Pith. sign in

REVIEW 4 major objections 4 minor 51 references

Large Language Models for Material Property Predictions: elastic constant tensor prediction and materials design

T0 review · 4 major / 4 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 Fine-tuned on natural-language descriptions, a large language model predicts elastic constant tensors with a 2.32 GPa mean absolute error and beats dedicated materials-science models.

desk verdict A serious LLM-for-properties benchmark with a real evaluation gap: the headline MAE is diluted by symmetry-zero tensor components, so per-component or nonzero-only errors are needed before the SOTA claim is credible. read the letter →

arxiv 2411.12280 v1 pith:D5ROK5BJ submitted 2024-11-19 cond-mat.mtrl-sci physics.comp-ph

Siyu Liu , Tongqi Wen , Beilin Ye , Zhuoyuan Li , David J. Srolovitz This is my paper · ORCID
classification cond-mat.mtrl-sciphysics.comp-ph
keywords elasticconstanttensorlargelanguagemodelsmaterialspropertypredictionProjectbulkmodulusretrieval-augmentedgenerationinversedesignnaturalprompts
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to show that a large language model can take the place of specialized machine-learning pipelines for predicting elastic constant tensors, and can do so from plain-language descriptions of a material's composition and crystal structure. The authors build ElaTBot-DFT, a fine-tuned LLM that reaches a mean absolute error of 2.32 GPa and $R^2=0.965$ on the average Voigt component of the 0 K test set, cutting the error of a domain-specific materials LLM by 33.1% and of an equivariant graph neural network by 10.4%. A second model, ElaTBot, extends the same recipe to finite-temperature bulk modulus and to generating candidate chemical formulas for a target stiffness, and retrieval-augmented generation lets it incorporate new data without retraining. A sympathetic reader would care because elastic constant data are scarce, experimental measurements are expensive, and the result suggests that non-specialists can get state-of-the-art predictions through natural-language prompts rather than hand-built descriptors.

What carries the argument

The load-bearing object is the input representation: a prompt that aligns a textual description of chemical composition with a textual description of crystal structure (the paper's prompt type 4), generated by standard crystallographic tools. The paper shows this combined text prompt outperforms JSON structure input, composition-only text, and structure-only text, reducing MAE by 29.7%, 13.1%, and 5.3% respectively, so the alignment of composition and structure carries the predictive power. The evaluation target is the scalar average $\overline{C_{ij}} = \frac{1}{36}\sum_{i,j} C_{ij}$ over the 36 Voigt components, and the training machinery is parameter-efficient LoRA+ fine-tuning of the Llama2-7b base model, followed by knowledge-fusion training across four tasks for the finite-temperature variant.

What would settle it

Run ElaTBot-DFT on the same 522-material test set and compute the mean absolute error separately for each Voigt component, especially the off-diagonal and shear entries; if those per-component errors are much larger than the 2.32 GPa average implies, or if the average itself cannot be reproduced with the released code and data, the central claim that the LLM reduces elastic tensor prediction error would be refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that a general LLM fine-tuned on text describing both composition and crystal structure can predict the full elastic constant tensor at 0 K more accurately than dedicated materials-science models trained on the same data: on the 522-material test set, ElaTBot-DFT achieves a mean absolute error of 2.32 GPa and $R^2 = 0.965$ for the average Voigt component $\overline{C_{ij}}$, a 33.1% reduction relative to the Darwin materials LLM and a 10.4% reduction relative to the MatTen equivariant graph neural network. The model also produces symmetry-correct zeros in 94% of test tensors under a $\pm 2$ GPa tolerance, versus 5% for MatTen and 7% for a random forest. The same text-to-text formulation, augmented with finite-temperature data and multi-task training, yields ElaTBot, which predicts the temperature-dependent bulk modulus of unseen multicomponent alloys and generates candidate chemical formulas for target stiffness, with retrieval-augmented generation cutting bulk modulus error on nine alloy-temperature cases from 27.49% to 0.95%.

