Pith. sign in

REVIEW 4 major objections 3 minor 60 references

Environment-dependent charges plus Coulomb terms let machine-learning potentials recover LO–TO splitting and dielectric constants from energies, forces, and stresses alone.

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 · grok-4.5

2026-07-15 13:52 UTC pith:PH3YHWHV

load-bearing objection Abstract promises a real MLIP advance (charge-conserving long-range MTP recovering NaCl LO–TO and dielectric constants from E/F/stress only), but the cached full text is a completely different paper, so the physics cannot be audited. the 4 major comments →

arxiv 2603.06396 v2 pith:PH3YHWHV submitted 2026-03-06 physics.comp-ph

Long-range machine-learning potentials with environment-dependent charges enable predicting LO-TO splitting and dielectric constants

Dmitry Korogod , Alexander V. Shapeev , Ivan S. Novikov This is my paper
classification physics.comp-ph PACS 63.20.D-77.22.-d31.15.xv71.15.Pd
keywords machine-learning potentialslong-range electrostaticsenvironment-dependent chargesMoment Tensor PotentialLO-TO splittingdielectric constantphonon spectraNaCl
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.

Local machine-learning potentials miss long-range electrostatics, so they struggle with charged molecular complexes and polar crystals. This paper adds explicit Coulomb interactions driven by point charges that depend on each atom’s local environment, and optionally conserves the system’s total charge, then couples those terms to a local Moment Tensor Potential. On organic anion–molecule dimers and NaCl, the hybrid models cut training error relative to pure local fits and produce qualitatively correct binding curves. A phonon method that uses only the long-range models fitted to energies, forces, and stresses recovers the correct LO–TO splitting at the Γ-point for isotropic NaCl; molecular-dynamics dipole fluctuations with the same model yield a dielectric constant close to experiment. The same approach also gives a PbTiO₃ phonon spectrum that matches density-functional theory, even though the phonon construction was formally derived for isotropic materials. The practical claim is that environment-dependent charges plus classical Coulomb electrostatics are enough to restore the polar and dielectric observables that pure short-range machine-learning potentials cannot capture.

Core claim

A long-range model whose point charges depend on local atomic environments and that conserves total charge, combined with a local Moment Tensor Potential and trained only on energies, forces, and stresses, predicts the correct LO–TO splitting at the Γ-point for isotropic NaCl and a dielectric constant from molecular-dynamics dipole fluctuations that agrees with experiment; the same long-range construction also yields a PbTiO₃ phonon spectrum consistent with density-functional theory.

What carries the argument

Environment-dependent point charges inside classical Coulomb interactions (with optional total-charge conservation), combined with a local Moment Tensor Potential: the charges supply the missing long-range electrostatics while the local potential handles short-range chemistry, enabling phonon spectra and dielectric response without fitting those observables directly.

Load-bearing premise

That environment-dependent point charges plus classical Coulomb terms are a sufficient and transferable stand-in for true long-range electrostatics across both charged organic dimers and ionic or ferroelectric crystals, and that a phonon method derived for isotropic materials remains reliable for uniaxial systems such as PbTiO₃.

What would settle it

Compute LO–TO splitting at Γ and the static dielectric constant for NaCl (and the full phonon spectrum for tetragonal PbTiO₃) with the charge-conserving long-range MTP and compare to high-quality density-functional theory or experiment: a large mismatch in the LO–TO gap or dielectric constant would refute the central claim.

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

If this is right

  • Machine-learning potentials trained only on energies, forces, and stresses can still target polar and dielectric observables if long-range electrostatics are modeled with environment-dependent charges.
  • LO–TO splitting in isotropic crystals becomes accessible without separate Born-effective-charge or dielectric fitting.
  • Binding curves of charged organic dimers improve when total charge is conserved and Coulomb interactions are explicit.
  • The same long-range construction appears usable beyond isotropic crystals, as shown by agreement with DFT phonons for uniaxial PbTiO₃.
  • Dielectric constants can be estimated from molecular-dynamics dipole fluctuations driven by the long-range potential rather than from separate response calculations.

Where Pith is reading between the lines

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

  • If environment-dependent charges generalize, many existing short-range machine-learning force fields could be retrofitted with a charge model rather than retrained from scratch for polar materials.
  • Failure modes will likely appear first in systems with strong charge transfer, metallic screening, or highly anisotropic Born charges where a classical point-charge Coulomb form is inadequate.
  • The charge-conserving variant may be the more transferable default for charged non-periodic systems and defective crystals, even when total charge is nominally zero.
  • A natural next test is whether the same models recover infrared intensities or Born effective charges without extra training data.

