Pith. sign in

REVIEW 3 major objections 5 minor 73 references

AutoPot: Automated and massively parallelized construction of Machine-Learning Potentials

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a fully automated active-learning workflow can now train quantum-accurate machine-learning interatomic potentials in a couple of hours, with no human in the loop and with every intermediate calculation archived.

desk verdict AutoPot is a useful, well-demonstrated workflow tool for automated MTP active learning; the stopping-criterion caveat and a bug in Algorithm 2.2 need fixing, but the software deserves peer review. read the letter →

arxiv 2601.01185 v2 pith:MSTS5UN7 submitted 2026-01-03 physics.comp-ph cond-mat.mtrl-sci

Max Hodapp , Guillaume Anciaux This is my paper · ORCID
classification physics.comp-phcond-mat.mtrl-sci
keywords activelearningmachine-learninginteratomicpotentialsmomenttensorD-optimalitycriterionextrapolationgradeworkfloworchestrationreproducibilityrandomalloys
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

This paper introduces AutoPot, a software platform that claims to automate the entire active-learning protocol for machine-learning interatomic potentials. The workflow selects uncertain atomic configurations from large candidate sets or from running molecular dynamics, launches density functional theory single-point calculations that add these configurations to the training set, retrains the potential, and repeats until no new configuration is deemed uncertain. The central claim is that a user can go from a small initial training set and a pool of candidates to a production-ready moment tensor potential within a couple of hours, with minimal manual input, because the event-driven orchestrator runs thousands of tasks in parallel and preserves every intermediate dataset for reproducibility. The paper demonstrates this for tungsten and for molybdenum-tantalum random alloys: the resulting potentials are trained from only 123 and 389 configurations, respectively, and predict elastic constants, stacking-fault energies, dislocation core structures, and Peierls barriers within a few percent of density functional theory reference values. A sympathetic reader would care because the bottleneck of building high-quality training sets, currently requiring expert attention and careful protocols, is removed.

What carries the argument

The extrapolation grade γ (defined in Eq. 6) is the scalar that carries the entire active-learning logic: for a candidate configuration, γ = max_i |c_i|, where c is the row vector of the candidate's derivative with respect to the MTP parameters multiplied by the inverse of the Jacobian matrix A of the active set. It classifies a configuration as interpolating (γ≤1), accurately extrapolating (1<γ≤2), still reliable (2<γ≤10), or risky (γ>10), and AutoPot aborts MD simulations or spawns DFT calculations accordingly. The maxvol algorithm selects the m most linearly independent neighborhoods to build A efficiently. Around this quantity, the paper builds an event-based asynchronous orchestrator th

What would settle it

Run a molecular dynamics simulation at, say, 2000 K for tungsten using the AutoPot-trained potential, compute the DFT energy of a few sampled snapshots, and check whether any snapshot with γ<2 deviates by more than, say, 50 meV/atom in energy. If such a snapshot exists, the D-optimality gate failed to select it.

Watch

Extended reading notes

Core claim

The central claim is that constructing a quantum-accurate machine-learning interatomic potential can be made fully automatic by tying together, in a parallelized event-driven workflow, the four steps of active learning—configuration selection, reference quantum-mechanical evaluation, retraining, and convergence checking—and by storing every dataset and metadata so the process can be queried, reused, and reproduced. The decision of which configurations need quantum-mechanical evaluation is driven by the extrapolation grade γ, a scalar derived from the D-optimality criterion: γ measures, effectively, how far a candidate atomic neighborhood is from the current training set. Configurations above

Load-bearing premise

The load-bearing premise is that the extrapolation grade γ reliably flags every atomic neighborhood where the potential is inaccurate; if a configuration has a low γ but the potential's energy and forces there are nevertheless far from quantum-mechanical results, AutoPot reports convergence and hands over an undertrained potential.

Editorial extensions

If this is right

  • A user can start from a handful of initial configurations and a set of candidate structures and receive a converged moment tensor potential in about two hours, without manually adjusting thresholds or restarting simulations.
  • Every task input, output, and parameter value is stored in a queryable object database, so a finished training run can be reconstructed exactly, or a branch of the workflow can be rerun without recomputing unchanged parts.
  • Because tasks are defined as plain Python functions, the same orchestrator can accommodate other potential classes and other quantum-chemistry packages, making the protocol a testbed for comparing active-learning strategies.
  • The on-the-fly sampling stage can monitor thousands of independent molecular dynamics runs, automatically requesting fresh quantum-mechanical reference data for any trajectory that begins to extrapolate, and aborting those that go too far.

