Pith. sign in

REVIEW 5 major objections 5 minor 51 references

Graph neural network framework for energy mapping of hybrid monte-carlo molecular dynamics simulations of Medium Entropy Alloys

T0 review · 5 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A graph convolutional network predicts potential energy in NiCoCr alloy simulations with R2 up to 0.98.

desk verdict A transparent but methodologically shaky GNN energy surrogate for NiCoCr; the velocity feature is a likely temperature proxy and the reported R2 is not fully trustworthy. read the letter →

arxiv 2411.13670 v1 pith:XDLCW56T submitted 2024-11-20 cond-mat.mtrl-sci cs.LG

Mashaekh Tausif Ehsan , Saifuddin Zafar , Apurba Sarker , Sourav Das Suvro , Mohammad Nasim Hasan This is my paper · ORCID
classification cond-mat.mtrl-scics.LG
keywords GraphneuralnetworkMedium-entropyalloyHybridMonte-CarlomoleculardynamicsPotentialenergypredictionLocalchemicalorderNiCoCrEAMMachinelearningsurrogate
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 claims that a graph convolutional neural network can predict the potential energy of a NiCoCr medium-entropy alloy during hybrid Monte-Carlo/molecular dynamics simulations, using a graph where each atom is a node connected to its 12 nearest neighbors and node features are atom type and absolute velocity. Across three case studies, the model reaches R2 values between 0.93 and 0.98 and mean absolute percentage errors between 0.018 and 0.050, including for configurations at annealing temperatures not seen during training. If correct, the GCNN serves as a fast surrogate for the embedded-atom-method potential and captures the local chemical order that controls properties of the alloy. The significance is a graph-based route to predicting energy-driven properties of medium- and high-entropy alloys from simulation snapshots.

What carries the argument

The central object is the graph representation of an alloy configuration: 13,500 nodes, one per atom, with node features (atom type, absolute velocity) and edges to the 12 nearest neighbors plus self-loops, with no edge features. The GCNN stacks GCNConv layers using the normalized adjacency with self-loops, followed by ReLU and dropout, and is trained with mean squared error to regress the per-atom normalized total potential energy. The 12-neighbor edge construction is the mechanism intended to encode the local chemical order measured by Warren-Cowley parameters, and the velocity feature is included on the basis of prior use in GNN energy prediction.

What would settle it

Build two atomic configurations of NiCoCr that have identical species and velocities on every atom and identical 12-nearest-neighbor environments but differ in the arrangement of atoms in the second coordination shell; if the GCNN assigns nearly identical energies to both while the EAM potential gives noticeably different energies, the 12-neighbor graph is not a faithful energy mapping.

Watch

Extended reading notes

Core claim

The discovery is that the energy profile of hybrid MC/MD simulations of the equiatomic NiCoCr MEA can be mapped by a GCNN that reads only atom identity, absolute velocity, and a 12-nearest-neighbor edge list. The model reproduces the potential-energy-versus-MC-step curves for individual annealing temperatures (450K, 650K, 950K) with R2 of 0.98, 0.93, and 0.96; trained on a combined set (450K, 650K, 850K) it predicts all three with R2 of 0.97, 0.97, and 0.96; and trained on 350K, 650K, 850K, 1150K it generalizes to held-out temperatures with the best performance at lower temperatures and a documented drop at higher temperatures where energy plateaus quickly.

Load-bearing premise

The load-bearing premise is that each atom's total potential energy under the EAM potential is fully determined by its atom type, its absolute velocity, and the species of its 12 nearest neighbors, even though the EAM embedding function sums over neighbors within a longer cutoff.

Editorial extensions

If this is right

  • If the central claim holds, one trained GCNN can replace repeated EAM energy evaluations during MC/MD annealing at the temperatures it was trained on, at a small fraction of the cost.
  • A single model trained on a few annealing temperatures covers intermediate temperatures with R2 >= 0.96, so the approach transfers across thermal conditions within the training range.
  • Generalization to unseen annealing temperatures is demonstrated for lower temperatures, establishing that graph representations of MEA configurations encode transferable structure-energy relationships.
  • Because the graph has no edge features, the framework is directly applicable to other MEAs and HEAs without per-system feature engineering.
  • The observed degradation at high unseen temperatures identifies the data-density limitation that future work would need to address.

