Pith. sign in

REVIEW 4 major objections 2 minor 1 cited by

Excavating The Ruins: an Ancient $z=2.675$ Galaxy Which Formed in the First 500 Myr

T0 review · 4 major / 2 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Eridu, a quiescent galaxy at z=2.675, appears to have assembled a stellar mass of about 10^11 solar masses within 300 million years of the Big Bang and to have stopped forming stars by z~10.

desk verdict The abstract describes an exciting ancient galaxy, but the full text is an unrelated materials-science benchmark, so the result is unverifiable as submitted. read the letter →

arxiv 2508.05752 v1 pith:SJKPYSA4 submitted 2025-08-07 astro-ph.GA

Ian McConachie , Jacqueline Antwi-Danso , Wenjun Chang , M. C. Cooper , Adit Edward , Ben Forrest , Percy Gomez , Han Lei
show 9 more authors
Zach J. Lewis Danilo Marchesini Michael V. Maseda Adam Muzzin Allison Noble Stephanie M. Urbano Stawinski Tracy Webb Gillian Wilson M. E. Wisz
This is my paper · ORCID
classification astro-ph.GA
keywords quiescentgalaxiesJWSTNIRSpecstarformationhistorieshigh-redshiftalphaenhancementSMILEScosmicdawn
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper reports a single galaxy, Eridu, at $z=2.675$ that is already dead: fitting its JWST/NIRSpec spectra together with $0.4$–$25\,\mu\mathrm{m}$ HST, NIRCam, and MIRI photometry gives a stellar mass of $\log(M_*/\mathrm{M_\odot})=10.96$ and an average star formation rate below $1\,\mathrm{M_\odot\,yr^{-1}}$ over the past 100 Myr. The consequential claim is the inferred history: several independent star formation history models place the assembly of this mass within roughly 300 Myr of the Big Bang, with quenching complete by $z\sim10$. That implies a progenitor of about $10^{11}\,\mathrm{M_\odot}$ at $z>10$, nearly two orders of magnitude above the most massive galaxies currently observed at cosmic dawn. The strongly $\alpha$-enhanced ratio $[\mathrm{Mg/Fe}]=+0.65$ independently supports rapid enrichment, and Eridu's location in a massive protostructure offers an environmental route to early quenching. A sympathetic reader would take the paper as evidence that fully grown, dead galaxies may exist much earlier than current models and observations anticipate.

What carries the argument

The argument is carried by stellar population synthesis (SPS) fitting of the JWST/NIRSpec G140M and G235M spectra together with HST, JWST/NIRCam, and JWST/MIRI photometry from $0.4$ to $25\,\mu\mathrm{m}$, giving stellar mass, age, and star formation history; the $[\mathrm{Mg/Fe}]$ ratio measured from the spectrum acts as an independent chemical clock. The star formation history inversion is the step that converts an already-quiescent $z=2.675$ galaxy into a claim about formation at $z>10$.

What would settle it

A deep rest-frame optical spectrum of Eridu with signal-to-noise high enough to measure velocity dispersion and [O II] emission: if the dynamical mass is inconsistent with $10^{11}\,\mathrm{M_\odot}$ at the factor-of-two level, or if [O II] implies a star formation rate above $1\,\mathrm{M_\odot\,yr^{-1}}$, the quiescent-mass and early-formation claims would fail. A second decisive check is whether any $z>10$ survey finds a $\sim10^{11}\,\mathrm{M_\odot}$ progenitor population; if none exists, the inferred formation history is unsupported.

Watch

Extended reading notes

Core claim

Eridu is a $z=2.675$ galaxy whose rest-frame spectrum and multiwavelength photometry, modeled simultaneously, support a stellar mass of $10^{10.96}\,\mathrm{M_\odot}$ with no significant star formation over the last 100 Myr. The paper's central claim is that this present-day quiescence is the surface of a more extreme history: the star formation histories recovered by different modeling codes converge on assembly within about 300 Myr of the Big Bang and shutdown of star formation at $z\sim10$. If correct, Eridu's progenitor was a fully assembled $\sim10^{11}\,\mathrm{M_\odot}$ stellar system during the cosmic dawn, exceeding by about two orders of magnitude the most massive $z>10$ galaxies f

Load-bearing premise

The load-bearing assumption is that the stellar population synthesis modeling of the NIRSpec spectra and the 0.4–25 micron photometry—including its IMF, dust attenuation, and metallicity treatment—recovers the true stellar mass and star formation history, so that the implied formation at z~10 is physical rather than a modeling artifact.

