pith. sign in

arxiv: 2606.07327 · v2 · pith:ZMZUDXGBnew · submitted 2026-06-05 · ❄️ cond-mat.mtrl-sci · cond-mat.dis-nn· physics.app-ph· physics.comp-ph

Six Open Questions in Machine-Learned Interatomic Potential Foundation Models

Isabel Creed , Tim Rein , Ingvars Vitenburgs , Wojciech G. Stark , Viktor Ellingsson , Ahmed Y. Ismail , Guangyu Liu , Yuchen Lou
show 16 more authors
Bradley A. A. Martin Cyprien Bone Matthew A. H. Walker Mueen Taj Shirui Wang Kelvin Wong Ruiqi Wu Prakriti Kayastha Bingqing Cheng Aditi Krishnapriyan Michele Ceriotti Marcel F. Langer Jarvist Moore Frost Alex M. Ganose Venkat Kapil Keith T. Butler
This is my paper

Pith reviewed 2026-06-27 21:26 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci cond-mat.dis-nnphysics.app-phphysics.comp-ph
keywords machine-learned interatomic potentialsfoundation modelsopen questionsmolecular modelingmaterials simulationMLIPsinteratomic potentials
0
0 comments X

The pith

Six open questions will shape foundational machine-learned interatomic potentials for years ahead.

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

The paper first offers a working definition of foundational MLIPs as models trained on large, diverse datasets that can handle new systems with little additional training. It then identifies and explores six specific open questions that the authors regard as the most important unresolved issues in this area. A reader would care because MLIPs aim to remove the traditional trade-off between simulation scale and accuracy in molecular and materials modeling. The authors argue that, even with fast model development, these questions remain central and will steer research priorities. The piece frames the questions explicitly around the definition to keep the discussion focused on broad applicability rather than narrow model tweaks.

Core claim

The authors develop a working definition of foundational MLIPs and use it to articulate six open questions; they claim that, despite rapid progress and proliferation of models, these questions constitute the fundamental challenges that will continue to define cutting-edge research in the field for years to come.

What carries the argument

The working definition of foundational MLIPs, which frames the selection and discussion of the six open questions.

If this is right

  • Progress on foundational MLIPs will require systematic attention to the six questions rather than isolated model improvements.
  • Models trained on large diverse datasets will need to demonstrate reliable performance on new systems with minimal retraining to qualify as foundational.
  • The tension between scale and accuracy in simulations will remain unresolved until the listed questions receive answers.
  • The field will continue to produce many models, but only those addressing the core questions will set the research agenda.

Where Pith is reading between the lines

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

  • Resolving the questions could allow a single pretrained model to replace many specialized potentials across different chemical systems.
  • The emphasis on minimal updates for new systems may push the community toward transfer-learning techniques that are currently underdeveloped for interatomic potentials.
  • If the six questions prove decisive, funding and publication priorities in materials modeling may shift toward broad benchmark suites rather than single-material case studies.

Load-bearing premise

The authors' choice of exactly these six questions, framed by their working definition, correctly identifies the load-bearing challenges rather than other unlisted issues.

What would settle it

Future research activity in MLIPs that concentrates overwhelmingly on problems outside the six listed questions would undermine the claim that these questions define the field's direction.

Figures

Figures reproduced from arXiv: 2606.07327 by Aditi Krishnapriyan, Ahmed Y. Ismail, Alex M. Ganose, Bingqing Cheng, Bradley A. A. Martin, Cyprien Bone, Guangyu Liu, Ingvars Vitenburgs, Isabel Creed, Jarvist Moore Frost, Keith T. Butler, Kelvin Wong, Marcel F. Langer, Matthew A. H. Walker, Michele Ceriotti, Mueen Taj, Prakriti Kayastha, Ruiqi Wu, Shirui Wang, Tim Rein, Venkat Kapil, Viktor Ellingsson, Wojciech G. Stark, Yuchen Lou.