Reading between the lines

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

  • Editorial inference: the inclusion of absolute velocity as a node feature means the model may be learning a temperature or kinetic-energy proxy in addition to the potential-energy surface; retraining with velocity withheld would reveal how much of the accuracy is structural.
  • Editorial inference: because edges are truncated at 12 nearest neighbors and unweighted, contributions to the EAM embedding energy from atoms beyond the first coordination shell are invisible to the model, so the mapping is likely an effective first-shell model rather than a complete EAM surrogate.
  • Editorial inference: a natural stress test is to apply the trained GCNN as the energy evaluator inside the Metropolis acceptance step of a fresh MC/MD run; if the MC trajectories remain stable and consistent, the surrogate is usable for accelerated sampling.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper proposes a graph convolutional neural network (GCNN) framework to predict the potential energy of a NiCoCr medium-entropy alloy from hybrid Monte-Carlo/molecular dynamics (MC/MD) simulation data. Atomic configurations from LAMMPS dump files are converted into graphs with atoms as nodes, edges to the 12 nearest neighbors, and node features consisting of atom type and absolute velocity. The model is trained on per-atom potential energy labels from the EAM potential. Three case studies are presented: training separate models for individual annealing temperatures (450K, 650K, 950K), training one model on a combined set of temperatures (450K, 650K, 850K), and training on a set of temperatures (350K, 650K, 850K, 1150K) to test on unseen temperatures. Reported R2 values range from 0.93 to 0.98 and MAPE from 0.018 to 0.050 for the first two cases, while the third case shows degraded performance at higher temperatures. The abstract claims strong performance on both training data and unseen configurations.

Significance. If validated, the proposed GCNN would provide a fast surrogate for the EAM potential in MC/MD energy mapping, potentially enabling efficient exploration of local chemical order in MEAs and HEAs. However, the current evidence is undermined by two critical methodological issues: the inclusion of absolute velocity as a node feature, which can act as a temperature proxy and decouple predictions from the actual atomic configuration, and the absence of a clearly described held-out test split in the first two case studies. The third case study, which does use held-out temperatures, reports degraded performance, weakening the central claim. The authors do provide data and code availability on GitHub, which is a strength for reproducibility. Substantial methodological revision is needed before the claims can be accepted.

major comments (5)
  1. [Section 3.4, Eq. (2)] The node features include absolute velocity, but the EAM potential in Eq. (2) is purely a function of atomic positions. Velocity is not a position-derived feature and directly encodes the instantaneous temperature via the Maxwell-Boltzmann distribution. In the hybrid MC/MD annealing simulations, potential energy is strongly correlated with temperature, so the model can output the correct energy level by reading the velocity feature rather than learning the configurational energy surface. In case study 3, test temperatures are not in the training set, making velocity a direct temperature tag. To support the claim that the model captures local chemical order, the authors should remove velocity from the features or perform a permutation test that shuffles velocities across configurations and show that prediction quality does not degrade.
  2. [Sections 3.5.1, 3.5.2, 4.2, 4.3] The first two case studies report high R2 and low MAPE values, but the text does not describe any held-out train/test split within each annealing temperature. The description states that the parameters with the lowest MSE are stored, implying that the metrics are computed on the training data themselves. The abstract's claim of performance on 'unseen configurations' is therefore not supported for these cases. The authors should specify a proper split (e.g., reserving a fraction of the 300 MC frames per temperature for validation and testing) and report metrics on the test set, not just the training set.
  3. [Section 3.5.1, Fig. 1] The architecture description is internally inconsistent and incomplete. The first GCNConv layer is stated to have an input channel count equal to the number of atoms (13500), yet Section 3.4 says that node features are atom type and absolute velocity, which would be a 2-dimensional feature vector per node. Additionally, GCNConv outputs per-node features, but the target is a single scalar potential energy per graph; no graph-level readout (e.g., global pooling) is described between the final GCNConv layer and the fully connected layer. Please clarify the exact data flow, including how per-node outputs are aggregated into the scalar energy prediction.
  4. [Section 4.4] The third case study is the only one that tests on genuinely unseen configurations (held-out annealing temperatures), and the text admits that R2 drops as annealing temperature increases and that the model 'struggles to predict the higher energy levels.' This directly contradicts the abstract and conclusion, which claim satisfactory results on unseen configurations. The authors should report the actual R2 and MAPE values for each temperature in both the training and test sets, and temper the summary claims to reflect the observed degradation.
  5. [Section 3.4] The graph representation includes edges only to the 12 nearest neighbors (the first coordination shell in FCC). However, the EAM potential (Eq. 2) includes a pair potential and an embedding functional that depend on the electron density, which may extend beyond the first shell for the chosen potential. Truncating the neighborhood to 12 neighbors could discard physically relevant interactions. The authors should justify this truncation or perform a sensitivity analysis with a larger neighbor count.
minor comments (5)
  1. [Fig. 2(c)] The figure caption lists '1350K' but the text uses '1150K' as the highest annealing temperature; please make the notation consistent.
  2. [Section 3.5.1] The statement that the first GCNConv layer has 'input channels equal to the number of atoms in the system (which is 13500)' is inconsistent with the node feature matrix described in Section 3.4; this is likely a typo and should be corrected to the node feature dimension.
  3. [Section 3.6] The text says R2 values range from 0 to 1, but R2 can be negative for poorly fitting models; the wording should be adjusted for mathematical accuracy.
  4. [Section 3.4] The justification for including velocity as a feature relies on reference [51], but that work uses velocity in a different context; a physical argument for why velocity is relevant to a position-dependent energy function is needed.
  5. [Data Availability] The data availability statement says all data and codes are available on GitHub, but the repository link is in the author contributions section; please include the URL in the Data Availability section as well.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the GCNN is a supervised surrogate trained on EAM-generated labels; no load-bearing claim reduces to its inputs by definition.

