Pith. sign in

REVIEW 3 major objections 4 minor 294 references

Six shared enrichment patterns reproduce 22-element abundances across 426 red giant stars.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 08:45 UTC pith:QPDBJ7BF

load-bearing objection Solid new HARPS/Korg red-giant abundance catalogue worth refereeing, but the latent-model 'generative' claim overstates an in-sample NMF reconstruction and the abstract's 0.02 dex precision figure should not be read as accuracy. the 3 major comments →

arxiv 2607.21001 v1 pith:QPDBJ7BF submitted 2026-07-23 astro-ph.SR astro-ph.GA

HARPS Abundances with Korg I: 22 Element Abundances for 426 Red Giant Stars

Sarah E. Aquilina , Melissa Ness , Adam J. Wheeler , Sven Buder This is my paper
classification astro-ph.SR astro-ph.GA
keywords stellar abundancesred giant starsHARPSspectral synthesisKorglatent variable modelnucleosynthesisr-process
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper aims to establish that the 22 measured element abundances of red giants are not 22 independent numbers: a generative model with just six shared enrichment channels can reproduce them per star, implying stellar chemical space is effectively six-dimensional. To reach that claim it builds a self-consistent, high-precision catalogue of stellar parameters and element abundances for 426 red giants from HARPS spectra, using the spectral-synthesis code Korg with a median internal precision of about 0.02 dex. The catalogue matters because it adds nine neutron-capture elements at high fidelity, and the inter-element gradients derived from it quantify relative production efficiencies and suggest multiple r-process production sites. If the low-dimensional structure is real, then measuring a well-chosen handful of elements per star can recover most of the enrichment information that large surveys currently chase with dozens.

Core claim

The central discovery is the collapse of 22 element abundances into a generative 6-parameter latent-variable model. Using non-negative matrix factorization, the abundances X are written as X = f × P, where P is a set of shared enrichment channels and f is the per-star fractional contribution; six channels reproduce the abundances of the 214 stars that have all 22 elements measured, with a median reduced chi-squared of about 6. This is presented as evidence that red-giant abundance space is low-dimensional, though the imperfect fit indicates that individual elements also encode higher-order information. The paper further claims that element-element gradients, measured relative to reference el

What carries the argument

The latent-variable factorization X = f × P (non-negative matrix factorization) is the central object that carries the low-dimensionality claim: it decomposes the 22 measured abundances into a small set of population-shared enrichment patterns, with per-star fractions, and the reduced chi-squared of its reconstruction is the paper's headline diagnostic. The second piece of machinery is Korg, a 1D LTE spectral-synthesis code that uses automatic differentiation to fit stellar parameters and line-by-line abundances to the HARPS spectra; this produces the catalogue values on which everything else depends. The third is the inter-element gradient analysis, fitting Δ[X/H]/Δ[ref/H] with an orthogona

Load-bearing premise

The catalogue's zero-points are accurate: the ~0.02 dex figure is repeat-observation precision, while systematic offsets of 0.1–0.25 dex due to 1D LTE modelling and atomic data are reported for several elements, and if those systematics are underestimated, the element gradients and the inferred six-channel structure are built on biased inputs.

What would settle it

Re-analyse a subset of these 426 stars with non-LTE and 3D model atmospheres, or with line lists whose oscillator strengths are independently measured; if the Na, Zn, Cr, and metallicity offsets disappear while the element-element trends shift by more than about 0.1 dex, the specific gradients and the latent channels would not be robust. Alternatively, fit the same six-channel factorization to an independent sample of main-sequence stars from the same instrument: if the median reduced chi-squared is far above 6, the low-dimensional claim is not universal.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If abundance space is truly six-dimensional, surveys can prioritize a sparse set of well-chosen elements (the paper highlights Ti, Fe, Zr, and Ba for their discriminating power) and still recover the bulk of the enrichment structure.
  • The catalogue provides a benchmark set of red-giant abundances that can be used to train data-driven models and to cross-check the abundance scales of larger surveys.
  • The inter-element gradients give empirical constraints on galactic chemical evolution models, including a direct indication that r-process material (Eu) is not produced in lockstep with a single nucleosynthetic family.
  • The nine neutron-capture elements at ~0.02 dex internal precision add leverage on s-process and r-process yields that moderate-resolution surveys cannot provide.
  • The six latent channels can be tentatively identified with physical sources (massive stars and Type II supernovae, low- and intermediate-mass AGB stars, Type Ia supernovae), allowing chemical evolution models to be tested pattern-by-pattern rather than element-by-element.

Where Pith is reading between the lines

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

  • If the six-channel model is universal, the same six patterns should reproduce the abundances of an independent sample, such as the main-sequence stars the authors say they will analyse in the second paper of the series; failure there would mean the low-dimensional structure is specific to red giants or to this sample.
  • The large gap between the formal parameter sensitivities (σ_log g ≈ 0.17–0.27 dex in Appendix B) and the tiny repeat-observation scatter implies that the catalogue is precise but its zero-points may drift; a non-LTE or 3D re-analysis of a subset could test whether the reported 0.1–0.25 dex offsets (Na, Zn, Cr, metallicity) are physical or modelling artifacts, and whether the six latent channels su
  • The gradient analysis is restricted to the 214 stars with all 22 elements, a subsample concentrated in the thin disk; the inferred production efficiencies and the claim of multiple r-process sites may not extend to thick-disk or metal-poor populations, so extending the analysis would test the universality of the pattern.
  • A concrete prediction of the multi-site r-process claim is that Eu should show residual scatter beyond what the alpha-element channels explain; comparing this residual against neutron-star-merger yield ratios in chemical evolution models would sharpen or refute the interpretation.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper presents a catalogue of stellar parameters and 22 element abundances for 426 red giants observed with HARPS, derived using the Korg spectral synthesis code. The authors describe their data selection, radial-velocity corrections, continuum normalisation, line-by-line abundance analysis, and uncertainty calibration via repeat observations. They compare their parameters and abundances with literature samples, report inter-element abundance gradients, and propose a non-negative matrix factorisation latent-variable model (Equation 6) in which 22 abundances are represented as a product of six shared patterns and per-star fractions. The central claims are that the catalogue achieves ~0.02 dex median internal precision and that the six-channel latent model 'accurately generates' the abundances with a median reduced chi-squared of 6, implying a low-dimensional chemical space.