Figure 1
Figure 1. Figure 1: The connections between definitions and open questions. The text of the article has been analysed using a cosine similarity of vectorized embeddings. On the left, we show how the definition criteria link to the questions, on the right, the relations between the questions (colour-coded as on the left) are displayed. The line widths reflect the similarity of the embeddings of the content. To ground this comp… view at source ↗
Figure 2
Figure 2. Figure 2: Log-log plot of the predictive error on the water data set from [59] using NequIP with rotation order L ∈ {0, 1, 2, 3} as a function of training set size, measured via the force MAE. Figure from Batzner et al. [32] Mattersim, Alexandria, and OMat24 [55, 61, 62, 63]. In recent work with the MatPES data-efficient sampling scheme [64], the authors show that MLIPs trained on a compact, representative 400K data… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of conceptual approaches to accurately capture long range interactions in GNNs while mitigating oversmoothing and oversquashing. Strategies include: Graph Rewiring (top left) for structural optimization; Dual Architectures (top right) for global-local separation; Hierarchical Approaches (bottom right) for multi-scale feature extraction; and incorporation of Physics-Informed Priors that encode the … view at source ↗
Figure 4
Figure 4. Figure 4: Different resolutions of lipid membranes. All-atom (AA) resolution explicitly considers all atoms. Coarse-grain (CG) resolution considers small atom groups and their associated hydrogens. Supra-CG resolution represents solvents implicitly and proteins and lipids as qualitative few-bead models. Implicit resolution further integrates out lipid molecules. Reprinted from [197]. data by augmenting the MLIP arch… view at source ↗
Figure 5
Figure 5. Figure 5: Combined performance score against the model size (number of model parameters) obtained with different foundation MLIPs, based on Matbench Discovery [38] benchmark website (https://matbench-discovery.materialsproject.org, June 2026). Large-scale atomistic simulations with MLIPs are already feasible, as demonstrated, for instance, by Musaelian et al. in modelling large, biological systems [256]. However, su… view at source ↗
Figure 6
Figure 6. Figure 6: Distribution of the metric rankings from the leaderboard of the Matbench Discovery [38] benchmark website (https://matbench-discovery.materialsproject.org, June 2026). An overarching question covering all of the subjects we have covered is: how do we know if a particular model can do what we want? We may want to do a large simulation of a well-known system, or explore some new exotic physics; perhaps our s… view at source ↗
read the original abstract

Machine-learned interatomic potentials (MLIPs) have had a profound impact on molecular modelling in recent years, promising to resolve the long-standing tension between the scale and accuracy of simulations. There has been a proliferation of new models and designs, and recently the paradigm of ``foundational'' MLIPs has become prevalent. Broadly speaking, foundation models are trained on large diverse datasets and promise to work well for new systems with minimal updates required. However, in such a new and fast moving field, there are many unanswered questions. In this article, we set out to articulate and explore what we see as the most important among these questions. We start by developing a working definition for foundational MLIPs and use this definition to frame the subsequent open questions. Despite the rapid progress in the field of MLIP models, we believe that these are fundamental questions which will continue to define cutting edge research in MLIPs in the years to come.

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

0 major / 2 minor

Summary. The manuscript develops a working definition of foundational MLIPs (models trained on large, diverse datasets that generalize to new systems with minimal updates) and uses this definition to identify and discuss six open questions that the authors argue will continue to shape cutting-edge research in the field.

Significance. As a perspective piece, the manuscript provides a structured framing that could help organize community discussion around generalization, data requirements, and architectural choices in MLIP development; its value lies in the clarity of the definitional starting point rather than in new empirical or theoretical results.

minor comments (2)
  1. [Abstract] The abstract states that the authors 'start by developing a working definition' but does not preview the six questions; adding a brief enumerated list would improve immediate readability for readers scanning the piece.
  2. Section headings for the six questions are not numbered in the provided text; consistent numbering (e.g., Question 1, Question 2) would make cross-references within the manuscript easier to follow.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive review and recommendation to accept. Their summary correctly identifies the manuscript as a perspective piece that proposes a working definition of foundational MLIPs and frames six open questions around generalization, data, and architecture.

