pith. machine review for the scientific record. sign in

arxiv: 2604.01642 · v2 · submitted 2026-04-02 · ❄️ cond-mat.mtrl-sci

Recognition: 2 theorem links

· Lean Theorem

Machine Learning Interatomic Potentials for Million-Atom Simulations of Multicomponent Alloys

Fei Shuang, Fengxian Liu, Kai Liu, Minqiang Jiang, Penghua Ying, Poulumi Dey, Zheyong Fan, Zixiong Wei

Authors on Pith no claims yet

Pith reviewed 2026-05-13 21:21 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci
keywords machine learning interatomic potentialsmulticomponent alloyshigh-entropy alloysneuroevolution potentialgraph atomic cluster expansionmolecular dynamicsuncertainty quantificationshock simulations
0
0 comments X

The pith

GRACE potentials train more efficiently with slightly better accuracy than NEP for multicomponent alloys, while NEP runs about sixty times faster for million-atom simulations.

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

This paper compares two machine learning frameworks for building interatomic potentials that can handle simulations of complex metal alloys with many different elements. It evaluates them across sixteen elemental metals and various multicomponent alloy compositions. GRACE with a Finnis-Sinclair formulation delivers higher training efficiency and modestly superior results on mechanical properties, thermal stability, and how well the model extrapolates to new chemical combinations. NEP stands out for its dramatically higher speed when running the simulations themselves, opening the door to systems with millions of atoms. The authors also test uncertainty estimates and find that averaging across multiple models gives reliable error predictions, which supports trustworthy large-scale dynamic runs under extreme conditions.

Core claim

GRACE potential with Finnis-Sinclair type shows substantially higher training efficiency and consistently, though only slightly, better accuracy for mechanical properties, thermal stability, and chemical extrapolation. In contrast, NEP achieves an approximately 60-fold higher inference speed, making it attractive for million-atom molecular dynamics simulations. Ensemble-based uncertainty quantification correlates robustly with model error, whereas D-optimality is less reliable, and this combination enables efficient and reliable nonequilibrium simulations of shock propagation.

What carries the argument

The neuroevolution potential (NEP) and graph atomic cluster expansion (GRACE) frameworks for machine learning interatomic potentials, tested head-to-head on training cost, predictive accuracy, and inference speed for alloy systems.

If this is right

  • GRACE becomes the practical choice when training data are limited or maximum accuracy on mechanical and thermal properties matters most.
  • NEP combined with ensemble uncertainty allows reliable molecular dynamics of shock waves and other extreme conditions in systems too large for slower potentials.
  • Ensemble uncertainty estimates can be used directly to flag regions of high error during large-scale runs without extra computational overhead.
  • The speed-accuracy trade-off guides framework selection depending on whether the simulation prioritizes system size or precision on new chemistries.

Where Pith is reading between the lines

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

  • NEP's inference advantage could extend simulations to alloy systems with even higher numbers of atoms or more dynamic processes than those demonstrated here.
  • The reliability of ensemble uncertainty might transfer to other machine learning potentials for alloys, providing a general route to trustworthy extrapolation.
  • Future work could test whether the same training-versus-speed pattern appears when the same frameworks are applied to non-metallic multicomponent materials.

Load-bearing premise

The performance differences between the two frameworks hold for the specific set of sixteen metals and alloy compositions tested.

What would settle it

Finding that GRACE loses its training or accuracy edge or that NEP loses its speed advantage when applied to a new set of alloy compositions or properties outside the studied group would challenge the central comparison.

Figures

Figures reproduced from arXiv: 2604.01642 by Fei Shuang, Fengxian Liu, Kai Liu, Minqiang Jiang, Penghua Ying, Poulumi Dey, Zheyong Fan, Zixiong Wei.

Figure 1
Figure 1. Figure 1: FIG. 1. Benchmarking the accuracy and efficiency of UNEP [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Uncertainty quantification for UNEP-v1 and GRACE-FS-M potentials. (a, b) Uncertainty estimates for UNEP-v1 [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Computational speed comparison between UNEP-v1 and GRACE-FS-M for (a) pure Cu and (b) the Al [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Assessing the thermal stability of HEAs via MD [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Assessing the chemical transferability of different [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 9
Figure 9. Figure 9: (a-c) presents the evolution of density, stress, and temperature during the shock process, as simulated by the primary UNEP-v1 model from Ref. [38]. The UNEP-v1 model properly captures the dynamic response of these quantities throughout both the shock compres￾sion and release stages, including wave propagation, re￾flection, and interaction. A spall fracture is initiated at approximately 30 ps, marked by a … view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10. Microstructural evolution during shock simulation. [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
read the original abstract

