pith. machine review for the scientific record. sign in

arxiv: 2604.16073 · v1 · submitted 2026-04-17 · 🌌 astro-ph.EP · astro-ph.IM· physics.chem-ph· physics.comp-ph

Recognition: unknown

Machine learning isotope shifts in molecular energy levels

Marco G. Barnfield , Oleg L. Polyansky , Sergei N. Yurchenko , Jonathan Tennyson
Authors on Pith no claims yet

Pith reviewed 2026-05-10 07:29 UTC · model grok-4.3

classification 🌌 astro-ph.EP astro-ph.IMphysics.chem-phphysics.comp-ph
keywords machine learningisotope shiftsmolecular energy levelsexoplanet atmospherestransfer learningspectroscopic line listscarbon dioxidecarbon monoxide
0
0 comments X

The pith

A neural network learns residual errors in isotopologue extrapolations from CO2 and transfers them to improve CO energy levels.

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

The paper develops a machine learning framework to correct inaccuracies in theoretical predictions of energy levels for minor isotopologues of molecules used in exoplanet atmosphere studies. A fully connected neural network trained on carbon dioxide data models the residual errors left by the isotopologue extrapolation method and reduces mean absolute error for more than 87 percent of levels against empirical benchmarks. A hybrid transfer learning architecture then propagates those learned correction patterns to the data-poor carbon monoxide system, achieving improvements for over 93 percent of CO samples. The resulting updated line lists for 11 CO2 isotopologues and predicted excited-state levels for CO demonstrate that isotopic correction factors can be generalized across chemically related molecules. This approach offers a scalable way to refine spectroscopic data where experimental measurements remain limited.

Core claim

A fully connected neural network architecture for carbon dioxide predicts energy corrections with high fidelity, reducing the mean absolute error relative to the original IE approach for more than 87 percent of the levels when benchmarked against empirical energies. A novel hybrid, molecule-aware transfer learning architecture successfully propagates correction patterns from the data-rich CO2 system to the data-poor CO system, yielding MAE improvements in over 93 percent of CO samples. Updated and improved line lists are presented for 11 CO2 isotopologues and energy levels for excited states of CO isotopologues are predicted.

What carries the argument

The hybrid molecule-aware transfer learning neural network that models residual errors of the isotopologue extrapolation method and applies learned corrections across related molecular systems.

If this is right

  • Line lists for 11 CO2 isotopologues become more accurate for use in atmospheric modeling.
  • Excited-state energy levels for CO isotopologues receive data-driven predictions where experiments are sparse.
  • The method provides a scalable route to refine other molecular line lists in large databases.
  • Minor isotopologue data gains reliability for tracing planetary formation and evolution.

Where Pith is reading between the lines

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

  • The same transfer-learning pattern could extend to other molecule pairs to boost predictions for rare isotopologues.
  • Limited experimental data across broader molecular databases could be leveraged more effectively through similar cross-molecule corrections.
  • Testing on molecules with different bonding or larger mass differences would reveal how far the generalization holds.

Load-bearing premise

Residual error patterns from the abundant CO2 data can be transferred to the scarce CO data without major loss of physical accuracy or interference from differences between the two molecules.

What would settle it

New high-resolution experimental energy measurements for CO isotopologues not seen during training, compared directly against the machine-learning-corrected predictions.

Figures

Figures reproduced from arXiv: 2604.16073 by Jonathan Tennyson, Marco G. Barnfield, Oleg L. Polyansky, Sergei N. Yurchenko.

Figure 1
Figure 1. Figure 1: Neural Network structure for CO2 IE corrections. The activation function used was the Gaussian Error Linear Unit (GELU) [37]. Unlike the standard Rectified Linear Unit (ReLU) [38] which has a sharp discontinuity at zero, GELU, is a smooth, probabilistic activation function. This smoothness is advantageous for regression tasks in physics, where the target function (in this case the energy correction) is con… view at source ↗
Figure 2
Figure 2. Figure 2: Neural Network structure for CO IE corrections. [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The mean absolute error (MAE) of the original Isotopologue Extrapolation (IE) method for each CO [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: 13C 16O residuals before and after ML correction. When 636 was included, the overall MAE improvement across isotopologues was 85.93 %, with 83.14 % of individual samples showing improvement. After removing 636, the overall MAE improvement increased to 89.27 %, and the proportion of samples showing improvement rose to 91.62 % over the original IE method. The overall model performance following this adjustme… view at source ↗
Figure 5
Figure 5. Figure 5: Mean absolute error (MAE) and root mean square error (RMSE) across minor CO [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Distribution of residuals for all CO2 isotopologues, representing the discrepancy between empirical Marvel energy levels and the IE-calculated energies before and after ML correction. 12 [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Residuals for individual CO2 isotopologues plotted against Marvel empirical energy levels before and after the ML correction. 3.2.1. Feature Analysis Feature importance was examined using the ablation approach described in Section 2.2, where individual features were removed and the corresponding change in MAE was recorded [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Mean absolute error (MAE) and root mean square error (RMSE) across minor CO isotopologues before and [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Distribution of residuals for CO isotopologues before and after the ML correction. [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Residuals for individual CO isotopologues plotted against empirical M [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
read the original abstract