full rationale

The paper's derivation chain is: hybrid MC/MD simulations with the EAM potential (Eq. 2) produce dump files and potential-energy labels; graphs are constructed from atomic positions and velocities with 12-nearest-neighbor edges; a GCNN is trained with MSE loss to reproduce those labels; and performance is reported via R2 and MAPE. This is a standard supervised regression task. The fact that training labels are generated by the same EAM potential that defines the target property is not circular: the paper does not claim to derive the EAM from first principles, nor is any input feature, graph edge, or network parameter defined in terms of the target energy. There is no load-bearing self-citation; reference [51] is external prior work, and no uniqueness theorem or ansatz is imported from the authors' own previous studies. The inclusion of absolute velocity as a node feature is a potential validity concern, because the EAM energy is position-only and velocity could act as a temperature proxy, but this is not circularity: the model's predictions are not forced by construction to equal the labels. Likewise, the paper's incomplete description of train/test splitting in the first two case studies may mean some reported metrics are fit quality rather than generalization, but that is a reporting issue, not a reduction of the prediction to its input by definition. Under the strict standard requiring a quotable Eq.-equals-Eq. or fitted-parameter-as-prediction reduction, no circular step is present.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities or forces. Its contributions are a graph representation, a choice of node features, and a GNN architecture, all of which rest on the assumptions listed above. The most fragile assumptions are the sufficiency of 12 nearest neighbors for the EAM potential and the physical validity of using velocity as an input feature.

free parameters (5)
  • number_of_gcn_layers = n=1 for case 1, n=3 for cases 2 and 3
    Chosen by the authors without ablation study; affects model capacity and reported accuracy.
  • hidden_channels = 100 (case 1), 300 (cases 2 and 3)
    Chosen by hand; no sensitivity analysis is provided.
  • learning_rate = 0.0001
    Fixed hyperparameter; not optimized or varied.
  • dropout_probability = 0.5
    Fixed regularization hyperparameter.
  • number_of_nearest_neighbors = 12
    Design choice to capture FCC first coordination shell; the paper does not test whether 12 neighbors are sufficient for the EAM potential.
assumptions (4)
  • domain assumption The EAM potential from Li et al. [40] accurately represents the interatomic interactions in NiCoCr.
    This potential generates the labels; if it is inaccurate, the model's predictions are meaningless for real alloys. Invoked in Section 3.1.
  • domain assumption A graph with edges to 12 nearest neighbors and no edge features contains sufficient information to determine the per-atom potential energy under the EAM potential.
    This is the central architectural premise, introduced in Section 3.4. The EAM energy depends on pair distances and local electron density, which may extend beyond the first shell.
  • ad hoc to paper Absolute atomic velocity is a valid input feature for predicting potential energy.
    Potential energy is a function of atomic positions, not velocities. Including velocity may allow the model to infer the annealing temperature, acting as a label leak. Used in Section 3.4.
  • domain assumption The configurations seen during training are representative of the unseen configurations at other annealing temperatures.
    Case study 3 assumes that a model trained at 350K, 650K, 850K, and 1150K can interpolate to temperatures in between. The paper itself notes degraded performance at high temperatures.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Graph neural network framework for energy mapping of hybrid monte-carlo molecular dynamics simulations of Medium Entropy Alloys." pith.science (2026). https://pith.science/paper/XDLCW56T