Significance. If the catalogue is accurate at the claimed level, it would be a valuable high-resolution benchmark for red-giant chemical abundances, complementing lower-resolution surveys and providing a self-consistent set of measurements for 22 elements including nine neutron-capture species. The paper's strengths include the use of high-quality HARPS spectra, a transparent line-selection procedure, repeat-observation precision estimates, and extensive comparisons with literature samples. The latent-model analysis is also a useful demonstration that much of the variance in the abundance matrix can be compressed to a few components, although—as discussed below—the 'generative' claim is not supported by the analysis as presented.

major comments (3)
  1. [§4.2, Eq. (6), Fig. 22] The claim that the six-channel model 'accurately generates' the abundances is circular as stated. The matrices P and f in X = f×P are fitted by NMF to the same 214-star abundance matrix that Figure 22 calls 'predicted'. This is a rank-6 reconstruction of the input data, not an independent generative model. The optimal number of channels m = 6 is also selected using the in-sample reduced chi-squared. Moreover, the median reduced chi-squared of 5.8 is not small in absolute terms; it indicates that the model leaves substantial variance unexplained given the quoted tiny uncertainties. To support the 'generative' and 'low-dimensional subspace' claims, the authors should either perform out-of-sample validation (e.g., train on a subset of stars and evaluate on held-out stars) or explicitly reframe the analysis as a lossy dimensionality-reduction exercise.
  2. [§2.6, Appendix B, §3.4] The catalogue's headline precision (~0.02 dex from repeat observations) measures internal repeatability, not accuracy. The paper's own Appendix B reports formal parameter sensitivities of σ_log g = 0.17–0.27 dex and abundance sensitivities above 0.04 dex for many neutron-capture species (e.g., Nd II up to 0.37 dex in Table B2). Section 3.4 documents zero-point offsets of 0.1–0.25 dex for [Fe/H], Na, Zn, Cr, and Ni, attributed to 1D LTE and atomic data. These systematics propagate directly into the inter-element gradients (Figure 20) and into the latent model, because both are fit to the reported abundances under the assumption of small uncertainties. The abstract and conclusions should not present 0.02 dex as the relevant error for the science claims without a systematic-error budget. At minimum, the authors should quantify how reported gradients and latent-space structure change under t
  3. [§4.1, Fig. 20] The inter-element gradients are derived with an orthogonal-distance regression that assumes the x- and y-axis uncertainties are uncorrelated. In this catalogue, both axes are [X/H] measurements from the same stars, same spectra, and same pipeline; the uncertainties are therefore strongly correlated (e.g., through the common stellar parameters and continuum normalisation). This can bias the reported slopes and make their uncertainties underestimated. A sensitivity test using correlated uncertainties, or at least a discussion of the expected sign and magnitude of the bias, is needed before the gradients can serve as quantitative nucleosynthetic constraints.
minor comments (4)
  1. [§3.1, Fig. 8 caption] The text says the median difference between [Fe/H] and [M/H] is 0.02 dex, while the Figure 8 caption says −0.01 dex. Please reconcile.
  2. [§2.6, Table 4] The text says elements with at least 5 lines are Ti I, V I, Fe I, and Y II, but the Table 4 caption includes Nd II. Table 1 lists only 2 Nd II lines. Correct the inconsistency.
  3. [§4.2, Fig. 23] The physical interpretation of the latent patterns is offered tentatively, which is appropriate, but the phrase 'non-physical signals' in channels 2 and 4 should be clarified: do the authors mean that the pattern has no obvious single nucleosynthetic origin, or that it is an artifact of the NMF algorithm?
  4. [§2.5.1, Table 1] For several elements only one or two lines are used. The per-star abundances for such elements should be flagged more prominently in the catalogue tables, since their line-to-line systematic errors cannot be assessed internally.

Circularity Check

2 steps flagged

The latent-model 'generation' is an in-sample NMF reconstruction of the same 214-star matrix; the six-channel generative claim reduces to a fit, while the catalogue itself is externally compared.

specific steps
  1. fitted input called prediction [Section 4.2 (Eq. 6, Figs 22–23); Abstract]
    "We used the element abundances of the 214 stars with all elements measured to construct a latent model representation with non negative matrix factorisation described in Ness et al. (submitted). This parameterises the abundances for the 214 stars as X=f×P (6) ... The generated abundances from the latent model in Equation 6, obtained by multiplying the solved matrix P by matrix f for a selection of our HARPS stars, is displayed in Figure 22."

    P and f are both solved from the same 214-star abundance matrix that is subsequently labelled 'generated'. The quoted median χ²_reduced=6 is the training residual of a rank-6 factorization of those exact data, with m=6 chosen from the in-sample χ². Hence 'this model accurately generates the abundances' is a restatement of the factorization used to fit the data, not an out-of-sample or independent prediction.

  2. self citation load bearing [Section 4.2 (choice of channel number; 'Ness et al. (submitted)')]
    "In Ness et al. (submitted), they found that m=4 latent patterns generates 16 element abundances ... with an overall reduced χ²_mode ∼1.2 ... At m=6, the χ² value drops below 1 in their study, indicating that using more latent variables than m=5 leads to overfitting."

    The paper's low-dimensionality claim rests on an unpublished manuscript with overlapping authorship (Ness), cited for the method and the overfitting threshold but not listed in the references, machine-checked, or externally reproduced. The present choice m=6 is then justified by the same in-sample χ² as the 'generation' test, so the central generative claim depends on an unverifiable self-citation chain for its framework.

full rationale

The catalogue construction (Sections 2–3) is not circular: parameters and abundances are externally compared (e.g., Adibekyan et al. 2012, 2015; Gaia benchmark stars; Delgado Mena et al. 2017), and the repeat-observation uncertainties are empirical scatter. The Appendix B scale-accuracy caveats (σ_logg = 0.17–0.27 dex; neutron-capture sensitivities >0.04 dex, Nd II to 0.37 dex) are a correctness/accuracy risk, not a circularity. The central circularity is the latent model in Section 4.2: equation (6) X=f×P is solved from the same 214-star matrix that is then called 'generated' (Figure 22), with m=6 selected from the in-sample reduced χ². The abstract's 'generative 6-parameter model accurately generates the abundances' therefore reduces to a rank-6 reconstruction of the input data, with no held-out or independent test. The framing also leans on an unpublished 'Ness et al. (submitted)' with overlapping authorship and no reference entry, so the low-dimensionality claim is not independently verifiable from the manuscript alone.

