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

Ali Hamedani; Flyura Djurabekova; Jesper Byggm\"astar; Kai Nordlund

REVIEW 1 major objections 6 minor 89 references

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

T0 review · 1 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read A machine-learned Gaussian Approximation Potential for tungsten reproduces bulk, surface, liquid, and defect-cluster properties at near-DFT accuracy, with a re-fitted screened-Coulomb short-range repulsion for collision cascades.

desk verdict A radiation-specific tungsten GAP with a well-designed training set and mostly honest validation; the one real soft spot is the unvalidated sub-1.1 Å many-body repulsion that governs high-energy cascade cores. read the letter →

arxiv 1908.07330 v2 pith:W4GWJS3R submitted 2019-08-20 physics.comp-ph cond-mat.mtrl-sci

Jesper Byggmästar , Ali Hamedani , Kai Nordlund , Flyura Djurabekova This is my paper · ORCID

classification physics.comp-phcond-mat.mtrl-sci

keywords tungstenmachine-learninginteratomicpotentialGaussianApproximationradiationdamagecollisioncascadesself-interstitialclustersvacancyfusionmaterials

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

Collision cascades in tungsten, the leading candidate for fusion-reactor armor, are normally simulated with analytical interatomic potentials whose fixed forms cannot simultaneously get short-range repulsion, liquid behavior, surface energies, and defect-cluster stabilities right. This paper introduces a machine-learned potential for tungsten, built in the Gaussian Approximation Potential framework, and trains it specifically so that all four of these properties are captured. The central claim is that the resulting potential reproduces molecular dynamics of radiation damage with essentially density-functional-theory accuracy, including the correct stability of self-interstitial clusters and the peculiar repulsive binding of the second-nearest-neighbor di-vacancy. If this holds, cascade simulations and defect-evolution studies can move from a regime limited by potential errors to one limited by the underlying DFT data.

What carries the argument

The total energy is a sum of an external pair potential and two Gaussian-process regression terms. The pair potential has the screened-Coulomb form with a screening function refitted to all-electron DFT data; it dominates below about 2.2 Å and fully controls dynamics below 1.1 Å, where the GAP is trained to contribute almost nothing. The machine-learned part uses a two-body squared-exponential kernel on interatomic distance for bond energies and the SOAP (Smooth Overlap of Atomic Positions) kernel, which compares atomic environments by the overlap of their smeared atomic densities, for many-body effects. Around 40,000 local environments from DFT make up the training set, deliberately including liquids, damaged surfaces, di-vacancy and di-self-interstitial structures, and short-range displaced-atom configurations.

What would settle it

Recompute the W-W dimer repulsion and several static displacement paths below 1.1 Å with an independent all-electron DFT implementation; if the forces differ from the fitted pair potential by more than a few percent in the 100 eV range, the claimed DFT-level short-range cascade dynamics would not hold.

Watch

Extended reading notes

Core claim

The paper claims, in its own words, that the potential enables molecular dynamics simulations of radiation damage in tungsten with unprecedented accuracy and captures a variety of tungsten properties with essentially DFT accuracy. The potential reproduces bulk elastic constants and phonons, melting near the DFT-based estimate, surface energies within about 5 meV/Å² of DFT including a (100) surface reconstruction, vacancy-cluster binding trends, and self-interstitial cluster formation energies aligned with DFT data. Its short-range many-body repulsion is set by an external screened-Coulomb pair potential fitted to all-electron DFT data, so the machine-learned part handles near-equilibrium energetics while the pair potential governs high-energy collisions.

Load-bearing premise

The load-bearing premise is that the re-fitted screened-Coulomb pair potential, built from a single set of all-electron DFT calculations, correctly describes W-W repulsion below about 1.1 Å, because the machine-learned part is deliberately trained to contribute almost nothing at those separations.

Editorial extensions

If this is right

Cascade simulations with this potential should produce primary damage statistics, such as Frenkel-pair production and cluster-size distributions, much closer to DFT than standard embedded-atom or bond-order potentials.
The long-standing disagreements among tungsten potentials over the relative stability of 1/2⟨111⟩ versus ⟨100⟩ dislocation loops can be revisited with a potential that matches DFT across the cluster-size range.
Surface-irradiation studies become feasible at near-DFT fidelity, since the potential reproduces surface energies, relaxations, and the (100) zigzag reconstruction even though that reconstruction was not explicitly in the training set.
The authors note that adatom migration barriers are underestimated by 20-30%, so quantitative surface-diffusion studies would require extending the training database with adatom structures.
The same training structures and fitting strategy can be reused for other non-magnetic bcc metals and as a base for tungsten-alloy potentials.

