth D., \"O stlin G., 2000, @doi [ ] 10.1007/s001590000005 , https://ui.adsabs.harvard.edu/abs/2000A&ARv..10....1K 10, 1

    2000 doi

  43. [55]

    L., Zhang W., Kong X., Zhang Y.-X., Shen S.-Y., Zhao Y.-H., 2023, @doi [ ] 10.3847/1538-4365/acd69c , https://ui.adsabs.harvard.edu/abs/2023ApJS..267...16L 267, 16

    Liu S., Luo A. L., Zhang W., Kong X., Zhang Y.-X., Shen S.-Y., Zhao Y.-H., 2023, @doi [ ] 10.3847/1538-4365/acd69c , https://ui.adsabs.harvard.edu/abs/2023ApJS..267...16L 267, 16

    2023 doi

  44. [56]

    Madau P., Pozzetti L., Dickinson M., 1998, @doi [ ] 10.1086/305523 , https://ui.adsabs.harvard.edu/abs/1998ApJ...498..106M 498, 106

    1998 doi

  45. [57]

    Marigo P., 2001, @doi [ ] 10.1051/0004-6361:20000247 , https://ui.adsabs.harvard.edu/abs/2001A&A...370..194M 370, 194

    2001 doi

  46. [58]

    Matsumoto A., et al., 2022, @doi [ ] 10.3847/1538-4357/ac9ea1 , https://ui.adsabs.harvard.edu/abs/2022ApJ...941..167M 941, 167

    2022 doi

  47. [59]

    McQuinn K. B. W., et al., 2015, @doi [ ] 10.1088/0004-637X/812/2/158 , https://ui.adsabs.harvard.edu/abs/2015ApJ...812..158M 812, 158

    2015 doi

  48. [60]

    Meynet G., Maeder A., 2002, @doi [ ] 10.1051/0004-6361:20020755 , https://ui.adsabs.harvard.edu/abs/2002A&A...390..561M 390, 561

    2002 doi

  49. [61]

    Micheva G., \"O stlin G., Bergvall N., Zackrisson E., Masegosa J., Marquez I., Marquart T., Durret F., 2013, @doi [ ] 10.1093/mnras/stt146 , https://ui.adsabs.harvard.edu/abs/2013MNRAS.431..102M 431, 102

    2013 doi

  50. [62]

    Nakajima K., et al., 2022, @doi [ ] 10.3847/1538-4365/ac7710 , https://ui.adsabs.harvard.edu/abs/2022ApJS..262....3N 262, 3

    2022 doi

  51. [63]

    Peimbert A., Peimbert M., Luridiana V., 2016, @doi [ ] 10.48550/arXiv.1608.02062 , https://ui.adsabs.harvard.edu/abs/2016RMxAA..52..419P 52, 419

    2016 doi arXiv

  52. [64]

    Pettini M., Pagel B. E. J., 2004, @doi [ ] 10.1111/j.1365-2966.2004.07591.x , https://ui.adsabs.harvard.edu/abs/2004MNRAS.348L..59P 348, L59

    2004

  53. [65]

    S., Thuan T

    Pilyugin L. S., Thuan T. X., V \' lchez J. M., 2003, @doi [ ] 10.1051/0004-6361:20021458 , https://ui.adsabs.harvard.edu/abs/2003A&A...397..487P 397, 487

    2003 doi

  54. [66]

    Planck Collaboration et al., 2020, @doi [ ] 10.1051/0004-6361/201833910 , https://ui.adsabs.harvard.edu/abs/2020A&A...641A...6P 641, A6

    2020 doi

  55. [67]

    X., et al., 2020a, pypeit/PypeIt: Release 1.0.0 , @doi 10.5281/zenodo.3743493

    Prochaska J. X., et al., 2020a, pypeit/PypeIt: Release 1.0.0 , @doi 10.5281/zenodo.3743493

    doi

  56. [68]

    X., et al., 2020b, @doi [Journal of Open Source Software] 10.21105/joss.02308 , 5, 2308

    Prochaska J. X., et al., 2020b, @doi [Journal of Open Source Software] 10.21105/joss.02308 , 5, 2308

    doi

  57. [70]

    Raimann D., Bica E., Storchi-Bergmann T., Melnick J., Schmitt H., 2000, @doi [ ] 10.1046/j.1365-8711.2000.03317.x , https://ui.adsabs.harvard.edu/abs/2000MNRAS.314..295R 314, 295

    2000

  58. [71]

    A., Krumholz M

    Roy A., Dopita M. A., Krumholz M. R., Kewley L. J., Sutherland R. S., Heger A., 2021, @doi [ ] 10.1093/mnras/stab376 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.502.4359R 502, 4359

    2021 doi

  59. [72]

    Ruiz-Escobedo F., Pe \ n a M., Hern \'a ndez-Mart \' nez L., Garc \' a-Rojas J., 2018, @doi [ ] 10.1093/mnras/sty2265 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.481..396R 481, 396

    2018 doi

  60. [73]

    E., 1955, @doi [ ] 10.1086/145971 , https://ui.adsabs.harvard.edu/abs/1955ApJ...121..161S 121, 161

    Salpeter E. E., 1955, @doi [ ] 10.1086/145971 , https://ui.adsabs.harvard.edu/abs/1955ApJ...121..161S 121, 161