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

4 major / 3 minor

Summary. The manuscript (as represented by its abstract) proposes two machine-learning interatomic potentials that add explicit long-range Coulomb electrostatics via environment-dependent point charges, one of which also enforces total-charge conservation. These long-range terms are combined with a local Moment Tensor Potential (MTP) and trained on energies, forces, and stresses for non-periodic organic dimers (CH3COO− with 4-methylphenol or 4-methylimidazole) and periodic NaCl. The abstract claims lower training errors than local MTP alone, qualitatively correct dimer binding curves, a phonon method for isotropic materials that recovers the correct LO–TO splitting at Γ for NaCl, a dielectric constant from MD dipole fluctuations in good agreement with experiment, and a phonon spectrum for uniaxial tetragonal PbTiO3 that matches DFT despite the method being formally derived for isotropic systems.

Significance. If the full results hold under scrutiny, the work would be a meaningful contribution to long-range ML potentials: recovering LO–TO splitting and a dielectric constant from models fitted only to energies, forces, and stresses (rather than to those observables) is a strong, falsifiable test of electrostatic representation. Combining environment-dependent charges with total-charge conservation and a local MTP is a clear, reusable design choice. The claimed extension from isotropic NaCl to uniaxial PbTiO3 would broaden impact if the phonon construction is shown to be controlled rather than fortuitous. These strengths cannot be confirmed from the abstract alone and require the actual methods, numbers, and baselines.

major comments (4)
  1. The supplied full-text body does not match arXiv:2603.06396. The cached manuscript is a different paper (schema-gated agentic AI workflows). Without the correct methods, equations for the charge models, phonon construction, tables of LO–TO/dielectric values, and PbTiO3 spectra, the central claims cannot be audited. A complete technical review requires the actual manuscript of 2603.06396.
  2. Abstract claim on LO–TO: the charge-conserving long-range model is said to predict the correct Γ-point LO–TO splitting for isotropic NaCl when fitted only to E/F/stress. This is the load-bearing result. The full paper must report the numerical LO and TO frequencies (model vs DFT/experiment), the non-analytic correction or equivalent construction used at Γ, and a local-MTP-only baseline showing that LO–TO is not recovered without the long-range term.
  3. Abstract claim on the phonon method: a method is introduced for phonon spectra of isotropic materials using only the long-range models, then applied to uniaxial tetragonal PbTiO3 with the statement that it is 'also perspective' for uniaxial materials. The formal isotropic derivation and the controlled approximation (or correction) used for uniaxial systems must be stated explicitly; qualitative DFT agreement alone does not establish that the isotropic construction is reliable for PbTiO3.
  4. Dielectric constant from MD dipole fluctuations: the abstract asserts good agreement with experiment for NaCl. The full paper must specify ensemble, boundary conditions, system size, fluctuation formula, and numerical value with uncertainty, plus comparison to experiment and to a local-only baseline, so that agreement is not under-specified.
minor comments (3)
  1. The abstract should state whether LO–TO and the dielectric constant were held out as pure predictions (never used in training or hyperparameter selection) and give at least one quantitative LO–TO and dielectric number for NaCl.
  2. Clarify early how total-charge conservation is enforced (soft penalty vs hard constraint) and whether charges are predicted per environment with a global correction, as this affects transferability between charged dimers and neutral crystals.
  3. Name the long-range summation scheme (Ewald/PME/etc.) and any dielectric boundary conditions used for periodic NaCl and PbTiO3, since these choices affect LO–TO and dielectric results.

Circularity Check

0 steps flagged

No circularity found in the claimed chain: LO–TO and dielectric constants are presented as post-fit predictions from potentials trained only on energies, forces, and stresses.