Reading between the lines

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

  • The paper's convergence claim leans on the extrapolation grade γ being a faithful error oracle; since the authors themselves note that γ is not a good quantification of the true error, a natural next test is to couple AutoPot with calibrated uncertainty estimates to see whether the automatically chosen training set still misses rare but important neighborhoods.
  • The orchestration pattern—event-based, parallel, data-preserving—appears portable to other active-learning problems, such as training continuum models from atomistic data, which the paper explicitly names as future work.
  • A concrete stress test of the 'fully automatic in a couple of hours' claim would be running the identical protocol on a ternary or magnetic alloy; the growth in configuration space and the need for multiple γ thresholds may push the runtime well beyond the two-hour mark, revealing how much of the claim depends on system simplicity.
  • The demonstrated benchmarks focus on ground-state and low-temperature properties; long high-temperature runs or radiation-damage conditions would be a stronger test of whether the automatically assembled training set ensures simulation stability far from the initial candidate distribution.
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

3 major / 5 minor

Summary. The paper presents AutoPot, an open-source workflow for automating the construction of Moment Tensor Potentials (MTPs) through active learning. AutoPot is built on the BlackDynamite parametric-study framework and the Motoko event-based orchestrator, and it implements two selection protocols: pre-selection from a large candidate set (Algorithm 2.1) and on-the-fly selection from molecular dynamics simulations (Algorithm 2.2), both using the D-optimality extrapolation grade γ. The workflow is demonstrated by constructing MTPs for bcc W and for Mo-Ta random alloys, with validation against DFT for lattice constants, elastic constants, stacking-fault energies, and, for W, screw-dislocation core structure and Peierls barrier. The central claim is that AutoPot enables fully automatic, reproducible construction of accurate MTPs within a couple of hours with minimal user input.

Significance. The automation and provenance-tracking of MLIP training protocols is a genuine need in computational materials science, and the paper's architecture (persistent task storage, asynchronous orchestration, Python-based tasks) is a reasonable response. If the workflow is robust, it lowers the barrier for non-experts and improves reproducibility. The two demonstrations show that the produced potentials match DFT or reference MTPs well for the tested properties, which is encouraging. The paper explicitly provides open-source repositories, which is a strength. However, the validity of the stopping criterion and the quantitative performance claim ('couple of hours') are not demonstrated, and one of the two presented algorithms contains a logical error. These issues need to be resolved before the paper's claims can be fully accepted.

major comments (3)
  1. [Algorithm 2.2, lines 7–11] The conditional in the MD selection loop is logically inverted. If γmin < γmax (the natural reading, also implied by the names 'min' and 'max'), then the condition 'γ* > γmax' is never reached for γ* > γmin, because any γ* > γmax already satisfies γ* > γmin and is captured by the first branch. Thus the abort-on-risky-configuration branch is unreachable. If instead γmin > γmax, the naming is confusing and the abort branch is reachable only for γmax < γ* ≤ γmin, again not the intended semantics. This is a load-bearing defect: the safety mechanism that prevents continuing MD in unsafe extrapolation regions does nothing as written. Even if the implementation in the repository is correct, the pseudocode misrepresents the algorithm and could mislead users.
  2. [Abstract and §5.2] The claim that AutoPot constructs 'highly accurate MTPs ... fully automatically within a couple of hours' is unsupported by any timing or scaling data. No wall-clock times, number of VASP single points, degree of parallelism, or speedups are reported. The paper mentions that the examples ran on VSC-5, but that is not sufficient to substantiate a quantitative performance claim. The authors should either provide timing measurements for the W and Mo-Ta cases (including the number of DFT calculations and the parallelization achieved) or soften the claim to something qualitative such as 'rapid'.
  3. [§2.3 and §6] The stopping criterion of the workflow is that no configuration has extrapolation grade γ above γmin (typically 1–2). The paper's own Section 6 states that γ 'is an error indicator and guarantees a stable potential but is not necessarily a good quantification of the true error.' This is a frank admission, but it means that the central 'fully automatic' claim rests on an uncalibrated proxy. The paper reports no test of whether γ ≤ γmin corresponds to acceptable energy/force errors on held-out configurations, nor any sensitivity analysis of the choice of γmin. Without such a calibration, a user cannot know whether AutoPot terminates with an under-trained potential in a new application. I recommend adding either (i) a calibration plot of γ against actual errors on a held-out set for the demonstrated systems, or (ii) a comparison of potentials produced with different γmin values (e.g., 1, 2,
minor comments (5)
  1. [§5.2, Table 1 and Figure 6] The text says 'The MTP training errors are given in Table 2' and 'the agreement with DFT ... as shown in Table 2', but Table 2 is the W property table; the training errors are in Table 1, and the Mo-Ta property comparison is in Figure 6. Please correct these cross-references.
  2. [Algorithm 2.1 and §2.3 text] The pseudocode of Algorithm 2.1 adds every candidate with γ > γthreshold to T_selected, while the preceding paragraph argues that in practice it suffices to add only the candidate with the highest extrapolation grade. This is a real algorithmic difference, not just a simplification. Clarify which behavior is implemented in AutoPot and, if only the highest is added, align the pseudocode with the implementation.
  3. [§2.2, after Eq. (5)] The sentence 'We remark that the training set typically contains many more configurations than parameters, making A overdetermined' is imprecise: A as defined in Eq. (5) is an m×m matrix built from m active configurations. What is meant is that the full training set is overdetermined and maxvol selects the m most representative rows. Suggest rewording.
  4. [Algorithm 2.2, line 15] The instruction 'Add the most representative configurations from T_cand to the training set' is vague. How is 'most representative' determined? Since Algorithm 2.1 is then invoked with γthreshold = γmin, and all configurations in T_cand have γ > γmin by construction, it is unclear how any filtering occurs. Please specify the criterion used.
  5. [§6, first paragraph] The sentence 'The D-optimality criterion is an error indicator and guarantees a stable potential' conflates two different levels of certainty. An 'error indicator' does not 'guarantee' stability in a mathematical sense. Given the same paragraph immediately notes that γ is not a good quantification of the true error, please soften 'guarantees' to 'is intended to support' or similar.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; minor self-citations and an admitted calibration limitation do not make the workflow self-deriving.