@misc{pith2026241113670,
  author       = {Pith},
  title        = {Pith review of: Graph neural network framework for energy mapping of hybrid monte-carlo molecular dynamics simulations of Medium Entropy Alloys},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XDLCW56T}},
  note         = {Machine review of arXiv:2411.13670}
}
read the original abstract

Machine learning (ML) methods have drawn significant interest in material design and discovery. Graph neural networks (GNNs), in particular, have demonstrated strong potential for predicting material properties. The present study proposes a graph-based representation for modeling medium-entropy alloys (MEAs). Hybrid Monte-Carlo molecular dynamics (MC/MD) simulations are employed to achieve thermally stable structures across various annealing temperatures in an MEA. These simulations generate dump files and potential energy labels, which are used to construct graph representations of the atomic configurations. Edges are created between each atom and its 12 nearest neighbors without incorporating explicit edge features. These graphs then serve as input for a Graph Convolutional Neural Network (GCNN) based ML model to predict the system's potential energy. The GCNN architecture effectively captures the local environment and chemical ordering within the MEA structure. The GCNN-based ML model demonstrates strong performance in predicting potential energy at different steps, showing satisfactory results on both the training data and unseen configurations. Our approach presents a graph-based modeling framework for MEAs and high-entropy alloys (HEAs), which effectively captures the local chemical order (LCO) within the alloy structure. This allows us to predict key material properties influenced by LCO in both MEAs and HEAs, providing deeper insights into how atomic-scale arrangements affect the properties of these alloys.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 46 canonical work pages

  1. [1]

    Y.J. Zhou, Y. Zhang, Y.L. Wang, G.L. Chen, Solid solution alloys of AlCoCrFeNiTix with excellent room-temperature mechanical properties, Appl. Phys. Lett. 90 (2007) 181904

    work page 2007

  2. [2]

    El-Atwani, N

    O. El-Atwani, N. Li, M. Li, A. Devaraj, J.K.S. Baldwin, M.M. Schneider, D. Sobieraj, J.S. Wróbel, D. Nguyen-Manh, S.A. Maloy, E. Martinez, Outstanding radiation resistance of tungsten-based high-entropy alloys, Sci Adv 5 (2019). https://doi.org/10.1126/sciadv.aav2002

    doi:10.1126/sciadv.aav2002 2019

  3. [3]

    J. Ge, C. Chen, R. Zhao, Q. Liu, Y. Long, J. Wang, Z. Ren, S. Yin, Strength-ductility synergy of CoCrNi medium-entropy alloy processed with laser powder bed fusion, Materials & Design 219 (2022) 110774. https://doi.org/10.1016/j.matdes.2022.110774

    work page arXiv 2022

  4. [4]

    H. Inui, K. Kishida, L. Li, A.M. Manzoni, S. Haas, U. Glatzel, Uniaxial mechanical properties of face-centered cubic single- and multiphase high-entropy alloys, MRS Bull 47 (2022) 168–174. https://doi.org/10.1557/s43577-022-00280-y

    work page doi:10.1557/s43577-022-00280-y 2022

  5. [5]

    Garcia Filho, R.O

    F.D.C. Garcia Filho, R.O. Ritchie, M.A. Meyers, S.N. Monteiro, Cantor-derived medium- entropy alloys: bridging the gap between traditional metallic and high-entropy alloys, J. Mater. Res. Technol. 17 (2022) 1868–1895

    work page 2022

  6. [7]

