Pith. sign in

REVIEW 3 major objections 5 minor 58 references

Thermodynamic Prediction Enabled by Automatic Dataset Building and Machine Learning

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read An LLM-powered extraction pipeline can build thermodynamic datasets from legacy papers, and the CatBoost model trained on them predicts mineral formation enthalpies with a mean absolute error of 46.5 kJ/mol.

desk verdict Useful data-mining pipeline, but the ML prediction claim is inflated by identity leakage from the random split and mineral-type features. read the letter →

arxiv 2507.07293 v1 pith:BQQOWBXT submitted 2025-07-09 cond-mat.mtrl-sci cs.LG

Juejing Liu , Haydn Anderson , Noah I. Waxman , Vsevolod Kovalev , Byron Fisher , Elizabeth Li , Xiaofeng Guo This is my paper · ORCID
classification cond-mat.mtrl-scics.LG
keywords thermodynamicsmachinelearninglargelanguagemodelsdataminingrare-earthmineralsenthalpyofformationCatBoostdatabaseconstruction
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 the two slowest parts of thermodynamic research—reading decades of literature and predicting properties for unmeasured minerals—can be automated and linked. It builds LMExt, an LLM-based pipeline that turns PDF papers into machine-readable datasets, and uses it to assemble a rare-earth-element mineral thermodynamic dataset. On that dataset it trains a CatBoost model that predicts standard enthalpy of formation with a mean absolute error of 46.5 kJ/mol, with 50.5% of test predictions within 15 kJ/mol of the reported values. If this holds, a researcher could obtain a first-pass thermodynamic estimate for a new REE mineral in minutes rather than weeks.

What carries the argument

The machinery is LMExt plus CatBoost. LMExt, the Language Model Extractor, converts PDF pages into high-resolution images, OCRs them into two markdown files (full text and tables only), then submits each to an LLM with few-shot prompts that require evidence for every extracted value. The two-pass table/text split and the evidence requirement are what make extraction work on messy legacy documents. CatBoost then trains on the resulting dataset, using element-property features and one-hot encoded mineral types to predict $\Delta H^\circ_f$.

What would settle it

Retrain the CatBoost model after removing every data point whose label came from a pre-2000 LMExt extraction, then test on independently measured enthalpies; if the MAE stays near 46.5 kJ/mol the claim survives, while a large error increase shows the model relied on biased labels.

Watch

Extended reading notes

Core claim

The paper's central claim is that a fully automatic pipeline can go from a pile of legacy PDFs to a trained thermodynamic predictor. The LMExt pipeline first renders pages at 400 DPI and OCRs them into markdown, separately exposing tables, then prompts an LLM to extract values such as $\Delta H^\circ_f$ while justifying each choice. Applied to rare-earth-element minerals, this produced a dataset of 3185 entries. On an 80/20 split, a CatBoost model predicts $\Delta H^\circ_f$ with mean absolute error 46.5 kJ/mol, and 50.5% of test predictions are within 15 kJ/mol—comparable to typical experimental uncertainty. The paper presents this as evidence that LLM-based dataset construction can remove the manual-data bottleneck in ML-driven thermodynamics.

Load-bearing premise

The weakest load-bearing premise is that the combined manual-plus-LMExt thermodynamic dataset is accurate enough to serve as ground truth; LMExt's extraction accuracy on pre-2000 papers is only 43.8%, and if those errors are systematic the trained model inherits biases that the test split cannot reveal.

Editorial extensions

If this is right

  • The pipeline turns a manual literature review that would take weeks into a few days of mostly automated extraction, so thermodynamic databases can grow faster than manual curation allows.
  • Predictions with errors comparable to experimental scatter can serve as a pre-experimental reference, guiding which minerals are worth synthesizing and measuring.
  • The evidence-prompting trick transfers: asking the LLM to justify each extraction materially improves success on legacy papers.
  • Because the extraction tool is field-agnostic, the same workflow can build structured datasets for medical or financial documents, not just chemistry.
  • Adding more curated data should push the model's accuracy further, since CatBoost handles small datasets and improves as the label set grows.