full rationale

The central derivation chain is workflow automation, not a mathematical prediction. The MTP is trained on DFT energies/forces/stresses via Eq. (4), and the property benchmarks (lattice constants, elastic constants, stacking fault energies) are computed from the trained MTP and compared with DFT values from Xu et al., Hu et al., Bijjala et al., and [70] — external references. The D-optimality extrapolation grade gamma (Eq. 6) is an internal selection/stopping heuristic; the paper does not define accuracy in terms of gamma and in fact concedes in Section 6 that gamma "is an error indicator and guarantees a stable potential but is not necessarily a good quantification of the true error." That is a calibration/robustness limitation, not a circular reduction: the convergence criterion is a heuristic, not a definitional guarantee of accuracy. Some self-citations appear (e.g., the reference MTP of Hodapp and Shapeev [28] for the W dislocation benchmarks, [51] for some Mo-Ta SFE DFT values, and [28] also for the W elastic-constant DFT values), but these are prior-published DFT or MTP values, not fitted outputs of the present workflow, and the benchmarks also rest on independent external DFT data elsewhere. The comparison to the authors' own reference MTP for dislocation cores is a transferability-style check rather than a case where the predicted quantity is defined by the cited reference. No equation in the paper reduces to its input by construction, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical quantities or free parameters in a derivation sense. The workflow relies on user-chosen thresholds and standard MTP hyperparameters, which affect the final training set but are not fitted to match the benchmark properties. The principal conceptual assumptions are the D-optimality uncertainty proxy and transferability from bulk/SF to dislocation neighborhoods.

free parameters (5)
  • select extrapolation grade thresholds (γ_threshold) = W: [100000, 10000, 2]; Mo-Ta: [100000, 10000, 2]
    Chosen by the user to control iterative pre-selection in Algorithm 2.1; not inferred from data. They control the size and composition of the final training set.
  • MD selection thresholds γ_min, γ_max
    User-defined abort/select thresholds in Algorithm 2.2; the paper does not state their values for the W/Mo-Ta runs.
  • loss weights w_e, w_f, w_s
    Weights in Eq. (4) balancing energy, force, and stress errors. Standard MTP hyperparameters, chosen by hand; not reported numerically.
  • strain bounds and displacement perturbation = W: ±2%, σ=0.02 Å; Mo-Ta: ±3%, σ=0.02 Å
    Generation hyperparameters for training candidates, chosen by the authors.
  • MD temperature = 100 K
    Langevin dynamics temperature in Section 5.1; chosen to be near ground state, no sensitivity analysis.
assumptions (5)
  • domain assumption The total energy decomposes into a sum of per-atom energies E(N) (Eq. 1)
    Locality assumption central to MTPs; standard for short-range interatomic potentials with cut-off radius 5 Å.
  • domain assumption The D-optimality extrapolation grade γ is a reliable indicator of extrapolation and thereby of the need for retraining
    Used throughout Sections 2.2-2.3 and Algorithms 2.1/2.2; the authors qualify in Section 6 that γ is not a true error estimate, so this is a substantive assumption.
  • standard math maxvol (Goreinov et al.) computes an active set that makes the Jacobian A well-conditioned
    Invoked in Section 2.2 to select m neighborhoods when the training set is overdetermined; taken as a known numerical tool.
  • domain assumption Training on bulk and 1/4[111] stacking-fault configurations transfers to dislocation core properties
    The W MTP is trained only on bulk and SF cells, yet is benchmarked on screw dislocation cores (Sec. 5.1). The reasonable agreement with the reference MTP supports this assumption for the tested case, but it is not guaranteed generally.
  • domain assumption DFT (PBE/VASP) is the reference truth for energies, forces, and stresses
    Standard practice; all labels come from VASP single-point calculations as described in Section 5.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AutoPot: Automated and massively parallelized construction of Machine-Learning Potentials." pith.science (2026). https://pith.science/paper/MSTS5UN7

@misc{pith2026260101185,
  author       = {Pith},
  title        = {Pith review of: AutoPot: Automated and massively parallelized construction of Machine-Learning Potentials},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MSTS5UN7}},
  note         = {Machine review of arXiv:2601.01185}
}
read the original abstract

Machine-learning potentials (MLIPs) have been a breakthrough for computational physics in bringing the accuracy of quantum mechanics to atomistic modeling. To achieve near-quantum accuracy, it is necessary that neighborhoods contained in the training set are rather close to the ones encountered during a simulation. Yet, constructing a single training set that works well for all applications is, and likely will remain, infeasible, so, one strategy is to supplement training protocols for MLIPs with additional learning methods, such as active learning, or fine-tuning. This strategy, however, yields very complex training protocols that are difficult to implement efficiently, and cumbersome to interpret, analyze, and reproduce. To address the above difficulties, we propose AutoPot, a software for automating the construction and archiving of MLIPs. AutoPot is based on BlackDynamite, a software operating parametric tasks, e.g., running simulations, or single-point ab initio calculations, in a highly-parallelized fashion, and Motoko, an event-based workflow manager for orchestrating interactions between the tasks. The initial version of AutoPot supports selection of training configurations from large training candidate sets, and on-the-fly selection from molecular dynamics simulations, using Moment Tensor Potentials as implemented in MLIP-2, and single-point calculations of the selected training configurations using VASP. Another strength of AutoPot is its flexibility: BlackDynamite tasks and orchestrators are Python functions to which own existing code can be easily added and manipulated without writing complex parsers. Therefore, it will be straightforward to add other MLIP and ab initio codes, and manipulate the Motoko orchestrators to implement other training protocols.