Load-bearing premise

The load-bearing premise is that the single scalar $\overline{C_{ij}}$, the average of all 36 Voigt components, is the right target for evaluating elastic tensor prediction; the paper reports no per-component errors, so if the model is accurate on the average but substantially worse on off-diagonal or shear components, its usefulness for anisotropy-sensitive materials design would not be established.

Editorial extensions

If this is right

  • Elastic constant tensor prediction can be posed as natural-language regression: a prompt combining composition and structure text yields a 33.1% lower MAE than the Darwin materials LLM and a 10.4% lower MAE than the MatTen graph neural network on the 0 K test set.
  • A single fine-tuned LLM can handle multiple tasks beyond regression, including finite-temperature bulk modulus prediction and generation of candidate chemical formulas for a target stiffness.
  • Retrieval-augmented generation lets the model absorb new data without retraining, reducing bulk modulus error from 27.49% to 0.95% on nine alloy-temperature combinations.
  • Pairing the domain-specific LLM with a general LLM can screen candidate materials for concrete application targets, such as bone-like modulus implants, soft-robot exoskeletons, corrosion-resistant alloys in saltwater, and battery protective layers.

Reading between the lines

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

  • An inference the paper does not draw: if per-component errors are comparable to the averaged MAE, LLM-based tensor prediction becomes a practical tool for anisotropy-sensitive design, but the aggregate metric reported here does not establish that.
  • A further inference: the model's tendency to output exact zeros where symmetry requires them suggests text-based training captures crystal symmetry implicitly, pointing toward future models that enforce tensor symmetries without explicit constraints.
  • The RAG results hint at an active-learning extension, not explored in the paper, where the LLM retrieves fresh DFT results and proposes new compounds to compute, closing a generation-prediction-verification loop.
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

4 major / 4 minor

Summary. This paper presents ElaTBot, a Llama2-7b-based LLM fine-tuned on natural-language descriptions of chemical composition and crystal structure, for predicting elastic constant tensors and related properties. The authors report that ElaTBot-DFT achieves a mean absolute error (MAE) of 2.32 GPa and R² = 0.965 on the average Voigt-matrix component Cij on a held-out Materials Project test set, and that it outperforms Darwin, MatTen, and a random forest model trained on the same splits by 33.1%, 10.4%, and 31.8%, respectively. The paper also demonstrates finite-temperature bulk modulus prediction, retrieval-augmented generation (RAG) for prediction, and composition generation combined with GPT-4o screening. The empirical protocol is a held-out comparison against baselines on a public dataset, with a systematic prompt-type ablation and a symmetry-accuracy check.

Significance. If the central claim were supported, the paper would be a valuable demonstration that fine-tuned LLMs with textual inputs can be competitive with graph neural networks and traditional ML for a tensorial property, while adding interactive and generative capabilities. The paper has several concrete strengths: a public benchmark with identical train/validation/test splits across models, a careful prompt-type comparison, a symmetry check, and a clearly described RAG pipeline. However, the headline accuracy is computed on a scalar average over all 36 Voigt entries, many of which are symmetry-enforced zeros, and no per-component or independent-constant metrics are reported. The significance of the claimed state-of-the-art result therefore cannot be assessed from the current manuscript, because the actual accuracy on the physically relevant nonzero elastic constants is unreported.

