Pith. sign in

REVIEW 3 major objections 6 minor 39 references

Accelerating point defect photo-emission calculations with machine learning interatomic potentials

T0 review · 3 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A pre-trained machine-learned potential reproduces defect photoluminescence spectra with near-DFT accuracy at over ten times lower cost.

desk verdict Solid application of universal MLIPs to defect photoluminescence with a large honest benchmark, but the headline speedup is never measured and the accuracy claims cover only magnetic defects. read the letter →

arxiv 2505.01403 v2 pith:7E3FCZ3M submitted 2025-05-02 cond-mat.mtrl-sci physics.atom-phphysics.comp-ph

Kartikeya Sharma , Antoine Loew , Haiyuan Wang , Fredrik A. Nilsson , Manjari Jain , Miguel A. L. Marques , Kristian S. Thygesen This is my paper · ORCID
classification cond-mat.mtrl-sciphysics.atom-phphysics.comp-ph
keywords photoluminescencespectraHuang-Rhysfactorpointdefectscolorcentersmachinelearninginteratomicpotentialsphononcalculationsdensityfunctionaltheoryelectron-phononcoupling
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

Photoluminescence spectra of point defects are expensive to compute because phonon modes of a 100+-atom defect supercell require hundreds of DFT force calculations. This paper argues that the phonon bottleneck can be replaced by a pre-trained universal machine learning interatomic potential without losing ab initio fidelity. Benchmarking against DFT emission data for 791 color centers in ten 2D host crystals, plus bulk defects and a molecule on a surface, the authors report Huang-Rhys factors with a mean absolute error of 0.79, about 12% overall and 9% for small-HR defects, and a 12-defect sample shows close agreement in the photoluminescence lineshapes. The replacement cuts the cost by more than an order of magnitude. For large-displacement cases where the MLIP fails, a hybrid scheme that computes force constants with DFT only for atoms within a 4-5 Å radius of the defect reduces the average Huang-Rhys error from 48.7% to 5.1%.

What carries the argument

The machinery is the electron-phonon spectral function $S(\omega)=\sum_k S_k \delta(\hbar\omega-\hbar\omega_k)$ built from partial Huang-Rhys factors $S_k=\omega_k Q_k^2/(2\hbar)$, where $Q_k$ is the projection of the mass-weighted ground-to-excited displacement $\Delta Q$ onto phonon mode $k$; the photoluminescence spectrum follows from this spectral function through a generating function. The computational replacement is the phonon force-constant matrix: instead of computing it with DFT for every displaced atom, a universal machine-learned interatomic potential provides the forces, and a hybrid variant uses DFT forces only for displacements of atoms within a cutoff radius $r_c$ of the defect and MLIP forces for all other displacements. The hybrid dynamical matrix therefore keeps DFT accuracy precisely in the region where the defect's phonon modes have large amplitude.

What would settle it

A controlled benchmark on defects with $S_{\mathrm{DFT}} \ge 4$ would settle it: if MLIP-only Huang-Rhys factors deviate by roughly a factor of two in a substantial fraction of such large-displacement cases, the claim of ab initio accuracy fails outside the small-reorganization regime. The paper's own hybrid results show that this is exactly the regime where DFT forces near the defect are needed to recover agreement.

Watch

Extended reading notes

Core claim

The central claim is that the electron-phonon coupling shaping a defect's photoluminescence, quantified by the Huang-Rhys factor $S$ and the spectral function $S(\omega)=\sum_k S_k\delta(\hbar\omega-\hbar\omega_k)$ with $S_k=\omega_k Q_k^2/(2\hbar)$, can be computed to ab initio accuracy using MLIP phonons in place of DFT phonons. The ground-state displacement vector $\Delta Q$ is still obtained from DFT relaxations of ground and excited states; only the normal modes and frequencies of the ground-state supercell come from the machine-learned potential. Against a dataset of 791 spin-defect color centers, the best of seven universal MLIPs yields a mean absolute error of 0.79 in the Huang-Rhys factor, and accuracy is essentially independent of the defect's charge state (errors 0.73-0.83 for $q=-1,0,+1$) and magnetic moment. The authors attribute this to the division of labor: the charge-sensitive displacement stays in DFT while the MLIP only supplies ground-state phonons. Large ground-to-excited displacements degrade accuracy, with worst cases off by about a factor of two, and the paper's hybrid force-constant scheme restores DFT accuracy by computing force constants with DFT for atoms within a cutoff radius around the defect.