Machine learning interatomic potentials (MLIPs) with broad chemical flexibility are important for atomistic simulations of compositionally complex materials such as high-entropy alloys. Here, we study two state-of-the-art MLIP frameworks, the neuroevolution potential (NEP) and the graph atomic cluster expansion (GRACE), for 16 elemental metals and multicomponent alloys. GRACE potential with Finnis-Sinclair type shows substantially higher training efficiency and consistently, though only slightly, better accuracy for mechanical properties, thermal stability, and chemical extrapolation. In contrast, NEP achieves an approximately 60-fold higher inference speed, making it attractive for million-atom molecular dynamics simulations. We further examine uncertainty quantification strategies and find that ensemble-based uncertainty correlates robustly with model error, whereas D-optimality is less reliable for the systems considered here. Large-scale nonequilibrium molecular dynamics simulations of shock propagation further show that NEP, combined with ensemble-based uncertainty quantification, enables efficient and reliable simulations under extreme dynamic conditions.

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

1 major / 2 minor

Summary. The manuscript compares the neuroevolution potential (NEP) and graph atomic cluster expansion (GRACE) frameworks for machine learning interatomic potentials across 16 elemental metals and multicomponent alloys. GRACE (Finnis-Sinclair type) is reported to offer higher training efficiency and marginally better accuracy for mechanical properties, thermal stability, and chemical extrapolation, while NEP provides an approximately 60-fold inference speedup that makes it suitable for million-atom molecular dynamics. The work also assesses uncertainty quantification, finding ensemble-based methods more reliable than D-optimality, and demonstrates nonequilibrium shock propagation simulations using NEP with ensemble UQ.

Significance. If the performance trade-offs and large-system applicability hold, the paper offers practical guidance for selecting MLIPs in simulations of high-entropy alloys and other compositionally complex materials. Strengths include the multi-system empirical comparisons, explicit evaluation of uncertainty quantification strategies, and the execution of million-atom nonequilibrium MD simulations under extreme conditions. These elements address real needs in the field for scalable and reliable atomistic modeling.

major comments (1)
  1. [Inference timing and large-scale MD sections] The central claim that NEP's ~60-fold inference advantage makes it attractive specifically for million-atom MD simulations requires explicit support. The manuscript does not state the system sizes used for the timing benchmarks, nor confirm that identical LAMMPS interfaces, neighbor-list implementations, and cutoff settings were employed for both models. Because inference costs shift toward neighbor-list construction, domain decomposition, and MPI communication at large N, timings measured on small cells may not translate directly, weakening the practical recommendation for million-atom shock simulations.
minor comments (2)
  1. [Abstract] The abstract summarizes results without providing details on the specific datasets, training protocols, validation metrics, or statistical significance of the reported accuracy differences, which limits immediate assessment of the strength of the comparative claims.
  2. [Results and figures] Ensure that all tables and figures reporting accuracy metrics explicitly label the 16 metals and the alloy compositions tested, and include error bars or standard deviations where ensemble methods are compared.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful and constructive review of our manuscript. We address the major comment below and will revise the manuscript accordingly to provide the requested details and strengthen the discussion of large-scale applicability.

read point-by-point responses

  1. Referee: [Inference timing and large-scale MD sections] The central claim that NEP's ~60-fold inference advantage makes it attractive specifically for million-atom MD simulations requires explicit support. The manuscript does not state the system sizes used for the timing benchmarks, nor confirm that identical LAMMPS interfaces, neighbor-list implementations, and cutoff settings were employed for both models. Because inference costs shift toward neighbor-list construction, domain decomposition, and MPI communication at large N, timings measured on small cells may not translate directly, weakening the practical recommendation for million-atom shock simulations.