Circularity Check

0 steps flagged

No significant circularity; purely discursive perspective piece

full rationale

The paper is a perspective article that develops a working definition of foundational MLIPs and lists six open questions. It contains no derivations, equations, predictions, or fitted quantities that could reduce to inputs by construction. The central claim is explicitly subjective framing of future research directions rather than a technical result. No self-citation chains or ansatzes are invoked as load-bearing steps. This is self-contained as a non-derivational discussion and receives the default low circularity score.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The perspective rests on the authors' judgment that the listed questions are the most important; no free parameters, mathematical axioms, or new entities are introduced.

axioms (1)
  • domain assumption Foundational MLIPs are trained on large diverse datasets and promise to work well for new systems with minimal updates.
    This is the working definition the paper develops to frame the questions.

pith-pipeline@v0.9.1-grok · 5809 in / 1151 out tokens · 21781 ms · 2026-06-27T21:26:38.633582+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

268 extracted references · 126 canonical work pages · 2 internal anchors

  1. [1]

    Metropolis N, Rosenbluth A W, Rosenbluth M N, Teller A H and Teller E 1953The Journal of Chemical Physics211087–1092 ISSN 1089-7690 URL http://dx.doi.org/10.1063/1.1699114

    work page doi:10.1063/1.1699114

  2. [2]

    Hastings W K 1970Biometrika5797–109 ISSN 0006-3444 URL http://dx.doi.org/10.1093/biomet/57.1.97

    work page doi:10.1093/biomet/57.1.97

  3. [3]

    Rahman A 1964Physical Review136A405–A411 ISSN 0031-899X URL http://dx.doi.org/10.1103/PhysRev.136.A405

    work page doi:10.1103/physrev.136.a405

  4. [4]

    Verlet L 1967Physical Review15998–103 ISSN 0031-899X URL http://dx.doi.org/10.1103/PhysRev.159.98

    work page doi:10.1103/physrev.159.98

  5. [5]

    Chandler D and Wolynes P G 1981The Journal of Chemical Physics744078–4095 ISSN 1089-7690 URLhttp://dx.doi.org/10.1063/1.441588

    work page doi:10.1063/1.441588

  6. [6]

    Parrinello M and Rahman A 1981Journal of Applied Physics527182–7190 ISSN 1089-7550 URLhttp://dx.doi.org/10.1063/1.328693

    work page doi:10.1063/1.328693

  7. [7]

    Ceperley D M 1995Reviews of Modern Physics67279–355 ISSN 1539-0756 URL http://dx.doi.org/10.1103/RevModPhys.67.279

    work page doi:10.1103/revmodphys.67.279

  8. [8]

    Series A, Containing Papers of a Mathematical and Physical Character106463–477 ISSN 2053-9150 URL http://dx.doi.org/10.1098/rspa.1924.0082

    Jones J E 1924Proceedings of the Royal Society of London. Series A, Containing Papers of a Mathematical and Physical Character106463–477 ISSN 2053-9150 URL http://dx.doi.org/10.1098/rspa.1924.0082

    work page doi:10.1098/rspa.1924.0082 2053

  9. [9]

    Alder B J and Wainwright T E 1957The Journal of Chemical Physics271208–1209 ISSN 1089-7690 URLhttp://dx.doi.org/10.1063/1.1743957

    work page doi:10.1063/1.1743957

  10. [10]

    Behler J and Parrinello M 2007Phys. Rev. Lett.98146401 URL https://link.aps.org/doi/10.1103/PhysRevLett.98.146401

    work page doi:10.1103/physrevlett.98.146401

  11. [11]