major comments (4)
  1. [Methods, definition of Cij and MAE/R2; Fig. 2] The headline metric is Cij = (1/36) Σ_i Σ_j Cij, the average over all 36 Voigt entries. For high-symmetry crystals, many of these entries are identically zero by symmetry (e.g., 27 of 36 entries are zero for cubic crystals), and a model that outputs zero for those positions receives zero error on them. The reported Cij MAE of 2.32 GPa versus the bulk-modulus MAE of 7.74 GPa (Fig. 2) is exactly the pattern expected if errors on the nonzero elastic constants are several times larger than the diluted average. Because the comparisons with Darwin, MatTen, and random forest are all made on this same diluted scalar, the claimed 33.1%, 10.4%, and 31.8% reductions do not establish that ElaTBot-DFT predicts the physically meaningful independent elastic constants better. I request per-component MAE and R² for the nonzero Voigt entries, or equivalently for the independent elastic constants, together with the same comparisons on those metrics. The symmetry check in Fig. 2(c) only tests that symmetry-required entries are near zero, so it also rewards trivial zero prediction and does not measure accuracy of nonzero entries.
  2. [Methods, Model training and evaluation; Fig. 2] No measure of run-to-run variability is reported. Each comparison appears to be a single fine-tuning run, and no random seeds, confidence intervals, or multiple-run statistics are given. The claimed improvements over MatTen are modest (10.4% MAE and 0.4% R²), and without variance estimates it is not possible to determine whether the difference is significant. At minimum, I request results over at least three seeds for the LLM variants (and for MatTen if its training is stochastic), or a paired bootstrap over test materials, so that the state-of-the-art claim is statistically grounded.
  3. [Elastic Constant Tensor Predictions; Methods, finite-temperature evaluation] The finite-temperature evaluation is performed on only three alloy compositions, with no held-out test split from the 1,266 finite-temperature samples, and the predicted points are linearly fitted before errors are computed. The linear fit is an additional model applied after prediction, so the reported errors and the comparison with the 0K test-set average do not characterize raw prediction accuracy. Please report the raw predicted values, the fitted-curve error separately, and ideally a held-out set of finite-temperature measurements.
  4. [Material Discovery; Fig. 4] The materials discovery examples in Fig. 4 and the accompanying text present GPT-4o-screened compositions as candidate materials for orthopedic, soft-robot, corrosion-resistant, and battery applications, but none of these proposed compositions is validated by DFT or experiment. As presented, this section is a pipeline demonstration, and the claims about discovering materials for specific applications should be framed accordingly, with explicit statements that the proposed compositions are unverified suggestions.
minor comments (4)
  1. [Abstract and throughout] There are several typographical issues, including 'rapid-evolution' in the abstract and 'V oigt' for 'Voigt' in multiple places; these should be corrected.
  2. [Methods, notation] The symbol Cij is used both for an individual Voigt component and for the average (1/36)ΣCij; using an overbar or a different symbol for the average would remove ambiguity.
  3. [Code and data availability] The statements 'will be publicly accessible on GitHub' and 'provided upon reasonable requests' are not reproducible commitments; please provide a repository URL or DOI, and make the trained model weights or exact training configurations available.
  4. [Fig. 2 caption] The caption text 'ElaTBot-DFT (Llama2-7b + prompt4): Not pre-trained with materials knowledge' is ambiguous because Llama2-7b is a pre-trained general LLM; clarify that it was not pre-trained on materials-specific text.

Circularity Check

1 steps flagged · score 1.0 of 10

Core elastic-constant benchmark is independent and externally anchored; a minor RAG evaluation leak (same material at nearby temperatures left in the retrieval database) is the only mild circularity.

  1. other [RAG Enhanced Predictions, paragraph 2 (Fig. 3(a))]
    "Since the finite temperature data was not in the original ElaTBot training set, the model automatically queries our external database, finds bulk modulus data for γ−TiAl at similar temperatures (the 170 K data was removed from the database for comparison purposes). The predicted value (110.77 GPa) differs by only 0.1% from the true value, whereas without RAG, the error increases to 2.4%."