    Authors: We agree that the manuscript should provide explicit details on the timing benchmarks to support the inference-speed claim. In the revised version we will state the exact system sizes used for the benchmarks, confirm that both NEP and GRACE employed identical LAMMPS interfaces, neighbor-list implementations, and cutoff radii, and add a short discussion of how per-atom inference costs compare with neighbor-list and communication overhead at large N. We note that the million-atom nonequilibrium shock simulations were successfully performed with NEP (and ensemble UQ), providing direct empirical evidence of its practicality at that scale; we will reference these results when discussing the translation of small-cell timings to large-system performance. revision: yes

Circularity Check

0 steps flagged

No circularity in empirical MLIP comparisons

full rationale

The paper reports direct training, validation, and benchmarking of two independent MLIP frameworks (NEP and GRACE) across 16 metals and alloys. All central claims—training efficiency, accuracy on mechanical/thermal/chemical tasks, inference speed, and ensemble uncertainty correlation—are obtained from standard held-out testing and timing measurements on the trained models. No step defines a quantity in terms of itself, renames a fitted parameter as a prediction, or reduces a result to a self-citation chain. The ~60-fold speed claim is a measured benchmark, not a derived quantity forced by prior equations. The work is self-contained against external benchmarks and falsifiable via independent retraining or timing runs.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claims rest on the assumption that the ML models generalize from training data to the tested properties and that the reference data (likely DFT) is accurate.

free parameters (1)
  • NEP and GRACE model parameters
    Hundreds to thousands of parameters in the ML models fitted to reference data from DFT calculations.
axioms (1)
  • domain assumption MLIPs trained on elemental and alloy data can extrapolate to multicomponent systems
    Central to the chemical extrapolation claims.

pith-pipeline@v0.9.0 · 5499 in / 1430 out tokens · 67873 ms · 2026-05-13T21:21:14.393495+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages

  1. [1]

    Zhang, G

    L. Zhang, G. Cs´ anyi, E. van der Giessen, and F. Maresca, Efficiency, accuracy, and transferability of machine learn- ing potentials: Application to dislocations and cracks in iron, Acta Materialia270, 119788 (2024)

    work page 2024

  2. [2]

    K. Ito, T. Yokoi, K. Hyodo, and H. Mori, Machine learn- ing interatomic potential with dft accuracy for general grain boundaries inα-fe, npj Computational Materials 10, 10.1038/s41524-024-01451-y (2024)

    work page doi:10.1038/s41524-024-01451-y 2024

  3. [3]

    J.-P. Du, P. Yu, S. Shinzato, F.-S. Meng, Y. Sato, Y. Li, Y. Fan, and S. Ogata, Chemical domain structure and its formation kinetics in crconi medium-entropy alloy, Acta Materialia240, 118314 (2022)

    work page 2022

  4. [4]

    H. Meng, Y. Liu, H. Yu, L. Zhuang, and Y. Chu, Machine-learning-potential-driven prediction of high- entropy ceramics with ultra-high melting points, Cell Re- ports Physical Science6, 102449 (2025)

    work page 2025

  5. [5]

    Meng, J.-P

    F.-S. Meng, J.-P. Du, S. Shinzato, H. Mori, P. Yu, K. Matsubara, N. Ishikawa, and S. Ogata, General- purpose neural network interatomic potential for the α-iron and hydrogen binary system: Toward atomic- scale understanding of hydrogen embrittlement, Physical Review Materials5, 10.1103/physrevmaterials.5.113606 (2021)

    work page doi:10.1103/physrevmaterials.5.113606 2021

  6. [6]

    F.-S. Meng, S. Shinzato, S. Zhang, K. Matsubara, J.-P. Du, P. Yu, W.-T. Geng, and S. Ogata, A highly transfer- able and efficient machine learning interatomic potentials study ofα-Fe–C binary system, Acta Materialia281, 120408 (2024)

    work page 2024

  7. [7]

    K. Ito, T. Otaki, T. Yokoi, K. Hyodo, and H. Mori, Ma- chine learning interatomic potential reveals hydrogen em- brittlement origins at general grain boundaries inα-iron, Communications Materials7, 10.1038/s43246-025-01042- 4 (2025)

    work page doi:10.1038/s43246-025-01042- 2025

  8. [8]

    Cosmological baryon and lepto n number in the presence of electroweak fermion number violation,

    J. Behler and M. Parrinello, Generalized neural-network representation of high-dimensional potential-energy sur- faces, Physical Review Letters98, 10.1103/phys- revlett.98.146401 (2007)

    work page doi:10.1103/phys- 2007

  9. [9]

    A. P. Bart´ ok, M. C. Payne, R. Kondor, and G. Cs´ anyi, Gaussian approximation potentials: The accuracy of quantum mechanics, without the electrons, Physi- cal Review Letters104, 10.1103/physrevlett.104.136403 (2010)

    work page doi:10.1103/physrevlett.104.136403 2010

  10. [10]