    Bart´ ok A P, Kondor R and Cs´ anyi G 2013Phys. Rev. B87184115 URL https://link.aps.org/doi/10.1103/PhysRevB.87.184115

    work page internal anchor Pith review doi:10.1103/physrevb.87.184115

  12. [12]

    Drautz R 2019Phys. Rev. B99014104 URL https://link.aps.org/doi/10.1103/PhysRevB.99.014104

    work page doi:10.1103/physrevb.99.014104

  13. [13]

    Batatia I, Kov´ acs D P, Simm G N C, Ortner C and Cs´ anyi G 2022NeurIPS3511423–11436 URLhttps://proceedings.neurips.cc/paper_files/paper/2022/hash/ 4a36c3c51af11ed9f34615b81edb5bbc-Abstract-Conference.html

    2022

  14. [14]

    Thomas N, Smidt T, Kearnes S, Yang L, Li L, Kohlhoff K and Riley P 2018 Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point clouds eprint: 1802.08219 URLhttps://arxiv.org/abs/1802.08219

    Pith/arXiv arXiv 2018

  15. [15]

    Geiger M, Smidt T, M A, Miller B K, Boomsma W, Dice B, Lapchevskyi K, Weiler M, Tyszkiewicz M, Batzner S, Madisetti D, Uhrin M, Frellsen J, Jung N, Sanborn S, Wen M, Rackers J, Rød M and Bailey M 2022 Euclidean neural networks: e3nn URL https://doi.org/10.5281/zenodo.6459381

    work page doi:10.5281/zenodo.6459381 2022

  16. [16]

    Deng B, Zhong P, Jun K, Riebesell J, Han K, Bartel C J and Ceder G 2023Nature Machine Intelligence51031–1041 URLhttps://www.nature.com/articles/s42256-023-00716-3

  17. [17]

    Bommasani R, Hudson D A, Adeli E, Altman R, Arora S, Arx S v, Bernstein M S, Bohg J, Bosselut A, Brunskill E, Brynjolfsson E, Buch S, Card D, Castellon R, Chatterji N, Chen A, Creel K, Davis J Q, Demszky D, Donahue C, Doumbouya M, Durmus E, Ermon S, Etchemendy J, Ethayarajh K, Fei-Fei L, Finn C, Gale T, Gillespie L, Goel K, Goodman N, Grossman S, Guha N, ...

    Pith/arXiv arXiv 2022

  18. [18]

    Lysogorskiy Y, Bochkarev A and Drautz R 2026npj Computational Materials

  19. [19]

    Kreiman T, Bai Y, Atieh F, Weaver E, Qu E and Krishnapriyan A S 2025arXiv preprint arXiv:2510.02259

    arXiv

  20. [20]

    Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez A N, Kaiser L and Polosukhin I 2017Advances in neural information processing systems30

  21. [21]

    Mater.10300 URL https://doi.org/10.1038/s41524-024-01486-1

    Devi R, Butler K T and Sai Gautam G 2024npj Comput. Mater.10300 URL https://doi.org/10.1038/s41524-024-01486-1

    work page doi:10.1038/s41524-024-01486-1

  22. [22]

    Chen C, Zuo Y, Ye W, Li X and Ong S P 2021Nat. Comput. Sci.146–53 URL https://doi.org/10.1038/s43588-020-00002-x

    work page doi:10.1038/s43588-020-00002-x

  23. [23]

    Kim J, Kim J, Kim J, Lee J, Park Y, Kang Y and Han S 2024J. Am. Chem. Soc.URL https://doi.org/10.1021/jacs.4c14455

    work page doi:10.1021/jacs.4c14455

  24. [24]

    Batatia I, Lin C, Hart J, Kasoar E, Elena A M, Norwood S W, Wolf T and Cs´ anyi G 2025 ArXiv:2510.25380 URLhttps://arxiv.org/abs/2510.25380

    arXiv 2025

  25. [25]

