Pith. sign in

REVIEW 3 major objections 5 minor 32 references

A flexible kinetic Monte Carlo framework for GaN molecular beam epitaxy with adaptive on-the-fly barrier evaluation

T0 review · 3 major / 5 minor · reviewed 2026-07-31 · deepseek-v4-flash

Pith's one-line read A lattice kinetic Monte Carlo framework for GaN growth now computes step-edge barriers on the fly, revealing environment-dependent corrections that change predicted island morphology.

desk verdict A solid framework paper whose main new feature—MLIP/BEP on-the-fly barriers—is plausible but unvalidated; the OTF morphology changes fall within the uncertainty of α and CHGNet. read the letter →

arxiv 2607.24871 v1 pith:TVI6KFHH submitted 2026-07-27 cond-mat.mtrl-sci

classification cond-mat.mtrl-sci
keywords kineticMonteCarloGaN(0001)growthmolecularbeamepitaxyon-the-flybarrierevaluationBrønsted–Evans–Polanyimachine-learnedinteratomicpotentialEhrlich–SchwoebelOstwaldripening
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that a lattice kinetic Monte Carlo (KMC) framework can simulate GaN(0001) molecular beam epitaxy across multiple growth regimes, and that adding adaptive on-the-fly barrier evaluation—where activation energies are computed from the local atomic environment rather than pulled from a fixed catalog—captures step-edge physics that nearest-neighbor-only catalogs miss. Using predefined barriers, the framework reproduces compact triangular islands, Ostwald ripening during growth interruptions, Ehrlich–Schwoebel-barrier-induced multilayer nucleation, and a desorption-driven 'island walking' regime at high temperature. With on-the-fly barriers, it finds that specific step-edge configurations carry distinctly different activation energies (for example, a Ga2N step-edge hop is 0.14 eV higher than the flat-terrace value), and these corrections lead to morphologically different predictions such as fewer, larger islands and extended AdGa-decorated step edges. The value of the work is in showing that environment-aware barriers are both computable and consequential for GaN growth morphology.

What carries the argument

The central mechanism is the nn1×nn2 species-resolved coordination fingerprint plus the on-the-fly barrier-evaluation workflow. The fingerprint encodes the three tetrahedral bonding sites (nn1) and six same-sublattice lateral neighbors (nn2) around a diffusing atom, allowing the framework to distinguish terrace, step-edge, and kink geometries that share the same nn1 coordination. When a previously unseen fingerprint is encountered, the activation barrier is estimated via the Brønsted–Evans–Polanyi relation E_a = E_0 + αΔE, with α = 0.5 and ΔE computed by the machine-learned potential CHGNet, then cached in an in-memory dictionary and persistent database for reuse. This BEP-plus-MLIP approach

What would settle it

Compute full nudged elastic band barriers directly for the Ga2N step-edge configuration (nn2Ga = 2) using a reliable DFT functional and compare with the predicted 1.59 eV; a discrepancy larger than roughly 0.1 eV would indicate the BEP-CHGNet estimate is not capturing the step-edge stabilization. Alternatively, measure island size distributions in GaN MBE growth under slightly Ga-rich conditions and compare against the predicted shift to fewer, larger islands.

Watch

Extended reading notes

Core claim

The paper's central claim is that extending the local coordination descriptor from the first-neighbor shell (nn1) to include the six same-sublattice lateral neighbors (nn2) creates a configuration space of up to 140 or 497 distinct fingerprints, and that evaluating these barriers on the fly—rather than precomputing them all—makes environment-aware simulation feasible. When novel configurations arise, the framework computes activation barriers using a Brønsted–Evans–Polanyi relation E_a = E_0 + αΔE with α = 0.5, using CHGNet machine-learned reaction energies ΔE and reference barriers E_0 from the DFT catalog. This reveals that most novel barriers cluster within ±0.01 eV of the predefined valu

Load-bearing premise

The load-bearing premise is that the BEP estimate E_a = E_0 + αΔE with α = 0.5, using CHGNet reaction energies, gives accurate activation barriers for previously unencountered step-edge configurations; if the machine-learned reaction energies or the linear BEP form are unreliable for these GaN(0001) edge environments, the OTF corrections and the resulting morphology changes are unsupported.