Figures

Figures reproduced from arXiv: 2601.01185 by the authors.

Figure 1
Figure 1. Schematic of the active learning workflow implemented in AutoPot: (1) Starting from an initial training set, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. yaml files of the tasks implemented in AutoPot. For the sp_calc version that runs a VASP calculation, pot_fname corresponds to the INCAR file. For the train taks, the id’s of sp_calc tasks can be provided as a list in order to add the calculated energies, forces, and stresses, from those sp_calc tasks to the training set. 4.2 Orchestration In order to complete the workflow description, the orchestration is given as … view at source ↗
Figure 3
Figure 3. Workflow stages implemented in AutoPot. The cyan boxes are [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Evolution of training set size of the MTPs for W and Mo-Ta as a function of the selection iteration [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: (a) Differential displacement plot [71] of the relaxed dislocation core using the MTP for W. (b) Minimum energy path of the MTP for W and of the reference MTP from Hodapp and Shapeev [28] that has been trained on configurations containing the dislocation cores that app…
Figure 6
Figure 6. Figure 6: Comparison of the elastic properties and the 1/4 [111] unstable stacking fault energy of the MTP for Mo-Ta [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 15 canonical work pages

  1. [1]

    Generalized Neural-Network Representation of High-Dimensional Potential- Energy Surfaces

    Jörg Behler and Michele Parrinello. “Generalized Neural-Network Representation of High-Dimensional Potential- Energy Surfaces”. In:Physical Review Letters98.14 (Apr. 2, 2007), p. 146401.doi:10.1103/PhysRevLett.98. 146401

    doi:10.1103/physrevlett.98 2007

  2. [2]

    Gaussian Approximation Potentials: The Accuracy of Quantum Mechanics, without the Electrons

    Albert P. Bartók et al. “Gaussian Approximation Potentials: The Accuracy of Quantum Mechanics, without the Electrons”. In:Physical Review Letters104.13 (Apr. 1, 2010), p. 136403.doi:10.1103/PhysRevLett.104. 136403

    doi:10.1103/physrevlett.104 2010

  3. [3]

    Spectral neighbor analysis method for automated generation of quantum-accurate inter- atomic potentials

    A.P. Thompson et al. “Spectral neighbor analysis method for automated generation of quantum-accurate inter- atomic potentials”. In:Journal of Computational Physics285 (Mar. 2015), pp. 316–330.doi:10.1016/j.jcp. 2014.12.018

    doi:10.1016/j.jcp 2015

  4. [4]

    Moment Tensor Potentials: A Class of Systematically Improvable Interatomic Poten- tials

    Alexander V. Shapeev. “Moment Tensor Potentials: A Class of Systematically Improvable Interatomic Poten- tials”. In:Multiscale Modeling & Simulation14.3 (Jan. 2016), pp. 1153–1173.doi:10.1137/15M1054183

    doi:10.1137/15m1054183 2016

  5. [5]

    Deep Potential Molecular Dynamics: A Scalable Model with the Accuracy of Quantum Mechanics

    Linfeng Zhang et al. “Deep Potential Molecular Dynamics: A Scalable Model with the Accuracy of Quantum Mechanics”. In:Physical Review Letters120.14 (Apr. 4, 2018), p. 143001.doi:10.1103/PhysRevLett.120. 143001

    doi:10.1103/physrevlett.120 2018

  6. [6]

    Atomic cluster expansion for accurate and transferable interatomic potentials

    Ralf Drautz. “Atomic cluster expansion for accurate and transferable interatomic potentials”. In:Physical Review B99.1 (Jan. 8, 2019), p. 014104.doi:10.1103/PhysRevB.99.014104

    doi:10.1103/physrevb.99.014104 2019

  7. [7]

    Physically informed artificial neural networks for atomistic modeling of materials

    G. P. Purja Pun et al. “Physically informed artificial neural networks for atomistic modeling of materials”. In: Nature Communications10.1 (May 28, 2019), p. 2339.doi:10.1038/s41467-019-10343-5

    work page doi:10.1038/s41467-019-10343-5 2019

  8. [8]

    Machine Learning Force Fields

    Oliver T. Unke et al. “Machine Learning Force Fields”. In:Chemical Reviews121.16 (Aug. 25, 2021), pp. 10142– 10186.doi:10.1021/acs.chemrev.0c01111

    doi:10.1021/acs.chemrev.0c01111 2021

Show all 73 references
  1. [9]

    E(3)-equivariant graph neural networks for data-efficient and accurate interatomic poten- tials