    Gumber S, Alzate-Vargas L, Nebgen B T, Veelen A v, Kadvani S, Gibson T and Messerly R ArXiv:2506.10211 URLhttp://arxiv.org/abs/2506.10211

    arXiv

  26. [26]

    Mazitov A, Chorna S, Fraux G, Bercx M, Pizzi G, De S and Ceriotti M ArXiv:2506.19674 URLhttp://arxiv.org/abs/2506.19674

    arXiv

  27. [27]

    Park S, Seong K, Yang S, G´ omez-Bombarelli R and Ahn S Learning Collective Variables from BioEmu with Time-Lagged Generation (Preprint2507.07390)

    arXiv

  28. [28]

    Chorna S, Tisi D, Malosso C, How W B, Ceriotti M and Chong S Comparing the latent features of universal machine-learning interatomic potentials (Preprint2512.05717)

    Pith/arXiv arXiv

  29. [29]

    Edamadaka S, Yang S, Li J and G´ omez-Bombarelli R Universally Converging Representations of Matter Across Scientific Foundation Models (Preprint2512.03750)

    arXiv

  30. [30]

    Sutton R 2019Incomplete Ideas (blog)URL http://www.incompleteideas.net/IncIdeas/BitterLesson.html

  31. [31]

    Kranmer K 2025Theory and practice (blog)URL https://theoryandpractice.org/2025/09/The%20Bittersweet%20Lesson/

    2025

  32. [32]

    Nature Communications 13(1) (2022) https://doi.org/10.1038/s41467-022-29939-5

    Batzner S, Musaelian A, Sun L, Geiger M, Mailoa J P, Kornbluth M, Molinari N, Smidt T E and Kozinsky B 2022Nat. Commun.132453 URL https://doi.org/10.1038/s41467-022-29939-5

    work page doi:10.1038/s41467-022-29939-5

  33. [33]

    Qu E and Krishnapriyan A S 2024 The importance of being scalable: Improving the speed and accuracy of neural network interatomic potentials across chemical domainsAdvances in Neural Information Processing Systemsvol 37 ed Globerson A, Mackey L, Belgrave D, Fan A, Paquet U, Tomczak J and Zhang C (Curran Associates, Inc.) pp 139030–139053 URL https://procee...

    2024

  34. [34]

    Smidt T E, Geiger M and Miller B K 2021Phys. Rev. Res.3L012002 URL https://link.aps.org/doi/10.1103/PhysRevResearch.3.L012002

    work page doi:10.1103/physrevresearch.3.l012002

  35. [35]

    Xie Y and Smidt T 2024 ArXiv:2402.02681 URLhttps://arxiv.org/abs/2402.02681 22 ML4AtomsSix Open Questions for MLIPsAuthoret al

    arXiv 2024

  36. [36]

    Hofgard E, Wang R, Walters R and Smidt T 2024 ArXiv:2407.20471 URL https://arxiv.org/abs/2407.20471

    arXiv 2024

  37. [37]

    Rhodes B, Vandenhaute S, ˇSimkus V, Gin J, Godwin J, Duignan T and Neumann M 2025 ArXiv:2504.06231 URLhttps://arxiv.org/abs/2504.06231

    arXiv 2025

  38. [38]

    Riebesell J, Goodall R E A, Benner P, Chiang Y, Deng B, Ceder G, Asta M, Lee A A, Jain A and Persson K A 2025Nat. Mach. Intell.7836–847 URL https://www.nature.com/articles/s42256-025-01055-1

  39. [39]

    Bigi F, Langer M F and Ceriotti M 2025 The dark side of the forces: assessing non-conservative force models for atomistic machine learningProceedings of the 42nd International Conference on Machine Learning(Proceedings of Machine Learning Research vol 267) ed Singh A, Fazel M, Hsu D, Lacoste-Julien S, Berkenkamp F, Maharaj T, Wagstaff K and Zhu J (PMLR) p...