    1955 doi

  61. [74]

    Sargent W. L. W., Searle L., 1970, @doi [ ] 10.1086/180644 , https://ui.adsabs.harvard.edu/abs/1970ApJ...162L.155S 162, L155

    1970 doi

  62. [75]

    D., et al., 2013, @doi [ ] 10.1088/0004-6256/146/1/3 , https://ui.adsabs.harvard.edu/abs/2013AJ....146....3S 146, 3

    Skillman E. D., et al., 2013, @doi [ ] 10.1088/0004-6256/146/1/3 , https://ui.adsabs.harvard.edu/abs/2013AJ....146....3S 146, 3

    2013 doi

  63. [76]

    Steigman G., 2007, @doi [Annual Review of Nuclear and Particle Science] 10.1146/annurev.nucl.56.080805.140437 , https://ui.adsabs.harvard.edu/abs/2007ARNPS..57..463S 57, 463

    2007

  64. [77]

    L., 1994, @doi [ ] 10.1086/174345 , https://ui.adsabs.harvard.edu/abs/1994ApJ...429..572S 429, 572

    Storchi-Bergmann T., Calzetti D., Kinney A. L., 1994, @doi [ ] 10.1086/174345 , https://ui.adsabs.harvard.edu/abs/1994ApJ...429..572S 429, 572

    1994 doi

  65. [78]

    Tachiev G., Froese Fischer C., 2001, @doi [Canadian Journal of Physics] 10.1139/p01-059 , https://ui.adsabs.harvard.edu/abs/2001CaJPh..79..955T 79, 955

    2001 doi

  66. [79]

    X., Izotov Y

    Thuan T. X., Izotov Y. I., 2005, @doi [ ] 10.1086/491657 , https://ui.adsabs.harvard.edu/abs/2005ApJS..161..240T 161, 240

    2005 doi

  67. [80]

    X., Izotov Y

    Thuan T. X., Izotov Y. I., Lipovetsky V. A., 1995, @doi [ ] 10.1086/175676 , https://ui.adsabs.harvard.edu/abs/1995ApJ...445..108T 445, 108

    1995 doi

  68. [81]

    X., Guseva N

    Thuan T. X., Guseva N. G., Izotov Y. I., 2022, @doi [ ] 10.1093/mnrasl/slac095 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.516L..81T 516, L81

    2022 doi

  69. [82]

    A., et al., 2004, @doi [ ] 10.1086/423264 , https://ui.adsabs.harvard.edu/abs/2004ApJ...613..898T 613, 898

    Tremonti C. A., et al., 2004, @doi [ ] 10.1086/423264 , https://ui.adsabs.harvard.edu/abs/2004ApJ...613..898T 613, 898

    2004 doi

  70. [83]

    Wang L.-L., et al., 2018, @doi [ ] 10.1093/mnras/stx2798 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.474.1873W 474, 1873

    2018 doi

  71. [84]

    H., Turk M

    Wise J. H., Turk M. J., Norman M. L., Abel T., 2012, @doi [ ] 10.1088/0004-637X/745/1/50 , https://ui.adsabs.harvard.edu/abs/2012ApJ...745...50W 745, 50

    2012 doi

  72. [85]

    Xu Y., et al., 2022, @doi [ ] 10.3847/1538-4357/ac5e32 , https://ui.adsabs.harvard.edu/abs/2022ApJ...929..134X 929, 134

    2022 doi

  73. [86]

    E., Wang J., 2017, @doi [ ] 10.3847/1538-4357/aa8809 , https://ui.adsabs.harvard.edu/abs/2017ApJ...847...38Y 847, 38

    Yang H., Malhotra S., Rhoads J. E., Wang J., 2017, @doi [ ] 10.3847/1538-4357/aa8809 , https://ui.adsabs.harvard.edu/abs/2017ApJ...847...38Y 847, 38

    2017 doi

  74. [87]

    Y., Liang Y

    Yin S. Y., Liang Y. C., Hammer F., Brinchmann J., Zhang B., Deng L. C., Flores H., 2007, @doi [ ] 10.1051/0004-6361:20065798 , https://ui.adsabs.harvard.edu/abs/2007A&A...462..535Y 462, 535

    2007 doi

  75. [88]

    Zibetti S., Charlot S., Rix H.-W., 2009, @doi [ ] 10.1111/j.1365-2966.2009.15528.x , https://ui.adsabs.harvard.edu/abs/2009MNRAS.400.1181Z 400, 1181

    2009

  76. [89]

    Zou H., et al., 2024, @doi [ ] 10.3847/1538-4357/ad1409 , https://ui.adsabs.harvard.edu/abs/2024ApJ...961..173Z 961, 173

    2024 doi

  77. [90]

    van Zee L., 2000, @doi [ ] 10.1086/318176 , https://ui.adsabs.harvard.edu/abs/2000ApJ...543L..31V 543, L31

    2000 doi

  78. [91]

    P., 2006, @doi [ ] 10.1086/498017 , https://ui.adsabs.harvard.edu/abs/2006ApJ...636..214V 636, 214

    van Zee L., Haynes M. P., 2006, @doi [ ] 10.1086/498017 , https://ui.adsabs.harvard.edu/abs/2006ApJ...636..214V 636, 214

    2006 doi

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.