Editorial extensions

If this is right

  • Predefined nn1-only barrier catalogs systematically miss environment-dependent effects at step edges and kinks; the on-the-fly extension resolves these, producing quantitatively different step-edge kinetics.
  • If the OTF corrections are right, GaN growth simulations will predict fewer and larger islands with more elongated perimeters under Ga-rich conditions, and more extended AdGa accumulation when ES barriers are active.
  • The framework can model realistic MBE shutter sequences by explicitly simulating growth interruptions, capturing Ostwald ripening and post-deposition relaxation that continuous-flux models omit.
  • At high temperature, desorption-driven island walking emerges as a distinct morphological regime, with island translation rates set by the local balance of N–Ga exchange and N2 desorption.
  • Because the fingerprint and OTF machinery are species-generic, the approach should transfer to other compound semiconductors and nonequilibrium growth problems beyond GaN.

Reading between the lines

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

  • The BEP sensitivity analysis shows that the predicted barrier corrections vary by 0.086–0.14 eV across α = 0.25–0.75, which is the same order as the corrections themselves; direct NEB validation on the novel step-edge configurations would sharpen the quantitative claims.
  • The caching strategy effectively makes the framework a self-learning KMC for compound semiconductors, where the barrier database grows as the simulation roughens—this suggests a path toward predictive growth simulation without exhaustive precomputation.
  • An experimentally testable extension is to compare the predicted island density and size distribution against in situ STM or post-growth AFM on GaN(0001) grown under slightly Ga-rich conditions; the framework predicts measurably larger, more elongated islands than a nearest-neighbor-only model.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 a lattice-based kinetic Monte Carlo framework for simulating GaN(0001) MBE growth. The framework operates in two modes: a predefined-barrier mode using a DFT-derived coordination catalog, and an on-the-fly (OTF) mode that computes activation barriers for local environments absent from the catalog using a Brønsted–Evans–Polanyi (BEP) relation with CHGNet reaction energies. Simulations with predefined barriers reproduce compact triangular islands, Ostwald ripening, Ehrlich–Schwoebel barrier-induced multilayer nucleation, and, at high temperature with desorption, an 'island walking' regime attributed to N–Ga exchange and preferential AdGa desorption. OTF simulations produce slightly fewer, larger islands and more extended AdGa patches than the predefined-barrier runs. The paper claims the framework is a flexible platform for predictive atomic-scale simulations of GaN epitaxy.

Significance. If the OTF barrier evaluation were properly validated, the framework would be a valuable contribution: it extends environment-aware KMC to a compound semiconductor, introduces a configurable descriptor that resolves step-edge vs. terrace environments, and demonstrates multiphenomenon morphology in a single code. The caching database and MPI-parallel architecture are practical strengths. However, the current manuscript is essentially a method demonstration with qualitative morphological outputs. The central novelty—adaptive OTF barrier evaluation—rests on a BEP correction that is not tested against DFT/NEB for the specific environments that drive the reported morphology changes. The established qualitative behaviors (triangular islands, ripening, ES mounds) are consistent with prior work and are not in themselves evidence of predictive power. The framework has potential, but the quantitative validation needed to support the 'predictive' claim is missing.