Axiom & Free-Parameter Ledger

6 free parameters · 8 axioms · 0 invented entities

No new physical particles, forces, or dimensions are introduced; the six latent channels are data-driven basis vectors, not claimed physical entities. The central modelling claims rest on fitted NMF factors, per-line temperature corrections, an error floor, and post-hoc selection thresholds, all of which are tuned on the same data that are later described as “generated.”

free parameters (6)
  • NMF latent patterns P (6×22) and per-star fractions f (214×6) = 6 channels; reduced χ² 6.8 (m=4) to 5.8 (m=6)
    Equation 6: X = f × P. Both matrices are solved from the same 214-star abundances, so the reported “generation” is in-sample reconstruction with hundreds of fitted coefficients.
  • Line-by-line abundance temperature-correction polynomials = Second-order polynomial per line; shift Δx_i = x_i,ref − x_i,poly
    Section 2.5.2 shifts all lines of an element to match the reference line's temperature trend. If the shared-trend assumption is wrong, per-element abundances and all derived gradients are biased.
  • Flux error inflation floor = 0.003 added in quadrature
    Equation 2; chosen by hand as a conservative continuum error. It caps the effective SNR at 333 and directly controls every χ² value and quoted parameter uncertainty.
  • NMF latent dimension m = 6
    The number of channels is selected by comparing reduced χ² on the same training data, with no cross-validation, so the “6-parameter” label is not a fixed-parameter model.
  • Empirical SNR-uncertainty models = Bins of width 30 (parameters) and 5 (lines); smoothed across 10 stars
    Section 2.6: all quoted uncertainties come from these smoothed repeat-observation scatters. Bin widths, smoothing windows, and the SNR>85 cut are hand-chosen and affect every σ used in gradients and latent-model χ².
  • Sample and line selection thresholds = SNR≥100, vsini<15 km/s, vmic<5 km/s, rms<0.1, |[X/Fe]|≤1, Mg I Teff<4750 K removed
    Post-hoc cuts in Sections 2.4, 2.5.1 and 3.4.4 shape the sample and abundance trends before the gradient and latent-model analyses are applied to them.
axioms (8)
  • domain assumption Korg's 1D LTE MARCS synthesis models the observed lines accurately enough for abundance zero-points.
    Invoked in every fit (Sections 2.4, 2.5). The paper itself attributes 0.1–0.25 dex offsets to LTE and atomic-data errors, and Appendix B shows model-sensitivity uncertainties larger than the quoted internal precision.
  • domain assumption Adopted GaiaESO/linemake atomic data (oscillator strengths, hyperfine splitting) are correct for the selected lines.
    The authors flag incorrect CH log(gf) values, a Na I gf differing from VALD by 0.048, a Zn I gf offset, and the use of a U-flag Mg I line. Abundance values are directly proportional to assumed gf values.
  • domain assumption Lines of one element should follow the same abundance-temperature trend, so per-line offsets can be shifted onto a reference line.
    Section 2.5.2 Equations 3–4. If lines have genuinely different physical responses, the correction removes real signal rather than only systematic offsets.
  • domain assumption Red giant surface abundances equal birth abundances for elements other than possibly Na, Al and Mg.
    Used when comparing red giants to main-sequence literature trends and when interpreting the element gradients as production-efficiency measures (Sections 1, 3.4).
  • domain assumption Inter-element abundance relations are linear over the sampled metallicity range.
    All gradients in Section 4.1 and Appendix C are linear fits; the paper visually checks linearity but does not model nonlinearities or breaks between thin- and thick-disk populations.
  • domain assumption The 214 stars with all 22 elements measured are representative for gradient and latent-model inference.
    Appendix C shows these 214 stars are mostly thin-disk stars; missing elements or stars are not random, and selection can imprint correlations that the NMF and gradients then recover.
  • domain assumption Non-negative matrix factorization channels correspond tentatively to physical enrichment sources.
    Section 4.2 explicitly cautions that the interpretation is tentative and that systematics, relative uncertainties, and outliers may drive the covariance structure.
  • domain assumption Repeat-observation scatter is the appropriate uncertainty for model comparison.
    Section 2.6 uses repeat-observation scatter as the primary uncertainty, while Appendix B shows that model-parameter sensitivities are larger for log g and many elements; χ²-based model selection changes depending on which uncertainty model is used.

pith-pipeline@v1.3.0-alltime-deepseek · 37324 in / 18683 out tokens · 209389 ms · 2026-08-01T08:45:22.818393+00:00 · methodology

0 comments
read the original abstract

Large stellar surveys have revealed the global abundance structure of the Milky Way, but small high-fidelity spectral samples offer a critical complement of nucleosynthetic depth. We aim to access the encoded information in an ensemble of abundances by leveraging highest-quality spectra. We used HARPS spectra (R=115,000) to determine (Teff, log(g), [M/H], vmic and vsini) and 22 element abundances (Na, Mg, Al, Si, K, Ca, Sc, Ti, V, Cr, Fe, Ni, Zn, Sr, Y, Zr, Mo, Ba, La, Ce, Nd and Eu) for 426 red giant stars at a median internal precision of $\sim$0.02 dex evaluated from analysing repeat observations of a subset of stars. Stellar parameters and line-by-line abundances were obtained using the modern spectral synthesis code Korg -- the first time it has been used for HARPS. Comparisons with the literature reveal good overall agreement. A minor 0.1 dex offset in metallicity and specific discrepancies in individual element abundances are attributed to local thermal equilibrium assumptions and inaccuracies in atomic data. We show that 22 individual elements can be collapsed into a generative 6-parameter latent-variable model of shared enrichment patterns expressed in different per-star fractions; this model accurately generates the abundances with a median $\chi_{reduced}^2 = 6$. We report element gradients with respect to selected elements from different nucleosynthetic families. These gradients are a measure of inter-element production efficiencies and indicate multiple r-process production sites. Our analysis shows that abundances occupy a low-dimensional subspace, but joint (gradient-based) information encodes nucleosynthetic signatures. We have developed a Korg-pipeline to apply across evolutionary states on high-resolution spectra to provide our precision catalogue to serve as empirical constraints on chemical evolution and as a set of benchmark red giant abundance measurements.