Reading between the lines

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

A generic recipe emerges for applying machine-learned potentials to radiation damage: pair a re-fitted screened-Coulomb repulsion with a machine-learning term forced to zero at short distances; the switching scheme sketched in the appendix is a template wherever the descriptor becomes numerically unstable.
If the accuracy claim holds, the practical bottleneck for cascade statistics shifts from potential fidelity to computational cost; the paper quotes a 2-3 order-of-magnitude slowdown, so large-scale studies will likely need optimized kernels or hybrid schemes that combine the GAP with cheaper potentials.
The 20-30% underestimation of adatom migration barriers is a concrete, testable gap: adding adatom configurations to the training database should bring surface-diffusion kinetics to the same accuracy as bulk and defect properties.
The same architecture should transfer to other non-magnetic bcc metals and to tungsten-hydrogen or tungsten-helium systems; training analogous databases and checking whether the defect-stability improvements survive with a second element would be a direct extension.

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.

Desk Editor's Note

read the letter

You should know two things about this paper. First, it delivers a new Gaussian Approximation Potential for tungsten aimed specifically at radiation damage, and the genuinely new content is the training database: SIAs, liquid phases, short-range displaced-atom configurations, plus a refitted screened Coulomb pair potential for the close-encounter regime. The prior tungsten GAP from Szlachta et al. did not have these, so the paper fills a real need. Second, the validation is more honest than most in this genre — the authors repeatedly point out where their potential misses, and they include genuine out-of-sample checks.

What it does well: the training strategy is sensible and clearly described. The out-of-sample results are convincing for surfaces (high-index and reconstructed), SIA clusters of size 3 and above, di-vacancy binding at 3–5NN, liquid k-fold cross-validation, and static displacement paths. The TDE map is a nice dynamic test. The authors ship the potential files and training database, which makes the work reproducible. They also flag the known discrepancies themselves: the systematic ~0.1 eV SIA formation energy offset and the 20–30% underestimate of adatom migration barriers. That is the right way to report a fitted potential.

The soft spots are real but not fatal. The stress-test note is fair: below roughly 1.1 Å the GAP is trained to contribute almost nothing, so the entire interaction is the fitted screened Coulomb pair potential. That region is not directly validated in many-body environments — the dimer data and static displacement paths only reach down to about 1.1 Å, and a 10 keV recoil can push separations below that. An all-electron displaced-atom check would settle this. It is a genuine gap, but the paper is transparent about it, and the accessible evidence (dimer fit, TDE agreement, short-range displacement paths) is consistent with the potential being adequate. The claim of "unprecedented accuracy" is based on property benchmarks rather than direct cascade defect statistics, and the 2–3 order-of-magnitude computational cost is a practical limitation, but neither undermines the core contribution.

Bottom line: this paper deserves a serious referee. It is a solid, reproducible construction of a useful potential for radiation damage simulations in tungsten, with honest validation and clear limitations. I would send it to review and likely accept after minor revisions — mainly asking for a more prominent caveat about the sub-1.1 Å regime and, if feasible, one additional check of many-body repulsion at those distances.

Referee Report

1 major / 6 minor

Summary. The paper introduces a Gaussian Approximation Potential (GAP) for tungsten aimed at radiation-damage simulations. The potential combines two-body and SOAP descriptors with an external screened-Coulomb pair potential fitted to all-electron DFT dimer data, so that the GAP controls equilibrium and near-equilibrium behavior while the pair potential dominates below about 1.1 Å. The training database includes distorted crystals, vacancies, self-interstitial atoms and clusters, surfaces, liquids, and short-range displaced-atom configurations. Validation covers bulk properties, phonons, thermal expansion, melting, surface energies and reconstructions, short-range displacement paths, threshold displacement energies, SIA and vacancy cluster energetics, and di-vacancy binding. The authors conclude that the potential captures a variety of tungsten properties with near-DFT accuracy and enables more reliable molecular dynamics simulations of radiation damage.

Significance. If the potential performs as claimed, it addresses long-standing deficiencies of analytical potentials in tungsten (surface energies, SIA cluster stability, di-vacancy binding) and is a useful community resource for fusion-relevant radiation-damage simulations. The paper's strengths include an extensive validation set with many out-of-sample tests (untrained high-index surfaces, SIA clusters of size 3 and above, 3-5 nearest-neighbor di-vacancies, volume-conserving deformation paths, and a threshold-displacement-energy map), an honest discussion of known discrepancies (adatom migration barriers, single-SIA formation energies), five-fold cross-validation of the liquid data, and public availability of the potential files and training database. The main residual risk is the unvalidated sub-1.1 Å many-body repulsion, which is load-bearing for high-energy cascade cores; if that region is as reliable as the dimer fit suggests, the central claim is credible.