major comments (3)
  1. [§3.2, SI S1, Eq. (BEP relation)] The OTF barrier is computed as E_a = E0 + αΔE with E0 taken from the same nn1 catalog and α=0.5 assumed. The only consistency check reported—that novel barriers with ΔE≈0 coincide with E0—is tautological because the BEP form forces this limit. The three configurations with nonzero ΔE (Ga2N +0.14 eV, Ga2AdGa −0.09 eV, Ga2N,1AdGa +0.11 eV) are exactly where the assumed α and CHGNet's unquantified surface error matter. SI Table S1 shows that varying α∈[0.25,0.75] changes these barriers by 0.086–0.14 eV, the same order as the claimed corrections. CHGNet is bulk-trained with no demonstrated accuracy for GaN(0001) adatoms, step edges, or N vacancies. Without DFT/NEB reference calculations for at least these configurations, the OTF corrections and the resulting morphological differences in Fig. 9 are unsupported. This undermines the central claim that OTF barriers capture environment-dependent
  2. [§3.1–3.2, Conclusions] The paper repeatedly describes the framework as 'predictive' and concludes that predicted morphologies are 'directly testable by in situ STM or post-growth AFM,' but no quantitative validation against experiments or against an independent benchmark is provided. All comparisons are qualitative (e.g., triangular islands 'consistent with' STM observations; ripening 'characteristic of' Ostwald ripening). There are no comparisons of island densities, island size distributions, or temperature-dependent scaling laws. Given the large number of hand-set parameters (ES barriers, desorption bases and increments, N2 desorption barrier, flux ratio, α), the absence of any quantitative benchmark makes it impossible to assess whether the model's predictions are accurate or merely plausible. At minimum, one quantitative comparison—e.g., island density vs. temperature or a known scaling law—is needed to s
  3. [§3.1.4, §2 (desorption and N2 model)] The N2 associative desorption rule ('whenever two adjacent surface N atoms are both free of Ga coverage above them, fixed 1.5 eV barrier') and the coordination-dependent desorption barrier hierarchy (N>Ga>AdGa with fixed per-neighbor increments) are introduced as model assumptions. If the 'island walking' regime and the claimed stoichiometric drain are offered as physical predictions, these assumptions need at least a sensitivity analysis or a comparison to dedicated experimental/DFT data. Without this, the walking regime should be framed as a model-dependent scenario rather than a validated prediction. This is especially relevant because the paper's abstract highlights the walking regime as one of its phenomena.
minor comments (5)
  1. [§3.2] Typo in 'Brønsted–Evansf–Polanyi'; should read 'Brønsted–Evans–Polanyi'.
  2. [§3.2, Fig. 8 caption] The phrase 'their coincidence with predefined barriers for most event types is a physical consequence of near-zero BEP reaction energies (ΔE≈0)' is misleading. This is a mathematical consequence of the BEP form, not a physical validation of the barrier catalog; please rephrase to avoid overclaiming.
  3. [Abstract, §2] The abstract and methods state that barriers are computed via NEB, PES, or BEP, but the simulations reported here use only the BEP route. Please clarify that the other methods are available but not exercised in this work.
  4. [§3.1.3] The ES barrier values (0.15/0.25 eV) are described as 'conservative lower bounds relative to previously reported values', but the references (7,30,31) are not specific about the ranges. Please give the literature values explicitly so the reader can see the factor being used.
  5. [SI S2] The configuration-space analysis is useful, but the tables would benefit from a short explanation of why the 'nn1 only' counts differ between Tables S2 (20) and S3 (29) and how the predefined catalog of 21 barriers arises from 20 fingerprints.

Circularity Check

1 steps flagged · score 2.0 of 10

One tautological consistency check; no load-bearing circularity

  1. self definitional [§3.2, Figure 8 caption and surrounding main text]
    "For most events, the novel barriers cluster within ±0.01 eV of their predefined counterparts, confirming the accuracy of the nn1 catalog on flat terraces. ... Their coincidence with predefined barriers for most event types is a physical consequence of near-zero BEP reaction energies (∆E≈0) for symmetric hop geometries, which maps the novel barrier back to E0 regardless of nn2 coordination."