    2025

  40. [40]

    Mater.11178 URLhttps://doi.org/10.1038/s41524-025-01650-1

    Loew A, Sun D, Wang H C, Botti S and Marques M A L 2025npj Comput. Mater.11178 URLhttps://doi.org/10.1038/s41524-025-01650-1

    work page doi:10.1038/s41524-025-01650-1

  41. [41]

    Ngo K and Ravanbakhsh S Scaling Laws and Symmetry, Evidence from Neural Force Fields URLhttps://arxiv.org/abs/2510.09768v1

    Pith/arXiv arXiv

  42. [42]

    Brehmer J, Behrends S, de Haan P and Cohen T Does equivariance matter at scale? (Preprint2410.23179)

    arXiv

  43. [43]

    Qu E and Krishnapriyan A 2024Advances in Neural Information Processing Systems37 139030–139053

  44. [44]

    Mazitov A, Bigi F, Kellner M, Pegolo P, Tisi D, Fraux G, Pozdnyakov S, Loche P and Ceriotti M ArXiv:2503.14118 URLhttp://arxiv.org/abs/2503.14118

    arXiv

  45. [46]

    Xie Y and Smidt T 2025arXiv preprint arXiv:2506.02269

    arXiv

  46. [47]

    Accessed 2025-11-10 URLhttps://github.com/NVIDIA/cuEquivariance

    NVIDIA Corporation 2025 cuequivariance cUDA kernels and APIs for equivariant neural networks. Accessed 2025-11-10 URLhttps://github.com/NVIDIA/cuEquivariance

    2025

  47. [49]

    Paszke A, Gross S, Massa F, Lerer A, Bradbury J, Chanan G, Killeen T, Lin Z, Gimelshein N, Antiga L, Desmaison A, K¨ opf A, Yang E, DeVito Z, Raison M, Tejani A, Chilamkurthy S, Steiner B, Fang L, Bai J and Chintala S 2019 Pytorch: An imperative style, high-performance deep learning libraryAdvances in Neural Information Processing Systems (NeurIPS)URLhttp...

    2019

  48. [50]

    TorchScript is deprecated; usetorch.export

    PyTorch Contributors 2025TorchScript (torch.jit) — PyTorch Documentationlast updated 2025-07-16. TorchScript is deprecated; usetorch.export. Accessed 2025-11-10 URL https://docs.pytorch.org/docs/stable/jit.html

    2025

  49. [51]

    Han K, Deng B, Farimani A B and Ceder G 2025 ArXiv:2506.02023 [cs] URL https://arxiv.org/abs/2506.02023

    arXiv 2025

  50. [52]

    Loshchilov I and Hutter F 2019 ArXiv:1711.05101 URL https://arxiv.org/abs/1711.05101

    Pith/arXiv arXiv 2019

  51. [53]

    Jordan K 2024 Muon: An optimizer for the hidden layers in neural networks accessed 2025-11-10 URLhttps://kellerjordan.github.io/posts/muon/

    2024

  52. [54]

    Chen C and Ong S P 2022Nat. Comput. Sci.2718–728 URL https://doi.org/10.1038/s43588-022-00349-3 23 ML4AtomsSix Open Questions for MLIPsAuthoret al

    work page doi:10.1038/s43588-022-00349-3

  53. [55]

    Merchant A, Batzner S, Schoenholz S S, Aykol M, Cheon G and Cubuk E D 2023Nature 62480–85 URLhttps://www.nature.com/articles/s41586-023-06735-9

  54. [56]

    Mater.11 1–11 URLhttps://doi.org/10.1038/s41524-025-01727-x

    Radova M, Stark W G, Allen C S, Maurer R J and Bart´ ok A P 2025npj Comput. Mater.11 1–11 URLhttps://doi.org/10.1038/s41524-025-01727-x

    work page doi:10.1038/s41524-025-01727-x

  55. [57]