    Zhang, G

    L. Zhang, G. Cs´ anyi, E. van der Giessen, and F. Maresca, Atomistic fracture in bcc iron revealed by active learning of gaussian approximation potential, npj Computational Materials9, 10.1038/s41524-023-01174-6 (2023)

    work page doi:10.1038/s41524-023-01174-6 2023

  11. [11]

    I. S. Novikov, K. Gubaev, E. V. Podryabinkin, and A. V. Shapeev, The mlip package: moment tensor potentials with mpi and active learning, Machine Learning: Science and Technology2, 025002 (2021)

    work page 2021

  12. [12]

    Zhang, J

    L. Zhang, J. Han, H. Wang, R. Car, and W. E, Deep potential molecular dynamics: A scalable model with the accuracy of quantum mechanics, Physical Review Letters 120, 10.1103/physrevlett.120.143001 (2018)

    work page doi:10.1103/physrevlett.120.143001 2018

  13. [13]

    Drautz, Atomic cluster expansion for accurate and transferable interatomic potentials, Physical Review B 99, 014104 (2019)

    R. Drautz, Atomic cluster expansion for accurate and transferable interatomic potentials, Physical Review B 99, 014104 (2019)

    work page 2019

  14. [14]

    X.-y. Meng, L. Zhang, H.-y. Wang, and L.-h. Dai, Atom- istic mechanism underlying shock-induced phase transi- tion in hfnbtatizr energetic high-entropy alloy, Energetic Materials Frontiers6, 103–111 (2025)

    work page 2025

  15. [15]

    B. Deng, P. Zhong, K. Jun, J. Riebesell, K. Han, C. J. Bartel, and G. Ceder, Chgnet as a pretrained universal neural network potential for charge-informed atomistic modelling, Nature Machine Intelligence5, 1031 (2023)

    work page 2023

  16. [16]

    Y.-L. Liao, B. Wood, A. Das, and T. Smidt, Equiformerv2: Improved equivariant transformer for scaling to higher-degree representations, arXiv preprint arXiv:2306.12059 (2023)

    work page arXiv 2023

  17. [17]

    Y. Park, J. Kim, S. Hwang, and S. Han, Scalable parallel algorithm for graph neural network interatomic poten- tials in molecular dynamics simulations, J. Chem. Theory Comput.20, 4857 (2024)

    work page 2024

  18. [18]

    Neumann, J

    M. Neumann, J. Gin, B. Rhodes, S. Bennett, Z. Li, H. Choubisa, A. Hussey, and J. Godwin, Orb: A fast, scalable neural network potential, arXiv preprint arXiv:2410.22570 (2024)

    work page arXiv 2024

  19. [19]

    Wines and K

    D. Wines and K. Choudhary, Chips-ff: Evaluating uni- versal machine learning force fields for material proper- ties, ACS Materials Letters7, 2105–2114 (2025). 13

    work page 2025

  20. [20]

    B. M. Wood, M. Dzamba, X. Fu, M. Gao, M. Shuaibi, L. Barroso-Luque, K. Abdelmaqsoud, V. Gharakhanyan, J. R. Kitchin, D. S. Levine, K. Michel, A. Sriram, T. Co- hen, A. Das, A. Rizvi, S. J. Sahoo, Z. W. Ulissi, and C. L. Zitnick, Uma: A family of universal models for atoms (2026), arXiv:2506.23971 [cs.LG]

    work page arXiv 2026

  21. [21]

    Z. Fan, Z. Zeng, C. Zhang, Y. Wang, K. Song, H. Dong, Y. Chen, and T. Ala-Nissila, Neuroevolution machine learning potentials: Combining high accuracy and low cost in atomistic simulations and application to heat transport, Physical Review B104, 104309 (2021)

    work page 2021

  22. [22]

    K. Xu, H. Bu, S. Pan, E. Lindgren, Y. Wu, Y. Wang, J. Liu, K. Song, B. Xu, Y. Li, T. Hainer, L. Svens- son, J. Wiktor, R. Zhao, H. Huang, C. Qian, S. Zhang, Z. Zeng, B. Zhang, B. Tang, Y. Xiao, Z. Yan, J. Shi, Z. Liang, J. Wang, T. Liang, S. Cao, Y. Wang, P. Ying, N. Xu, C. Chen, Y. Zhang, Z. Chen, X. Wu, W. Jiang, E. Berger, Y. Li, S. Chen, A. J. Gabouri...