Load-bearing premise

The central assumption is that a pre-trained machine-learned potential, which has no charge or spin input and may have seen few defect geometries, produces ground-state phonon modes and frequencies for defect supercells that faithfully reproduce DFT phonons.

Editorial extensions

If this is right

  • Photoluminescence spectra and Huang-Rhys factors for point defects can be computed at more than an order of magnitude lower cost, with an overall error of about 12% and under 10% for the small-HR defects most relevant to applications.
  • The method is not biased by charge or magnetic state, so screening can safely cover neutral, charged, and magnetic color centers without special handling.
  • Defects with large ground-to-excited reorganization are the failure regime; on the 50 worst cases the hybrid DFT/MLIP force-constant scheme cuts the average Huang-Rhys error from 48.7% to 5.1% using a DFT region of only 16-20 atoms.
  • The same workflow applies to molecular emitters on surfaces, reproducing the Huang-Rhys factor and lineshape of terrylene on hexagonal boron nitride including low-energy molecule-substrate modes.
  • Removing the phonon bottleneck makes high-throughput screening of defect-engineered materials computationally tractable.

Reading between the lines

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

  • Because accuracy is tied to the size of the ground-to-excited displacement, a screening pipeline could decide on the fly whether to trust MLIP-only phonons or switch to the hybrid scheme based on $\Delta Q$.
  • The charge-state independence suggests the MLIP's ground-state phonon subspace transfers across charge states; a direct test would compare MLIP and DFT phonons on identical supercells in different charge states.
  • The hybrid cutoff behavior implies an error estimator based on phonon mode amplitude near the defect could replace the current empirical $r_c$ choice.
  • The 12% mean error in the Huang-Rhys factor is comparable to reported method-to-method variations from supercell size and exchange-correlation functional, so MLIP-accelerated spectra may be accurate enough for relative ranking even where absolute values carry systematic offsets.
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 / 6 minor

Summary. The paper proposes a workflow that replaces DFT phonon-mode calculations with universal machine-learning interatomic potential (MLIP) phonon calculations when computing Huang-Rhys factors and photoluminescence spectra of point defects. The ground-state and excited-state relaxations, as well as the displacement vector ΔR, are still obtained from DFT, while the phonons are computed with the MLIP. Seven universal MLIPs are benchmarked against a dataset of 791 two-dimensional point defects, with MatterSim-v1-5M identified as the best performer (MAE 0.79 in the Huang-Rhys factor). Full PL lineshapes are compared for 12 selected defects, a molecular emitter (terrylene) in gas phase and on hBN, and a hybrid DFT/MLIP force-constant approach is introduced to correct the worst cases, reducing the average error on S from 48.7% to 5.1% for the 50 worst structures. The abstract and discussion claim speed improvements exceeding an order of magnitude and 'ab initio accuracy' for the MLIP-accelerated spectra.

Significance. If the claimed speedup and accuracy hold, this would be a practically valuable tool for high-throughput screening of defect-based quantum emitters. The paper has notable strengths: a large and openly available benchmark dataset of 791 defect PL spectra, a systematic comparison of seven MLIPs, a clear analysis of error trends with ΔQ, and a hybrid scheme with a large demonstrated error reduction. The availability of data and code is also a positive feature. However, the central performance claim of 'speed improvements exceeding an order of magnitude' is never directly measured, and the accuracy claims are established only for a benchmark that is restricted to magnetic defects, with lineshape validation on a hand-picked subset. These gaps are load-bearing for the stated purpose of the paper.