Editorial extensions

If this is right

  • If Eridu's early assembly is correct, the early universe could build roughly $10^{11}\,\mathrm{M_\odot}$ of stars within about 300 Myr of the Big Bang, a rate current galaxy formation models do not naturally produce.
  • A quenching age of $z\sim10$ pushes the end of star formation into the epoch now being directly observed by JWST, making Eridu a concrete test case for environmental quenching inside a massive protostructure.
  • The $[\mathrm{Mg/Fe}]=+0.65$ measurement provides a chemical-clock consistency check: any successful model must simultaneously explain rapid assembly and strong alpha enhancement.
  • If the inferred progenitor mass at $z>10$ is representative, the bright end of the cosmic-dawn galaxy population must be revised upward; if Eridu is an outlier, star formation histories must explain why one system deviates so strongly.

Reading between the lines

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

  • The quoted mass uncertainty of $\pm0.01$ dex is smaller than SED-fitting systematics usually allow; with realistic dust, IMF, and metallicity priors the uncertainty is likely several times larger, and a lower true mass would soften the tension without necessarily removing the early-formation signal.
  • A natural independent test is dynamical: measuring the velocity dispersion from Eridu's absorption lines would yield a mass that can be compared with the photometric $10^{11}\,\mathrm{M_\odot}$; a large discrepancy would point to problems in the SPS assumptions.
  • If Eridu-like galaxies are common, deep surveys should already be finding massive, red, quiescent systems at $z\sim3$; counting such systems across the full survey footprint would quantify how unusual Eridu actually is.
  • The result also suggests that alpha-element ratios of quiescent galaxies at $z\sim3$ can serve as a fossil record of cosmic-dawn enrichment, linking early enrichment timescales to later quenching histories.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 2 minor

Summary. The submission presents an astrophysics abstract (arXiv:2508.05752, astro-ph.GA) claiming a detailed analysis of a z=2.675 quiescent galaxy ('Eridu') using JWST/NIRSpec spectroscopy and 0.4--25 micron photometry, with derived stellar mass log(M*/M_sun)=10.96, SFR<1 M_sun/yr over the last 100 Myr, formation within ~300 Myr of the Big Bang, quenching by z~10, and [Mg/Fe]=+0.65. The full text supplied, however, is an entirely different manuscript, arXiv:2508.05762v3 (cond-mat.mtrl-sci), titled 'UniFFBench: Evaluating Universal Machine Learning Force Fields Against Experimental Measurements.' This body contains no galaxy data, no NIRSpec reduction, no SED fitting, no star-formation-history modeling, and no analysis of Eridu. The abstract's central claims are therefore unsupported by the manuscript as submitted.

Significance. If the abstract's claims were substantiated, the result would be significant: a ~10^11 M_sun quiescent galaxy at z=2.675 that formed within 300 Myr of the Big Bang and quenched by z~10 would challenge current models of early massive galaxy assembly and provide a striking test for stellar population synthesis at high redshift. The abstract's [Mg/Fe] enhancement and environmental interpretation would add further interest. However, none of this analysis is present in the submitted full text. The unrelated materials-science body does contain reproducible code and datasets in its own domain, but that is not the claimed astrophysical contribution. As submitted, the manuscript provides no verifiable scientific content relevant to its abstract, so its potential significance cannot be assessed.