    Gludovatz, A

    B. Gludovatz, A. Hohenwarter, D. Catoor, E.H. Chang, E.P. George, R.O. Ritchie, A fracture-resistant high-entropy alloy for cryogenic applications, Science (1979) 345 (2014) 1153–1158

    work page 1979

  7. [8]

    Gludovatz, A

    B. Gludovatz, A. Hohenwarter, H. Thurston Keli V S and Bei, Z. Wu, E.P. George, R.O. Ritchie, Exceptional damage-tolerance of a medium-entropy alloy CrCoNi at cryogenic temperatures, Nat. Commun. 7 (2016) 10602

    work page 2016

  8. [9]

    B. Uzer, S. Picak, J. Liu, T. Jozaghi, D. Canadinc, I. Karaman, Y.I. Chumlyakov, I. Kireeva, On the mechanical response and microstructure evolution of NiCoCr single crystalline medium entropy alloys, Mater. Res. Lett. 6 (2018) 442–449

    work page 2018

Show all 51 references
  1. [10]

    Q. Ding, Y. Zhang, X. Chen, D. Fu Xiaoqian and Chen, S. Chen, L. Gu, F. Wei, H. Bei, Y. Gao, M. Wen, J. Li, Z. Zhang, T. Zhu, R.O. Ritchie, Q. Yu, Tuning element distribution, structure and properties by composition in high-entropy alloys, Nature 574 (2019) 223–227

    2019

  2. [11]

    X. Chen, Q. Wang, Z. Cheng, H. Zhu Mingliu and Zhou, P. Jiang, L. Zhou, F. Xue Qiqi and Yuan, J. Zhu, X. Wu, E. Ma, Direct observation of chemical short-range order in a medium-entropy alloy, Nature 592 (2021) 712–716

    2021

  3. [12]

    Zhang, S

    R. Zhang, S. Zhao, J. Ding, T. Chong Yan and Jia, C. Ophus, M. Asta, R.O. Ritchie, A.M. Minor, Short-range order and its impact on the CrCoNi medium-entropy alloy, Nature 581 (2020) 283–287

    2020

  4. [13]

    W. Ji, M.S. Wu, Retainable short-range order effects on the strength and toughness of NbMoTaW refractory high-entropy alloys, Intermetallics (Barking) 150 (2022) 107707

    2022

  5. [14]

    W. Ji, M.S. Wu, Inhibiting the inverse Hall-Petch behavior in CoCuFeNiPd high-entropy alloys with short-range ordering and grain boundary segregation, Scr. Mater. 221 (2022) 114950

    2022

  6. [15]

    Li, J.-P

    Y. Li, J.-P. Du, P. Yu, R. Li, S. Shinzato, Q. Peng, S. Ogata, Chemical ordering effect on the radiation resistance of a CoNiCrFeMn high-entropy alloy, Comput. Mater. Sci. 214 (2022) 111764

    2022

  7. [16]

    Arkoub, M

    H. Arkoub, M. Jin, Impact of chemical short-range order on radiation damage in Fe-Ni-Cr alloys, Scr. Mater. 229 (2023) 115373

    2023

  8. [17]

    Zhang, Z

    Z. Zhang, Z. Su, B. Zhang, Q. Yu, J. Ding, T. Shi, C. Lu, R.O. Ritchie, E. Ma, Effect of local chemical order on the irradiation-induced defect evolution in CrCoNi medium- entropy alloy, Proc. Natl. Acad. Sci. U. S. A. 120 (2023) e2218673120

    2023

  9. [18]

    Chen, S.-W

    Y. Chen, S.-W. Reng, J. Peng, X.-B. Liu, Chemical short range order and deformation mechanism of a refractory high entropy alloy HfNbTaZr under nanoindentation: An atomistic study, J. Mater. Res. Technol. 24 (2023) 3588–3598

    2023

  10. [19]

    Chen, Z.H

    S. Chen, Z.H. Aitken, Z. Pattamatta Subrahmanyam and Wu, Z.G. Yu, D.J. Srolovitz, P.K. Liaw, Y.-W. Zhang, Short-range ordering alters the dislocation nucleation and propagation in refractory high-entropy alloys, Mater. Today (Kidlington) 65 (2023) 14– 25