Recent advances in the use of High-Resolution Cross-Correlation Spectroscopy (HRCCS) to detect molecular species in exoplanet atmospheres, presents a new challenge for the accuracy of reference spectroscopic line lists. While parent isotopologues of key atmospheric tracers are often well-characterized, minor isotopologues, crucial for diagnosing planetary formation histories and evolution, suffer from a scarcity of experimental data, often leading to reliance on less accurate theoretical predictions. In this work, a comprehensive machine learning framework is designed to mitigate these inaccuracies by modelling the residual errors of the isotopologue extrapolation (IE) method used within the ExoMol project. A fully connected neural network architecture for carbon dioxide (CO$_2$) is shown to predict energy corrections with high fidelity, reducing the mean absolute error (MAE) relative to the original IE approach for more than 87\% of the levels when benchmarked against empirical (\Marvel) energies. Furthermore, development of a novel hybrid, molecule-aware transfer learning architecture is presented that successfully propagates correction patterns from the data-rich CO$_2$ system to the data-poor carbon monoxide (CO) system. This transfer learning approach yields MAE improvements in over 93\% of CO samples, demonstrating that physical correction factors related to isotopic substitution can be generalized across chemically related molecular systems. Updated and improved line lists are presented for 11 CO$_2$ isotopologues and energy levels for excited states of CO isotopologues are predicted. The methodology establishes a scalable, data-driven paradigm for refining molecular line lists, helping to bridge the gap between theoretical calculations and experimental precision.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 3 minor

Summary. The paper introduces a machine learning framework to correct residual errors in the isotopologue extrapolation (IE) method for molecular energy levels. For CO2, a fully connected neural network is trained on Marvel empirical energies to predict corrections, reducing MAE relative to the original IE approach for more than 87% of levels. A novel hybrid molecule-aware transfer learning architecture then propagates these correction patterns to the data-poor CO system, yielding MAE improvements for over 93% of CO samples. Updated line lists for 11 CO2 isotopologues and predictions for excited states of CO isotopologues are provided, establishing a scalable data-driven approach for refining spectroscopic data relevant to exoplanet atmospheres.

Significance. If the central claims hold after addressing validation details, the work provides a practical, scalable method to improve accuracy of minor isotopologue line lists where experimental data are scarce. This directly supports HRCCS applications in exoplanet science by bridging theoretical IE predictions and empirical precision. The transfer learning component, if robustly validated, could generalize to other molecular systems and represents a strength in leveraging data-rich to data-poor domains. The benchmarking against Marvel energies and provision of updated lists are positive elements.