major comments (1)

[Sec. IV and V.C] The short-range claim in Sec. V.C that "the GAP reproduces any short-range forces and energies encountered in cascade simulations with DFT accuracy" is stronger than the evidence presented. Because the GAP is deliberately trained to contribute negligibly below about 1.1 Å (Sec. IV), Eqs. (3)-(5) are the entire interaction in the close-encounter regime, and the validation in Fig. 9 and the TDE map in Fig. 10 sample only down to about 1.1 Å / 100-200 eV. A 10 keV recoil can easily produce separations below 1.1 Å, so an error in Vpair would feed directly into the cascade core. The all-electron dimer data from Ref. [38] constrain only the pairwise repulsion curve, not a displaced atom surrounded by neighbors. Please add a direct all-electron (or equivalent) test of a short displacement path in bulk tungsten with nearest-neighbor distances in the 0.6-1.1 Å range, or explicitly restrict the abstract and conclusion claims to the validated range.

minor comments (6)

[Table II and Sec. V.A] Table II reports 46 liquid structures in the training database, while Sec. V.A states that the k-fold cross-validation splits "the 45 liquid structures" into five subsamples; please reconcile the count.
[Sec. II, Fig. 2, Ref. [37]] There are several typographical issues: "The total energy of an atomi" should be "atom"; the vertical axis label in Fig. 2 appears as "V olume" with an extra space; and "Ziegler-Biersack-Littmarck" should spell "Littmark" (also in Ref. [37]).
[Fig. 3(c)] The legend entry "BCT BCC BCCGAP DFT" lacks a space between "BCC" and "GAP" and should be cleaned up.
[Sec. V.B] The adatom migration barriers are underestimated by 20-30% (e.g., 0.6 eV versus 0.87 eV for the main hop on (110)); since the abstract emphasizes surface properties, the text should state explicitly that the quantitative accuracy for surface transport is not at the same level as the surface energies.
[Sec. V.E and Fig. 15] The vacancy-loop-to-planar-void crossover for 1/2<111> clusters is reported at about 25 vacancies, while the comparison DFT value from Ref. [87] is about 45 vacancies; calling this "roughly consistent" needs justification, especially because Ref. [87] is a private communication.
[Sec. VI] The phrase "unprecedented accuracy" would be more convincing if accompanied by a quantitative comparison with at least one widely used analytical potential on the same validation set, rather than only qualitative statements.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the central radiation-damage claim rests on out-of-sample validation against independent DFT data and experiments.

full rationale

This paper is a machine-learning interatomic potential fitted to a DFT training database, and its validation strategy is transparent about which quantities are in-sample. Section V A explicitly says that bulk properties are 'well-represented by the training database, and therefore in close agreement with DFT', so the paper does not disguise fitted quantities as predictions. The radiation-damage claim is supported by genuinely out-of-sample tests: deformation paths beyond the training strains (Fig. 3), phonon dispersion compared with experiments and independent DFT (Fig. 4), surface energies for surfaces not in the training database (Fig. 7), self-interstitial cluster sizes 3 and larger (Fig. 13 caption: 'only sizes 1 and 2 were fit, so all the other data points serve as tests'), di-vacancy separations beyond 2NN (Fig. 14), and threshold displacement energies compared with experimental values (Fig. 10). The short-range repulsion is fitted to all-electron DFT-DMol data from Ref. [38], which is input data rather than a claimed prediction, and Appendix A explicitly checks that the GAP contribution remains negligible at short distances. Self-citations such as Refs. [38], [39], and [54] provide independent computational data, experimental validation, or prior potential versions, but the central argument does not reduce to an unverified self-citation and no uniqueness claim is imported from the authors' own prior work. The derivation chain is therefore self-contained against external benchmarks, and the honest finding is no significant circularity.

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

The central claim is an empirical fit: a machine-learned potential with fitted coefficients, hand-chosen hyperparameters, a fitted repulsive curve, and a representative training set. No new physical entities are postulated, but the accuracy claim rests on the completeness and fidelity of the DFT reference data and on the representativeness of the training structures.

free parameters (5)