major comments (3)
  1. [III. Discussion and Abstract] The abstract and Section III state that the approach achieves 'speed improvements exceeding an order of magnitude' and is 'faster by an order of magnitude', but the manuscript contains no wall-clock timing measurements of either the full DFT workflow or the MLIP-accelerated workflow. Since the MLIP replaces only the phonon step while DFT ground-state and excited-state relaxations are still required, and since the hybrid scheme reintroduces DFT forces for atoms within r_c, the asserted speedup is not established. Please provide measured timings for the phonon stage and for the end-to-end workflow, state the hardware and parallelization settings, and report the effective speedup of the hybrid scheme including the N_c/N DFT cost.
  2. [II.C, II.D, II.E] Section II.C states that the 791-defect benchmark 'contains only defects with finite magnetic moments' because non-magnetic defects were filtered out; the only non-magnetic system is the molecular emitter TRL in Section II.G. Consequently, the claim in Sections II.D and III that accuracy is independent of charge and magnetic state is demonstrated only for magnetic point defects, not for non-magnetic point defects. In addition, the PL lineshape agreement is shown for 12 hand-picked defects in Fig. 4, and the failure cases in Fig. 5 are also selected by hand; no systematic lineshape metric over the full dataset is provided. Please either add non-magnetic point defects to the benchmark or explicitly qualify the generality claim, and add a systematic lineshape metric (for example, spectral overlap or a sideband-position error) evaluated over the whole dataset.
  3. [II.E and II.F] The paper reports that the MLIP relaxation must start from the DFT-relaxed geometry, that large ground-to-excited displacements (large ΔQ) lead to errors of up to a factor of two in S (Fig. 5), and that the hybrid correction was tested only on the 50 worst errors with S_DFT<10. These are load-bearing limitations for the stated goal of high-throughput screening: the practical workflow still requires DFT relaxations, and the hybrid scheme reduces the phonon speedup by a factor N_c/N without a quantitative estimate of the residual speedup. Please quantify how often large-ΔQ failures occur in the full 791-defect dataset, measure the actual speedup of the hybrid approach as a function of r_c, and provide a practical criterion for deciding when the hybrid correction is needed.
minor comments (6)
  1. [References] The reference by Alkauskas et al. appears twice, as [10] and as [31]; the duplicate should be merged.
  2. [Eq. (6)] In the definition of the hybrid force-constant matrix, the notation F^α_i and u^β_j should be defined explicitly, and the symmetry of the resulting matrix should be discussed.
  3. [IV.A] The Methods section reports a 'k-point density of 3 Å'; this should read 3 Å^-1 or specify the actual Monkhorst-Pack grid used.
  4. [Throughout] The model name is written inconsistently as 'MatterSim-v1-5M', 'Mattersim-V1-5M', and 'MtS'; one convention should be used consistently.
  5. [Fig. 7 caption] The caption states that the ZPL positions are shifted to 0 and the ZPL intensity is set to unity, but the horizontal axis is labelled in absolute energy; the plotting convention should be clarified.
  6. [Section II.G] The version or checkpoint of MatterSim used should be stated precisely, since the model is maintained externally and the reproducibility of the central results depends on a fixed model version.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the predicted Huang-Rhys factors are not inputs to the fixed MatterSim model, and the DFT benchmark is an out-of-sample comparison for the MLIP.

full rationale

The central quantity S is computed from the DFT-derived displacement ΔR (Eqs. 1–3) projected onto phonon modes (Eq. 4). In the DFT reference, those phonons come from DFT force constants; in the MLIP workflow, they come from the fixed, pre-trained MatterSim-v1-5M network of Ref. [29]. No parameter of MatterSim is fitted to the 791-defect Huang-Rhys values, and the DFT benchmark is therefore a genuine external comparison of the phonon component while ΔR is held fixed. The hybrid phonon scheme (Eq. 6) intentionally converges to DFT as the cutoff radius grows, but the paper presents it as a corrective strategy, not as evidence for the pure MLIP result. The self-citations to Ref. [25] for model preselection and Ref. [30] for spin purification are not load-bearing: the model choice is independently re-tested in Table I and Fig. 1, and the spin-purification expression is a standard documented procedure. The paper itself states limitations, including the need to start MLIP relaxation from the DFT-relaxed structure and degraded accuracy for large ΔQ (Section II.E, Fig. 2); these are accuracy limitations, not circular steps. The absence of wall-clock timing measurements weakens the 'order of magnitude' speed claim, but that is a missing-evidence and correctness concern, not a circularity concern. The derivation is self-contained in the sense that every predicted quantity is computed from inputs that are not themselves the predicted quantity.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The paper relies on standard approximations in defect photoluminescence theory and treats PBE-DFT as the accuracy target. No new physical entities are introduced. The hybrid cutoff radius is a hand-chosen convergence parameter rather than a fitted physical constant.