major comments (4)
  1. [Full Text, Sections 1-4 and Supplementary Information] The submitted body is not the paper described in the abstract. It is a materials-science benchmark on universal machine-learning force fields, with a different title, author list, and subject matter. It contains no mention of Eridu, JWST, NIRSpec, SMILES, JADES, or z=2.675. The abstract's quantitative claims are therefore entirely unsupported by the manuscript text. This is a load-bearing verifiability failure: no equation, figure, table, or derivation in the body corresponds to the claimed galaxy analysis.
  2. [Abstract, first sentence] The abstract states that stellar mass, SFR, star formation history, and [Mg/Fe] are derived by simultaneously modeling NIRSpec G140M/G235M spectra and 0.4-25 micron photometry. The manuscript provides none of the required information: no data reduction, no spectral extraction, no definition of the SED-fitting or stellar-population-synthesis code, no treatment of dust attenuation, IMF, or metallicity priors, and no uncertainty covariance. In particular, the quoted stellar-mass uncertainty of +/-0.01 dex is implausible for SED-based estimates absent a detailed description of calibration and template systematics; with the manuscript as it stands, this uncertainty has no evidentiary basis.
  3. [Abstract, SFH claim] The abstract attributes the early-formation result to 'star formation histories inferred from various models,' but the manuscript gives no model definitions, prior choices, or comparison sets. The claim that formation occurred within ~300 Myr of the Big Bang and quenching by z~10 can only be evaluated if the authors show that their SFH parametrization allowed late or extended formation scenarios and that the result is robust across template libraries, IMF assumptions, and dust laws. Absent these details, the abstract's central inference is unfalsifiable in this submission. Note also that the abstract's own wording ('inferred from various models') indicates the result is model-dependent, and no model details are provided.
  4. [Abstract, progenitor mass and environment] The statement that the progenitor reached M* ~ 10^11 M_sun at z>10 depends on extrapolating the derived SFH to early times, and the claim that Eridu inhabits a massive protostructure is asserted without supporting observations or analysis. Neither the SFH integration nor any environmental characterization (e.g., galaxy density, merger indicators) appears in the manuscript. These are part of the paper's headline conclusions and require quantitative support.
minor comments (2)
  1. [Abstract] The abstract contains a duplicated word: 'the the JADES+SMILES photometric catalogs.'
  2. [References] The reference list is entirely from the unrelated materials-science manuscript; there are no references to JWST/NIRSpec, SMILES, JADES, stellar population synthesis codes, or high-redshift quiescent galaxy studies. The abstract therefore cannot be placed in context even at the level of citations.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity assessable: the supplied full text is a different paper; Eridu's derivation is absent.

full rationale

The target paper (arXiv:2508.05752, the Eridu galaxy analysis) is represented only by its abstract. The full text provided is arXiv:2508.05762v3, a materials-science benchmark paper on universal machine learning force fields (UniFFBench). None of the Eridu derivation chain — NIRSpec reduction, 0.4–25 micron SED fitting, SFH inversion, [Mg/Fe] measurement, or the quoted ±0.01 dex mass uncertainty — appears in the submitted manuscript body. The abstract's statement that star formation histories are 'inferred from various models' indicates model-dependence, but without the actual model equations, priors, or fitting procedures there is no quoted reduction by which any claimed result becomes equivalent to its inputs. Hard rule 1 requires quoting the paper and exhibiting a specific reduction (e.g., Eq. X = Eq. Y by construction) before flagging circularity; no such reduction can be exhibited for the Eridu results. The noted mismatch is a verifiability failure, not a demonstrated circular step. Accordingly, the score is 0, meaning no significant circularity is established from the available material, with the explicit caveat that the central claim cannot be checked until the actual Eridu paper body is supplied.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The abstract only, no equations or data. All numbers are model-dependent.

free parameters (2)
  • Star formation history parameters (onset time, timescale, burst fractions) = not stated
    The SFH 'inferred from various models' is fit to the SED; results depend on these parameters.
  • Stellar population synthesis template set, IMF, and dust law = not stated
    Stellar mass, age, and [Mg/Fe] inference depend on these choices.
assumptions (3)
  • domain assumption The spectroscopic redshift z=2.675 is accurate and the galaxy is a single object
    All derived quantities scale with redshift; the abstract gives no confirmation of line identification.
  • domain assumption The stellar population synthesis models and abundance diagnostics used are valid at z=2.7 and for the stellar populations of Eridu
    Mass, SFH, and [Mg/Fe] are inferred through models; if the models' priors exclude very early formation, the derived formation time is forced.
  • domain assumption The [Mg/Fe] measurement from G140M and G235M spectra is not degenerate with age, metallicity, or dust
    The abstract reports [Mg/Fe]=+0.65 with small uncertainties; this relies on spectral synthesis assumptions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Excavating The Ruins: an Ancient $z=2.675$ Galaxy Which Formed in the First 500 Myr." pith.science (2026). https://pith.science/paper/SJKPYSA4