    Under the paper's BEP relation Ea = E0 + αΔE, setting ΔE≈0 gives Ea=E0 identically. The observed clustering of OTF barriers around E0 is therefore not an empirical confirmation of the catalog; it is a mathematical consequence of the evaluation formula. The main-text sentence 'confirming the accuracy' uses a construction-forced coincidence as evidence for the input catalog. The nonzero corrections (Ga2N, etc.) are not forced this way—they depend on CHGNet ΔE and the assumed α=0.5—so the circularity is confined to the consistency-check statement and is not load-bearing for the claimed morphology differences.

full rationale

The central OTF claims are not circular by construction. E0 and α are prescribed rather than fitted to the target morphologies, and ΔE is obtained from an external pretrained MLIP (CHGNet). The +0.14 eV Ga2N correction and the resulting differences in island size and AdGa patches are genuine model outputs, not re-statements of the input catalog. The paper's lack of DFT/NEB validation for the specific CHGNet reaction energies is a correctness/robustness concern, not a circularity. The only identifiable circular step is the claim that novel barriers clustering within ±0.01 eV of E0 'confirms the accuracy' of the nn1 catalog: this is forced by the BEP formula whenever ΔE≈0, as the paper's own Figure 8 caption acknowledges. That step is descriptive and not load-bearing for the main conclusions, which rest on the non-zero CHGNet-derived corrections. Accordingly, the paper has one minor construction-forced consistency statement, but no significant circularity in the derivation chain.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

No new physical entities are invented: AdGa is an established surface species in the GaN literature. The central claims rest instead on hand-set kinetic parameters (desorption, ES barriers, alpha) and on the assumed validity of the CHGNet/BEP pipeline for novel step-edge environments.

free parameters (6)
  • Ehrlich-Schwoebel barriers = E_Ga_ES=0.15 eV, E_AdGa_ES=0.15 eV, E_N_ES=0.25 eV
    Hand-chosen as 'conservative lower bounds' in §3.1.3; directly control the multilayer nucleation claims.
  • Desorption barrier bases and neighbor increments = Ga: 2.0 eV base, +0.5 eV/N, +0.2 eV/AdGa; N: 2.5 eV base, +0.6 eV/Ga; AdGa: 1.5 eV base, +0.4 eV/Ga
    Set in §3.1.4 to produce the stability hierarchy N > Ga > AdGa; these parameters drive the island-walking regime.
  • N2 associative desorption barrier = 1.5 eV, 1.0 eV below isolated-N barrier
    Activated whenever two adjacent surface N atoms are both free of overlying Ga; an ad hoc modeling choice in §3.1.4.
  • BEP transfer coefficient alpha = 0.5
    Assumed near-symmetric transition state in §3.2; sensitivity analysis in SI S1 shows 0.086-0.14 eV variation across alpha in 0.25-0.75.
  • Flux ratio = 0.535 (Ga-rich)
    Representative simulation condition in §3; not varied systematically.
  • Deposition flux = 0.2 ML/s
    Simulation condition in §3; affects all rate comparisons.
assumptions (6)
  • domain assumption The predefined DFT barrier catalogue of Chugh and Ranganathan (Ref. 11) is accurate for GaN(0001) terrace diffusion, attachment, and as E0 in the BEP relation.
    Used as the baseline for all predefined-mode results and as the reference barrier in OTF BEP; the paper does not revalidate it.
  • domain assumption CHGNet, a pretrained universal MLIP, gives reliable relaxed geometries and reaction energies for GaN(0001) step-edge configurations.
    All OTF ΔE values come from CHGNet without DFT/NEB benchmark for the specific novel environments.
  • domain assumption The linear BEP relation E_a = E0 + αΔE holds for these surface hops with a single transfer coefficient α = 0.5.
    Invoked in §3.2 for all novel configurations; sensitivity analysis shows 0.1 eV-scale dependence on α.
  • domain assumption A lattice KMC model with count-based nn1/nn2 fingerprints captures the essential physics of GaN MBE growth.
    The framework excludes off-lattice, long-range strain, electrostatic, and explicit defect-formation degrees of freedom; assumed adequate in §2.
  • ad hoc to paper N2 associative desorption occurs whenever two adjacent surface N atoms are free of overlying Ga, with a fixed 1.5 eV barrier.
    Introduced in §3.1.4 as a model rule, justified only qualitatively by the experimentally known dominance of associative desorption.
  • ad hoc to paper The hand-set desorption/ES barrier hierarchy quantitatively represents GaN surface kinetics.
    These values are not fitted to experimental desorption rates or validated against DFT; they are chosen to illustrate regimes.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A flexible kinetic Monte Carlo framework for GaN molecular beam epitaxy with adaptive on-the-fly barrier evaluation." pith.science (2026). https://pith.science/paper/TVI6KFHH