free parameters (1)
  • Hybrid cutoff radius r_c = 4-5 Å (16-20 atoms)
    Chosen in the hybrid phonon approach to reduce the average HR-factor error on the 50 worst-case defects from 48.7% to 5.1%. It is a convergence parameter, not fitted to the test set, but it is a hand-selected setting that affects the reported hybrid improvement.
assumptions (4)
  • domain assumption Born-Oppenheimer approximation
    Stated in Methods C(i), justifies separation of electronic and nuclear motion; standard for semiconductor emitters with band gaps above 1 eV.
  • domain assumption Harmonic approximation and identical ground/excited state potential energy surfaces
    Stated in Methods C(iii). The displacement Delta R is projected onto ground-state phonon modes, assuming mode shapes and frequencies are unchanged by excitation and anharmonicity is negligible.
  • domain assumption Single-determinant excited states via DO-MOM
    Stated in Methods C(ii). Neglects explicit many-body correlation in the excited state, a standard approximation in this workflow.
  • domain assumption DFT-PBE reference is the accuracy target
    The MLIP predictions are measured against PBE results; the paper claims 'DFT accuracy' which is tied to PBE reliability and known functional sensitivity, acknowledged in Section II.D.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Accelerating point defect photo-emission calculations with machine learning interatomic potentials." pith.science (2026). https://pith.science/paper/7E3FCZ3M

@misc{pith2026250501403,
  author       = {Pith},
  title        = {Pith review of: Accelerating point defect photo-emission calculations with machine learning interatomic potentials},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7E3FCZ3M}},
  note         = {Machine review of arXiv:2505.01403}
}
read the original abstract

We introduce a computational framework leveraging universal machine learning interatomic potentials (MLIPs) to dramatically accelerate the calculation of photoluminescence (PL) spectra of atomic or molecular emitters with ab initio accuracy. By replacing the costly density functional theory (DFT) computation of phonon modes with much faster MLIP phonon mode calculations, our approach achieves speed improvements exceeding an order of magnitude with minimal precision loss. We benchmark the approach using a dataset comprising ab initio emission spectra of 791 color centers spanning various types of crystal point defects in different charge and magnetic states. The method is also applied to a molecular emitter adsorbed on a hexagonal boron nitride surface. Across all the systems, we find excellent agreement for both the Huang-Rhys factor and the PL lineshapes. This application of universal MLIPs bridges the gap between computational efficiency and spectroscopic fidelity, opening pathways to high-throughput screening of defect-engineered materials. Our work not only demonstrates accelerated calculation of PL spectra with DFT accuracy, but also makes such calculations tractable for more complex materials.

Figures

Figures reproduced from arXiv: 2505.01403 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7 [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 16 canonical work pages

  1. [1]

    biomag2d

    The spectral region beyond 45 meV is enhanced by a factor of 2 for clarity. (c) Vibrational modes corresponding to the peaks highlighted by the blue circle in (b). while maintaining maximum overlap with ground-state orbitals. The DO-MOM algorithm uses a nested opti- mization approach: an inner loop finds stationary points in the subspace of occupied and v...

    work page 2022

  2. [2]

    Wolfowicz, F

    G. Wolfowicz, F. J. Heremans, C. P. Anderson, S. Kanai, H. Seo, A. Gali, G. Galli, and D. D. Awschalom, Quan- tum guidelines for solid-state spin defects, Nat. Rev. Mater.6, 906–925 (2021)

    work page 2021

  3. [3]

    G. Wang, Y. Yang, D. Han, and Y. Li, Oxygen defective metal oxides for energy conversion and storage, Nano To- day13, 23 (2017)

    work page 2017

  4. [4]

    Sarkar, L

    A. Sarkar, L. Velasco, D. Wang, Q. Wang, G. Ta- lasila, L. de Biasi, C. Kübel, T. Brezesinski, S. S. Bhat- tacharya, H. Hahn, and B. Breitung, High entropy ox- ides for reversible energy storage, Nat. Commun.9, 10.1038/s41467-018-05774-5 (2018)

    doi:10.1038/s41467-018-05774-5 2018

  5. [5]

    Jhuria, V

    K. Jhuria, V. Ivanov, D. Polley, Y. Zhiyenbayev, W. Liu, A. Persaud, W. Redjem, W. Qarony, P. Parajuli, Q. Ji, A. J. Gonsalves, J. Bokor, L. Z. Tan, B. Kanté, and T. Schenkel, Programmable quantum emitter formation in silicon, Nature Communications15, 10.1038/s41467- 024-48714-2 (2024)

    doi:10.1038/s41467- 2024

  6. [6]

    Jelezko and J

    F. Jelezko and J. Wrachtrup, Single defect centres in di- amond: A review, Phys. Status Solidi A203, 3207–3225 (2006)

    work page 2006

  7. [7]

    M. W. Doherty, N. B. Manson, P. Delaney, F. Jelezko, J. Wrachtrup, and L. C. L. Hollenberg, The nitrogen-vacancy colour centre in diamond, arXiv 10.48550/ARXIV.1302.3288 (2013)

    arXiv doi:10.48550/arxiv.1302.3288 2013

  8. [8]