    Messerly M, Matin S, Allen A E A, Nebgen B, Barros K, Smith J S, Lubbers N and Messerly R 2025arXiv preprint arXiv:2505.01590ArXiv:2505.01590 URL https://arxiv.org/abs/2505.01590

    arXiv

  56. [58]

    Koker T, Kotak M and Smidt T 2025 ArXiv:2508.16067 URL https://arxiv.org/abs/2508.16067

    arXiv 2025

  57. [59]

    Cheng B, Engel E A, Behler J, Dellago C and Ceriotti M 2019Proc. Natl. Acad. Sci. U.S.A. 1161110–1115 URLhttps://www.pnas.org/doi/abs/10.1073/pnas.1815117116

    work page doi:10.1073/pnas.1815117116

  58. [60]

    Commun.114895 URL https://doi.org/10.1038/s41467-020-18556-9

    von Lilienfeld O A and Burke K 2020Nat. Commun.114895 URL https://doi.org/10.1038/s41467-020-18556-9

    work page doi:10.1038/s41467-020-18556-9

  59. [61]

    Yang H, Hu C, Zhou Y, Liu X, Shi Y, Li J, Li G, Chen Z, Chen S, Zeni C, Horton M, Pinsler R, Fowler A, Z¨ ugner D, Xie T, Smith J, Sun L, Wang Q, Kong L, Liu C, Hao H and Lu Z 2024 ArXiv:2405.04967 URLhttps://arxiv.org/abs/2405.04967

    Pith/arXiv arXiv 2024

  60. [62]

    Today Phys.48101560 URL https://doi.org/10.1016/j.mtphys.2024.101560

    Schmidt J, Cerqueira T F, Romero A H, Loew A, J¨ ager F, Wang H C, Botti S and Marques M A 2024Mater. Today Phys.48101560 URL https://doi.org/10.1016/j.mtphys.2024.101560

    work page doi:10.1016/j.mtphys.2024.101560 2024

  61. [63]

    Barroso-Luque L, Shuaibi M, Fu X, Wood B M, Dzamba M, Gao M, Rizvi A, Zitnick C L and Ulissi Z W 2024arXiv preprint arXiv:2410.12771ArXiv:2410.12771 URL https://doi.org/10.48550/arXiv.2410.12771

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2410.12771

  62. [64]

    Kaplan A D, Liu R, Qi J, Ko T W, Deng B, Riebesell J, Ceder G, Persson K A and Ong S P 2025 ArXiv:2503.04070 URLhttps://doi.org/10.48550/arXiv.2503.04070

    work page doi:10.48550/arxiv.2503.04070 2025

  63. [65]

    Qi J, Ko T W, Wood B C, Pham T A and Ong S P 2024npj Computational Materials1043 URLhttps://www.nature.com/articles/s41524-024-01227-4

  64. [66]

    Mazitov A, Chorna S, Fraux G, Bercx M, Pizzi G, De S and Ceriotti M 2025Scientific data 121857

  65. [67]

    Mater.7185 URL https://doi.org/10.1038/s41524-021-00650-1

    Choudhary K and DeCost B 2021npj Comput. Mater.7185 URL https://doi.org/10.1038/s41524-021-00650-1

    work page doi:10.1038/s41524-021-00650-1

  66. [68]

    Gibson J B, Janicki T D, Hire A C, Bishop C, Lane J M D and Hennig R G 2024 ArXiv:2409.07610 URLhttps://doi.org/10.48550/arXiv.2409.07610

    work page doi:10.48550/arxiv.2409.07610 2024

  67. [69]

    Nakkiran P, Kaplun G, Bansal Y, Yang T, Barak B and Sutskever I 2019 ArXiv:1912.02292 URLhttps://doi.org/10.48550/arXiv.1912.02292

    work page doi:10.48550/arxiv.1912.02292 2019

  68. [70]