    The RAG 'prediction' for γ-TiAl at 170 K is evaluated after deleting only the exact target row, while retaining rows for the same material at nearby temperatures. Since bulk modulus is a smooth function of temperature, the retrieved nearest-temperature value for the same composition essentially fixes the answer, so the reported 0.1% error is an interpolation/retrieval result rather than an independent model prediction. The large drop from 27.49% to 0.95% is thus partly by construction of the retrieval database, not by ElaTBot's learned reasoning. This is a mild data-leakage/evaluation-protocol issue, not a self-definitional derivation, and it does not affect the 0 K ElaTBot-DFT benchmark.

full rationale

The paper's central claim—that ElaTBot-DFT reduces MAE by 33.1% (average Cij) and 31.8% (bulk modulus) versus Darwin and outperforms MatTen/random forest—is an empirical comparison on a fixed Materials Project test set with all models trained on the same data. No derivation-from-assumptions or fitted-parameter-renamed-as-prediction step appears in that benchmark: the LLM is trained with cross-entropy token loss, and the evaluation metric is applied symmetrically to all models. The self-citations in the paper (e.g., Ref. [13] for prompt engineering) are contextual, not load-bearing. The use of Cij_bar = (1/36) * sum(Cij), which includes symmetry-enforced zeros, is a metric-validity concern that could inflate apparent accuracy, but it is not a circularity because the same diluted scalar is used for every model and is not the training objective. The only mild circular/leaky element is the RAG demonstration, where retrieved data for the same material at nearby temperatures makes the 'prediction' nearly a lookup; this is acknowledged in the text and is peripheral to the main benchmark, so the overall circularity score is low.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claims rest on the fidelity of public DFT and handbook data, on the representativeness of the random split, and on the adequacy of the scalar average Cij_bar as a stand-in for full tensor quality. The free parameters are training hyperparameters and the linear-fit post-processing used in the finite-temperature evaluation. No new physical entities are introduced; ElaTBot and ElaTBot-DFT are trained models, not postulated objects.

free parameters (3)
  • LoRA+ hyperparameters (rank, alpha, learning rates)
    Chosen by validation performance; central training configuration; exact values only in supplementary material, not auditable here.
  • Prompt template type 4
    Hand-designed textual description combining composition and structure; selected over types 1-3 by validation MAE; all downstream results depend on this representation.
  • Linear-fit coefficients for finite-temperature evaluation
    Predicted bulk modulus points are linearly fitted before comparing with experiment, smoothing the model's discrete outputs; the reported finite-T agreement depends on this post-processing.
assumptions (4)
  • domain assumption Materials Project DFT elastic constants are treated as exact 0 K ground truth.
    All 0K training and test labels come from MP DFT; systematic DFT errors are inherited by the model and by the benchmark.
  • domain assumption The random 9498/500/522 split is representative and has no material leakage.
    No deduplication of similar compositions or structures across train/test is reported; leakage would inflate the claimed MAE and R-squared.
  • domain assumption Textual descriptions preserve the elasticity-relevant information in the crystal structure.
    ElaTBot sees only text generated by pymatgen and robocrystallographer; any lost symmetry or geometric detail cannot be recovered, which matters for the comparison with structure-based GNNs.
  • domain assumption Experimental elastic data from Simmons and Wang is reliable enough to serve as finite-temperature ground truth.
    This handbook supplies the 1,266 finite-temperature training points; old handbook measurements may contain inconsistencies, and the comparison alloys are assessed against later experiments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Large Language Models for Material Property Predictions: elastic constant tensor prediction and materials design." pith.science (2026). https://pith.science/paper/D5ROK5BJ

@misc{pith2026241112280,
  author       = {Pith},
  title        = {Pith review of: Large Language Models for Material Property Predictions: elastic constant tensor prediction and materials design},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/D5ROK5BJ}},
  note         = {Machine review of arXiv:2411.12280}
}
read the original abstract

Efficient and accurate prediction of material properties is critical for advancing materials design and applications. The rapid-evolution of large language models (LLMs) presents a new opportunity for material property predictions, complementing experimental measurements and multi-scale computational methods. We focus on predicting the elastic constant tensor, as a case study, and develop domain-specific LLMs for predicting elastic constants and for materials discovery. The proposed ElaTBot LLM enables simultaneous prediction of elastic constant tensors, bulk modulus at finite temperatures, and the generation of new materials with targeted properties. Moreover, the capabilities of ElaTBot are further enhanced by integrating with general LLMs (GPT-4o) and Retrieval-Augmented Generation (RAG) for prediction. A specialized variant, ElaTBot-DFT, designed for 0 K elastic constant tensor prediction, reduces the prediction errors by 33.1% compared with domain-specific, material science LLMs (Darwin) trained on the same dataset. This natural language-based approach lowers the barriers to computational materials science and highlights the broader potential of LLMs for material property predictions and inverse design.