full rationale

From the supplied abstract of arXiv:2603.06396, the load-bearing chain is: (i) environment-dependent point-charge Coulomb models (one with total-charge conservation) combined with local MTP are fitted to energies, forces, and stresses on organic dimers and NaCl; (ii) a phonon method for isotropic materials is then applied using only those fitted long-range models; (iii) LO–TO splitting at Γ for NaCl and a dielectric constant from MD dipole-moment fluctuations are reported as predictions, with the latter compared to experiment; (iv) the same approach is applied to uniaxial PbTiO3 phonons and compared to DFT. None of these steps is self-definitional: LO–TO and ε are not fitting targets, so they are not forced by construction from the loss. There is no uniqueness theorem, no load-bearing self-citation chain, and no renaming of a known empirical pattern as the central result. Residual risk that charge models or training sets were informally tuned until LO–TO matched cannot be verified from the abstract and is not circularity under the hard rules (no speculation without a quoted reduction). Note: the CACHEABLE full-manuscript prefix is a different paper (arXiv:2603.06394, schema-gated agentic workflows), so equations, charge-model definitions, and numerical tables for 2603.06396 were unavailable; the assessment is therefore limited to the abstract’s stated derivation structure, which is self-contained against external benchmarks and shows no circular reduction.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 2 invented entities

Abstract-only ledger. The work rests on classical Coulomb electrostatics with environment-dependent point charges, optional global charge conservation, local MTP as the short-range model, and a phonon construction for isotropic materials that is asserted to extend to uniaxial PbTiO3. Free parameters (charge-network weights, MTP basis, cutoffs, training weights) are inevitable in ML potentials but not quantified here. No new physical particles are introduced; the invented entities are the two charge models and the phonon method.

free parameters (3)
  • Environment-dependent charge model parameters (network/weights mapping local environment to atomic charges)
    Charges are learned from local environments; their functional form and fitted coefficients are free parameters that the LO–TO and dielectric predictions depend on. Values not given in abstract.
  • Local MTP basis size, cutoffs, and training loss weights on energy/forces/stresses
    Standard MTP hyperparameters control short-range accuracy and how much the long-range term is needed; not specified in abstract.
  • Long-range electrostatic summation settings (Ewald/PME parameters, real-space cutoffs, dielectric boundary conditions)
    Coulomb evaluation in periodic systems requires numerical parameters that affect LO–TO and dielectric response; not stated in abstract.
axioms (4)
  • domain assumption Long-range electrostatics can be represented as classical Coulomb interactions between environment-dependent point charges (plus optional total-charge constraint).
    Core modeling choice stated in the abstract; multipoles, charge transfer beyond point charges, and electronic polarization are not included.
  • domain assumption Fitting only to energies, forces, and stresses is sufficient for the long-range model to predict LO–TO splitting and dielectric constants.
    Abstract claims phonon spectra and dielectric constant without fitting those observables directly; this is a strong transfer assumption.
  • ad hoc to paper A phonon method formally derived for isotropic materials remains applicable to uniaxial tetragonal PbTiO3.
    Abstract acknowledges formal isotropic derivation yet reports DFT-matching spectrum for PbTiO3; extension is empirical, not derived.
  • domain assumption Local Moment Tensor Potential plus the proposed long-range terms is an adequate short-range + long-range decomposition for the tested systems.
    Standard MLIP decomposition; success is demonstrated only on two dimers, NaCl, and PbTiO3 in the abstract.
invented entities (2)
  • Two long-range ML models with environment-dependent point charges (one with total-charge conservation) no independent evidence
    purpose: Supply explicit Coulomb electrostatics missing from local MTP and enable LO–TO and dielectric predictions.
    Architectural constructs of the paper; independent evidence would be open code, charge-validation on held-out systems, and published charge distributions—not available in abstract.
  • Method for phonon spectra of isotropic materials using only long-range models fitted to E/F/stress no independent evidence
    purpose: Compute LO–TO splitting and related lattice dynamics without separate non-analytic correction pipelines.
    Methodological contribution claimed in abstract; formal steps and assumptions not provided in available text.

pith-pipeline@v1.1.0-grok45 · 31224 in / 3194 out tokens · 34044 ms · 2026-07-15T13:52:01.295692+00:00 · methodology

0 comments
read the original abstract

We present two models with explicit long-range electrostatics in the form of Coulomb interactions. Both models include point charges depending on their local atomic environments, and the second model also conserves a total charge of an atomic system. We combine the proposed long-range models with local Moment Tensor Potential and demonstrate that they reduce the training errors of the MTP models fitted on the same training sets including the CH$_3$COO$^-$+4-methylphenol and CH$_3$COO$^-$+4-methylimidazole organic dimers (non-periodic systems) and the NaCl crystal (periodic system). For the organic dimers, the proposed models also give qualitatively correct predictions of the binding curves. Furthermore, in this study we introduce a method for calculating phonon spectra of isotropic materials only via these long-range models fitted to energies, forces, and stresses. The developed long-range model with point charges dependent on atomic environments and conserving total charge is capable of predicting the correct value of the LO-TO splitting in the $\Gamma$-point in the isotropic NaCl. For this system, we also predict dielectric constant from dipole moment fluctuations calculated with molecular dynamics simulations conducted with the developed long-range model. The calculated dielectric constant is in good agreement with experiment. Finally, we demonstrate the broader applicability of the introduced approach by computing the phonon spectrum of uniaxial tetragonal PbTiO$_3$. Although the method is formally derived for isotropic materials, we show that it is also perspective for uniaxial materials (e.g., PbTiO$_3$) as the spectrum obtained with our long-range interatomic potential corresponds to the one calculated with density functional theory.