GAP regression coefficients alpha_s (4020 sparse basis functions) = Determined by regularized least squares from DFT training data
The potential itself is a machine-learned regression; these coefficients are fitted to the training energies, forces, and virials, so the model's predictions inherit their values.
Screened Coulomb screening function coefficients = 0.32825 exp(-2.54931x)+0.09219 exp(-0.29182x)+0.58110 exp(-0.59231x)
Fitted to all-electron DFT-DMol W-W repulsion data to control short-range cascade dynamics; GAP contributes little at these distances by design.
Pair potential crossover radii r1 and r2 = r1=1 Å, r2=2.2 Å
Chosen so the screened Coulomb smoothly joins the machine-learned GAP near equilibrium; sets where short-range repulsion begins to dominate.
GAP hyperparameters (Rcut, Rdelta_cut, delta, Nsparse, nmax, lmax, sigma_atom, zeta) = Tab. I values, e.g. Rcut=5 Å, delta_SOAP=2 eV, delta_2b=10 eV
Selected by convergence tests and experience; they control descriptor resolution, sparsity, and the trade-off between smoothness and accuracy.
Regularization weights sigma_nu per structure type = Varies: 10 meV/atom + 0.4 eV/Å (liquids), 1 meV/atom + 0.04 eV/Å (crystals)
Chosen to reflect expected errors in each training subset; they determine how hard the fit tries to match each structure type.

assumptions (5)

domain assumption PBE-GGA density functional theory (VASP PAW) is an accurate reference for W energetics.
The entire training and validation hierarchy uses VASP-PBE as ground truth; any systematic DFT error, such as the melting point being 3450 K versus 3687 K experimental, is inherited by the potential.
domain assumption All-electron DFT-DMol data (Ref. [38]) correctly describes W-W repulsion below about 1.1 Å, where PAW-DFT diverges.
Invoked in Section II and IV when the screening function is fitted; this curve controls high-energy collision dynamics.
domain assumption SOAP and two-body descriptors with a 5 Å cutoff encode all interactions relevant to tungsten cascades.
The GAP's transferability depends on the descriptor representation; the 5 Å cutoff is chosen in Section II, Table I, and no longer-ranged contributions are modeled.
ad hoc to paper MD snapshots generated with early versions of the GAP are a representative sample of cascade-relevant configurations.
Liquid, disordered surface, and defect structures are generated iteratively using preliminary GAPs (Section IV); if those runs miss important geometries, the final potential inherits the bias.
ad hoc to paper Property-level validation transfers to actual cascade damage statistics.
No full collision cascade is simulated or compared to damage production; the paper assumes that matching DFT on short-range forces, TDEs, defects, and liquid behavior is sufficient.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Machine-learning interatomic potential for radiation damage and defects in tungsten." pith.science (2026). https://pith.science/paper/W4GWJS3R

@misc{pith2026190807330,
  author       = {Pith},
  title        = {Pith review of: Machine-learning interatomic potential for radiation damage and defects in tungsten},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W4GWJS3R}},
  note         = {Machine review of arXiv:1908.07330}
}

read the original abstract

We introduce a machine-learning interatomic potential for tungsten using the Gaussian Approximation Potential framework. We specifically focus on properties relevant for simulations of radiation-induced collision cascades and the damage they produce, including a realistic repulsive potential for the short-range many-body cascade dynamics and a good description of the liquid phase. Furthermore, the potential accurately reproduces surface properties and the energetics of vacancy and self-interstitial clusters, which have been long-standing deficiencies of existing potentials. The potential enables molecular dynamics simulations of radiation damage in tungsten with unprecedented accuracy.

Figures

Figures reproduced from arXiv: 1908.07330 by the authors.

**Figure 2.** FIG. 2: Energy-volume curves of the various crystal [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** FIG. 3: Volume-conserving deformation paths of bulk [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Figures from the paper (10 more)

**Figure 4.** Figure 4: FIG. 4: Phonon dispersion of bcc W as given by the [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 6.** Figure 6: FIG. 6: Elastic constants and bulk modulus of bcc W [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

**Figure 8.** Figure 8: FIG. 8: Reconstruction of the (1 0 0) surface. Top left: [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

**Figure 9.** Figure 9: FIG. 9: Total energy difference (left) and force components (right) for step-wise movement of one atom along [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]

**Figure 10.** Figure 10: FIG. 10: Threshold displacement energies obtained with [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]

**Figure 11.** Figure 11: FIG. 11: Formation energies of self-interstitial atoms in [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]

**Figure 13.** Figure 13: FIG. 13: Formation energies of self-interstitial clusters [PITH_FULL_IMAGE:figures/full_fig_p012_13.png]

**Figure 12.** Figure 12: FIG. 12: Top: Energy difference for a [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]

**Figure 14.** Figure 14: FIG. 14: Binding energies of a di-vacancy in bcc W at [PITH_FULL_IMAGE:figures/full_fig_p013_14.png]

**Figure 16.** Figure 16: FIG. 16: The energies and forces given by the GAP at [PITH_FULL_IMAGE:figures/full_fig_p015_16.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