Figures

Figures reproduced from arXiv: 2411.12280 by the authors.

Figure 1
Figure 1. FIG. 1. Datasets and overview of the ElaTBot for predicting elastic properties. (a) Comparison of the number of materials in the Materials [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Prediction abilities of ElaTBot-DFT and ElaTBot. (a, b) Performance comparison of the Llama2-7b model using different prompt [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Integration of Retrieval-Augmented Generation (RAG) with ElaTBot for enhanced prediction. (a) The steps involved in enabling [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Integration of a domain-specific LLM (ElaTBot) with a general LLM (GPT-4o) for materials discovery. The process begins by [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Dataset and model architecture for ElaTBot and ElaTBot-DFT. (a, b) Data distribution for Materials Project (MP) 0K DFT dataset [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 34 canonical work pages

  1. [1]

    L. Hu, P. L. Chee, S. Sugiarto, Y . Yu, C. Shi, R. Yan, Z. Yao, X. Shi, J. Zhi, D. Kai, H.-D. Yu, and W. Huang, Hydrogel-based flexible electronics, Advanced Materials 35, 2205326 (2023)

    work page 2023

  2. [2]

    J. Luo, D. Zou, Y . Wang, S. Wang, and L. Huang, Battery ther- mal management systems (btms) based on phase change ma- terial (pcm): A comprehensive review, Chemical Engineering Journal 430, 132741 (2022)

    work page 2022

  3. [3]

    Chaves, J

    A. Chaves, J. G. Azadani, H. Alsalman, D. R. da Costa, R. Frisenda, A. J. Chaves, S. H. Song, Y . D. Kim, D. He, J. Zhou, A. Castellanos-Gomez, F. M. Peeters, Z. Liu, C. L. Hinkle, S.-H. Oh, P. D. Ye, S. J. Koester, Y . H. Lee, P. Avouris, X. Wang, and T. Low, Bandgap engineering of two-dimensional semiconductor materials, npj 2D Materials and Application...

    work page 2020

  4. [4]

    Wu, H.-T

    C.-T. Wu, H.-T. Chang, C.-Y . Wu, S.-W. Chen, S.-Y . Huang, M. Huang, Y .-T. Pan, P. Bradbury, J. Chou, and H.-W. Yen, Ma- chine learning recommends affordable new ti alloy with bone- like modulus, Materials Today 34, 41 (2020)

    work page 2020

  5. [5]

    M. Fan, T. Wen, S. Chen, Y . Dong, and C.-A. Wang, Per- spectives toward damage-tolerant nanostructure ceramics, Ad- vanced Science 11, 2309834 (2024)

    work page 2024

  6. [6]

    M. Du, Y . Chen, Z. Wang, L. Nie, and D. Zhang, Large lan- guage models for automatic equation discovery of nonlinear dynamics, Physics of Fluids 36, 097121 (2024)

    work page 2024

  7. [7]

    Zheng, O

    Z. Zheng, O. Zhang, H. L. Nguyen, N. Rampal, A. H. Alawadhi, Z. Rong, T. Head-Gordon, C. Borgs, J. T. Chayes, and O. M. Yaghi, Chatgpt research group for optimizing the crystallinity of mofs and cofs, ACS Central Science 9, 2161 (2023)

    work page 2023

  8. [8]