@misc{pith2026250805752,
  author       = {Pith},
  title        = {Pith review of: Excavating The Ruins: an Ancient $z=2.675$ Galaxy Which Formed in the First 500 Myr},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SJKPYSA4}},
  note         = {Machine review of arXiv:2508.05752}
}
abstract

We present the analysis of an ancient galaxy at $z=2.675$ which we dub ``Eridu.'' Simultaneously modeling the JWST/NIRSpec G140M and G235M spectra from the SMILES program and $0.4-25\ \mu\mathrm{m}$ HST, JWST/NIRCam, and JWST/MIRI photometry from the the JADES+SMILES photometric catalogs shows that Eridu is massive and quiescent with stellar mass $\log(M_*/\mathrm{M_\odot})=10.96^{+0.01}_{-0.01}$ and average star formation rate $<1\ \mathrm{M_\odot\ yr^{-1}}$ over the last 100 Myr. Star formation histories inferred from various models produce disconcertingly early and fast formation within $\sim300$ Myr of the Big Bang and quenching 2 Gyr prior to observation ($z\sim10$). This stellar mass assembly implies that the progenitor of Eridu had $M_*\approx10^{11}\ \mathrm{M_\odot}$ at $z>10$, nearly two orders of magnitude more than the most massive current high redshift observations. From Eridu's spectrum we infer $\mathrm{[Mg/Fe]} =+0.65^{+0.20}_{-0.19}$, indicating its stellar population is extremely $\alpha$-enhanced, which is consistent with the rapid formation timescale inferred from its star formation history. Eridu inhabits a massive protostructure which offers additional explanations for rapid mass assembly and quenching via environmental mechanisms, e.g. major mergers. Though its inferred formation is at odds with observations of the brightest cosmic dawn galaxies, we anticipate that future high-redshift galaxy formation models and sophisticated stellar population modeling codes will unearth how Eridu formed at the dawn of time.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. SQuIGG$\vec{L}$E: Buried star formation cannot explain the rapidly fading CO(2-1) luminosity in massive, $z\sim0.7$ post-starburst galaxies

    astro-ph.GA 2025-08 conditional novelty 6.0 of 10

    Buried star formation cannot explain the rapid CO fading in z~0.7 post-starburst galaxies, so gas-rich recently quenched galaxies may rejuvenate instead of directly becoming quiescent.

Reference graph

Works this paper leans on

23 extracted references · 22 canonical work pages · cited by 1 Pith paper

  1. [1]

    Chgnet as a pretrained universal neural network potential for charge-informed atomistic modelling.������ ������� ������������, 5(9):1031–1041, 2023

    Bowen Deng, Peichen Zhong, KyuJung Jun, Janosh Riebesell, Kevin Han, Christopher J Bartel, and Gerbrand Ceder. Chgnet as a pretrained universal neural network potential for charge-informed atomistic modelling.������ ������� ������������, 5(9):1031–1041, 2023

    work page 2023

  2. [2]

    A universal graph deep learning interatomic potential for the periodic table.������ ������������� �������, 2(11):718–728, 2022

    Chi Chen and Shyue Ping Ong. A universal graph deep learning interatomic potential for the periodic table.������ ������������� �������, 2(11):718–728, 2022

    work page 2022

  3. [3]

    A foundation model for atomistic materials chemistry.��� ������� �� �������� �������, 163(18), 2025

    Ilyes Batatia, Philipp Benner, Yuan Chiang, Alin M Elena, Dávid P Kovács, Janosh Riebesell, Xavier R Advincula, Mark Asta, Matthew Avaylon, William J Baldwin, et al. A foundation model for atomistic materials chemistry.��� ������� �� �������� �������, 163(18), 2025

    work page 2025

  4. [4]

    MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures

    Han Yang, Chenxi Hu, Yichi Zhou, Xixian Liu, Yu Shi, Jielan Li, Guanzhi Li, Zekun Chen, Shuizhou Chen, Claudio Zeni, Matthew Horton, Robert Pinsler, Andrew Fowler, Daniel Zügner, Tian Xie, Jake Smith, Lixin Sun, Qian Wang, Lingyu Kong, Chang Liu, Hongxia Hao, and Ziheng Lu. MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures

    work page

  5. [5]