    Simon Batzner et al. “E(3)-equivariant graph neural networks for data-efficient and accurate interatomic poten- tials”. In:Nature Communications13.1 (May 4, 2022), p. 2453.doi:10.1038/s41467-022-29939-5

    2022 doi

  2. [10]

    MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields

    Ilyes Batatia et al. “MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields”. In:Advances in Neural Information Processing Systems. Ed. by S. Koyejo et al. Vol. 35. Curran Associates, Inc., 2022, pp. 11423–11436

    2022

  3. [11]

    GPUMD: A package for constructing accurate machine-learned potentials and performing highly efficient atomistic simulations

    Zheyong Fan et al. “GPUMD: A package for constructing accurate machine-learned potentials and performing highly efficient atomistic simulations”. In:The Journal of Chemical Physics157.11 (Sept. 21, 2022), p. 114801. doi:10.1063/5.0106617

    2022 doi

  4. [12]

    Equivariant tensor network potentials

    M Hodapp and A Shapeev. “Equivariant tensor network potentials”. In:Machine Learning: Science and Tech- nology5.3 (Sept. 1, 2024), p. 035075.doi:10.1088/2632-2153/ad79b5

    2024 doi

  5. [13]

    E(n)-Equivariant cartesian tensor message passing interatomic potential

    Junjie Wang et al. “E(n)-Equivariant cartesian tensor message passing interatomic potential”. In:Nature Com- munications15.1 (Sept. 1, 2024), p. 7607.doi:10.1038/s41467-024-51886-6

    2024 doi

  6. [14]

    Recent Advances in Machine Learning-Assisted Multiscale Design of Energy Materials

    Bohayra Mortazavi. “Recent Advances in Machine Learning-Assisted Multiscale Design of Energy Materials”. In:Advanced Energy Materials15.9 (Mar. 2025), p. 2403876.doi:10.1002/aenm.202403876

    2025 doi

  7. [15]

    Modelingrefractoryhigh-entropyalloyswithefficientmachine- learned interatomic potentials: Defects and segregation

    J.Byggmästar,K.Nordlund,andF.Djurabekova.“Modelingrefractoryhigh-entropyalloyswithefficientmachine- learned interatomic potentials: Defects and segregation”. In:Physical Review B104.10 (Sept. 3, 2021), p. 104101. doi:10.1103/PhysRevB.104.104101

    2021 doi

  8. [16]

    Ab initio framework for deciphering trade-off relationships in multi-component alloys

    Franco Moitzi et al. “Ab initio framework for deciphering trade-off relationships in multi-component alloys”. In: npj Computational Materials10.1 (July 16, 2024), p. 152.doi:10.1038/s41524-024-01342-2

    2024 doi

  9. [17]

    General-purpose machine-learned potential for 16 elemental metals and their alloys

    Keke Song et al. “General-purpose machine-learned potential for 16 elemental metals and their alloys”. In:Nature Communications15.1 (Nov. 25, 2024), p. 10208.doi:10.1038/s41467-024-54554-x

    2024 doi

  10. [18]

    Atomistic fracture in bcc iron revealed by active learning of Gaussian approximation potential

    Lei Zhang et al. “Atomistic fracture in bcc iron revealed by active learning of Gaussian approximation potential”. In:npj Computational Materials9.1 (Dec. 8, 2023), p. 217.doi:10.1038/s41524-023-01174-6

    2023 doi

  11. [19]

    Machine learning interatomic potential with DFT accuracy for general grain boundaries in α-Fe

    Kazuma Ito et al. “Machine learning interatomic potential with DFT accuracy for general grain boundaries in α-Fe”. In:npj Computational Materials10.1 (Nov. 13, 2024), p. 255.doi:10.1038/s41524-024-01451-y

    2024 doi

  12. [20]

    Machine-learning potentials for nanoscale simulations of tensile deformation and fracture in ceramics

    Shuyao Lin et al. “Machine-learning potentials for nanoscale simulations of tensile deformation and fracture in ceramics”. In:npj Computational Materials10.1 (Apr. 2, 2024), p. 67.doi:10.1038/s41524-024-01252-3. 15

    2024 doi

  13. [21]

    Learning from models: high-dimensional analyses on the performance of machine learning interatomic potentials

    Yunsheng Liu and Yifei Mo. “Learning from models: high-dimensional analyses on the performance of machine learning interatomic potentials”. In:npj Computational Materials10.1 (July 20, 2024), p. 159.doi:10.1038/ s41524-024-01333-3

    2024

  14. [22]

    Erhard et al.How Realistic are Idealized Copper Surfaces? A Machine Learning Study of Rough Copper- Water Interfaces

    Linus C. Erhard et al.How Realistic are Idealized Copper Surfaces? A Machine Learning Study of Rough Copper- Water Interfaces. Sept. 22, 2025.doi:10.48550/arXiv.2509.17833. arXiv:2509.17833[cond-mat]

    2025 doi

  15. [23]

    Modeling extensive defects in metals through classical potential-guided sampling and auto- mated configuration reconstruction

    Fei Shuang et al. “Modeling extensive defects in metals through classical potential-guided sampling and auto- mated configuration reconstruction”. In:npj Computational Materials11.1 (May 3, 2025), p. 118.doi:10.1038/ s41524-025-01599-1