    A. M. Bran, S. Cox, O. Schilter, C. Baldassari, A. D. White, and P. Schwaller, Augmenting large language models with chem- istry tools, Nature Machine Intelligence 6, 525 (2024)

    work page 2024

Show all 51 references
  1. [9]

    L. M. Antunes, K. T. Butler, and R. Grau-Crespo, Crystal struc- ture generation with autoregressive large language modeling, arXiv:2307.04340 (2024)

    2024 arXiv

  2. [10]

    Verma, M.-H

    P. Verma, M.-H. Van, and X. Wu, Beyond human vision: The role of large vision language models in microscope image anal- ysis, arXiv:2405.00876 (2024)

    2024 arXiv

  3. [11]

    Z. Ren, Z. Ren, Z. Zhang, T. Buonassisi, and J. Li, Autonomous experiments using active learning and ai, Nature Reviews Ma- terials 8, 563 (2023)

    2023

  4. [12]

    Dagdelen, A

    J. Dagdelen, A. Dunn, S. Lee, N. Walker, A. S. Rosen, G. Ceder, K. A. Persson, and A. Jain, Structured information extraction from scientific text with large language models, Nature Com- munications 15, 1418 (2024)

    2024

  5. [13]

    S. Liu, T. Wen, A. S. Pattamatta, and D. J. Srolovitz, A prompt- engineered large language model, deep learning workflow for materials classification, Materials Today, In Press (2024)

    2024

  6. [14]

    Ni and M

    B. Ni and M. J. Buehler, Mechagents: Large language model multi-agent collaborations can solve mechanics problems, gen- erate new data, and integrate knowledge, Extreme Mechanics Letters 67, 102131 (2024)

    2024

  7. [15]

    Zhang, K

    Q. Zhang, K. Ding, T. Lyv, X. Wang, Q. Yin, Y . Zhang, J. Yu, Y . Wang, X. Li, Z. Xiang, et al. , Scientific large lan- guage models: A survey on biological & chemical domains, arXiv:2401.14656 (2024)

    2024 arXiv

  8. [16]

    Ouyang, Z

    S. Ouyang, Z. Zhang, B. Yan, X. Liu, Y . Choi, J. Han, and L. Qin, Structured chemistry reasoning with large language models, arXiv:2311.09656 (2024)

    2024 arXiv

  9. [17]

    S. Yu, N. Ran, and J. Liu, Large-language models: The game- changers for materials science research, Artificial Intelligence Chemistry 2, 100076 (2024)

    2024

  10. [18]

    A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, S. Dacek, S. Cholia, D. Gunter, D. Skinner, G. Ceder, and K. A. Persson, Commentary: The Materials Project: A materi- als genome approach to accelerating materials innovation, APL Materials 1, 011002 (2013)

    2013

  11. [19]

    Jacobs, M

    R. Jacobs, M. P. Polak, L. E. Schultz, H. Mahdavi, V . Honavar, and D. Morgan, Regression with large language models for materials and molecular property prediction, arXiv:2409.06080 (2024)

    2024 arXiv

  12. [20]

    K. M. Jablonka, P. Schwaller, A. Ortega-Guerrero, and B. Smit, Leveraging large language models for predictive chemistry, Na- ture Machine Intelligence 6, 161 (2024)

    2024

  13. [21]

    P. R. Spackman, A. Grosjean, S. P. Thomas, D. P. Karothu, P. Naumov, and M. A. Spackman, Quantifying mechanical properties of molecular crystals: A critical overview of exper- imental elastic tensors, Angewandte Chemie International Edi- tion 61, e202110716 (2022)

    2022

  14. [22]

    Levine, J

    J. Levine, J. Betts, J. Garrett, S. Guo, J. Eng, A. Migliori, and R. Kaner, Full elastic tensor of a crystal of the superhard com- pound reb2, Acta Materialia 58, 1530 (2010)

    2010

  15. [23]