    2023

  11. [20]

    L. Liu, X. Liu, Q. Du, H. Wang, Y. Wu, S. Jiang, Z. Lu, Local chemical ordering and its impact on radiation damage behavior of multi-principal element alloys, J. Mater. Sci. Technol. 135 (2023) 13–25

    2023

  12. [21]

    Y. Liu, T. Zhao, W. Ju, S. Shi, Materials discovery and design using machine learning, J. Mater. 3 (2017) 159–177

    2017

  13. [22]

    Schmidt, M.R.G

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

    2019

  14. [23]

    C.H. Chan, M. Sun, B. Huang, Application of machine learning for advanced material prediction and design, EcoMat (2022)

    2022

  15. [24]

    Behler, Constructing high-dimensional neural network potentials: A tutorial review, Int

    J. Behler, Constructing high-dimensional neural network potentials: A tutorial review, Int. J. Quantum Chem. 115 (2015) 1032–1050

    2015

  16. [25]

    Zhang, J

    L. Zhang, J. Han, H. Wang, R. and W. Car, Deep potential molecular dynamics: A scalable model with the accuracy of quantum mechanics, Phys. Rev. Lett. 120 (2018)

    2018

  17. [26]

    Ahmed, Y

    M. Ahmed, Y. Li, W. Chen, E.-P. Li, Diffusion Barrier Prediction of Graphene and Boron Nitride for Copper Interconnects by Deep Learning, IEEE Access 8 (2020) 210542– 210549. https://doi.org/10.1109/access.2020.3039257

    2020

  18. [27]

    Z. Cao, Y. Dan, Z. Xiong, C. Niu, X. Li, S. Qian, J. Hu, Convolutional Neural Networks for Crystal Material Property Prediction Using Hybrid Orbital-Field Matrix and Magpie Descriptors, Crystals (Basel) 9 (2019) 191. https://doi.org/10.3390/cryst9040191

    2019 doi

  19. [28]

    Cho, L.-C

    E.H. Cho, L.-C. Lin, Nanoporous Material Recognition via 3D Convolutional Neural Networks: Prediction of Adsorption Properties, J Phys Chem Lett 12 (2021) 2279–2285. https://doi.org/10.1021/acs.jpclett.1c00293

    2021 doi

  20. [29]

    Sathana, M

    V. Sathana, M. Mathumathi, K. Makanyadevi, Prediction of material property using optimized augmented graph-attention layer in GNN, Mater. Today 69 (2022) 1419–1424

    2022

  21. [30]

    S. Chen, A. Wulamu, Q. Zou, L. Zheng Han and Wen, X. Guo, H. Chen, T. Zhang, Y. Zhang, MD-GNN: A mechanism-data-driven graph neural network for molecular properties prediction and new material discovery, J. Mol. Graph. Model. 123 (2023) 108506

    2023

  22. [31]

    Y. Cui, J. Zhu, W. Zhou, H. Zang, Y. Ren, J. Xu, S. Zhan, W. Ma, SA-GNN: Prediction of material properties using graph neural network based on multi-head self-attention optimization, AIP Adv. 14 (2024)

    2024

  23. [32]

    C. Chen, W. Ye, Y. Zuo, C. Zheng, S.P. Ong, Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals, Chemistry of Materials 31 (2019) 3564–

    2019

  24. [33]

    Rahaman, A

    O. Rahaman, A. Gagliardi, Deep Learning Total Energies and Orbital Energies of Large Organic Molecules Using Hybridization of Molecular Fingerprints, J Chem Inf Model 60 (2020) 5971–5983. https://doi.org/10.1021/acs.jcim.0c00687

    2020 doi

  25. [34]

    Cervino, L

    J. Cervino, L. Ruiz, A. Ribeiro, Training Stable Graph Neural Networks Through Constrained Learning, in: ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, 2022: pp. 4223–4227. https://doi.org/10.1109/icassp43922.2022.9746912

    2022

  26. [35]

    C.W. Park, M. Kornbluth, J. Vandermause, C. Wolverton, B. Kozinsky, J.P. Mailoa, Accurate and scalable graph neural network force field and molecular dynamics with direct force architecture, Npj Comput. Mater. 7 (2021)