Figures

Figures reproduced from arXiv: 2607.21001 by Adam J. Wheeler, Melissa Ness, Sarah E. Aquilina, Sven Buder.

Figure 1
Figure 1. Figure 1: Colour-magnitude diagram of stars observed by HARPS using Gaia DR3 colours and reddening corrections. Points are coloured by their corresponding [Fe/H] in Gaia DR3. Selected red giant stars reside in the red box and FGK main sequence stars in the blue box from the 6488 HARPS stars in our crossmtached sample. The main sequence stars will be included in paper II of this series. giant stars from 113 different… view at source ↗
Figure 2
Figure 2. Figure 2: Top: Kiel diagram showing our derived log(𝑔), 𝑇eff and [M/H] for 426 red giant stars after selection cuts with the corresponding uncertainties. Parameters for unflagged GALAH DR4 stars are shown in grey to serve as comparison and validation these fall in appropriate 𝑇eff-log(𝑔) regions. Bot￾tom: Kiel diagram showing our derived log(𝑔) and 𝑇eff by the corresponding program ID of each star. There are 113 pro… view at source ↗
Figure 3
Figure 3. Figure 3: A comparison of our metallicity distribution of red giant stars observed by HARPS in pink to that of main sequence stars observed by HARPS in Delgado Mena et al. (2017) in blue. This highlights the fewer observed metal-rich ([Fe/H] > 0.1 dex) red giant stars compared to the main sequence stars in HARPS. deviation 500 pixels. Smoothing ensured that the synthetic and observed spectra of each observation had … view at source ↗
Figure 4
Figure 4. Figure 4: Top: Spectrum of 𝜆 Pyxidis after stacking 5 observations. Bottom: Continuum normalised spectrum of 𝜆 Pyxidis using suppnet. There is a CCD gap from 5300 Å to 5330 Å. vacuum wavelengths as well as a line list, wavelength windows of fitting regions, an initial guess for each parameter and resolution, we used korg’s fit_spectrum method to fit a synthesised spectrum to the observed spectrum via 𝜒 2 minimisatio… view at source ↗
Figure 5
Figure 5. Figure 5: Reference line fits in 𝜆 Pyxidis for all 22 elements including an additional 2 ionisation states. Elements are listed by increasing atomic number. The wavelength of the line is shown as a red dashed line, the green line is the fit, the black line is the continuum normalised HARPS spectrum, and the flux errors are the black shaded region. The optimised wavelength region is indicated by the dashed green line… view at source ↗
Figure 6
Figure 6. Figure 6: Top: Distribution of [Zn I/Fe] measurements across stellar tem￾perature for each of the three Zn I lines as indicated in different colours (reference line at 4811 Å in crosses). Bottom: Distribution of the three Zn I line abundances after applying corrections. All wavelengths are in the vacuum rest frame. We expect some scatter in line abundances since higher [Zn I/Fe] abundances occur with decreasing meta… view at source ↗
Figure 7
Figure 7. Figure 7: Difference between the Zr II 4963 Å line abundances of individual observations and the combined spectrum of the 33 randomly selected stars with respect to the SNR of those observations. A smoothed 1-𝜎 standard deviation of these differences across 10 stars provided a line abundance uncertainty model. We only consider stars SNR > 85 since all stars have SNR > 100 after selection cuts. stars shown as grey po… view at source ↗
Figure 8
Figure 8. Figure 8: The [Fe/H] abundance measurements compared to the metallicity ([M/H]) measurements, coloured by 𝑇eff. The median difference between [M/H] and [Fe/H] is −0.01 dex with a standard deviation of 0.02 dex. Uncertainties are included but are on the order of 0.01 dex in both [Fe/H] and [M/H]. Most points lie along the black dashed one-to-one line demonstrating that they are correlated across our parameter space. … view at source ↗
Figure 9
Figure 9. Figure 9: Top: Variation in our 𝑣 sin 𝑖 with log(𝑔) derived from HARPS spectra. Bottom: Variation in our 𝑣mic with log(𝑔) derived from HARPS spectra. All points are coloured by the effective temperature. The decreasing 𝑣 sin 𝑖 with log(𝑔) highlights their evolution across the red giant branch. Stars with 𝑣 sin 𝑖 > 5 km/s are likely an artifact from imperfect model spectra and optimisation. Uncertainties in 𝑣 sin 𝑖 a… view at source ↗
Figure 10
Figure 10. Figure 10: Top: A direct comparison of our derived 𝑇eff, log(𝑔), and [Fe/H] for 176 stars in common with Luck (2015) with a 1:1 dashed line drawn for reference. We include residuals (Luck (2015) − This Work) in sub-panels at the bottom for each parameter. Bottom: A direct comparison of our derived 𝑇eff, log(𝑔), and [Fe/H] for 35 Gaia benchmark stars (Jofré et al. 2014; Heiter et al. 2015; Soubiran et al. 2024) with … view at source ↗
Figure 11
Figure 11. Figure 11: A direct comparison of our derived 𝑇eff, log(𝑔), and [Fe/H] for 103 stars in common with Alves et al. (2015) from the Tsantaki et al. (2013) line list with a 1:1 dashed line drawn for reference. We include residuals (Alves et al. (2015) − This Work) in sub-panels at the bottom for each parameter. Parameters from Alves et al. (2015) were adopted by Adibekyan et al. (2015). same as in the Neves et al. (2009… view at source ↗
Figure 12
Figure 12. Figure 12: Comparison of our individual element abundances in common with Adibekyan et al. (2015) for 103 red giant stars. The points in the abundance comparison are coloured by the difference between our log(𝑔) and that derived from the Tsantaki et al. (2013) line list in Alves et al. (2015) to highlight its impact on Ti II compared to the other neutral species (see Section 3.4.2). is expected with the increase of … view at source ↗
Figure 13
Figure 13. Figure 13: Element abundance trends of odd-Z elements with respect to [Fe/H] and coloured by effective temperature. LTE abundances of FGK main sequence stars for Al I from Adibekyan et al. (2012) and from Zhao et al. (2016) for K I are shown as grey points. Uncertainties in [Fe/H] and [X/Fe] are included for stars in our sample [PITH_FULL_IMAGE:figures/full_fig_p016_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Element abundance trends of 𝛼-elements with respect to [Fe/H] and coloured by effective temperature. Adibekyan et al. (2012) HARPS LTE abundances of FGK main sequence stars are shown in grey for each 𝛼-element. Uncertainties in [Fe/H] and [X/Fe] are included for all stars in our sample. their temperature corrections. The temperature dependence of V I at 𝑇eff < 5300 K in their sample is not present in our … view at source ↗
Figure 15
Figure 15. Figure 15: Star abundances of iron-peak elements with respect to [Fe/H] and coloured by effective temperature. Adibekyan et al. (2012) HARPS LTE abundances of FGK main sequence stars are shown in grey for Sc I, V I, Cr II, Ni I. We compare to the HARPS abundances of Delgado Mena et al. (2017) for Zn I. Uncertainties in [Fe/H] and [X/Fe] are included for all stars in our sample [PITH_FULL_IMAGE:figures/full_fig_p017… view at source ↗
Figure 16
Figure 16. Figure 16: Element abundance trends of light s-process elements with respect to [Fe/H] and coloured by effective temperature. Delgado Mena et al. (2017) HARPS LTE abundances of FGK main sequence stars are shown in grey for Sr I, Y II, Zr I, Zr II. Uncertainties in [Fe/H] and [X/Fe] are included for all stars in our sample. Ni I: Our Ni I abundance trend across [Fe/H] ( [PITH_FULL_IMAGE:figures/full_fig_p017_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Element abundance trends of heavy s-process elements with respect to [Fe/H] and coloured by effective temperature. We include abundances from Delgado Mena et al. (2017) for Ba II, Ce II and Nd II as well as Battistini & Bensby (2016) for La II in grey. Uncertainties in [Fe/H] and [X/Fe] are included for all stars in our sample [PITH_FULL_IMAGE:figures/full_fig_p018_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Element abundance trends of r-process elements with respect to [Fe/H] and coloured by effective temperature. For Mo I we compare to the main sequence and giant stars of Mishenina et al. (2019) shown in grey. Eu II abundances are compared to Delgado Mena et al. (2017). Uncertainties in [Fe/H] and [X/Fe] are included for all stars in our sample. Note that while we group Mo I as a r-process element it also h… view at source ↗
Figure 19
Figure 19. Figure 19: Top: Distribution of the median uncertainty across all stars with recorded abundances per element. Bottom: Distribution of the discriminating power across all stars with recorded abundances per element. The discrim￾inating power is the ratio of the standard deviation of abundances and the median uncertainty. The number of lines used to calculate the overall abun￾dance for each element is indicated above t… view at source ↗
Figure 20
Figure 20. Figure 20: Gradients between each elements and a chosen reference element for each family: Al I (odd-Z), Si I (explosive 𝛼), Ni I (iron-peak), Y II (light s-process), Ce II (heavy s-process) and Eu II (r-process). Gradients of each element with respect to the reference elements were calculated using 214 stars with abundances recorded for every element. Uncertainties in the gradients are included but are not visible … view at source ↗
Figure 21
Figure 21. Figure 21: Pearson correlation coefficients between all elements using the 105 stars with abundances recorded for each of the 22 elements, including the additional 2 ionisation states. where 𝑃 is a subset of latent sources (shared between the popula￾tion) and 𝑓 the fractional contribution of each source (per star). This model uses non-negative matrix factorisation of the [X/H] abundances to solve for the per-star fr… view at source ↗
Figure 22
Figure 22. Figure 22: Measured abundances ([X/H]) for 22 elements compared to predicted abundances from the latent model in Ness et al. (submitted), with 6 latent channels that are shared among stars (and with fractional contributions that vary for each star). All elements in the catalogue are included. The eight stars were randomly selected to span our parameter space. This shows that a latent basis of 𝑚 = 6 can reconstruct t… view at source ↗
Figure 23
Figure 23. Figure 23: Normalised contribution to every element from each of the 6 channels in the latent model. The contributions are determined from the 214 stars with all elements reported. Darker segments indicate higher channel contributions for particular production sites and element families. models. Scatter in the line-by-line abundances could be reduced by using adaptive window sizes across our parameter space. This wo… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