    M. Wen, M. K. Horton, J. M. Munro, P. Huck, and K. A. Persson, An equivariant graph neural network for the elasticity tensors of all seven crystal systems, Digital Discovery 3, 869 (2024)

    2024

  16. [24]

    T. Xie, Y . Wan, W. Huang, Z. Yin, Y . Liu, S. Wang, Q. Linghu, C. Kit, C. Grazian, W. Zhang, I. Razzak, and B. Hoex, Dar- win series: Domain specific large language models for natural 11 science, arXiv:2308.13565 (2023)

    2023 arXiv

  17. [25]

    Wallow, G

    F. Wallow, G. Neite, W. Schr ¨oer, and E. Nembach, Stiffness constants, dislocation line energies, and tensions of ni3al and of the gamma’-phases of nimonic 105 and of nimonic pe16, Physica Status Solidi (A) 99, 483 (1987)

    1987

  18. [26]

    Tanaka, Single-crystal elastic constants of gamma-tial, Philosophical Magazine Letters 73, 71 (1996)

    K. Tanaka, Single-crystal elastic constants of gamma-tial, Philosophical Magazine Letters 73, 71 (1996)

    1996

  19. [27]

    Y . Li, L. V o ˇcadlo, and J. P. Brodholt, Elast: A toolkit for thermoelastic calculations, Computer Physics Communications 273, 108280 (2022)

    2022

  20. [28]

    V . Revi, S. Kasodariya, A. Talapatra, G. Pilania, and A. Alankar, Machine learning elastic constants of multi- component alloys, Computational Materials Science 198, 110671 (2021)

    2021

  21. [29]

    Vazquez, P

    G. Vazquez, P. Singh, D. Sauceda, R. Couperthwaite, N. Britt, K. Youssef, D. D. Johnson, and R. Arr´oyave, Efficient machine- learning model for fast assessment of elastic properties of high- entropy alloys, Acta Materialia 232, 117924 (2022)

    2022

  22. [30]

    Pakornchote, A

    T. Pakornchote, A. Ektarawong, and T. Chotibut, Straintensor- net: Predicting crystal structure elastic properties using se(3)- equivariant graph neural networks, Physical Review Research 5, 043198 (2023)

    2023

  23. [31]

    Lewis, E

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V . Karpukhin, N. Goyal, H. K ¨uttler, M. Lewis, W.-t. Yih, T. Rockt ¨aschel, et al., Retrieval-augmented generation for knowledge-intensive nlp tasks, Advances in Neural Information Processing Systems 33, 9459 (2020)

    2020

  24. [32]

    Gruver, A

    N. Gruver, A. Sriram, A. Madotto, A. G. Wilson, C. L. Zitnick, and Z. Ulissi, Fine-tuned language models generate stable inor- ganic materials as text, arXiv:2402.04379 (2024)

    2024 arXiv

  25. [33]

    L. Ward, A. Agrawal, A. Choudhary, and C. Wolverton, A general-purpose machine learning framework for predicting properties of inorganic materials, npj Computational Materials 2, 16028 (2016)

    2016

  26. [34]

    T. Xie, Y . Wan, W. Huang, Y . Zhou, Y . Liu, Q. Linghu, S. Wang, C. Kit, C. Grazian, W. Zhang, et al., Large language models as master key: unlocking the secrets of materials science with gpt, arXiv:2304.02213 (2023)

    2023 arXiv

  27. [35]

    Li and H.-Y

    C.-A. Li and H.-Y . Lee, Examining forgetting in continual pre- training of aligned large language models, arXiv:2401.03129 (2024)

    2024 arXiv

  28. [36]

    Gorejov ´a, L

    R. Gorejov ´a, L. Haverov ´a, R. Ori ˇnakov´a, A. Ori ˇnak, and M. Ori ˇnak, Recent advancements in fe-based biodegradable materials for bone repair, Journal of Materials Science54, 1913 (2019)

    2019

  29. [37]

    T. E. ToolBox, Metals and alloys - bulk modulus elasticity (2008), accessed: 26 September 2024

    2008

  30. [38]