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

60 extracted references · 21 canonical work pages

  1. [1]

    Strickland, B

    J. Strickland, B. Nenchev, K. Tassenberg, S. Perry, G. Sheppard, H. Dong, R. Zhang, G. Burca, and N. D’Souza. On the origin of mosaicity in directionally solidified ni-base superalloys.Acta Materialia, 217:117180, 2021. doi: 10.1016/j.actamat.2021.117180

    work page doi:10.1016/j.actamat.2021.117180 2021

  2. [2]

    Strickland, B

    J. Strickland, B. Nenchev, S. Perry, K. Tassenberg, S. Gill, C. Panwisawas, H. Dong, N. D’Souza, and S. Irwin. On the nature of hexagonality within the solidification structure of single crystal alloys: Mechanisms and applications.Acta Materialia, 200:417–431, 2020. doi: 10.1016/j.actamat.2020.09.019

    work page doi:10.1016/j.actamat.2020.09.019 2020

  3. [3]

    Strickland, M

    J. Strickland, M. Ghisoni, H. Marshall, T. Whitehead, B. Nenchev, B. Pellegrini, C. Phillips, K. Tassenberg, G. Conduit, S. Davey, S. Dorman, D. Ferguson, and J. Sol. Degrees of uncer- tainty: conformal deep learning for non-invasive core body temperature prediction in extreme environments.Communications Engineering, 4:219, 2025. doi: 10.1038/s44172-025-00548-6

    work page doi:10.1038/s44172-025-00548-6 2025

  4. [4]

    Cohen-Boulakia, K

    S. Cohen-Boulakia, K. Belhajjame, O. Collin, J. Chopard, C. Froidevaux, A. Gaignard, K. Hinsen, P. Larmande, Y. Le Bras, F. Lemoine, and F. Mareuil. Scientific workflows for computational reproducibility in the life sciences: Status, challenges and opportunities. Future Generation Computer Systems, 75:284–298, 2017. doi: 10.1016/j.future.2017.01.012

    doi:10.1016/j.future.2017.01.012 2017

  5. [5]

    T. Hope, D. Downey, D. S. Weld, O. Etzioni, and E. Horvitz. A computational inflection for scientific discovery.Communications of the ACM, 66(8):62–73, 2023. doi: 10.1145/3576896

    doi:10.1145/3576896 2023

  6. [6]

    Foundation models for scientific discovery: From paradigm enhancement to paradigm transition

    Fan Liu, Jindong Han, Tengfei Lyu, Weijia Zhang, Zhe-Rui Yang, Lu Dai, Cancheng Liu, and Hao Liu. Foundation models for scientific discovery: From paradigm enhancement to paradigm transition. arXiv preprint arXiv:2510.15280, 2025. Preprint. 32

    arXiv 2025

  7. [7]

    S. P. Huber, E. Bosoni, M. Bercx, J. Bröder, A. Degomme, V. Dikan, K. Eimre, E. Flage- Larsen, A. Garcia, L. Genovese, and D. Gresch. Common workflows for computing material properties using different quantum engines.npj Computational Materials, 7(1):136, 2021. doi: 10.1038/s41524-021-00594-6

    doi:10.1038/s41524-021-00594-6 2021

  8. [8]

    A. M. Ganose, H. Sahasrabuddhe, M. Asta, K. Beck, T. Biswas, A. Bonkowski, J. Busta- mante, X. Chen, Y. Chiang, D. C. Chrzan, and J. Clary. Atomate2: Modular workflows for materials science.Digital Discovery, 4(7):1944–1973, 2025. doi: 10.1039/D5DD00019J

    doi:10.1039/d5dd00019j 1944

  9. [9]