294 extracted references · 104 canonical work pages · 5 internal anchors

  1. [1]

    , keywords =

    The Gaia-ESO Survey: Homogenisation of stellar parameters and elemental abundances. , keywords =. doi:10.1051/0004-6361/202345910 , archivePrefix =. 2304.07720 , primaryClass =

    Pith/arXiv arXiv

  2. [2]

    Astronomy & Astrophysics , author =

    Partition functions and equilibrium constants for diatomic molecules and atoms of astrophysical interest , volume =. Astronomy & Astrophysics , author =. 2016 , pages =. doi:10.1051/0004-6361/201526961 , abstract =

    doi:10.1051/0004-6361/201526961 2016

  3. [3]

    The Journal of Chemical Physics , author =

    Gaussian‐2 theory for molecular energies of first‐ and second‐row compounds , volume =. The Journal of Chemical Physics , author =. 1991 , pages =. doi:10.1063/1.460205 , language =

    doi:10.1063/1.460205 1991

  4. [4]

    Revista Mexicana de Astronomia y Astrofisica , author =

    Molecular. Revista Mexicana de Astronomia y Astrofisica , author =. 1992 , pages =. doi:1992RMxAA..23..151C , urldate =

    1992

  5. [6]

    Astronomy & Astrophysics , author =

    Carbon, oxygen, and iron abundances in disk and halo stars:. Astronomy & Astrophysics , author =. 2019 , pages =. doi:10.1051/0004-6361/201936265 , abstract =

    doi:10.1051/0004-6361/201936265 2019

  6. [7]

    Monthly Notices of the Royal Astronomical Society , author =

    Detailed chemical compositions of planet-hosting stars:. Monthly Notices of the Royal Astronomical Society , author =. 2022 , pages =. doi:10.1093/mnras/stac1119 , abstract =

    doi:10.1093/mnras/stac1119 2022

  7. [8]

    Astronomy & Astrophysics , author =

    Stellar dating using chemical clocks and. Astronomy & Astrophysics , author =. 2022 , pages =. doi:10.1051/0004-6361/202141125 , number =

    work page doi:10.1051/0004-6361/202141125 2022

  8. [9]