major comments (3)
  1. [§3.2] §3.2 (Neural network for CO2): The description of the dataset partitioning is insufficient; no details are given on the train/test split ratios, whether the Marvel energies were randomly partitioned or stratified by vibrational/rotational quantum numbers, or any cross-validation procedure used to obtain the >87% MAE reduction figure. This directly affects whether the reported improvement demonstrates generalization or risks overfitting to the training distribution.
  2. [§4.3] §4.3 (Hybrid transfer learning to CO): The hybrid molecule-aware architecture's implementation lacks explicit controls for domain shift between CO2 (triatomic) and CO (diatomic). It is unclear how the 93% MAE improvement on CO samples was computed (e.g., on fully held-out CO levels never seen during transfer, or including any CO data used in architecture design), and no comparison is provided against independent high-level ab initio calculations for CO levels outside the Marvel set. This is load-bearing for the generalization claim.
  3. [Table 2 and Table 4] Table 2 (CO2 MAE results) and Table 4 (CO results): The tables report aggregate percentages (87% and 93%) but do not break down performance by isotopologue, energy range, or quantum number regime. Without these, it is difficult to assess whether improvements are uniform or concentrated in well-sampled regions, undermining the claim of broad applicability.
minor comments (3)
  1. [Abstract] The abstract states 'presents a new challenge' but should read 'present' for grammatical agreement.
  2. [Figure 3] Figure 3 caption does not specify the exact loss function or optimizer hyperparameters used in the NN training, which would aid reproducibility.
  3. [§4.1] Notation for the molecule-aware embedding in the transfer architecture is introduced without a clear equation reference; adding an explicit definition would improve clarity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their thorough review and constructive feedback on our manuscript. Their comments have identified areas where additional clarity and detail will strengthen the presentation of our methods and results. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (Neural network for CO2): The description of the dataset partitioning is insufficient; no details are given on the train/test split ratios, whether the Marvel energies were randomly partitioned or stratified by vibrational/rotational quantum numbers, or any cross-validation procedure used to obtain the >87% MAE reduction figure. This directly affects whether the reported improvement demonstrates generalization or risks overfitting to the training distribution.

    Authors: We agree that the original description in §3.2 was insufficiently detailed. The Marvel energies were randomly partitioned using an 80/20 train/test split with no stratification by vibrational or rotational quantum numbers. To demonstrate generalization, we have now conducted 5-fold cross-validation, and the reported >87% MAE reduction is consistent across folds (average improvement 86.4%). In the revised manuscript we will expand §3.2 to explicitly state the split ratios, the random partitioning procedure, and the cross-validation results. revision: yes

  2. Referee: [§4.3] §4.3 (Hybrid transfer learning to CO): The hybrid molecule-aware architecture's implementation lacks explicit controls for domain shift between CO2 (triatomic) and CO (diatomic). It is unclear how the 93% MAE improvement on CO samples was computed (e.g., on fully held-out CO levels never seen during transfer, or including any CO data used in architecture design), and no comparison is provided against independent high-level ab initio calculations for CO levels outside the Marvel set. This is load-bearing for the generalization claim.

    Authors: We appreciate the emphasis on rigorous validation of the transfer step. The hybrid architecture employs molecule-specific embeddings and separate decoder heads to explicitly handle domain differences between triatomic CO2 and diatomic CO; these controls are described in §4.3 but will be expanded for clarity. The 93% MAE improvement was evaluated exclusively on a fully held-out CO test set that was never used during architecture design, pre-training, or fine-tuning. No CO data entered the transfer process. We did not include comparisons against independent high-level ab initio calculations for CO levels outside Marvel, as our benchmark focused on empirical Marvel energies where available; we will add an explicit discussion of this limitation in the revised text. revision: partial

  3. Referee: [Table 2 and Table 4] Table 2 (CO2 MAE results) and Table 4 (CO results): The tables report aggregate percentages (87% and 93%) but do not break down performance by isotopologue, energy range, or quantum number regime. Without these, it is difficult to assess whether improvements are uniform or concentrated in well-sampled regions, undermining the claim of broad applicability.

    Authors: We agree that aggregate percentages alone limit assessment of uniformity. In the revised manuscript we will augment Tables 2 and 4 (or add supplementary tables) with breakdowns by isotopologue, energy range (e.g., low- vs. high-lying vibrational levels), and quantum-number regime (e.g., ranges of J). These additions will allow readers to evaluate whether improvements are broadly distributed or localized to well-sampled regions. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain.

full rationale

The paper trains fully connected and hybrid transfer-learning neural networks on external empirical Marvel energies to model residuals of an independent IE (isotopologue extrapolation) method from the ExoMol project. Reported MAE reductions (87% of CO2 levels, 93% of CO samples) are benchmarked against held-out or separate empirical data rather than being fitted inputs renamed as predictions. No self-definitional equations, load-bearing self-citations that force uniqueness, or ansatz smuggling appear in the provided abstract or claims; the central results rest on standard supervised learning with cross-system validation.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that residual errors in the IE method are learnable from Marvel data and that those patterns transfer across molecules; the ledger counts standard neural-network training assumptions plus the domain-specific transferability premise.