    Kaplan J, McCandlish S, Henighan T, Brown T B, Chess B, Child R, Gray S, Radford A, Wu J and Amodei D 2020arXiv preprint arXiv:2001.08361ArXiv:2001.08361 URL https://arxiv.org/abs/2001.08361

    Pith/arXiv arXiv 2001

  69. [71]

    Frey N C, Soklaski R, Axelrod S, Samsi S, G´ omez-Bombarelli R, Coley C W and Gadepally V 2023Nat. Mach. Intell.51297–1305 URLhttps://doi.org/10.1038/s42256-023-00740-3

    work page doi:10.1038/s42256-023-00740-3

  70. [72]

    Kuryla D, Berger F, Cs´ anyi G and Michaelides A 2025arXiv preprint arXiv:2510.19774 ArXiv:2510.19774 URLhttps://doi.org/10.48550/arXiv.2510.19774

    work page doi:10.48550/arxiv.2510.19774

  71. [73]

    Perdew J P, Burke K and Ernzerhof M 1996Phys. Rev. Lett.77(18) 3865–3868 URL https://link.aps.org/doi/10.1103/PhysRevLett.77.3865

    work page doi:10.1103/physrevlett.77.3865

  72. [74]

    Meggiolaro D and De Angelis F 2018ACS Energy Letters32206–2222 ISSN 2380-8195 URL http://dx.doi.org/10.1021/acsenergylett.8b01212

    work page doi:10.1021/acsenergylett.8b01212

  73. [75]

    Taheri A, Da Silva C and Amon C H 2018Journal of Applied Physics123ISSN 1089-7550 URLhttp://dx.doi.org/10.1063/1.5027619 24 ML4AtomsSix Open Questions for MLIPsAuthoret al

    work page doi:10.1063/1.5027619

  74. [76]

    Stephens P J, Devlin F J, Chabalowski C F and Frisch M J 1994The Journal of Physical Chemistry9811623–11627 (Preprinthttps://doi.org/10.1021/j100096a001) URL https://doi.org/10.1021/j100096a001

    work page doi:10.1021/j100096a001

  75. [77]

    Krukau A V, Vydrov O A, Izmaylov A F and Scuseria G E 2006The Journal of Chemical Physics125224106 ISSN 0021-9606 (Preprinthttps://pubs.aip.org/aip/jcp/ article-pdf/doi/10.1063/1.2404663/13263224/224106_1_online.pdf) URL https://doi.org/10.1063/1.2404663

    work page doi:10.1063/1.2404663/13263224/224106_1_online.pdf

  76. [78]

    Deng B, Choi Y, Zhong P, Riebesell J, Anand S, Li Z, Jun K, Persson K A and Ceder G 2025 npj Computational Materials119 URL https://www.nature.com/articles/s41524-024-01500-6

    2025

  77. [79]

    Fu X, Wood B M, Barroso-Luque L, Levine D S, Gao M, Dzamba M and Zitnick C L 2025 arXiv preprint arXiv:2502.12147ArXiv:2502.12147 URL https://doi.org/10.48550/arXiv.2502.12147

    work page doi:10.48550/arxiv.2502.12147 2025

  78. [80]

    Mater.11URL https://doi.org/10.1038/s41524-025-01550-4

    Ko T W and Ong S P 2025npj Comput. Mater.11URL https://doi.org/10.1038/s41524-025-01550-4

    work page doi:10.1038/s41524-025-01550-4

  79. [81]

    Learn.: Sci

    Oerder R, Schmieden G and Hamaekers J 2025Mach. Learn.: Sci. Technol.6045004 URL https://doi.org/10.1088/2632-2153/ae0d41

    work page doi:10.1088/2632-2153/ae0d41

  80. [82]

    Schmitz N F, Ploumhans B and Herbst M F126 ISSN 2057-3960

    2057

Showing first 80 references.