89 extracted references · 60 canonical work pages

[38]

Nordlund, N

K. Nordlund, N. Runeberg, and D. Sundholm, Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms 132, 45 (1997)

work page 1997
[1]

Rieth, S

M. Rieth, S. L. Dudarev, S. M. Gonzalez de Vicente, J. Aktaa, T. Ahlgren, S. Antusch, D. E. J. Arm- strong, M. Balden, N. Baluc, M. F. Barthe, W. W. Basuki, M. Battabyal, C. S. Becquart, D. Blagoeva, H. Boldyryeva, J. Brinkmann, M. Celino, L. Ciupinski, J. B. Correia, A. De Backer, C. Domain, E. Gaganidze, C. Garc´ ıa-Rosales, J. Gibson, M. R. Gilbert, S....

2013
[2]

Zinkle and L

S. Zinkle and L. Snead, Annu. Rev. Mater. Res. 44, 241 (2014)

2014
[3]

Nordlund, S

K. Nordlund, S. J. Zinkle, A. E. Sand, F. Granberg, R. S. Averback, R. E. Stoller, T. Suzudo, L. Malerba, F. Banhart, W. J. Weber, F. Willaime, S. L. Dudarev, and D. Simeone, Journal of Nuclear Materials 512, 450 (2018)

2018
[4]

A. E. Sand, S. L. Dudarev, and K. Nordlund, EPL 103, 46003 (2013)

2013
[5]

Byggm¨ astar, F

J. Byggm¨ astar, F. Granberg, A. E. Sand, A. Pirttikoski, R. Alexander, M.-C. Marinica, and K. Nordlund, J. Phys.: Condens. Matter 31, 245402 (2019)

2019
[6]

Nordlund, M

K. Nordlund, M. Ghaly, R. S. Averback, M. Caturla, T. Diaz de la Rubia, and J. Tarus, Phys. Rev. B 57, 7556 (1998)

1998
[7]

A. E. Sand, J. Dequeker, C. S. Becquart, C. Domain, and K. Nordlund, Journal of Nuclear Materials 470, 119 (2016)

2016

Show all 89 references

[8]

Byggm¨ astar, F

J. Byggm¨ astar, F. Granberg, and K. Nordlund, Journal of Nuclear Materials 508, 530 (2018)

2018
[9]

R. E. Stoller, A. Tamm, L. K. B´ eland, G. D. Samolyuk, G. M. Stocks, A. Caro, L. V. Slipchenko, Y. N. Osetsky, A. Aabloo, M. Klintenberg, and Y. Wang, J. Chem. Theory Comput. 12, 2871 (2016)

2016
[10]

M. S. Daw and M. I. Baskes, Phys. Rev. B 29, 6443 (1984)

1984
[11]

Tersoﬀ, Phys

J. Tersoﬀ, Phys. Rev. B 37, 6991 (1988)

1988
[12]

G. J. Ackland and R. Thetford, Philos. Mag. A 56, 15 (1987)

1987
[13]

P. M. Derlet, D. Nguyen-Manh, and S. L. Dudarev, Phys. Rev. B 76, 054107 (2007)

2007
[14]

Marinica, L

M.-C. Marinica, L. Ventelon, M. R. Gilbert, L. Proville, S. L. Dudarev, J. Marian, G. Bencteux, and F. Willaime, J. Phys.: Condens. Matter 25, 395502 (2013)

2013
[15]

D. R. Mason, D. Nguyen-Manh, and C. S. Becquart, J. Phys.: Condens. Matter 29, 505501 (2017)

2017
[16]

Chen, Y.-H

Y. Chen, Y.-H. Li, N. Gao, H.-B. Zhou, W. Hu, G.-H. Lu, F. Gao, and H. Deng, Journal of Nuclear Materials 502, 141 (2018)

2018
[17]

Juslin, P

N. Juslin, P. Erhart, P. Tr¨ askelin, J. Nord, K. O. E. Hen- riksson, K. Nordlund, E. Salonen, and K. Albe, J. Appl. Phys. 98, 123520 (2005)

2005
[18]

Ahlgren, K

T. Ahlgren, K. Heinola, N. Juslin, and A. Kuronen, Jour- nal of Applied Physics 107, 033516 (2010)

2010
[19]

Fikar, R

J. Fikar, R. Sch¨ aublin, D. R. Mason, and D. Nguyen- Manh, Nuclear Materials and Energy 16, 60 (2018)

2018
[20]

Fellman, A

A. Fellman, A. E. Sand, J. Byggm¨ astar, and K. Nord- lund, J. Phys.: Condens. Matter 31, 405402 (2019)

2019
[21]