free parameters (1)
  • Neural network architecture hyperparameters
    Layer count, neuron counts, learning rate and other choices selected or tuned on CO2 data to achieve the reported MAE reductions.
axioms (2)
  • domain assumption Residual errors of the isotopologue extrapolation method are systematic and can be modeled by a fully connected neural network trained on empirical energies.
    Invoked when the authors state the network predicts energy corrections with high fidelity.
  • ad hoc to paper Correction patterns learned on CO2 generalize to CO via a molecule-aware transfer learning architecture.
    Central premise of the hybrid transfer step; no independent physical derivation is given.

pith-pipeline@v0.9.0 · 5604 in / 1510 out tokens · 60107 ms · 2026-05-10T07:29:54.614876+00:00 · methodology

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

42 extracted references · 40 canonical work pages · 4 internal anchors

  1. [1]

    , year = 1995, month = nov, volume =

    M. Mayor, D. Queloz, A jupiter-mass companion to a solar-type star, Nature 378 (1995) 355–359. doi:10.1038/378355a0

    work page doi:10.1038/378355a0 1995

  2. [2]

    W. J. Borucki, D. Koch, G. Basri, N. Batalha, T. Brown, D. Caldwell, J. Caldwell, J. Christensen-Dalsgaard, W. D. Cochran, E. DeV ore, E. W. Dunham, A. K. Dupree, T. N. Gautier, J. C. Geary, R. Gilliland, A. Gould, S. B. Howell, J. M. Jenkins, Y . Kondo, D. W. Latham, G. W. Marcy, S. Meibom, H. Kjeldsen, J. J. Lissauer, D. G. Monet, D. Mor- rison, D. Sass...

    work page doi:10.1126/science.1185402 2010

  3. [3]

    G. R. Ricker, J. N. Winn, R. Vanderspek, D. W. Latham, G. A. Bakos, J. L. Bean, Z. K. Berta-Thompson, T. M. Brown, L. Buchhave, N. R. Butler, R. P. Butler, W. J. Chaplin, D. Charbonneau, J. Christensen-Dalsgaard, M. Clampin, D. Deming, J. Doty, N. De Lee, C. Dressing, E. W. Dunham, M. Endl, F. Fressin, J. Ge, T. Henning, M. J. Holman, A. W. Howard, S. Ida...

    work page internal anchor Pith review doi:10.1117/1.jatis.1.1.014003 2014

  4. [4]

    E. L. Rice, T. Barman, I. S. Mclean, L. Prato, J. D. Kirkpatrick, Physical properties of young brown dwarfs and very low mass stars inferred from high-resolution model spectra, Astrophys. J. Suppl. Ser. 186 (2010) 63. doi:10.1088/0067-0049/186/1/63

    work page doi:10.1088/0067-0049/186/1/63 2010

  5. [5]

    Madhusudhan, Exoplanetary Atmospheres: Key Insights, Challenges, and Prospects, Annu

    N. Madhusudhan, Exoplanetary Atmospheres: Key Insights, Challenges, and Prospects, Annu. Rev. Astron. Astrophys 57 (2019) 617–663. doi:10.1146/annurev-astro-081817- 051846

    work page doi:10.1146/annurev-astro-081817- 2019

  6. [7]

    S. N. Yurchenko, J. Tennyson, M. Brogi, High-resolution spectroscopy of exoplanets: data challenges and prospects, Nat. Rev. Phys. (2025). doi:10.1038/s42254-025-00839-z

    work page doi:10.1038/s42254-025-00839-z 2025

  7. [8]

    J. K. Barstow, S. Aigrain, P. G. J. Irwin, S. Kendrew, L. N. Fletcher, Transit spectroscopy with James Webb Space Telescope: systematics, starspots and stitching, Mon. Not. Roy. Astron. Soc. 448 (2015) 2546–2561. doi:10.1093/mnras/stv186. 20

    work page doi:10.1093/mnras/stv186 2015

  8. [9]

    L. S. Wiser, T. J. Bell, M. R. Line, E. Schlawin, T. G. Beatty, L. Welbanks, T. P. Greene, V . Parmentier, M. M. Murphy, J. J. Fortney, K. Arnold, N. Mehta, K. Ohno, S. Mukher- jee, A precise metallicity and carbon-to-oxygen ratio for a warm giant exoplanet from its panchromatic JWST emission spectrum, Proc. Nat. Acad. Sci. U. S. A. 122 (2024). doi:10.107...

    work page doi:10.1073/pnas.2416193122 2024

  9. [10]