@misc{pith2026260724871,
  author       = {Pith},
  title        = {Pith review of: A flexible kinetic Monte Carlo framework for GaN molecular beam epitaxy with adaptive on-the-fly barrier evaluation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TVI6KFHH}},
  note         = {Machine review of arXiv:2607.24871}
}
read the original abstract

We present a lattice-based kinetic Monte Carlo (KMC) framework for simulating GaN(0001) growth by molecular beam epitaxy. The framework captures the key microscopic processes governing epitaxial growth, including temperature-dependent surface diffusion, flux-driven deposition, Ehrlich--Schwoebel (ES) step-edge barriers, Ostwald ripening, and species-specific desorption, within a scalable architecture that enables systematic exploration of experimentally relevant growth conditions. In addition to predefined activation-energy catalogs, the framework supports adaptive on-the-fly barrier evaluation using machine-learned interatomic potentials. When previously unencountered local atomic configurations arise, activation barriers are computed via nudged elastic band, potential energy scans, or Br{\o}nsted--Evans--Polanyi methods, and cached for reuse. Predefined-barrier simulations reproduce compact triangular island formation, and further capture Ostwald ripening during growth interruptions and ES barrier-induced multilayer nucleation. At elevated temperatures, desorption drives an island ``walking'' regime, in which N--Ga exchange generates weakly bound Ga adatoms (AdGa) at trailing edges; preferential desorption of AdGa leads to asymmetric edge retreat and net island translation. Our KMC framework provides a flexible platform for predictive simulations of GaN epitaxy at the atomic scale and, more broadly, non-equilibrium growth of compound semiconductors.

Figures

Figures reproduced from arXiv: 2607.24871 by the authors.

Figure 1
Figure 1. FIG. 1. Schematic comparison of kinetic Monte Carlo (KMC) workflows. (a) Conventional lattice KMC using predefined or tabulated [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Final snapshot of the GaN(0001) surface morphology from [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. GaN(0001) surface morphology at the end of deposition and [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Surface morphologies obtained with the ES barrier activated. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Mean island drift from nucleation site versus deposition cy [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. GaN(0001) surface morphology after 2450 deposition cycles [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. Schematic top view of the GaN(0001) wurtzite surface show [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8. Complete predefined barrier catalog (21 events, light-blue bars) taken from the DFT-based tabulation of Chugh [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: FIG. 9. OTF-KMC surface morphologies of GaN(0001) at 0 [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 1 canonical work pages

  1. [1]

    author author T. D. \ Moustakas \ and\ author R. Paiella ,\ @noop journal journal Rep. Progr. Phys. \ volume 80 ,\ pages 106501 ( year 2017 ) NoStop

  2. [2]

    Nicholls , author L

    author author J. Nicholls , author L. Anderson , author W. Lee , author J. J. S. \ Ahn , author A. Baskaran , author H. Bang , author M. Belloeil , author Y. Cai , author J. Campbell , author J. Chai , et al. ,\ @noop journal journal Appl. Phys. Lett. \ volume 123 ( year 2023 ) NoStop

  3. [3]

    Piva , author C

    author author F. Piva , author C. De Santi , author M. Deki , author M. Kushimoto , author H. Amano , author H. Tomozawa , author N. Shibata , author G. Meneghesso , author E. Zanoni ,\ and\ author M. Meneghini ,\ @noop journal journal Photonics Res. \ volume 8 ,\ pages 1786 ( year 2020 ) NoStop

  4. [4]

    author author A. F. \ Voter ,\ @noop journal journal Radiation Effects in Solids \ ,\ pages 1 ( year 2007 ) NoStop