    The Astrophysical Journal , author =

    Detailed chemical abundances for a benchmark sample of. The Astrophysical Journal , author =. 2022 , pages =. doi:10.3847/1538-4357/ac4891 , number =

    doi:10.3847/1538-4357/ac4891 2022

  9. [10]

    and Rosenfeld, L

    Cambresier, Y. and Rosenfeld, L. , year =. On the dissociation of molecules in the atmospheres of the stars of the main sequence , volume =. doi:10.1093/mnras/93.9.710 , journal =

    work page doi:10.1093/mnras/93.9.710

  10. [11]

    Bond dissociation energies in simple molecules , volume =

    Darwent, B deB , year =. Bond dissociation energies in simple molecules , volume =

  11. [12]

    Publications of the Astronomical Society of the Pacific , author =

    Chroma+. Publications of the Astronomical Society of the Pacific , author =. 2021 , note =. doi:10.1088/1538-3873/ac0410 , abstract =

    work page doi:10.1088/1538-3873/ac0410 2021

  12. [13]

    The Astrophysical Journal Supplement Series , author =

    A set of partition functions and equilibrium constants for 300 diatomic molecules of astrophysical interest , volume =. The Astrophysical Journal Supplement Series , author =. 1984 , pages =. doi:10.1086/190980 , language =

    work page doi:10.1086/190980 1984

  13. [14]

    2010 , pages =

    The Astrophysical Journal , author =. 2010 , pages =. doi:10.1088/0004-637X/711/1/239 , language =

    work page doi:10.1088/0004-637x/711/1/239 2010

  14. [15]

    2013 , pages =

    Revista Mexicana de Astronomia y Astrofisica , author =. 2013 , pages =

    2013

  15. [16]

    Revista Mexicana de Astronomia y Astrofisica , author =

    Analytic partition function for plasmas , volume =. Revista Mexicana de Astronomia y Astrofisica , author =. 2005 , pages =

    2005

  16. [17]

    Astronomy & Astrophysics , author =

    Refined diatomic partition functions , volume =. Astronomy & Astrophysics , author =. 1987 , pages =

    1987

  17. [18]

    Astronomy & Astrophysics , author =

    Partition functions:. Astronomy & Astrophysics , author =. 2016 , pages =. doi:10.1051/0004-6361/201527209 , abstract =

    doi:10.1051/0004-6361/201527209 2016

  18. [19]

    Journal of Quantitative Spectroscopy and Radiative Transfer , author =

    Total internal partition sums for the. Journal of Quantitative Spectroscopy and Radiative Transfer , author =. 2021 , pages =. doi:10.1016/j.jqsrt.2021.107713 , language =

    arXiv 2021

  19. [20]

    Astronomy & Astrophysics , author =

    The partition functions of. Astronomy & Astrophysics , author =. 1988 , pages =

    1988

  20. [21]

    Journal of Chemical Theory and Computation , author =

    Theoretical. Journal of Chemical Theory and Computation , author =. 2009 , pages =. doi:10.1021/ct900391p , language =

    work page doi:10.1021/ct900391p 2009

  21. [22]

    Journal of Quantitative Spectroscopy and Radiative Transfer , author =

    High-temperature partition functions, specific heats and spectral radiative properties of diatomic molecules with an improved calculation of energy levels , volume =. Journal of Quantitative Spectroscopy and Radiative Transfer , author =. 2018 , pages =. doi:10.1016/j.jqsrt.2018.02.004 , language =

    work page doi:10.1016/j.jqsrt.2018.02.004 2018

  22. [23]

    The Journal of Chemical Physics , author =

    Monte. The Journal of Chemical Physics , author =. 2014 , pages =. doi:10.1063/1.4866609 , language =

    work page doi:10.1063/1.4866609 2014

  23. [24]

    Journal of Chemical Theory and Computation , author =

    Bond. Journal of Chemical Theory and Computation , author =. 2017 , pages =. doi:10.1021/acs.jctc.6b00970 , number =

    work page doi:10.1021/acs.jctc.6b00970 2017

  24. [25]

    Science in China Series G: Physics, Mechanics and Astronomy volume , author =

    Accurate studies on dissociation energies of diatomic molecules , volume =. Science in China Series G: Physics, Mechanics and Astronomy volume , author =. 2007 , pages =. doi:10.1007/s11433-007-0065-3 , number =

    work page doi:10.1007/s11433-007-0065-3 2007

  25. [26]

    MethodsX , author =

    Spectroscopy learning:. MethodsX , author =. 2020 , pages =. doi:10.1016/j.mex.2020.101127 , language =

    arXiv 2020

  26. [27]

    Science China Physics, Mechanics & Astronomy , author =

    Method for studying diatomic rovibrational spectra at a given vibrational state , volume =. Science China Physics, Mechanics & Astronomy , author =. 2019 , pages =. doi:10.1007/s11433-018-9311-3 , language =

    work page doi:10.1007/s11433-018-9311-3 2019

  27. [28]

    The Journal of Chemical Physics , author =

    Gaussian‐1 theory:. The Journal of Chemical Physics , author =. 1989 , pages =. doi:10.1063/1.456415 , language =

    work page doi:10.1063/1.456415 1989

  28. [30]

    The Journal of Physical Chemistry , author =

    Intrinsic. The Journal of Physical Chemistry , author =. 1995 , pages =. doi:10.1021/j100012a005 , language =

    work page doi:10.1021/j100012a005 1995

  29. [31]

    Accounts of Chemical Research , author =

    Predissociation. Accounts of Chemical Research , author =. 2019 , pages =. doi:10.1021/acs.accounts.8b00526 , language =

    work page doi:10.1021/acs.accounts.8b00526 2019

  30. [32]

    The Astrophysical Journal , author =

    Diatomic. The Astrophysical Journal , author =. 1963 , pages =. doi:10.1086/147685 , abstract =

    work page doi:10.1086/147685 1963

  31. [33]

    The dissociation energies of

  32. [34]

    Journal of Molecular Spectroscopy , author =

    Comparison of actual and linearly extrapolated dissociation energies of diatomic molecules , volume =. Journal of Molecular Spectroscopy , author =. 1971 , pages =. doi:10.1016/0022-2852(71)90254-2 , language =

    work page doi:10.1016/0022-2852(71)90254-2 1971

  33. [35]