Ventelon, F

L. Ventelon, F. Willaime, C.-C. Fu, M. Heran, and I. Gi- noux, Journal of Nuclear Materials Microstructure Prop- erties of Irradiated Materials, 425, 16 (2012)

2012
[22]

Heinola, F

K. Heinola, F. Djurabekova, and T. Ahlgren, Nucl. Fu- sion 58, 026004 (2018)

2018
[23]

Bonny, D

G. Bonny, D. Terentyev, A. Bakaev, P. Grigorev, and D. Van Neck, Model. Simul. Mater. Sci. Eng. 22, 053001 (2014)

2014
[24]

Behler and M

J. Behler and M. Parrinello, Phys. Rev. Lett. 98 (2007), 10.1103/PhysRevLett.98.146401

2007 doi
[26]

A. P. Thompson, L. P. Swiler, C. R. Trott, S. M. Foiles, and G. J. Tucker, Journal of Computational Physics285, 316 (2015)

2015
[27]

A. V. Shapeev, Multiscale Model. Simul. 14, 1153 (2016), arXiv:1512.06054

2016 arXiv
[28]

Glielmo, C

A. Glielmo, C. Zeni, and A. De Vita, Phys. Rev. B 97 (2018), 10.1103/PhysRevB.97.184307

2018 doi
[29]

Zhang, J

L. Zhang, J. Han, H. Wang, R. Car, and W. E, Phys. Rev. Lett. 120 (2018), 10.1103/PhysRevLett.120.143001

2018 doi
[30]

Behler, J

J. Behler, J. Chem. Phys. 145, 170901 (2016)

2016
[31]

Y. Zuo, C. Chen, X. Li, Z. Deng, Y. Chen, J. Behler, G. Cs´ anyi, A. V. Shapeev, A. P. Thompson, M. A. Wood, and S. P. Ong, ArXiv190608888 Cond-Mat Physicsphysics (2019), arXiv:1906.08888 [cond-mat, physics:physics]

2019 arXiv
[32]

M. A. Caro, Phys. Rev. B 100, 024112 (2019)

2019
[33]

A. P. Bart´ ok, R. Kondor, and G. Cs´ anyi, Phys. Rev. B 87 (2013), 10.1103/PhysRevB.87.184115

2013 doi
[34]

A. P. Bart´ ok and G. Cs´ anyi, Int. J. Quantum Chem.115, 1051 (2015)

2015
[35]

C. E. Rasmussen and C. K. I. Williams, Gaussian Pro- cesses for Machine Learning (The MIT Press, Cam- bridge, Massachusetts, 2006)

2006
[36]

A. P. Bart´ ok, J. Kermode, N. Bernstein, and G. Cs´ anyi, Phys. Rev. X 8 (2018), 10.1103/PhysRevX.8.041048

2018 doi
[37]

J. F. Ziegler, J. P. Biersack, and U. Littmarck, inTreatise on Heavy-Ion Science (Pergamon, New York, 1985) pp. 93–129

1985
[39]

A. N. Zinoviev and K. Nordlund, Nuclear Instruments and Methods in Physics Research Section B: Beam In- teractions with Materials and Atoms The 27th Interna- tional Conference on Atomic Collisions in Solids, 406, 511 (2017)

2017
[40]

Perriot, X

R. Perriot, X. Gu, Y. Lin, V. V. Zhakhovsky, and I. I. Oleynik, Phys. Rev. B 88, 064101 (2013)

2013
[41]

Kresse and J

G. Kresse and J. Hafner, Phys. Rev. B 47, 558 (1993)

1993
[42]

Kresse and J

G. Kresse and J. Hafner, Phys. Rev. B 49, 14251 (1994)

1994
[43]

Kresse and J

G. Kresse and J. Furthm¨ uller, Computational Materials Science 6, 15 (1996)

1996
[44]

Kresse and J

G. Kresse and J. Furthm¨ uller, Phys. Rev. B 54, 11169 (1996)

1996
[45]

J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996)

1996
[46]

P. E. Bl¨ ochl, Phys. Rev. B50, 17953 (1994)

1994
[47]

Kresse and D

G. Kresse and D. Joubert, Phys. Rev. B 59, 1758 (1999)

1999
[48]

H. J. Monkhorst and J. D. Pack, Phys. Rev. B 13, 5188 (1976)

1976
[49]

Methfessel and A

M. Methfessel and A. T. Paxton, Phys. Rev. B 40, 3616 (1989)

1989
[50]

QUIP - QUantum mechanics and Interatomic Po- tentials,

“QUIP - QUantum mechanics and Interatomic Po- tentials,” https://github.com/libAtoms/QUIP, www. libatoms.org. 17
[51]

Plimpton, Journal of Computational Physics 117, 1 (1995), http://lammps.sandia.gov

S. Plimpton, Journal of Computational Physics 117, 1 (1995), http://lammps.sandia.gov

1995
[52]

A. H. Larsen, J. J. Mortensen, J. Blomqvist, I. E. Castelli, R. Christensen, Marcin Du lak, J. Friis, M. N. Groves, B. Hammer, C. Hargus, E. D. Hermes, P. C. Jennings, P. B. Jensen, J. Kermode, J. R. Kitchin, E. L. Kolsbjerg, J. Kubal, Kristen Kaasbjerg, S. Lysgaard, J. B. Mar...