    Snellen, R

    I. Snellen, R. de Kok, J. L. Birkby, B. Brandl, M. Brogi, C. Keller, M. Kenworthy, H. Schwarz, R. Stuik, Combining high-dispersion spectroscopy with high contrast imag- ing: Probing rocky planets around our nearest neighbors, Astron. Astrophys. 576 (2015) A59. doi:10.1051/0004-6361/201425018

    work page doi:10.1051/0004-6361/201425018 2015

  10. [11]

    I. A. Snellen, Exoplanet atmospheres at high spectral resolution, Annu. Rev. Astron. Astro- phys 63 (2025) 83–125. doi:10.1146/annurev-astro-052622-031342

    work page doi:10.1146/annurev-astro-052622-031342 2025

  11. [12]

    Machine behaviour

    Y . Zhang, I. A. G. Snellen, A. J. Bohn, P. Molliere, C. Ginski, H. J. Hoeijmakers, M. A. Kenworthy, E. E. Mamajek, T. Meshkat, M. Reggiani, F. Snik, The13CO-rich atmosphere of a young accreting super-Jupiter, Nature 595 (7867) (2021) 370–372. doi:10.1038/s41586- 021-03616-x

    work page doi:10.1038/s41586- 2021

  12. [13]

    Esparza-Borges, M

    E. Esparza-Borges, M. López-Morales, J. I. Adams Redai, E. Pallé, J. Kirk, N. Casasayas- Barris, N. E. Batalha, B. V . Rackham, J. L. Bean, S. L. Casewell, L. Decin, L. A. Dos Santos, A. G. Muñoz, J. Harrington, K. Heng, R. Hu, L. Mancini, K. Molaverdikhani, G. Morello, N. K. Nikolov, M. C. Nixon, S. Redfield, K. B. Stevenson, H. R. Wakeford, M. K. Alam, ...

    work page doi:10.3847/2041-8213/acf27b 2023

  13. [14]

    Brogi, M

    M. Brogi, M. R. Line, Retrieving Temperatures and Abundances of Exoplanet Atmo- spheres with High-resolution Cross-correlation Spectroscopy, Astrophys. J. 157 (2019)

    2019

  14. [15]

    doi:10.3847/1538-3881/aaffd3

    work page doi:10.3847/1538-3881/aaffd3

  15. [16]

    2001, MNRAS, 322, 2,

    J. Tennyson, S. N. Yurchenko, ExoMol: molecular line lists for exoplanet and other atmospheres, Mon. Not. Roy. Astron. Soc. 425 (2012) 21–33. doi:10.1111/j.1365- 2966.2012.21440.x

    work page doi:10.1111/j.1365- 2012

  16. [17]

    Furtenbacher, A

    T. Furtenbacher, A. G. Császár, J. Tennyson, MARVEL: measured active rotational-vibrational energy levels, J. Mol. Spectrosc. 245 (2007) 115–125. doi:10.1016/j.jms.2007.07.005

    work page doi:10.1016/j.jms.2007.07.005 2007

  17. [18]

    Tennyson, S

    J. Tennyson, S. N. Yurchenko, J. Zhang, C. A. Bowesman, R. P. Brady, J. Buldyreva, K. L. Chubb, R. R. Gamache, M. N. Gorman, E. R. Guest, C. Hill, K. Kefala, A. E. Lynas-Gray, T. M. Mellor, L. K. McKemmish, G. B. Mitev, I. I. Mizus, A. Owens, Z. Peng, A. N. Perri, M. Pezzella, O. L. Polyansky, Q. Qu, M. Semenov, O. Smola, A. ov, W. Somogyi, A. Upadhyay, S...

    work page doi:10.1016/j.jqsrt.2024.109083 2024

  18. [19]

    O. L. Polyansky, A. A. Kyuberis, L. Lodi, J. Tennyson, R. I. Ovsyannikov, N. Zobov, ExoMol molecular line lists XIX: high accuracy computed line lists for H217O and H218O, Mon. Not. Roy. Astron. Soc. 466 (2017) 1363–1371. doi:10.1093/mnras/stw3125. 21

    work page doi:10.1093/mnras/stw3125 2017

  19. [20]

    L. K. McKemmish, C. A. Bowesman, K. Kefala, A. N. Perri, A. M. Syme, S. N. Yurchenko, J. Tennyson, A hybrid approach to generating diatomic line lists for high resolution studies of exoplanets and other hot astronomical objects: Updates to ExoMol MgO, VO and TiO line lists, RAS Tech. Instr. 3 (2024) 565–583. doi:10.1093/rasti/rzae037

    work page doi:10.1093/rasti/rzae037 2024

  20. [21]