  5. [5]

    author author K. A. \ Fichthorn \ and\ author W. H. \ Weinberg ,\ https://doi.org/10.1063/1.461138 journal journal J. Chem. Phys. \ volume 95 ,\ pages 1090 ( year 1991 ) NoStop

  6. [6]

    Evans , author P

    author author J. Evans , author P. Thiel ,\ and\ author M. C. \ Bartelt ,\ @noop journal journal Surf. Sci. Rep. \ volume 61 ,\ pages 1 ( year 2006 ) NoStop

  7. [7]

    author author N. A. \ Kaufmann , author L. Lahourcade , author B. Hourahine , author D. Martin ,\ and\ author N. Grandjean ,\ @noop journal journal J. Cryst. Growth \ volume 433 ,\ pages 36 ( year 2016 ) NoStop

  8. [8]

    Bruno , author M

    author author G. Bruno , author M. Losurdo , author T.-H. \ Kim ,\ and\ author A. Brown ,\ @noop journal journal Phys. Rev. B \ volume 82 ,\ pages 075326 ( year 2010 ) NoStop

Show all 32 references
  1. [9]

    Lymperakis , author K

    author author L. Lymperakis , author K. Lymperakis ,\ and\ author E. Iliopoulos ,\ @noop journal journal J. Appl. Phys. \ volume 136 ( year 2024 ) NoStop

  2. [10]

    author author P. W. \ Voorhees ,\ @noop journal journal J. Stat. Phys. \ volume 38 ,\ pages 231 ( year 1985 ) NoStop

  3. [11]

    Chugh \ and\ author M

    author author M. Chugh \ and\ author M. Ranganathan ,\ @noop journal journal Appl. Surf. Sci. \ volume 422 ,\ pages 1120 ( year 2017 ) NoStop

  4. [12]

    Ranganathan ,\ @noop journal journal Phys

    author author Razia \ and\ author M. Ranganathan ,\ @noop journal journal Phys. Chem. Chem. Phys. \ volume 27 ,\ pages 5317 ( year 2025 ) NoStop

  5. [13]

    Yokaichiya , author T

    author author T. Yokaichiya , author T. Ikeda , author K. Muraoka ,\ and\ author A. Nakayama ,\ @noop journal journal J. Chem. Phys. \ volume 160 ( year 2024 ) NoStop

  6. [14]

    Thomas , author R

    author author G. Thomas , author R. A. \ Ferreyra ,\ and\ author M. A. \ Quiroga ,\ https://doi.org/10.1016/j.apsusc.2024.160558 journal journal Appl. Surf. Sci. \ volume 670 ,\ pages 160558 ( year 2024 ) NoStop

  7. [15]

    author author R. A. \ Ferreyra \ and\ author M. A. \ Quiroga ,\ https://doi.org/10.1016/j.apsusc.2021.149132 journal journal Appl. Surf. Sci. \ volume 546 ,\ pages 149132 ( year 2021 ) NoStop

  8. [16]

    Thomas , author R

    author author G. Thomas , author R. A. \ Ferreyra ,\ and\ author M. A. \ Quiroga ,\ https://doi.org/10.1016/j.apsusc.2025.166361 journal journal Appl. Surf. Sci. \ volume 730 ,\ pages 166361 ( year 2025 ) NoStop

  9. [17]

    Trushin , author A

    author author O. Trushin , author A. Karim , author A. Kara ,\ and\ author T. S. \ Rahman ,\ https://doi.org/10.1103/PhysRevB.72.115401 journal journal Phys. Rev. B \ volume 72 ,\ pages 115401 ( year 2005 ) NoStop

  10. [18]

    author author R. A. \ Van Santen , author M. Neurock ,\ and\ author S. G. \ Shetty ,\ @noop journal journal Chem. Rev. \ volume 110 ,\ pages 2005 ( year 2009 ) NoStop

  11. [19]

    Henkelman \ and\ author H

    author author G. Henkelman \ and\ author H. J \'o nsson ,\ @noop journal journal J. Chem. Phys. \ volume 113 ,\ pages 9978 ( year 2000 ) NoStop