    D. C. Elton, Z. Boukouvalas, M. S. Butrico, M. D. Fuge, and P. W. Chung. Applying machine learning techniques to predict the properties of energetic materials.Scientific Reports, 8(1):9059, 2018. doi: 10.1038/s41598-018-27344-x

    work page doi:10.1038/s41598-018-27344-x 2018

  10. [10]

    R. A. Patel, C. H. Borca, and M. A. Webb. Featurization strategies for polymer sequence or composition design by machine learning.Molecular Systems Design & Engineering, 7(6): 661–676, 2022. doi: 10.1039/D1ME00160D

    work page doi:10.1039/d1me00160d 2022

  11. [11]

    A. L. Lamprecht, M. Palmblad, J. Ison, V. Schwämmle, M. S. Al Manir, I. Altintas, C. J. Baker, A. B. H. Amor, S. Capella-Gutierrez, P. Charonyktakis, and M. R. Crusoe. Perspectives on automated composition of workflows in the life sciences.F1000Research, 10:897, 2021. doi: 10.12688/f1000research.54159.1

    work page doi:10.12688/f1000research.54159.1 2021

  12. [12]

    Singh, S

    A. Singh, S. Purawat, A. Rao, and I. Altintas. Modular performance prediction for scientific workflows using machine learning.Future Generation Computer Systems, 114:1–14, 2021. doi: 10.1016/j.future.2020.04.048

    work page doi:10.1016/j.future.2020.04.048 2021

  13. [13]

    O. J. R. Gustafsson, S. R. Wilkinson, F. Bacall, S. Soiland-Reyes, S. Leo, L. Pireddu, S. Owen, N. Juty, J. M. Fernández, T. Brown, and H. Ménager. Workflowhub: a registry for computational workflows.Scientific Data, 12(1):837, 2025. doi: 10.1038/s41597-025-04786-3

    work page doi:10.1038/s41597-025-04786-3 2025

  14. [14]

    J. C. J. Strickland, P. F. D. Woolston, and T. M. Whitehead. Adaptive experimental design. InThe Digital Transformation of Product Formulation. CRC Press, 2024. doi: 10.1201/9781003385974-15

    work page doi:10.1201/9781003385974-15 2024

  15. [15]

    Farshidi, X

    S. Farshidi, X. Liao, N. Li, D. Goldfarb, B. Magagna, M. Stocker, K. Jeffery, P. Thijsse, C. Pichot, A. Petzold, and Z. Zhao. Knowledge sharing and discovery across heterogeneous research infrastructures.Open Research Europe, 1:68, 2023. doi: 10.12688/openreseurope. 13677.3

    doi:10.12688/openreseurope 2023

  16. [16]

    L. C. McInnes, D. Arnold, P. Balaprakash, M. Bernhardt, B. Cerny, A. Dubey, R. Giles, D. W. Hood, M. A. Leung, V. Lopez-Marrero, and P. Messina. Report of the 2025 workshop on next-generation ecosystems for scientific computing: Harnessing community, software, and ai for cross-disciplinary team science. arXiv preprint arXiv:2510.03413, 2025. arXiv

    arXiv 2025

  17. [17]

    Cambridge University Press, 2006

    S.OliveiraandD.E.Stewart.Writing Scientific Software: A Guide to Good Style. Cambridge University Press, 2006. ISBN 9780521675956

    2006

  18. [18]

    K. Alam, B. Roy, C. K. Roy, and K. Mittal. An empirical investigation on the challenges in scientific workflow systems development.Empirical Software Engineering, 30(5):151, 2025. doi: 10.1007/s10664-025-10705-2

    work page doi:10.1007/s10664-025-10705-2 2025

  19. [19]

    B. D. Lee. Ten simple rules for documenting scientific software.PLOS Computational Biology, 14(12):e1006561, 2018. doi: 10.1371/journal.pcbi.1006561. 33

    work page doi:10.1371/journal.pcbi.1006561 2018

  20. [20]

    J. Wang, G. Xiao, S. Zhang, H. Lei, Y. Liu, and Y. Sui. Compatibility issues in deep learning systems: Problems and opportunities. InProceedings of the 31st ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE ’23), pages 476–488, nov 2023. doi: 10.1145/3611643.3616321

    doi:10.1145/3611643.3616321 2023

  21. [21]

    A. D. Pimentel, T. Stefanov, H. Nikolov, M. Thompson, S. Polstra, and E. F. Deprettere. Tool integration and interoperability challenges of a system-level design flow: A case study. InEmbedded Computer Systems: Architectures, Modeling, and Simulation (SAMOS 2008), Lecture Notes in Computer Science, volume 5114, pages 167–176. Springer, jul 2008. doi: 10.1...