    Y . V . Pavlenko, S. N. Yurchenko, L. K. McKemmish, J. Tennyson, Analysis of the TiO iso- topologues in stellar optical spectra, Astron. Astrophys. 42 (2020) A77. doi:10.1051/0004- 6361/202037863

    work page doi:10.1051/0004- 2020

  21. [22]

    D. W. Schwenke, Beyond the potential energy surface: Ab initio corrections to the born-oppenheimer approximation for h2o, J. Phys. Chem. A 105 (2001) 2352–2360. doi:10.1021/jp0032513

    work page doi:10.1021/jp0032513 2001

  22. [23]

    Hansen, F

    K. Hansen, F. Biegler, R. Ramakrishnan, W. Pronobis, O. von Lilienfeld, K.-R. Müller, A. Tkatchenko, Machine learning predictions of molecular properties: Accurate many-body potentials and nonlocality in chemical space, JOURNAL OF PHYSICAL CHEMISTRY LETTERS 6 (2015) 2326–2331. doi:10.1021/acs.jpclett.5b00831

    work page doi:10.1021/acs.jpclett.5b00831 2015

  23. [24]

    Westermayr, P

    J. Westermayr, P. Marquetand, Machine learning spectroscopy to advance computation and analysis, Chemical Science 46 (16) (2025) 21660–21676. doi:10.1039/d5sc05628d

    work page doi:10.1039/d5sc05628d 2025

  24. [25]

    E. R. Guest, J. Tennyson, S. N. Yurchenko, Modelling the Rotational Dependence of Line Broadening using Machine Learning, J. Mol. Spectrosc. 401 (2024) 111901. doi:10.1016/j.jms.2024.111901

    work page doi:10.1016/j.jms.2024.111901 2024

  25. [26]

    S. N. Yurchenko, M. G. Barnfield, C. A. Bowesman, R. P. Brady, E. R. Guest, K. Kefala, Q.-H. Ni, A. N. Perri, O. A. Smola, A. Solokov, C. Tao, J. Tennyson, ExoMol line lists – LXIII: ExoMol line lists for 12 isotopologues of CO 2, Mon. Not. Roy. Astron. Soc. 545 (2026) staf2135. doi:10.1093/mnras/staf2135

    work page doi:10.1093/mnras/staf2135 2026

  26. [27]

    M. T. I. Ibrahim, D. Alatoom, T. Furtenbacher, A. G. Császár, S. N. Yurchenko, A. A. A. Azzam, J. Tennyson, MARVEL analysis of high-resolution rovibrational spec- tra of 13C16O2, J. Comput. Chem. 45 (2024) 969–984. doi:10.1002/jcc.27266

    work page doi:10.1002/jcc.27266 2024

  27. [28]

    Alatoom, M

    D. Alatoom, M. T. I. Ibrahim, T. Furtenbacher, A. G. Császár, M. Alghizzawi, S. N. Yurchenko, A. A. A. Azzam, J. Tennyson, MARVEL analysis of high-resolution rovibra- tional spectra of 16O12C18O, J. Comput. Chem. 45 (2024) 2558. doi:10.1002/jcc.27453

    work page doi:10.1002/jcc.27453 2024

  28. [29]

    A. A. A. Azzam, S. A. A. Azzam, K. A. A. Aburumman, J. Tennyson, S. N. Yurchenko, A. G. Császár, T. Furtenbacher, MARVEL analysis of high-resolution rovibrational spectra of 18O12C18O, 17O12C18O and 18O13C18O isotopologues of carbon dioxide, J. Mol. Spec- trosc. 405 (2024) 111947. doi:10.1016/j.jms.2024.111947

    work page doi:10.1016/j.jms.2024.111947 2024

  29. [30]

    A. A. A. Azzam, B. M. J. Abou Doud, M. Q. A. Shersheer, B. K. M. Almasri, C. N. M. Bader, A. M. H. A. Baraa O. A. KH. Musleh and, A. W. M. Al Shatarat, B. I. M. Qat- tan, L. H. M. Hamamsy, A. O. G. Saafneh, M. N. A. ALso’ub, M. M. A. Alkhashash- neh, H. O. M. Al-Zawahra, D. Alatoom, M. T. I. Ibrahim, J. Tennyson, S. N. Yurchenko, T. Furtenbacher, A. G. Cs...