    Jones and P

    R. Jones and P. Briddon, Chapter 6 the ab initio cluster method and the dynamics of defects in semiconductors, inIdentification of Defects in Semiconductors, Semicon- ductors and Semimetals, Vol. 51, edited by M. Stavola (Elsevier, 1998) pp. 287–349

    work page 1998

Show all 39 references
  1. [9]

    P. Deák, B. Aradi, T. Frauenheim, and A. Gali, Chal- lenges for ab initio defect modeling, Mater. Sci. Eng., B 154-155, 187 (2008), front-End Junction and Contact Formation in Future Silicon/Germanium Based Devices

    2008

  2. [10]

    Gali, Recent advances in the ab initio theory of solid- state defect qubits, Nanophotonics12, 359–397 (2023)

    A. Gali, Recent advances in the ab initio theory of solid- state defect qubits, Nanophotonics12, 359–397 (2023)

    2023

  3. [12]

    Behler, Perspective: Machine learning potentials for atomistic simulations, J

    J. Behler, Perspective: Machine learning potentials for atomistic simulations, J. Chem. Phys.145, 170901 (2016)

    2016

  4. [13]

    Schmidt, M

    J. Schmidt, M. R. G. Marques, S. Botti, and M. A. L. Marques, Recent advances and applications of machine learning in solid-state materials science, npj Comput. Mater.5, 83 (2019)

    2019

  5. [14]

    Graser, S

    J. Graser, S. K. Kauwe, and T. D. Sparks, Machine learning and energy minimization approaches for crystal structure predictions: A review and new horizons, Chem. Mater.30, 3601–3612 (2018)

    2018

  6. [15]

    O. T. Unke, S. Chmiela, H. E. Sauceda, M. Gastegger, I. Poltavsky, K. T. Schütt, A. Tkatchenko, and K.-R. Müller, Machine learning force fields, Chem. Rev.121, 10142–10186 (2021)

    2021

  7. [16]

    C. Chen, W. Ye, Y. Zuo, C. Zheng, and S. P. Ong, Graph networks as a universal machine learning framework for molecules and crystals, Chem. Mater.31, 3564–3572 (2019)

    2019

  8. [17]

    Chen and S

    C. Chen and S. P. Ong, A universal graph deep learning interatomic potential for the periodic table, Nat. Com- put. Sci.2, 718–728 (2022)

    2022

  9. [18]

    Batatia, D

    I. Batatia, D. P. Kovacs, G. Simm, C. Ortner, and G. Csanyi, MACE: Higher order equivariant message passing neural networks for fast and accurate force fields, inAdvances in Neural Information Processing Systems, Vol. 35, edited by S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrav...

  10. [19]

    B. Deng, P. Zhong, K. Jun, J. Riebesell, K. Han, C. J. Bartel, and G. Ceder, CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling, Nat. Mach. Intell.5, 1031–1041 (2023)

    2023

  11. [20]

    Neumann, J

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

    2024 arXiv

  12. [21]

    Y. Park, J. Kim, S. Hwang, and S. Han, Scalable parallel algorithm for graph neural network interatomic poten- tialsinmoleculardynamicssimulations,J.Chem.Theory Comput.20, 4857–4868 (2024)

    2024

  13. [22]

    Y.-L. Liao, B. M. Wood, A. Das, and T. Smidt, EquiformerV2: Improved equivariant transformer for scaling to higher-degree representations, inThe Twelfth International Conference on Learning Representations (2024)

    2024

  14. [23]

    Choudhary and B

    K. Choudhary and B. DeCost, Atomistic line graph neu- ral network for improved materials property predictions, npj Comput. Mater.7, 185 (2021)

    2021

  15. [24]

    Chmiela, V

    S. Chmiela, V. Vassilev-Galindo, O. T. Unke, A. Kabylda, H. E. Sauceda, A. Tkatchenko, and K.-R. Müller, Accurate global machine learning force fields for molecules with hundreds of atoms, Sci. Adv.9, eadf0873 (2023)

    2023

  16. [25]

    Stocker, J

    S. Stocker, J. Gasteiger, F. Becker, S. Günnemann, and J. T. Margraf, How robust are modern graph neural net- work potentials in long and hot molecular dynamics sim- ulations?, Machine Learning: Science and Technology3, 045010 (2022)

    2022

  17. [26]