    2021

  27. [36]

    T.N. Kipf, M. Welling, Semi-supervised classification with graph convolutional networks, (2016)

    2016

  28. [37]

    Kingma, J

    D.P. Kingma, J. Ba, Adam: A method for stochastic optimization, (2014)

    2014

  29. [38]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, others, Pytorch: An imperative style, high-performance deep learning library, Adv Neural Inf Process Syst 32 (2019)

    2019

  30. [39]

    Fey, J.E

    M. Fey, J.E. Lenssen, Fast graph representation learning with PyTorch Geometric, (2019)

    2019

  31. [40]

    Q.-J. Li, H. Sheng, E. Ma, Strengthening in multi-principal element alloys with local- chemical-order roughened dislocation pathways, Nat. Commun. 10 (2019) 3563

    2019

  32. [41]

    S. Ma, J. Zhang, B. Xu, Y. Xiong, W. Shao, S. Zhao, Chemical short-range ordering regulated dislocation cross slip in high-entropy alloys, J. Alloys Compd. 911 (2022) 165144

    2022

  33. [42]

    Jarlöv, W

    A. Jarlöv, W. Ji, R. Babicheva, Y. Tian, Z. Hu, H.L. Seet, L. Tan, F. Liu, Y. Liu, M.L.S. Nai, U. Ramamurty, K. Zhou, Tailoring short-range order and dislocation evolution in Cr– Co–Ni medium-entropy alloys: A molecular dynamics study, Mater. Des. 240 (2024) 112840

    2024

  34. [43]

    Huang, X

    G. Huang, X. Zhang, Z. Xie, W.-R. Jian, R. Zhang, X. Yao, Effects of lattice distortion and chemical short-range order on creep behavior of medium-entropy alloy CoCrNi, Mech. Mater. 177 (2023) 104549

    2023

  35. [44]

    Naghdi, K

    A.H. Naghdi, K. Karimi, A.E. Poisvert, A. Esfandiarpour, R. Alvarez, P. Sobkowicz, M. Alava, S. Papanikolaou, Dislocation plasticity in equiatomic NiCoCr alloys: Effect of short-range order, Phys. Rev. B. 107 (2023)

    2023

  36. [45]

    Hirel, Atomsk: A tool for manipulating and converting atomic data files, Comput

    P. Hirel, Atomsk: A tool for manipulating and converting atomic data files, Comput. Phys. Commun. 197 (2015) 212–219

    2015

  37. [46]

    Plimpton, Fast parallel algorithms for short-range molecular dynamics, J

    S. Plimpton, Fast parallel algorithms for short-range molecular dynamics, J. Comput. Phys. 117 (1995) 1–19

    1995

  38. [47]

    Chen, Z.H

    S. Chen, Z.H. Aitken, Z. Pattamatta Subrahmanyam and Wu, Z.G. Yu, D.J. Srolovitz, P.K. Liaw, Y.-W. Zhang, Simultaneously enhancing the ultimate strength and ductility of high- entropy alloys via short-range ordering, Nat. Commun. 12 (2021) 4953

    2021

  39. [48]

    Stukowski, Visualization and analysis of atomistic simulation data with OVITO–the Open Visualization Tool, Model

    A. Stukowski, Visualization and analysis of atomistic simulation data with OVITO–the Open Visualization Tool, Model. Simul. Mat. Sci. Eng. 18 (2010) 15012

    2010

  40. [49]

    Cowley, An approximate theory of order in alloys, Phys

    J.M. Cowley, An approximate theory of order in alloys, Phys. Rev. 77 (1950) 669–675

    1950

  41. [50]

    Sheriff, Y

    K. Sheriff, Y. Cao, T. Smidt, R. Freitas, Quantifying chemical short-range order in metallic alloys, (n.d.). https://doi.org/10.1073/pnas.2322962121

    doi

  42. [51]

    K. Noda, Y. Shibuta, Prediction of potential energy profiles of molecular dynamic simulation by graph convolutional networks, Comput. Mater. Sci. 229 (2023) 112448. List of Figure Captions Fig. 1 Diagram of the machine learning model architecture built on Graph Convolutional N...

    2023

  43. [3572]

    https://doi.org/10.1021/acs.chemmater.9b01294

    doi

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.