    work page doi:10.1007/978-3-540-70550-5_19 2008

  22. [22]

    G. K. Sandve, A. Nekrutenko, J. Taylor, and E. Hovig. Ten simple rules for reproducible computational research.PLOS Computational Biology, 9(10):e1003285, 2013. doi: 10.1371/ journal.pcbi.1003285

    2013

  23. [23]

    Stodden and S

    V. Stodden and S. Miguez. Best practices for computational science: Software infrastructure and environments for reproducible and extensible research.Journal of Open Research Software, 2(1):e21, 2014. doi: 10.5334/jors.ay

    work page doi:10.5334/jors.ay 2014

  24. [24]

    R. C. Jiménez, M. Kuzak, M. Alhamdoosh, M. Barker, B. Batut, M. Borg, S. Capella- Gutierrez, N. C. Hong, M. Cook, M. Corpas, and M. Flannery. Four simple recommendations to encourage best practices in research software.F1000Research, 6:ELIXIR–876, 2017. doi: 10.12688/f1000research.11407.1

    doi:10.12688/f1000research.11407.1 2017

  25. [25]

    Freire, D

    J. Freire, D. Koop, E. Santos, and C. T. Silva. Provenance for computational tasks: A survey. Computing in Science & Engineering, 10(3):11–21, 2008. doi: 10.1109/MCSE.2008.79

    work page doi:10.1109/mcse.2008.79 2008

  26. [26]

    Missier, S

    P. Missier, S. Dey, K. Belhajjame, V. Cuevas-Vicenttín, and B. Ludäscher. D-PROV: Extending the PROV provenance model with Workflow structure. In5th USENIX Workshop on the Theory and Practice of Provenance (TaPP 13), 2013. URL������������������� �������������������������������������������������������������

    2013

  27. [27]

    Moreau, B

    L. Moreau, B. Clifford, J. Freire, J. Futrelle, Y. Gil, P. Groth, N. Kwasnikowska, S. Miles, P. Missier, J. Myers, and B. Plale. The open provenance model core specification (v1.1). Future Generation Computer Systems, 27(6):743–756, 2011. doi: 10.1016/j.future.2010.07. 005

    work page doi:10.1016/j.future.2010.07 2011

  28. [28]

    Boettiger

    C. Boettiger. An introduction to docker for reproducible research.ACM SIGOPS Operating Systems Review, 49(1):71–79, 2015. doi: 10.1145/2723872.2723882

    doi:10.1145/2723872.2723882 2015

  29. [29]

    Grüning, J

    B. Grüning, J. Chilton, J. Köster, R. Dale, N. Soranzo, M. Van Den Beek, J. Goecks, R. Backofen, A. Nekrutenko, and J. Taylor. Practical computational reproducibility in the life sciences.Cell Systems, 6(6):631–635, 2018. doi: 10.1016/j.cels.2018.03.014

    work page doi:10.1016/j.cels.2018.03.014 2018

  30. [30]

    J. E. Hannay, C. MacLeod, J. Singer, H. P. Langtangen, D. Pfahl, and G. Wilson. How do scientists develop and use scientific software? In2009 ICSE Workshop on Software Engineering for Computational Science and Engineering, pages 1–8. IEEE, may 2009. doi: 10.1109/SECSE.2009.5069155

    doi:10.1109/secse.2009.5069155 2009

  31. [31]

    Hettrick, M

    S. Hettrick, M. Antonioletti, L. Carr, N. Chue Hong, S. Crouch, D. C. De Roure, I. Emsley, C. Goble, A. Hay, D. Inupakutika, and M. Jackson. Uk research software survey 2014. Zenodo dataset, 2014. Dataset

    2014

  32. [32]

    M. S. Krafczyk, A. Shi, A. Bhaskar, D. Marinov, and V. Stodden. Learning from re- producing computational results: introducing three principles and the reproduction pack- age.Philosophical Transactions of the Royal Society A, 379(2197):20200069, 2021. doi: 10.1098/rsta.2020.0069. 34

    work page doi:10.1098/rsta.2020.0069 2021

  33. [33]

    Ziemann, P

    M. Ziemann, P. Poulain, and A. Bora. The five pillars of computational reproducibility: bioinformatics and beyond.Briefings in Bioinformatics, 24(6):bbad375, 2023. doi: 10.1093/ bib/bbad375