    2025

  16. [24]

    Active Learning Literature Survey

    Burr Settles. “Active Learning Literature Survey”. In: (2009)

    2009

  17. [25]

    Active learning of linearly parametrized interatomic po- tentials

    Evgeny V. Podryabinkin and Alexander V. Shapeev. “Active learning of linearly parametrized interatomic po- tentials”. In:Computational Materials Science140 (Dec. 2017), pp. 171–180.doi:10.1016/j.commatsci.2017. 08.031

    2017 doi

  18. [26]

    Active learning of uniformly accurate interatomic potentials for materials simulation

    Linfeng Zhang et al. “Active learning of uniformly accurate interatomic potentials for materials simulation”. In: Physical Review Materials3.2 (Feb. 25, 2019), p. 023804.doi:10.1103/PhysRevMaterials.3.023804

    2019 doi

  19. [27]

    On-the-fly active learning of interpretable Bayesian force fields for atomistic rare events

    Jonathan Vandermause et al. “On-the-fly active learning of interpretable Bayesian force fields for atomistic rare events”. In:npj Computational Materials6.1 (Mar. 18, 2020), p. 20.doi:10.1038/s41524-020-0283-z

    2020 doi

  20. [28]

    Inoperandoactivelearningofinteratomicinteractionduringlarge-scalesimulations

    MHodappandAShapeev.“Inoperandoactivelearningofinteratomicinteractionduringlarge-scalesimulations”. In:Machine Learning: Science and Technology1.4 (Dec. 1, 2020), p. 045005.doi:10.1088/2632-2153/aba373

    2020 doi

  21. [29]

    Nanohardness from First Principles with Active Learning on Atomic Environ- ments

    Evgeny V. Podryabinkin et al. “Nanohardness from First Principles with Active Learning on Atomic Environ- ments”. In:Journal of Chemical Theory and Computation18.2 (Feb. 8, 2022), pp. 1109–1121.doi:10.1021/ acs.jctc.1c00783

    2022

  22. [30]

    May 11, 2024.doi:10.48550/arXiv.2405.07105

    Bowen Deng et al.Overcoming systematic softening in universal machine learning interatomic potentials by fine-tuning. May 11, 2024.doi:10.48550/arXiv.2405.07105. arXiv:2405.07105[cond-mat]

    2024 doi arXiv

  23. [31]

    Actively trained magnetic moment tensor potentials for mechanical, dynamical, and thermal properties of paramagnetic CrN

    Alexey S. Kotykhov et al. “Actively trained magnetic moment tensor potentials for mechanical, dynamical, and thermal properties of paramagnetic CrN”. In:Physical Review B111.9 (Mar. 27, 2025), p. 094438.doi: 10.1103/PhysRevB.111.094438. [32]FAIR Principles. https://www.go-fair...

    2025 doi

  24. [39]

    https://www.materialscloud.org/home

    The Materials Cloud Team.Materials Cloud. https://www.materialscloud.org/home. Text

  25. [40]

    AiiDA: automated interactive infrastructure and database for computational science

    Giovanni Pizzi et al. “AiiDA: automated interactive infrastructure and database for computational science”. In: Computational Materials Science111 (2016), pp. 218–230.doi:https://doi.org/10.1016/j.commatsci. 2015.09.013. [41]Custodian. http://materialsproject.github.io/custodian/

    2016 doi

  26. [42]

    The Potential of Atomistic Simulations and the Knowledgebase of Interatomic Models

    E. B. Tadmor et al. “The Potential of Atomistic Simulations and the Knowledgebase of Interatomic Models”. In: JOM63.7 (July 2011), pp. 17–17.doi:10.1007/s11837-011-0102-6

    2011 doi

  27. [43]

    ColabFit exchange: Open-access datasets for data-driven interatomic potentials

    Joshua A. Vita et al. “ColabFit exchange: Open-access datasets for data-driven interatomic potentials”. In:The Journal of Chemical Physics159.15 (Oct. 21, 2023), p. 154802.doi:10.1063/5.0163882

    2023 doi

  28. [44]

    Atomate: A high-level interface to generate, execute, and analyze computational materials science workflows

    Kiran Mathew et al. “Atomate: A high-level interface to generate, execute, and analyze computational materials science workflows”. In:Computational Materials Science139 (2017), pp. 140–152.doi:https://doi.org/10. 1016/j.commatsci.2017.07.030

    2017

  29. [45]

    pyiron: An integrated development environment for computational materials science

    Jan Janssen et al. “pyiron: An integrated development environment for computational materials science”. In: Computational Materials Science163 (2019), pp. 24–36.doi:https://doi.org/10.1016/j.commatsci.2018. 07.043. [46]Motoko: a data preserving asyncronous orchestrator. https:...