    Scalable parallel algorithm for graph neural network interatomic potentials in molecular dynamics simulations.�� ����� ������ �������, 20(11):4857–4868, 2024

    Yutack Park, Jaesun Kim, Seungwoo Hwang, and Seungwu Han. Scalable parallel algorithm for graph neural network interatomic potentials in molecular dynamics simulations.�� ����� ������ �������, 20(11):4857–4868, 2024

    work page 2024

  6. [6]

    A. P. Thompson, H. M. Aktulga, R. Berger, D. S. Bolintineanu, W. M. Brown, P. S. Crozier, P. J. in ’t Veld, A. Kohlmeyer, S. G. Moore, T. D. Nguyen, R. Shan, M. J. Stevens, J. Tranchida, C. Trott, and S. J. Plimpton. LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales.����� ����� �����, 271:1...

    work page 2022

  7. [7]

    Orb: A fast, scalable neural network potential, 2024

    Mark Neumann, James Gin, Benjamin Rhodes, Steven Bennett, Zhiyi Li, Hitarth Choubisa, Arthur Hussey, and Jonathan Godwin. Orb: A fast, scalable neural network potential, 2024

    work page 2024

  8. [8]

    Wood, Misko Dzamba, Xiang Fu, Meng Gao, Muhammed Shuaibi, Luis Barroso-Luque, Kareem Abdelmaqsoud, Vahe Gharakhanyan, John R

    Brandon M. Wood, Misko Dzamba, Xiang Fu, Meng Gao, Muhammed Shuaibi, Luis Barroso-Luque, Kareem Abdelmaqsoud, Vahe Gharakhanyan, John R. Kitchin, Daniel S. Levine, Kyle Michel, Anuroop Sriram, Taco Cohen, Abhishek Das, Ammar Rizvi, Sushree Jagriti Sahoo, Zachary W. Ulissi, and C. Lawrence Zitnick. Uma: A family of universal models for atoms, 2025

    work page 2025

Show all 23 references
  1. [9]

    Materials Graph Library, June 2021

    Tsz Wai Ko, Marcel Nassar, Santiago Miret, Elliott Liu, Ji Qi, and Shyue Ping Ong. Materials Graph Library, June 2021

    2021

  2. [10]

    Uma: A family of universal models for atoms.����� �������� ����������������, 2025

    Brandon M Wood, Misko Dzamba, Xiang Fu, Meng Gao, Muhammed Shuaibi, Luis Barroso-Luque, Kareem Abdelmaqsoud, Vahe Gharakhanyan, John R Kitchin, Daniel S Levine, et al. Uma: A family of universal models for atoms.����� �������� ����������������, 2025

    2025

  3. [11]

    Better without u: Impact of selective hubbard u correction on foundational mlips.����� �������� ����������������, 2026

    Thomas Warford, Fabian L Thiemann, and Gåbor Csånyi. Better without u: Impact of selective hubbard u correction on foundational mlips.����� �������� ����������������, 2026

    2026

  4. [12]

    Energy & force regression on dft trajectories is not enough for universal machine learning interatomic potentials

    Santiago Miret, Kin Long Kelvin Lee, Carmelo Gonzales, Sajid Mannan, and NM Krishnan. Energy & force regression on dft trajectories is not enough for universal machine learning interatomic potentials. ����� �������� ����������������, 2025

    2025

  5. [13]

    Learning smooth and expressive interatomic potentials for physical property prediction.����� �������� ����������������, 2025

    Xiang Fu, Brandon M Wood, Luis Barroso-Luque, Daniel S Levine, Meng Gao, Misko Dzamba, and C Lawrence Zitnick. Learning smooth and expressive interatomic potentials for physical property prediction.����� �������� ����������������, 2025

    2025

  6. [14]

    Open catalyst 2020 (oc20) dataset and community challenges.��� ���������, 11(10):6059–6072, 2021

    Lowik Chanussot, Abhishek Das, Siddharth Goyal, Thibaut Lavril, Muhammed Shuaibi, Morgane Riviere, Kevin Tran, Javier Heras-Domingo, Caleb Ho, Weihua Hu, et al. Open catalyst 2020 (oc20) dataset and community challenges.��� ���������, 11(10):6059–6072, 2021. 43