    work page doi:10.1002/mgea.70028 2025

  23. [23]

    H. Dong, Y. Shi, P. Ying, K. Xu, T. Liang, Y. Wang, Z. Zeng, X. Wu, W. Zhou, S. Xiong,et al., Molecular dynamics simulations of heat transport using machine- learned potentials: A mini-review and tutorial on gpumd with neuroevolution potentials, Journal of Ap- plied Physics135(2024)

    work page 2024

  24. [24]

    P. Ying, W. Zhou, L. Svensson, E. Berger, E. Frans- son, F. Eriksson, K. Xu, T. Liang, J. Xu, B. Song,et al., Highly efficient path-integral molecular dynamics simula- tions with gpumd using neuroevolution potentials: Case studies on thermal properties of materials, The Journal of Chemical Physics162(2025)

    work page 2025

  25. [25]

    Liang, K

    T. Liang, K. Xu, E. Lindgren, Z. Chen, R. Zhao, J. Liu, E. Berger, B. Tang, B. Zhang, Y. Wang, K. Song, P. Ying, N. Xu, H. Dong, S. Chen, P. Erhart, Z. Fan, T. Ala- Nissila, and J. Xu, Nep89: Universal neuroevolution po- tential for inorganic and organic materials across 89 ele- ments (2025)

    work page 2025

  26. [26]

    Bochkarev, Y

    A. Bochkarev, Y. Lysogorskiy, and R. Drautz, Graph atomic cluster expansion for semilocal interactions be- yond equivariant message passing, Phys. Rev. X14, 021036 (2024)

    work page 2024

  27. [27]

    Barroso-Luque et al., Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models

    L. Barroso-Luque, M. Shuaibi, X. Fu, B. M. Wood, M. Dzamba, M. Gao, A. Rizvi, C. L. Zitnick, and Z. W. Ulissi, Open materials 2024 (omat24) inorganic materi- als dataset and models (2024), arXiv:2410.12771 [cond- mat.mtrl-sci]

    work page arXiv 2024

  28. [28]

    Schmidt, N

    J. Schmidt, N. Hoffmann, H. Wang, P. Borlido, P. J. M. A. Carri¸ co, T. F. T. Cerqueira, S. Botti, and M. A. L. Marques, Machine-learning-assisted determination of the global zero-temperature phase diagram of materials, Ad- vanced Materials35, 10.1002/adma.202210788 (2023)

    work page doi:10.1002/adma.202210788 2023

  29. [29]

    Lysogorskiy, A

    Y. Lysogorskiy, A. Bochkarev, and R. Drautz, Graph atomic cluster expansion for foundational machine learn- ing interatomic potentials, npj Computational Materials 10.1038/s41524-026-01979-1 (2026)

    work page doi:10.1038/s41524-026-01979-1 2026

  30. [30]

    A. P. Thompson, H. M. Aktulga, R. Berger, D. S. Bolin- tineanu, 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, Comp. Phys. Comm.271,...

    work page 2022

  31. [31]

    S. Zhao, G. M. Stocks, and Y. Zhang, Stacking fault en- ergies of face-centered cubic concentrated solid solution alloys, Acta Materialia134, 334 (2017)

    work page 2017

  32. [32]

    X. Chen, Q. Wang, Z. Cheng, M. Zhu, H. Zhou, P. Jiang, L. Zhou, Q. Xue, F. Yuan, J. Zhu,et al., Direct observa- tion of chemical short-range order in a medium-entropy alloy, Nature592, 712 (2021)

    work page 2021

  33. [33]

    Cao, Maximum strength and dislocation patterning in multi–principal element alloys, Science advances8, eabq7433 (2022)

    P. Cao, Maximum strength and dislocation patterning in multi–principal element alloys, Science advances8, eabq7433 (2022)

    work page 2022

  34. [34]

    S. K. Nemani, M. Torkamanzadeh, B. C. Wyatt, V. Presser, and B. Anasori, Functional two-dimensional high-entropy materials, Communications Materials4, 16 (2023)

    work page 2023

  35. [35]

    Zhang and M

    R.-Z. Zhang and M. J. Reece, Review of high entropy ce- ramics: design, synthesis, structure and properties, Jour- nal of Materials Chemistry A7, 22148 (2019)

    work page 2019

  36. [36]

    X. Zhou, J. Marian, F. Zhou, and V. V. Bulatov, Probing multi-dimensional composition spaces in search of strong metallic alloys, npj Computational Materials (2026)

    work page 2026

  37. [37]