    A. Loew, D. Sun, H.-C. Wang, S. Botti, and M. A. L. Marques, Universal machine learning interatomic poten- tials are ready for phonons, arXiv , 2412.16551 (2024)

    2024 arXiv

  18. [27]

    Riebesell, R

    J. Riebesell, R. E. A. Goodall, P. Benner, Y. Chi- ang, B. Deng, A. A. Lee, A. Jain, and K. A. Persson, Matbench discovery – a framework to evaluate machine learning crystal stability predictions, arXiv , 2308.14920 (2023)

    2023 arXiv

  19. [28]

    Batzner, A

    S. Batzner, A. Musaelian, L. Sun, M. Geiger, J. P. Mailoa, M. Kornbluth, N. Molinari, T. E. Smidt, and B. Kozinsky, E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials, Nat. Commun.13, 2453 (2022)

    2022

  20. [29]

    Barroso-Luque, M

    L. Barroso-Luque, M. Shuaibi, X. Fu, B. M. Wood, M. Dzamba, M. Gao, A. Rizvi, C. L. Zitnick, and Z. W. Ulissi, Open materials 2024 (OMat24) inorganic materi- als dataset and models, arXiv , 2410.12771 (2024)

    2024 arXiv

  21. [30]

    H. Yang, C. Hu, Y. Zhou, X. Liu, Y. Shi, J. Li, G. Li, Z. Chen, S. Chen, C. Zeni, M. Horton, R. Pinsler, A. Fowler, D. Zügner, T. Xie, J. Smith, L. Sun, Q. Wang, L. Kong, C. Liu, H. Hao, and Z. Lu, Mattersim: A deep learning atomistic model across elements, temperatures and pr...

    2024 arXiv

  22. [31]

    H. Wang, N. Stenger, P. Lyngby, M. Kuisma, J. Schiøtz, and K. S. Thygesen, Two-dimensional materials as ideal substrates for molecular quantum emitters, Nano Letters 25, 9952 (2025)

    2025

  23. [32]

    Alkauskas, B

    A. Alkauskas, B. B. Buckley, D. D. Awschalom, and C. G. Van de Walle, First-principles theory of the luminescence lineshape for the triplet transition in diamond nv centres, New J. Phys.16, 073026 (2014)

    2014

  24. [33]

    J. Lee, H. Wang, K.-Y. Park, S. Huh, D. Kim, M. Yu, C. Kim, K. S. Thygesen, and J. Lee, Room temperature quantum emitters in van der waalsα-moo3, Nano Lett. 25, 1142 (2025)

    2025

  25. [34]

    R. Smit, A. Tebyani, J. Hameury, S. J. van der Molen, and M. Orrit, Sharp zero-phonon lines of single organic molecules on a hexagonal boron-nitride surface, Nature Communications14, 7960 (2023)

    2023

  26. [35]

    J. J. Mortensen, A. H. Larsen, M. Kuisma, A. V. Ivanov, A. Taghizadeh, A. Peterson, A. Haldar, A. O. Dohn, C. Schäfer, E. O. Jónsson, E. D. Hermes, F. A. Nilsson, G. Kastlunger, G. Levi, H. Jónsson, H. Häkkinen, J. Fojt, J. Kangsabanik, J. Sødequist, J. Lehtomäki, J. Heske, J....

    2024

  27. [36]

    J. P. Perdew, K. Burke, and M. Ernzerhof, Generalized gradient approximation made simple, Phys. Rev. Lett. 77, 3865 (1996)

    1996

  28. [37]

    A. V. Ivanov, G. Levi, E. Ö. Jónsson, and H. Jónsson, Method for calculating excited electronic states using density functionals and direct orbital optimization with real space grid or plane-wave basis set, Journal of Chem- ical Theory and Computation17, 5034 (2021), pMID: 342...

    2021 doi

  29. [38]

    Bitzek, P

    E. Bitzek, P. Koskinen, F. Gähler, M. Moseler, and P. Gumbsch, Structural relaxation made simple, Phys. Rev. Lett.97, 170201 (2006)

    2006

  30. [39]

    A. Togo, L. Chaput, and I. Tanaka, Distributions of phonon lifetimes in Brillouin zones, Phys. Rev. B91, 094306 (2015)

    2015

  31. [40]

    Togo, First-principles phonon calculations with Phonopy and Phono3py, J

    A. Togo, First-principles phonon calculations with Phonopy and Phono3py, J. Phys. Soc. Jpn.92, 012001 (2023)

    2023

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.