    2023

  34. [34]

    de Oliveira Andrade

    R. de Oliveira Andrade. Huge reproducibility project fails to validate dozens of biomedical studies.Nature, 641(8062):293–294, 2025. doi: 10.1038/d41586-025-01266-x

    work page doi:10.1038/d41586-025-01266-x 2025

  35. [35]

    Yildiz and T

    O. Yildiz and T. Peterka. Do large language models speak scientific workflows? arXiv preprint arXiv:2412.10606, 2024. arXiv

    arXiv 2024

  36. [36]

    From prompt to pipeline: Large language models for scientific workflow development in bioinformatics

    Khairul Alam and Banani Roy. From prompt to pipeline: Large language models for scientific workflow development in bioinformatics. arXiv preprint arXiv:2507.20122, 2025. arXiv

    Pith/arXiv arXiv 2025

  37. [37]

    Jiang, W

    X. Jiang, W. Wang, S. Tian, H. Wang, T. Lookman, and Y. Su. Applications of natural language processing and large language models in materials discovery.npj Computational Materials, 11(1):79, 2025. doi: 10.1038/s41524-025-01554-0

    doi:10.1038/s41524-025-01554-0 2025

  38. [38]

    S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. Narasimhan, and Y. Cao. React: Synergizing reasoning and acting in language models. arXiv preprint arXiv:2210.03629, 2022

    Pith/arXiv arXiv 2022

  39. [39]

    Executable code actions elicit better LLM agents

    Xingyao Wang, Yangyi Chen, Lifan Yuan, Yizhe Zhang, Yunzhu Li, Hao Peng, and Heng Ji. Executable code actions elicit better LLM agents. InProceedings of the 41st International Conference on Machine Learning, volume 235 ofProceedings of Machine Learning Research, pages 50208–50232, 2024. URL�����������������������������������������������

    2024

  40. [40]

    A. M. Bran, S. Cox, O. Schilter, C. Baldassari, A. D. White, and P. Schwaller. Augmenting large language models with chemistry tools.Nature Machine Intelligence, 6:525–535, 2024. doi: 10.1038/s42256-024-00832-8

    doi:10.1038/s42256-024-00832-8 2024

  41. [41]

    W. Shin, R. Souza, D. Rosendo, F. Suter, F. Wang, P. Balaprakash, and R. F. da Silva. The (r) evolution of scientific workflows in the agentic ai era: Towards autonomous science. arXiv preprint arXiv:2509.09915, 2025

    Pith/arXiv arXiv 2025

  42. [42]

    Morishige and R

    M. Morishige and R. Koshihara. Ensuring reproducibility in generative ai systems for general use cases: A framework for regression testing and open datasets. arXiv preprint arXiv:2505.02854, 2025

    Pith/arXiv arXiv 2025

  43. [43]

    Hosseini, S

    M. Hosseini, S. P. Horbach, K. Holmes, and T. Ross-Hellauer. Open science at the generative ai turn: An exploratory analysis of challenges and opportunities.Quantitative Science Studies, 6:22–45, 2025. doi: 10.1162/qss_a_00337

    doi:10.1162/qss_a_00337 2025

  44. [44]

    Baltes, F

    S. Baltes, F. Angermeir, C. Arora, M. M. Barón, C. Chen, L. Böhme, F. Calefato, N. Ernst, D. Falessi, B. Fitzgerald, and D. Fucci. Evaluation guidelines for empirical studies in software engineering involving large language models. arXiv preprint arXiv:2508.15503,

    Pith/arXiv arXiv

  45. [45]

    Brucks and O

    M. Brucks and O. Toubia. Prompt architecture induces methodological artifacts in large language models.PLOS ONE, 20(4):e0319159, 2025. doi: 10.1371/journal.pone.0319159

    doi:10.1371/journal.pone.0319159 2025

  46. [46]

    Kervadec, F

    C. Kervadec, F. Franzon, and M. Baroni. Unnatural language processing: How do language models handle machine-generated prompts? arXiv preprint arXiv:2310.15829, 2023

    Pith/arXiv arXiv 2023

  47. [47]

    Z. Ji, N. Lee, R. Frieske, T. Yu, D. Su, Y. Xu, E. Ishii, Y. J. Bang, A. Madotto, and P. Fung. Survey of hallucination in natural language generation.ACM Computing Surveys, 55(12): 1–38, 2023. doi: 10.1145/3571730. 35

    doi:10.1145/3571730 2023

  48. [48]