    2019 doi

  30. [48]

    Machine learning of molecular prop- erties: Locality and active learning

    Konstantin Gubaev, Evgeny V. Podryabinkin, and Alexander V. Shapeev. “Machine learning of molecular prop- erties: Locality and active learning”. In:The Journal of Chemical Physics148.24 (June 28, 2018), p. 241727. doi:10.1063/1.5005095. 16

    2018 doi

  31. [49]

    How to Find a Good Submatrix

    S. A. Goreinov et al. “How to Find a Good Submatrix”. In: Vadim Olshevsky and Eugene Tyrtyshnikov.Matrix Methods: Theory, Algorithms and Applications. WORLD SCIENTIFIC, Apr. 2010, pp. 247–256.doi:10.1142/ 9789812836021_0015

    2010

  32. [50]

    The MLIP package: moment tensor potentials with MPI and active learning

    Ivan S Novikov et al. “The MLIP package: moment tensor potentials with MPI and active learning”. In:Machine Learning: Science and Technology2.2 (Jan. 1, 2021), p. 025002.doi:10.1088/2632-2153/abc9fe

    2021 doi

  33. [51]

    Machine-learning potentials enable predictive and tractable high-throughput screening of random alloys

    M. Hodapp and A. Shapeev. “Machine-learning potentials enable predictive and tractable high-throughput screening of random alloys”. In:Physical Review Materials5.11 (Nov. 22, 2021), p. 113802.doi:10.1103/ PhysRevMaterials.5.113802

    2021

  34. [52]

    AI-accelerated materials informatics method for the discovery of ductile alloys

    I. Novikov et al. “AI-accelerated materials informatics method for the discovery of ductile alloys”. In:Journal of Materials Research37.21 (Nov. 14, 2022), pp. 3491–3504.doi:10.1557/s43578-022-00783-z

    2022 doi

  35. [53]

    Apache License 2.0, open-source project

    Flyte Contributors.Flyte: Cloud-Native Workflow Orchestration Platform.https://github.com/flyteorg/ flyte. Apache License 2.0, open-source project. 2025

    2025

  36. [54]

    Apache License 2.0

    Kubeflow Community.Kubeflow Pipelines.https://github.com/kubeflow/pipelines. Apache License 2.0. 2025

    2025

  37. [55]

    Apache License 2.0

    MLflow Contributors.MLflow.https://github.com/mlflow/mlflow. Apache License 2.0. 2025

    2025

  38. [56]

    Apache License 2.0

    Apache Software Foundation.Apache Airflow.https://github.com/apache/airflow. Apache License 2.0. 2025

    2025

  39. [57]

    Apache License 2.0

    ZenML Contributors.ZenML.https://github.com/zenml-io/zenml. Apache License 2.0. 2025

    2025

  40. [58]

    Apache License 2.0

    Netflix and Metaflow Community.Metaflow.https://github.com/Netflix/metaflow. Apache License 2.0. 2025

    2025

  41. [59]

    BSD 3-Clause License

    Scikit-learn Developers.Scikit-learn: Machine Learning in Python.https://github.com/scikit-learn/ scikit-learn. BSD 3-Clause License. 2025

    2025

  42. [60]

    https://gitlab.com/ganciaux/blackdynamite

    Guillaume Anciaux and the BlackDynamite contributors.BlackDynamite: A Scientific Parametric-Study Tool. https://gitlab.com/ganciaux/blackdynamite. Version 1.3.3, GPL-3.0-or-later. 2025

    2025

  43. [61]

    Version 5.x

    Zope Foundation.ZODB: A Native Object Database for Python.https://www.zodb.org/. Version 5.x. 2024

    2024

  44. [62]

    Networked Storage Server for ZODB

    Zope Foundation.ZEO: Zope Enterprise Objects.https://github.com/zopefoundation/ZEO. Networked Storage Server for ZODB. 2024

    2024

  45. [63]

    The atomic simulation environment—a Python library for working with atoms

    Ask Hjorth Larsen et al. “The atomic simulation environment—a Python library for working with atoms”. In: Journal of Physics: Condensed Matter29.27 (July 12, 2017), p. 273002.doi:10.1088/1361-648X/aa680e

    2017 doi

  46. [64]

    Efficient iterative schemes forab initiototal-energy calculations using a plane- wave basis set

    G. Kresse and J. Furthmüller. “Efficient iterative schemes forab initiototal-energy calculations using a plane- wave basis set”. In:Physical Review B54.16 (Oct. 15, 1996), pp. 11169–11186.doi:10.1103/PhysRevB.54. 11169

    1996 doi

  47. [65]

    Projector augmented-wave method

    P. E. Blöchl. “Projector augmented-wave method”. In:Physical Review B50.24 (Dec. 15, 1994), pp. 17953–17979. doi:10.1103/PhysRevB.50.17953

    1994 doi

  48. [66]

    From ultrasoft pseudopotentials to the projector augmented-wave method

    G. Kresse and D. Joubert. “From ultrasoft pseudopotentials to the projector augmented-wave method”. In: Physical Review B59.3 (Jan. 15, 1999), pp. 1758–1775.doi:10.1103/PhysRevB.59.1758

    1999 doi

  49. [67]

    Generalized Gradient Approximation Made Simple

    John P. Perdew, Kieron Burke, and Matthias Ernzerhof. “Generalized Gradient Approximation Made Simple”. In:Physical Review Letters77.18 (Oct. 28, 1996), pp. 3865–3868.doi:10.1103/PhysRevLett.77.3865

    1996 doi

  50. [68]

    Automated atomistic simulations of dissociated dislocations withab initioaccu- racy