    The Journal of Chemical Physics , author =

    Determination of. The Journal of Chemical Physics , author =. 1968 , pages =. doi:10.1063/1.1670082 , language =

    work page doi:10.1063/1.1670082 1968

  34. [36]

    Astronomy & Astrophysics , author =

    Photodissociation and photoionisation of atoms and molecules of astrophysical interest , volume =. Astronomy & Astrophysics , author =. 2017 , pages =. doi:10.1051/0004-6361/201628742 , abstract =

    doi:10.1051/0004-6361/201628742 2017

  35. [37]

    Monthly Notices of the Royal Astronomical Society , author =

    An investigation of. Monthly Notices of the Royal Astronomical Society , author =. 1991 , pages =. doi:10.1093/mnras/253.3.549 , number =

    work page doi:10.1093/mnras/253.3.549 1991

  36. [38]

    The Journal of Chemical Physics , author =

    A comparison of techniques for predicting higher order correlation effects:. The Journal of Chemical Physics , author =. 1999 , pages =. doi:10.1063/1.479202 , language =

    work page doi:10.1063/1.479202 1999

  37. [39]

    Reviews of Modern Physics , author =

    Coupled-cluster theory in quantum chemistry , volume =. Reviews of Modern Physics , author =. 2007 , pages =. doi:10.1103/RevModPhys.79.291 , language =

    doi:10.1103/revmodphys.79.291 2007

  38. [40]

    The Journal of Physical Chemistry , author =

    Theoretical. The Journal of Physical Chemistry , author =. 2001 , pages =. doi:10.1021/jp010309k , number =

    work page doi:10.1021/jp010309k 2001

  39. [41]

    Molecular Physics , author =

    A mathematical and computational review of. Molecular Physics , author =. 2007 , pages =. doi:10.1080/00268970701757875 , language =

    doi:10.1080/00268970701757875 2007

  40. [42]

    The Journal of Chemical Physics , author =

    Gaussian-3(. The Journal of Chemical Physics , author =. 1998 , pages =. doi:10.1063/1.477422 , number =

    work page doi:10.1063/1.477422 1998

  41. [43]

    Wiley interdisciplinary reviews , author =

    Møller-. Wiley interdisciplinary reviews , author =. 2011 , pages =. doi:10.1002/wcms.58 , number =

    doi:10.1002/wcms.58 2011

  42. [44]

    The Journal of Chemical Physics , author =

    Extension of. The Journal of Chemical Physics , author =. 2001 , pages =. doi:10.1063/1.1366337 , number =

    work page doi:10.1063/1.1366337 2001

  43. [45]

    The Journal of Physical Chemistry A , author =

    The. The Journal of Physical Chemistry A , author =. 2021 , pages =. doi:10.1021/acs.jpca.1c02741 , language =

    work page doi:10.1021/acs.jpca.1c02741 2021

  44. [46]

    The Journal of Chemical Physics , author =

    Gaussian-4 theory , volume =. The Journal of Chemical Physics , author =. 2007 , pages =. doi:10.1063/1.2436888 , language =

    work page doi:10.1063/1.2436888 2007

  45. [47]

    The Journal of Chemical Physics , author =

    Gaussian-4 theory using reduced order perturbation theory , volume =. The Journal of Chemical Physics , author =. 2007 , pages =. doi:10.1063/1.2770701 , language =

    work page doi:10.1063/1.2770701 2007

  46. [48]

    Astronomy & Astrophysics , author =

    Higher metal abundances do not solve the solar problem , volume =. Astronomy & Astrophysics , author =. 2023 , note =. doi:10.1051/0004-6361/202245448 , abstract =

    work page doi:10.1051/0004-6361/202245448 2023

  47. [49]

    and Phillips, Andrew C

    Faber, Sandra M. and Phillips, Andrew C. and Kibrick, Robert I. and Alcott, Barry and Allen, Steven L. and Burrous, Jim and Cantrall, T. and Clarke, De and Coil, Alison L. and Cowley, David J. and Davis, Marc and Deich, William T. S. and Dietsch, Ken and Gilmore, David K. and Harper, Carol A. and Hilyard, David F. and Lewis, Jeffrey P. and McVeigh, Molly ...

    2003

  48. [50]

    2002 , pages =

    Publications of the Astronomical Society of Japan , author =. 2002 , pages =. doi:10.1093/pasj/54.6.819 , abstract =

    doi:10.1093/pasj/54.6.819 2002

  49. [51]

    2011 , pages =

    Astronomy & Astrophysics , author =. 2011 , pages =. doi:10.1051/0004-6361/201015435 , abstract =

    doi:10.1051/0004-6361/201015435 2011

  50. [52]

    The Astronomical Journal , author =

    The. The Astronomical Journal , author =. 2013 , note =. doi:10.1088/0004-6256/146/2/32 , abstract =

    doi:10.1088/0004-6256/146/2/32 2013

  51. [53]

    Bulletin of the American Astronomical Society , author =

    State of the. Bulletin of the American Astronomical Society , author =. 2019 , pages =

    2019

  52. [54]

    Nature , author =

    A higher-than-predicted measurement of iron opacity at solar interior temperatures , volume =. Nature , author =. 2015 , pages =. doi:10.1038/nature14048 , language =

    work page doi:10.1038/nature14048 2015

  53. [55]

    The Journal of Chemical Physics , author =

    An augmented coupled cluster method and its application to the first‐row homonuclear diatomics , volume =. The Journal of Chemical Physics , author =. 1985 , pages =. doi:10.1063/1.448718 , language =

    work page doi:10.1063/1.448718 1985

  54. [56]

    Pramana , author =

    Relation between spectroscopic constants with limited. Pramana , author =. 2005 , pages =. doi:10.1007/BF02705289 , number =

    work page doi:10.1007/bf02705289 2005

  55. [57]

    EPL (Europhysics Letters) , author =

    Morse potential derived from first principles , volume =. EPL (Europhysics Letters) , author =. 2013 , note =. doi:10.1209/0295-5075/101/10009 , abstract =

    work page doi:10.1209/0295-5075/101/10009 2013

  56. [58]

    Opacity calculations for stellar astrophysics , url =

    Pain, Jean-Christophe and Gilleron, Franck , month = nov, year =. Opacity calculations for stellar astrophysics , url =. doi:10.5281/zenodo.1590773 , abstract =

    work page doi:10.5281/zenodo.1590773

  57. [59]