    F. Jia, T. Wu, X. Qin, and A. Squicciarini. The task shield: Enforcing task alignment to defend against indirect prompt injection in llm agents. InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 29680–29697, jul 2025. URL��������������������������������������������

    2025

  49. [49]

    Hines, G

    K. Hines, G. Lopez, M. Hall, F. Zarfati, Y. Zunger, and E. Kiciman. Defending against indirect prompt injection attacks with spotlighting. arXiv preprint arXiv:2403.14720, 2024

    Pith/arXiv arXiv 2024

  50. [50]

    Y. Hu, S. Wang, T. Nie, Y. Zhao, and H. Wang. Understanding large language model supply chain: Structure, domain, and vulnerabilities. arXiv preprint arXiv:2504.20763, 2025

    Pith/arXiv arXiv 2025

  51. [51]

    B. C. Das, M. H. Amini, and Y. Wu. Security and privacy challenges of large language models: A survey.ACM Computing Surveys, 57(6):1–39, 2025. doi: 10.1145/3712001

    doi:10.1145/3712001 2025

  52. [52]

    Goecks, A

    J. Goecks, A. Nekrutenko, J. Taylor, and Galaxy Team. Galaxy: a comprehensive approach for supporting accessible, reproducible, and transparent computational research in the life sciences.Genome Biology, 11(8):R86, 2010. doi: 10.1186/gb-2010-11-8-r86

    work page doi:10.1186/gb-2010-11-8-r86 2010

  53. [53]

    Köster and S

    J. Köster and S. Rahmann. Snakemake—a scalable bioinformatics workflow engine.Bioin- formatics, 28(19):2520–2522, 2012. doi: 10.1093/bioinformatics/bts480

    doi:10.1093/bioinformatics/bts480 2012

  54. [54]

    Di Tommaso, M

    P. Di Tommaso, M. Chatzou, E. W. Floden, P. P. Barja, E. Palumbo, and C. Notredame. Nextflow enables reproducible computational workflows.Nature Biotechnology, 35(4): 316–319, 2017. doi: 10.1038/nbt.3820

    doi:10.1038/nbt.3820 2017

  55. [55]

    Mölder, K

    F. Mölder, K. P. Jablonski, B. Letcher, M. B. Hall, P. C. van Dyken, C. H. Tomkins-Tinch, V. Sochat, J. Forster, F. G. Vieira, C. Meesters, and S. Lee. Sustainable data analysis with snakemake.F1000Research, 10:33, 2021. doi: 10.12688/f1000research.29032.3

    doi:10.12688/f1000research.29032.3 2021

  56. [56]

    Kanwal, F

    S. Kanwal, F. Z. Khan, A. Lonie, and R. O. Sinnott. Investigating reproducibility and tracking provenance—a genomic workflow case study.BMC Bioinformatics, 18(1):337, 2017. doi: 10.1186/s12859-017-1747-0

    work page doi:10.1186/s12859-017-1747-0 2017

  57. [57]

    Using thematic analysis in psychology.Qualitative Research in Psychology, 3(2):77–101, 2006

    Virginia Braun and Victoria Clarke. Using thematic analysis in psychology.Qualitative Research in Psychology, 3(2):77–101, 2006. doi: 10.1191/1478088706qp063oa

    doi:10.1191/1478088706qp063oa 2006

  58. [58]

    Model context protocol specification, 2024

    Anthropic. Model context protocol specification, 2024. URL �������� �������������������������������������. Accessed: 2025-06-01

    2024

  59. [59]

    Ewels, Alexander Peltzer, Sven Fillinger, Harshil Patel, Johannes Alneberg, Andreas Wilm, Maxime Ulysse Garcia, Paolo Di Tommaso, and Sven Nahnsen

    Philip A. Ewels, Alexander Peltzer, Sven Fillinger, Harshil Patel, Johannes Alneberg, Andreas Wilm, Maxime Ulysse Garcia, Paolo Di Tommaso, and Sven Nahnsen. The nf-core framework for community-curated bioinformatics pipelines.Nature Biotechnology, 38(3): 276–278, 2020. doi: 10.1038/s41587-020-0439-x

    work page doi:10.1038/s41587-020-0439-x 2020

  60. [60]

    B. D. Conduit, N. G. Jones, H. J. Stone, and G. J. Conduit. Design of a nickel-base superalloy using a neural network.Materials & Design, 131:358–365, 2017. doi: 10.1016/j. matdes.2017.06.007. 36

    doi:10.1016/j 2017