    work page doi:10.1038/s41597- 2025

  30. [31]

    A. A. A. Azzam, J. Tennyson, S. N. Yurchenko, T. Furtenbacher, A. G. Császár, MARVEL analysis of high-resolution rovibrational spectra of16O13C18O, J. Comput. Chem. 46 (2025) e27541. doi:10.1002/jcc.27541

    work page doi:10.1002/jcc.27541 2025

  31. [32]

    S. A. M. Obaidata, A. A. A. Azzam, J. Tennyson, S. N. Yurchenko, T. Furtenbacher, A. G. Császár, MARVEL analysis of high-resolution rovibrational spectra of 16O12C17O, J. Mol. Spectrosc. 340 (2025) 109444. doi:10.1016/j.jqsrt.2025.109444

    work page doi:10.1016/j.jqsrt.2025.109444 2025

  32. [33]

    M. H. I. Mansour, A. A. A. Azzam, J. Tennyson, S. N. Yurchenko, T. Furtenbacher, A. G. Császár, MARVEL analysis of high-resolution rovibrational spectra of 16O13C17O and 17O12C17O, Mol. Phys. (2025) e2550568doi:10.1080/00268976.2025.2550568

    work page doi:10.1080/00268976.2025.2550568 2025

  33. [34]

    A. A. A. Azzam, J. M. A. AlAlawin, J. Tennyson, S. N. Yurchenko, T. Furten- bacher, A. G. Császár, MARVEL analysis of high-resolution rovibrational spectra of 17O13C18O and 17O13C17O, J. Quant. Spectrosc. Radiat. Transf. 343 (2025) 109485. doi:10.1016/j.jqsrt.2025.109485

    work page doi:10.1016/j.jqsrt.2025.109485 2025

  34. [35]

    M., Freire, P

    S. Mahmoud, N. El-Kork, N. Abu Elkher, M. Almehairbi, M. S. Khalil, T. Furtenbacher, O. P. Yurchenko, S. N. Yurchenko, J. Tennyson, MARVEL Analysis of the Measured High- resolution Spectra of 12C16O, Astrophys. J. Suppl. Ser. 276 (2025) 66. doi:10.3847/1538- 4365/ada3c9

    work page doi:10.3847/1538- 2025

  35. [36]

    Grigorev, Y

    T. Grigorev, Y . Dai, M. Potter, X. Xiang, K. Zhang, J. Tennyson, MARVEL Analysis of the Measured High-resolution Spectra of CO Isotopologues, Astrophys. J. Suppl. Ser. 283 (2026) 39. doi:10.3847/1538-4365/ae40f0

    work page doi:10.3847/1538-4365/ae40f0 2026

  36. [37]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Köpf, E. Yang, Z. DeVito, M. Raison, A. Te- jani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, S. Chintala, Pytorch: An imperative style, high-performance deep learning library, in: Advances in Neural Information Processing Syst...

    2019

  37. [38]

    Gaussian Error Linear Units (GELUs)

    D. Hendrycks, K. Gimpel, Gaussian error linear units (gelus), arXiv (2023). doi:10.48550/arXiv.1606.08415

    work page Pith review doi:10.48550/arxiv.1606.08415 2023

  38. [39]

    A. F. Agarap, Deep learning using rectified linear units (relu), arXiv (2019). doi:10.48550/arXiv.1803.08375

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1803.08375 2019

  39. [40]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization, arXiv (2017). doi:10.48550/arXiv.1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1412.6980 2017

  40. [41]

    The Annals of Mathematical Statistics , author =

    P. Huber, Robust estimation of a location parameter, The Annals of Mathematical Statistics 35 (1964) 73–101. doi:10.1214/aoms/1177703732

    work page doi:10.1214/aoms/1177703732 1964

  41. [42]

    J. L. Ba, J. R. Kiros, G. E. Hinton, Layer normalization, arXiv (2016). doi:10.48550/arXiv.1607.06450

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1607.06450 2016

  42. [43]

    L. S. Rothman, L. D. G. Young, Infrared energy levels and intensities of carbon dioxide-II, J. Quant. Spectrosc. Radiat. Transf. 25 (1981) 505–524. doi:10.1016/0022-4073(81)90026- . URLhttps://doi.org/10.1016/0022-4073(81)90026-1 23

    work page doi:10.1016/0022-4073(81)90026- 1981