    The Astrophysical Journal , author =

    Effects of. The Astrophysical Journal , author =. 2019 , pages =. doi:10.3847/1538-4357/aaf6b7 , abstract =

    doi:10.3847/1538-4357/aaf6b7 2019

  58. [60]

    Astronomy & Astrophysics , author =

    Line-by-line opacity stellar model atmospheres , volume =. Astronomy & Astrophysics , author =. 2004 , pages =. doi:10.1051/0004-6361:20034169 , abstract =

    work page doi:10.1051/0004-6361:20034169 2004

  59. [61]

    Astronomy & Astrophysics , author =

    Microturbulent velocity from stellar spectra: a comparison between different approaches , volume =. Astronomy & Astrophysics , author =. 2011 , pages =. doi:10.1051/0004-6361/201015814 , abstract =

    doi:10.1051/0004-6361/201015814 2011

  60. [62]

    and Monier, R

    Gebran, M. and Monier, R. and Royer, F. and Lobel, A. and Blomme, R. , month = dec, year =. Microturbulence in

  61. [63]

    Monthly Notices of the Royal Astronomical Society , author =

    Non-. Monthly Notices of the Royal Astronomical Society , author =. 2012 , pages =. doi:10.1111/j.1365-2966.2012.21686.x , abstract =

    arXiv 2012

  62. [64]

    Astronomy & Astrophysics , author =

    Gaia. Astronomy & Astrophysics , author =. 2017 , note =. doi:10.1051/0004-6361/201629833 , abstract =

    doi:10.1051/0004-6361/201629833 2017

  63. [65]

    Astronomy & Astrophysics , author =

    Opacity distribution functions for stellar spectra synthesis , volume =. Astronomy & Astrophysics , author =. 2019 , pages =. doi:10.1051/0004-6361/201935723 , abstract =

    work page doi:10.1051/0004-6361/201935723 2019

  64. [66]

    Astronomy & Astrophysics , author =

    An accurate and self-consistent chemical abundance catalogue for the. Astronomy & Astrophysics , author =. 2016 , pages =. doi:10.1051/0004-6361/201628812 , abstract =

    doi:10.1051/0004-6361/201628812 2016

  65. [67]

    Monthly Notices of the Royal Astronomical Society , author =

    Non-. Monthly Notices of the Royal Astronomical Society , author =. 2016 , note =. doi:10.1093/mnras/stw2077 , abstract =

    doi:10.1093/mnras/stw2077 2016

  66. [68]

    Astronomy & Astrophysics , author =

  67. [69]

    The Astrophysical Journal , author =

    Chemical. The Astrophysical Journal , author =. 2004 , pages =. doi:10.1086/381683 , abstract =

    work page doi:10.1086/381683 2004

  68. [70]

    2019 , note =

    Monthly Notices of the Royal Astronomical Society , author =. 2019 , note =. doi:10.1093/mnras/stz2517 , abstract =

    doi:10.1093/mnras/stz2517 2019

  69. [71]

    Monthly Notices of the Royal Astronomical Society , author =

    A. Monthly Notices of the Royal Astronomical Society , author =. 2008 , note =. doi:10.1111/j.1365-2966.2008.13642.x , abstract =

    arXiv 2008

  70. [72]

    Physica Scripta , author =

    Estimation of inaccuracies in oscillator strength calculations , volume =. Physica Scripta , author =. 1996 , pages =. doi:10.1088/0031-8949/1996/T65/014 , urldate =

    work page doi:10.1088/0031-8949/1996/t65/014 1996

  71. [73]

    Brooke, James S. A. and Bernath, Peter F. and Western, Colin M. and Sneden, Christopher and Afşar, Melike and Li, Gang and Gordon, Iouli E. , month = apr, year =. Line

  72. [74]

    Journal of Quantitative Spectroscopy and Radiative Transfer , author =

    Line strengths of rovibrational and rotational transitions in the. Journal of Quantitative Spectroscopy and Radiative Transfer , author =. 2016 , pages =. doi:10.1016/j.jqsrt.2015.07.021 , abstract =

    doi:10.1016/j.jqsrt.2015.07.021 2016

  73. [75]

    Physical Review Letters , author =

    New. Physical Review Letters , author =. 2022 , note =. doi:10.1103/PhysRevLett.129.245001 , abstract =

    work page doi:10.1103/physrevlett.129.245001 2022

  74. [76]

    Canadian Journal of Physics , author =

    Oscillator. Canadian Journal of Physics , author =. 2011 , note =. doi:10.1139/p11-030 , abstract =

    doi:10.1139/p11-030 2011

  75. [77]

    Canadian Journal of Physics , author =

    Laboratory measurements of oscillator strengths and their astrophysical applications1 1. Canadian Journal of Physics , author =. 2011 , pages =. doi:10.1139/p11-044 , number =

    work page doi:10.1139/p11-044 2011

  76. [78]

    EAS Publications Series , author =

    Oscillator. EAS Publications Series , author =. 2010 , pages =. doi:10.1051/eas/1043007 , abstract =

    arXiv 2010

  77. [79]

    Journal of Physics B: Atomic, Molecular and Optical Physics , author =

    Landé g-factors along the sixth row of the periodic table , volume =. Journal of Physics B: Atomic, Molecular and Optical Physics , author =. 2010 , pages =. doi:10.1088/0953-4075/43/7/074010 , number =

    work page doi:10.1088/0953-4075/43/7/074010 2010

  78. [80]

    The Astrophysical Journal , author =

    Atomic. The Astrophysical Journal , author =. 2021 , pages =. doi:10.3847/1538-4357/ac2cca , abstract =

    work page doi:10.3847/1538-4357/ac2cca 2021

  79. [81]

    2019 , pages =

    Journal of Quantitative Spectroscopy and Radiative Transfer , author =. 2019 , pages =. doi:10.1016/j.jqsrt.2019.106639 , language =

    arXiv 2019

  80. [82]

    Astronomy & Geophysics , author =

    Annie. Astronomy & Geophysics , author =. 2016 , pages =. doi:10.1093/astrogeo/atw096 , language =

    work page doi:10.1093/astrogeo/atw096 2016

Showing first 80 references.