2017
[53]

Togo and I

A. Togo and I. Tanaka, Scripta Materialia 108, 1 (2015)

2015
[54]

W. J. Szlachta, A. P. Bart´ ok, and G. Cs´ anyi, Phys. Rev. B 90, 104108 (2014)

2014
[55]

Marinica, F

M.-C. Marinica, F. Willaime, and J.-P. Crocombette, Phys. Rev. Lett. 108, 025501 (2012)

2012
[56]

D´ ezerald, M

L. D´ ezerald, M. C. Marinica, L. Ventelon, D. Rodney, and F. Willaime, Journal of Nuclear Materials 449, 219 (2014)

2014
[57]

Rumble, ed., CRC Handbook of Chemistry and Physics, 100th Edition , 100th ed

J. Rumble, ed., CRC Handbook of Chemistry and Physics, 100th Edition , 100th ed. (CRC Press, Boca Ra- ton, Fla., 2019)

2019
[58]

Dragoni, T

D. Dragoni, T. D. Daﬀ, G. Cs´ anyi, and N. Marzari, Phys. Rev. Materials 2, 013808 (2018)

2018
[59]

W. R. Tyson and W. A. Miller, Surface Science 62, 267 (1977)

1977
[60]

Rasch, R

K.-D. Rasch, R. W. Siegel, and H. Schultz, Philos. Mag. A 41, 91 (1980)

1980
[61]

Heikinheimo, K

J. Heikinheimo, K. Mizohata, J. R¨ ais¨ anen, T. Ahlgren, P. Jalkanen, A. Lahtinen, N. Catarino, E. Alves, and F. Tuomisto, APL Mater. 7, 021103 (2019)

2019
[62]

Ma and S

P.-W. Ma and S. L. Dudarev, Phys. Rev. Materials 3, 063601 (2019)

2019
[63]

Ma and S

P.-W. Ma and S. L. Dudarev, Phys. Rev. Mater.3 (2019), 10.1103/PhysRevMaterials.3.013605

2019 doi
[64]

Ma and S

P.-W. Ma and S. L. Dudarev, Phys. Rev. Mater.3 (2019), 10.1103/PhysRevMaterials.3.043606

2019 doi
[65]

L. G. Wang, A. van de Walle, and D. Alf` e, Phys. Rev. B 84 (2011), 10.1103/PhysRevB.84.092102

2011 doi
[66]

A. T. Paxton, P. Gumbsch, and M. Methfessel, Philos. Mag. Lett. 63, 267 (1991)

1991
[67]

Paidar, L

V. Paidar, L. G. Wang, M. Sob, and V. Vitek, Model. Simul. Mater. Sci. Eng. 7, 369 (1999)

1999
[68]

Mrovec, R

M. Mrovec, R. Gr¨ oger, A. G. Bailey, D. Nguyen-Manh, C. Els¨ asser, and V. Vitek, Phys. Rev. B 75 (2007), 10.1103/PhysRevB.75.104119

2007 doi
[69]

Ravelo, T

R. Ravelo, T. C. Germann, O. Guerrero, Q. An, and B. L. Holian, Phys. Rev. B 88 (2013), 10.1103/Phys- RevB.88.134101

2013 doi
[70]

Einarsdotter, B

K. Einarsdotter, B. Sadigh, G. Grimvall, and V. Ozoli¸ nˇ s, Phys. Rev. Lett. 79, 2073 (1997)

1997
[71]

Larose and B

A. Larose and B. N. Brockhouse, Can. J. Phys. 54, 1819 (1976)

1976
[72]

G. K. White and M. L. Minges, Int J Thermophys 18, 1269 (1997)

1997
[73]

G. K. White and S. J. Collocott, J. Phys. Chem. Ref. Data 13, 1251 (1984)

1984
[74]

Lowrie and A

R. Lowrie and A. M. Gonas, Journal of Applied Physics 38, 4505 (1967)

1967
[75]