    Z. Deng, Z. Wang, I.-H. Chu, J. Luo, and S. P. Ong, Elastic properties of alkali superionic conductor electrolytes from first principles calculations, Journal of The Electrochemical Society 163, A67 (2015)

    2015

  31. [39]

    Kaplan, S

    J. Kaplan, S. McCandlish, T. Henighan, T. B. Brown, B. Chess, R. Child, S. Gray, A. Radford, J. Wu, and D. Amodei, Scaling laws for neural language models, arXiv:2001.08361 (2020)

    2020 arXiv

  32. [40]

    Ghafarollahi and M

    A. Ghafarollahi and M. J. Buehler, Atomagents: Alloy design and discovery through physics-aware multi-modal multi-agent artificial intelligence, arXiv:2407.10022 (2024)

    2024 arXiv

  33. [41]

    Golkar, M

    S. Golkar, M. Pettee, M. Eickenberg, A. Bietti, M. Cranmer, G. Krawezik, F. Lanusse, M. McCabe, R. Ohana, L. Parker, B. R.-S. Blancard, T. Tesileanu, K. Cho, and S. Ho, xval: A continuous number encoding for large language models, arXiv:2309.02989 (2023)

    2023 arXiv

  34. [42]

    S. P. Ong, W. D. Richards, A. Jain, G. Hautier, M. Kocher, S. Cholia, D. Gunter, V . L. Chevrier, K. A. Persson, and G. Ceder, Python materials genomics (pymatgen): A robust, open-source python library for materials analysis, Computa- tional Materials Science 68, 314 (2013)

    2013

  35. [43]

    A. M. Ganose and A. Jain, Robocrystallographer: automated crystal structure text descriptions and analysis, MRS Commu- nications 9, 874 (2019)

    2019

  36. [44]

    L. Ward, A. Dunn, A. Faghaninia, N. E. Zimmermann, S. Bajaj, Q. Wang, J. Montoya, J. Chen, K. Bystrom, M. Dylla, K. Chard, M. Asta, K. A. Persson, G. J. Snyder, I. Foster, and A. Jain, Matminer: An open source toolkit for materials data mining, Computational Materials Science ...

    2018

  37. [45]

    Simmons and H

    G. Simmons and H. Wang, Single crystal elastic constants and calculated aggregate properties: a handbook (M.I.T. Press, Cambridge, Mass., 1971)

    1971

  38. [46]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, A. M ¨uller, J. Nothman, G. Louppe, P. Prettenhofer, R. Weiss, V . Dubourg, J. Vander- plas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and ´Edouard Duchesnay, Scikit-learn: Machine l...

    2018 arXiv

  39. [47]

    Hayou, N

    S. Hayou, N. Ghosh, and B. Yu, Lora+: Efficient low rank adap- tation of large models, arXiv:2402.12354 (2024)

    2024 arXiv

  40. [48]

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, and W. Chen, Lora: Low-rank adaptation of large lan- guage models, arXiv:2106.09685 (2021)

    2021 arXiv

  41. [49]

    Zheng, R

    Y . Zheng, R. Zhang, J. Zhang, Y . Ye, and Z. Luo, Llamafac- tory: Unified efficient fine-tuning of 100+ language models, arXiv:2403.13372 (2024)

    2024 arXiv

  42. [50]

    A. Abid, A. Abdalla, A. Abid, D. Khan, A. Alfozan, and J. Zou, Gradio: Hassle-free sharing and testing of ml models in the wild, arXiv:1906.02569 (2019)

    2019 arXiv

  43. [51]

    Pandya and M

    K. Pandya and M. Holia, Automating customer service using langchain: Building custom open-source gpt chatbot for orga- nizations, arXiv:2310.05421 (2023). Acknowledgments This work is supported by Research Grants Council, Hong Kong SAR through the General Research Fund (172107...

    2023 arXiv

Pith tools

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