    L. A. Zepeda-Ruiz, A. Stukowski, T. Oppelstrup, and V. V. Bulatov, Probing the limits of metal plasticity with molecular dynamics simulations, Nature550, 492 (2017)

    work page 2017

  38. [38]

    K. Song, R. Zhao, J. Liu, Y. Wang, E. Lindgren, Y. Wang, S. Chen, K. Xu, T. Liang, P. Ying,et al., General-purpose machine-learned potential for 16 ele- mental metals and their alloys, Nature Communications 15, 10208 (2024)

    work page 2024

  39. [39]

    Kresse and J

    G. Kresse and J. Furthm¨ uller, Efficient iterative schemes forab initiototal-energy calculations using a plane-wave basis set, Physical Review B54, 11169–11186 (1996)

    work page 1996

  40. [40]

    J. P. Perdew, K. Burke, and M. Ernzerhof, Generalized gradient approximation made simple, Physical Review Letters77, 3865–3868 (1996)

    work page 1996

  41. [41]

    P. E. Bl¨ ochl, Projector augmented-wave method, Physi- cal Review B50, 17953–17979 (1994)

    work page 1994

  42. [42]

    V. Wang, N. Xu, J.-C. Liu, G. Tang, and W.-T. Geng, Vaspkit: A user-friendly interface facilitating high- throughput computing and analysis using vasp code, Computer Physics Communications267, 108033 (2021)

    work page 2021

  43. [43]

    Fan, Dataset for UNEP-v1, 10.5281/zenodo.11533864 (2024)

    Z. Fan, Dataset for UNEP-v1, 10.5281/zenodo.11533864 (2024)

    work page doi:10.5281/zenodo.11533864 2024

  44. [44]

    J. F. Ziegler and J. P. Biersack, The stopping and range of ions in matter, inTreatise on Heavy-Ion Science: Vol- ume 6: Astrophysics, Chemistry, and Condensed Mat- ter, edited by D. A. Bromley (Springer US, Boston, MA,

    work page

  45. [45]

    Lysogorskiy, A

    Y. Lysogorskiy, A. Bochkarev, M. Mrovec, and R. Drautz, Active learning strategies for atomic cluster expansion models, Physical Review Materials7, 043801 (2023)

    work page 2023

  46. [46]

    E. V. Podryabinkin and A. V. Shapeev, Active learning of linearly parametrized interatomic potentials, Compu- tational Materials Science140, 171 (2017). [47]https://github.com/psn417/nep_extrapolation

    work page 2017

  47. [47]

    A. Stukowski, Visualization and analysis of atomistic simulation data with ovito–the open visualization tool, Modelling and Simulation in Materials Science and En- gineering18, 015012 (2009). 14

    work page 2009

  48. [48]

    Shuang, Z

    F. Shuang, Z. Wei, K. Liu, W. Gao, and P. Dey, Model accuracy and data heterogeneity shape uncertainty quan- tification in machine learning interatomic potentials, Machine Learning: Science and Technology7, 025002 (2026)

    work page 2026

  49. [49]

    Leimeroth, L

    N. Leimeroth, L. C. Erhard, K. Albe, and J. Rohrer, Machine-learning interatomic potentials from a users per- spective: a comparison of accuracy, speed and data ef- ficiency, Modelling and Simulation in Materials Science and Engineering33, 065012 (2025)

    work page 2025

  50. [50]

    S. Pan, J. Shi, Z. Liang, C. Liu, J. Wang, Y. Wang, H.-T. Wang, D. Xing, and J. Sun, Shock compression pathways to pyrite silica from machine learning simulations, Phys- ical Review B110, 10.1103/physrevb.110.224101 (2024)

    work page doi:10.1103/physrevb.110.224101 2024

  51. [51]

    Shuang, Z

    F. Shuang, Z. Wei, K. Liu, W. Gao, and P. Dey, Uni- versal machine learning interatomic potentials poised to supplant dft in modeling general defects in metals and random alloys, Machine Learning: Science and Technol- ogy6, 030501 (2025)

    work page 2025

  52. [52]

    Huang, J

    H. Huang, J. Peng, K. Li, J. Zhou, and Z. Sun, Effi- cient gpu-accelerated training of a neuroevolution poten- tial with analytical gradients, Computer Physics Com- munications320, 109994 (2026)

    work page 2026