    2020

  7. [15]

    Wood, Misko Dzamba, Meng Gao, Ammar Rizvi, C

    Luis Barroso-Luque, Muhammed Shuaibi, Xiang Fu, Brandon M. Wood, Misko Dzamba, Meng Gao, Ammar Rizvi, C. Lawrence Zitnick, and Zachary W. Ulissi. Open materials 2024 (omat24) inorganic materials dataset and models, 2024

    2024

  8. [16]

    The open molecules 2025 (omol25) dataset, evaluations, and models.����� �������� ����������������, 2025

    Daniel S Levine, Muhammed Shuaibi, Evan Walter Clark Spotte-Smith, Michael G Taylor, Muhammad R Hasyim, Kyle Michel, Ilyes Batatia, Gábor Csányi, Misko Dzamba, Peter Eastman, et al. The open molecules 2025 (omol25) dataset, evaluations, and models.����� �������� ����������������, 2025

    2025

  9. [17]

    The open catalyst 2022 (oc22) dataset and challenges for oxide electrocatalysts.��� ���������, 13(5):3066–3084, 2023

    Richard Tran, Janice Lan, Muhammed Shuaibi, Brandon M Wood, Siddharth Goyal, Abhishek Das, Javier Heras-Domingo, Adeesh Kolluru, Ammar Rizvi, Nima Shoghi, et al. The open catalyst 2022 (oc22) dataset and challenges for oxide electrocatalysts.��� ���������, 13(5):3066–3084, 2023

    2022

  10. [18]

    Foundation models for atomistic simulation of chemistry and materials.����� �������� ����������������, 2025

    Eric C-Y Yuan, Yunsheng Liu, Junmin Chen, Peichen Zhong, Sanjeev Raja, Tobias Kreiman, Santiago Vargas, Wenbin Xu, Martin Head-Gordon, Chao Yang, et al. Foundation models for atomistic simulation of chemistry and materials.����� �������� ����������������, 2025

    2025

  11. [19]

    A framework to evaluate machine learning crystal stability predictions.������ ������� ������������, 7(6):836–847, 2025

    Janosh Riebesell, Rhys EA Goodall, Philipp Benner, Yuan Chiang, Bowen Deng, Gerbrand Ceder, Mark Asta, Alpha A Lee, Anubhav Jain, and Kristin A Persson. A framework to evaluate machine learning crystal stability predictions.������ ������� ������������, 7(6):836–847, 2025

    2025

  12. [20]

    Forces are not enough: Benchmark and critical evaluation for machine learning force fields with molecular simulations, 2023

    Xiang Fu, Zhenghao Wu, Wujie Wang, Tian Xie, Sinan Keten, Rafael Gomez-Bombarelli, and Tommi Jaakkola. Forces are not enough: Benchmark and critical evaluation for machine learning force fields with molecular simulations, 2023

    2023

  13. [21]

    Egraffbench: evaluation of equivariant graph neural network force fields for atomistic simulations.������� ���������, 3(4):759–768, 2024

    Vaibhav Bihani, Sajid Mannan, Utkarsh Pratiush, Tao Du, Zhimin Chen, Santiago Miret, Matthieu Micoulaut, Morten M Smedskjaer, Sayan Ranu, and NM Anoop Krishnan. Egraffbench: evaluation of equivariant graph neural network force fields for atomistic simulations.������� ���������...

    2024

  14. [22]

    Force-field optimization by end-to-end differentiable atomistic simulation.������� �� �������� ������ ��� �����������, 21(12):5867–5879, 2025

    Abhijeet Sadashiv Gangan, Ekin Dogus Cubuk, Samuel S Schoenholz, Mathieu Bauchy, and NM Anoop Krishnan. Force-field optimization by end-to-end differentiable atomistic simulation.������� �� �������� ������ ��� �����������, 21(12):5867–5879, 2025

    2025

  15. [23]

    Learning neural network potentials from experimental data via differentiable trajectory reweighting.������ ��������������, 12(1):6884, 2021

    Stephan Thaler and Julija Zavadlav. Learning neural network potentials from experimental data via differentiable trajectory reweighting.������ ��������������, 12(1):6884, 2021. 44

    2021

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.