    L. Mismetti and M. Hodapp. “Automated atomistic simulations of dissociated dislocations withab initioaccu- racy”. In:Physical Review B109.9 (Mar. 27, 2024), p. 094120.doi:10.1103/PhysRevB.109.094120

    2024 doi

  51. [69]

    Machine-learning interatomic potential for radiation damage and defects in tungsten

    J. Byggmästar et al. “Machine-learning interatomic potential for radiation damage and defects in tungsten”. In: Physical Review B100.14 (Oct. 17, 2019), p. 144105.doi:10.1103/PhysRevB.100.144105

    2019 doi

  52. [70]

    Frank-Read source operation in six body-centered cubic refractory metals

    Shuozhi Xu et al. “Frank-Read source operation in six body-centered cubic refractory metals”. In:Journal of the Mechanics and Physics of Solids141 (Aug. 2020), p. 104017.doi:10.1016/j.jmps.2020.104017

    2020

  53. [71]

    Theory of the core structures of dislocations in body-centered-cubic metals

    V. Vitek. “Theory of the core structures of dislocations in body-centered-cubic metals”. In:Crystal Lattice Defects 5 (1974), pp. 1–34

    1974

  54. [72]

    Complex strengthening mechanisms in the NbMoTaW multi-principal element alloy

    Xiang-Guo Li et al. “Complex strengthening mechanisms in the NbMoTaW multi-principal element alloy”. In: npj Computational Materials6.1 (June 2, 2020), p. 70.doi:10.1038/s41524-020-0339-0

    2020 doi

  55. [73]

    Exact average many-body interatomic interaction model for random alloys

    M. Hodapp. “Exact average many-body interatomic interaction model for random alloys”. In:Computational Materials Today5 (Mar. 2025), p. 100018.doi:10.1016/j.commt.2024.100018

    2025

  56. [74]

    Ab-initio grain boundary thermodynamics beyond the dilute limit

    Tobias Spitaler et al. “Ab-initio grain boundary thermodynamics beyond the dilute limit”. In:Acta Materialia 286 (Mar. 2025), p. 120725.doi:10.1016/j.actamat.2025.120725. 17

    2025

  57. [75]

    Screening of generalized stacking fault energies, surface energies and intrinsic ductile potency of refractory multicomponent alloys

    Yong-Jie Hu et al. “Screening of generalized stacking fault energies, surface energies and intrinsic ductile potency of refractory multicomponent alloys”. In:Acta Materialia210 (May 2021), p. 116800.doi:10.1016/j.actamat. 2021.116800

    2021

  58. [76]

    Bijjala, Susan R

    Surya T. Bijjala, Susan R. Atlas, and Pankaj Kumar.Elastic tensor-derived properties of composition-dependent disordered refractory binary alloys using DFPT. Nov. 4, 2025.doi:10.48550/arXiv.2501.00127. arXiv: 2501.00127[cond-mat]

    2025 doi

  59. [77]

    Uncertainty-driven dynamics for active learning of interatomic potentials

    Maksim Kulichenko et al. “Uncertainty-driven dynamics for active learning of interatomic potentials”. In:Nature Computational Science3.3 (Mar. 6, 2023), pp. 230–239.doi:10.1038/s43588-023-00406-5

    2023 doi

  60. [78]

    Calibration of uncertainty in the active learning of machine learning force fields

    Adam Thomas-Mitchell, Glenn Hawe, and Paul L A Popelier. “Calibration of uncertainty in the active learning of machine learning force fields”. In:Machine Learning: Science and Technology4.4 (Dec. 1, 2023), p. 045034. doi:10.1088/2632-2153/ad0ab5

    2023 doi

  61. [79]

    Cheuk Hin Ho, Christoph Ortner, and Yangshuai Wang.Flexible Uncertainty Calibration for Machine-Learned Interatomic Potentials. Oct. 1, 2025.doi:10.48550/arXiv.2510.00721. arXiv:2510.00721[physics]

    2025 doi

  62. [80]

    Data-driven modeling of dislocation mobility from atomistics using physics-informed machine learning

    Yifeng Tian et al. “Data-driven modeling of dislocation mobility from atomistics using physics-informed machine learning”. In:npj Computational Materials10.1 (Sept. 14, 2024), p. 219.doi:10.1038/s41524-024-01394-4

    2024 doi

  63. [81]

    matscipy: materials science at the atomic scale withPython

    Petr Grigorev et al. “matscipy: materials science at the atomic scale withPython”. In:Journal of Open Source Software9.93 (Jan. 28, 2024), p. 5668.doi:10.21105/joss.05668

    2024 doi

  64. [82]

    Structural Relaxation Made Simple

    Erik Bitzek et al. “Structural Relaxation Made Simple”. In:Physical Review Letters97.17 (Oct. 27, 2006), p. 170201.doi:10.1103/PhysRevLett.97.170201

    2006 doi

  65. [83]

    Improved tangent estimate in the nudged elastic band method for finding minimum energy paths and saddle points

    Graeme Henkelman and Hannes Jónsson. “Improved tangent estimate in the nudged elastic band method for finding minimum energy paths and saddle points”. In:The Journal of Chemical Physics113.22 (Dec. 8, 2000), pp. 9978–9985.doi:10.1063/1.1323224. 18

    doi

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.