  12. [20]

    author author B. D. \ Lubachevsky ,\ @noop journal journal Complex Syst. \ volume 1 ,\ pages 1099 ( year 1988 ) NoStop

  13. [21]

    Shim \ and\ author J

    author author Y. Shim \ and\ author J. G. \ Amar ,\ https://doi.org/10.1103/PhysRevB.71.125432 journal journal Phys. Rev. B \ volume 71 ,\ pages 125432 ( year 2005 ) NoStop

  14. [22]

    Deng , author P

    author author B. Deng , author P. Zhong , author K. Jun , author J. Riebesell , author K. Han , author C. J. \ Bartel ,\ and\ author G. Ceder ,\ @noop journal journal Nat. Mach. Intell. \ volume 5 ,\ pages 1031 ( year 2023 ) NoStop

  15. [23]

    Batatia , author D

    author author I. Batatia , author D. P. \ Kovacs , author G. Simm , author C. Ortner ,\ and\ author G. Cs \'a nyi ,\ @noop journal journal Advances in neural information processing systems \ volume 35 ,\ pages 11423 ( year 2022 ) NoStop

  16. [24]

    Chen \ and\ author S

    author author C. Chen \ and\ author S. P. \ Ong ,\ @noop journal journal Nat. Computat. Sci. \ volume 2 ,\ pages 718 ( year 2022 ) NoStop

  17. [25]

    Takamoto , author C

    author author S. Takamoto , author C. Shinagawa , author D. Motoki , author K. Nakago , author W. Li , author I. Kurata , author T. Watanabe , author Y. Yayama , author H. Iriguchi , author Y. Asano , et al. ,\ @noop journal journal Nat. Commun. \ volume 13 ,\ pages 2991 ( yea...

  18. [26]

    Batzner , author A

    author author S. Batzner , author A. Musaelian , author L. Sun , author M. Geiger , author J. P. \ Mailoa , author M. Kornbluth , author N. Molinari , author T. E. \ Smidt ,\ and\ author B. Kozinsky ,\ @noop journal journal Nat. Commun. \ volume 13 ,\ pages 2453 ( year 2022 ) NoStop

  19. [27]

    Drautz ,\ @noop journal journal Phys

    author author R. Drautz ,\ @noop journal journal Phys. Rev. B \ volume 99 ,\ pages 014104 ( year 2019 ) NoStop

  20. [28]

    author author A. P. \ Thompson , author L. P. \ Swiler , author C. R. \ Trott , author S. M. \ Foiles ,\ and\ author G. J. \ Tucker ,\ @noop journal journal J. Comput. Phys. \ volume 285 ,\ pages 316 ( year 2015 ) NoStop

  21. [29]

    Xie , author S

    author author M. Xie , author S. Seutter , author W. Zhu , author L. Zheng , author H. Wu ,\ and\ author S. Tong ,\ @noop journal journal Phys. Rev. Lett. \ volume 82 ,\ pages 2749 ( year 1999 ) NoStop

  22. [30]

    Zywietz , author J

    author author T. Zywietz , author J. Neugebauer ,\ and\ author M. Scheffler ,\ @noop journal journal Appl. Phys. Lett. \ volume 73 ,\ pages 487 ( year 1998 ) NoStop

  23. [31]

    Turski , author F

    author author H. Turski , author F. Krzy \.z ewski , author A. Feduniewicz- \.Z muda , author P. Wolny , author M. Siekacz , author G. Muziol , author C. Cheze , author K. Nowakowski-Szukudlarek , author H. G. \ Xing , author D. Jena , et al. ,\ @noop journal journal Appl. Sur...

  24. [32]

    author author S. Y. \ Karpov , author R. Talalaev , author Y. N. \ Makarov , author N. Grandjean , author J. Massies ,\ and\ author B. Damilano ,\ @noop journal journal Surf. Sci. \ volume 450 ,\ pages 191 ( year 2000 ) NoStop

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.