Grizzi, M

O. Grizzi, M. Shi, H. Bu, J. W. Rabalais, and P. Hochmann, Phys. Rev. B 40, 10127 (1989)

1989
[76]

M. K. Debe and D. A. King, Phys. Rev. Lett. 39, 708 (1977)

1977
[77]

M. S. Altman, P. J. Estrup, and I. K. Robinson, Phys. Rev. B 38, 5211 (1988)

1988
[78]

Heinola and T

K. Heinola and T. Ahlgren, Phys. Rev. B 81 (2010), 10.1103/PhysRevB.81.073409

2010 doi
[79]

Chen and N

Z. Chen and N. Ghoniem, Phys. Rev. B 88, 035415 (2013)

2013
[80]

Maury, M

F. Maury, M. Biget, P. Vajda, A. Lucasson, and P. Lu- casson, Radiat. Eﬀ. 38, 53 (1978)

1978
[81]

Nordlund, J

K. Nordlund, J. Wallenius, and L. Malerba, Nuclear In- struments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms 246, 322 (2006)

2006
[82]

Alexander, M.-C

R. Alexander, M.-C. Marinica, L. Proville, F. Willaime, K. Arakawa, M. R. Gilbert, and S. L. Dudarev, Phys. Rev. B 94, 024103 (2016)

2016
[83]

X. Yi, M. L. Jenkins, M. Briceno, S. G. Roberts, Z. Zhou, and M. A. Kirk, Philos. Mag. 93, 1715 (2013)

2013
[84]

X. Yi, M. L. Jenkins, M. A. Kirk, Z. Zhou, and S. G. Roberts, Acta Materialia 112, 105 (2016)

2016
[85]

C. S. Becquart and C. Domain, Nuclear Instruments and Methods in Physics Research Section B: Beam Interac- tions with Materials and Atoms Computer Simulation of Radiation Eﬀects in Solids, 255, 23 (2007)

2007
[86]

M. R. Gilbert, S. L. Dudarev, P. M. Derlet, and D. G. Pettifor, J. Phys.: Condens. Matter 20, 345214 (2008)

2008
[87]

Ma and S

P.-W. Ma and S. L. Dudarev, (private communication)
[88]

https://gitlab.com/acclab/gap-data
[89]

Bj¨ orkas and K

C. Bj¨ orkas and K. Nordlund, Nuclear Instruments and Methods in Physics Research Section B: Beam Interac- tions with Materials and Atoms 259, 853 (2007)

2007
[90]

H. Wang, X. Guo, L. Zhang, H. Wang, and J. Xue, Appl. Phys. Lett. 114, 244101 (2019)

2019

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [38]

Nordlund, N

K. Nordlund, N. Runeberg, and D. Sundholm, Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms 132, 45 (1997)

work page 1997

[2] [1]

Rieth, S

M. Rieth, S. L. Dudarev, S. M. Gonzalez de Vicente, J. Aktaa, T. Ahlgren, S. Antusch, D. E. J. Arm- strong, M. Balden, N. Baluc, M. F. Barthe, W. W. Basuki, M. Battabyal, C. S. Becquart, D. Blagoeva, H. Boldyryeva, J. Brinkmann, M. Celino, L. Ciupinski, J. B. Correia, A. De Backer, C. Domain, E. Gaganidze, C. Garc´ ıa-Rosales, J. Gibson, M. R. Gilbert, S....

2013

[3] [2]

Zinkle and L

S. Zinkle and L. Snead, Annu. Rev. Mater. Res. 44, 241 (2014)

2014

[4] [3]

Nordlund, S

K. Nordlund, S. J. Zinkle, A. E. Sand, F. Granberg, R. S. Averback, R. E. Stoller, T. Suzudo, L. Malerba, F. Banhart, W. J. Weber, F. Willaime, S. L. Dudarev, and D. Simeone, Journal of Nuclear Materials 512, 450 (2018)

2018

[5] [4]

A. E. Sand, S. L. Dudarev, and K. Nordlund, EPL 103, 46003 (2013)

2013

[6] [5]

Byggm¨ astar, F

J. Byggm¨ astar, F. Granberg, A. E. Sand, A. Pirttikoski, R. Alexander, M.-C. Marinica, and K. Nordlund, J. Phys.: Condens. Matter 31, 245402 (2019)

2019

[7] [6]

Nordlund, M

K. Nordlund, M. Ghaly, R. S. Averback, M. Caturla, T. Diaz de la Rubia, and J. Tarus, Phys. Rev. B 57, 7556 (1998)

1998

[8] [7]

A. E. Sand, J. Dequeker, C. S. Becquart, C. Domain, and K. Nordlund, Journal of Nuclear Materials 470, 119 (2016)

2016