Reading between the lines

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

  • The reported MAE is probably a lower bound on true error because the test labels themselves contain extraction noise; a model trained on fully cleaned labels would likely show a larger but more honest error.
  • A useful next step is uncertainty quantification: flag predictions for mineral classes whose labels came mostly from pre-2000 extractions, so users know which values to trust.
  • The evidence-prompting mechanism resembles chain-of-thought reasoning and could be pushed further with self-consistency voting or a separate verification pass to lift the 43.8% pre-2000 extraction rate.
  • The dataset and model together form a benchmark that future LLM-based text-mining pipelines for chemistry can be measured against.
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

3 major / 5 minor

Summary. The manuscript reports two connected contributions: (1) LMExt, an LLM-based text data mining pipeline that converts PDF literature into machine-readable thermodynamic datasets, and (2) a CatBoost regression model trained on the resulting rare-earth-element (REE) mineral dataset to predict standard enthalpy of formation. The authors report near-perfect extraction accuracy for IUPAC stability constants, lower accuracy for pre-2000 thermodynamic papers (43.8% vs. 84.2% post-2000), and a test-set MAE of 46.5 kJ/mol for the CatBoost model, with 50.5% of test predictions within 15 kJ/mol. The paper also compares several LLMs on five selected papers and demonstrates LMExt on medical and financial documents.

Significance. If the claims hold, the work would be useful in two ways: it provides a practical recipe for LLM-based literature mining with evidence-prompting to handle legacy OCR-degraded documents, and it demonstrates a supervised model for mineral formation enthalpies built from automatically curated data. The paper is commendably concrete about the data sources and makes data and code links available. The extraction accuracy numbers, especially the pre-2000 failure rate, are honestly reported. However, the central quantitative claim of the ML section—that the model predicts enthalpies of minerals not in the training set—is not supported by the current evaluation protocol because the random split and one-hot mineral-type features permit identity leakage. The dataset-quality issues from the 43.8% pre-2000 extraction accuracy further weaken the label-grounding assumption. The strengths are the systematic workflow, the comparison across LLMs, and the clear statement of limitations, but the ML evaluation needs substantial revision.

major comments (3)
  1. [Results and Discussion, ML-based Thermodynamic Property Prediction; Method, Training ML model] The reported test MAE of 46.5 kJ/mol does not establish generalization to unseen minerals because the evaluation uses a random 80/20 split with no grouping by mineral species or chemical formula, while the input features include one-hot encoded general and specific mineral types. If the same mineral species appears in both training and test splits, CatBoost can memorize a mineral-specific average enthalpy, so the test error measures recall of known mineral identities rather than prediction for new compositions. The authors should split by mineral species/formula groups, report grouped cross-validation, and provide confidence intervals across multiple random seeds. This is load-bearing for the central claim that the model predicts minerals 'not included in its training set.'
  2. [Table 1 and Results and Discussion, Data Mining by LMExt] The training dataset is described as a mixture of manual review and LMExt output, and LMExt's extraction accuracy for pre-2000 thermodynamic papers is only 43.8%. If extraction errors are systematic (e.g., unit conversion mistakes, misassigned formulas, or OCR-induced value substitutions), the supervised model inherits those biases, and a held-out test split cannot reveal them because the test labels come from the same noisy pipeline. The authors should quantify label noise, report model performance separately on manually verified and LMExt-only entries, and assess sensitivity to plausible label errors. This is essential for interpreting the 46.5 kJ/mol MAE as a prediction of physical ground truth rather than of the pipeline's output.
  3. [Table 1 and Table 3] The extraction-accuracy evaluation rests on small samples: 50 randomly selected pairs for stability constants and five selected papers for the LLM comparison. These samples are too small to support the strong claims of '100% accuracy' and the ranking of LLMs, and no confidence intervals or per-paper breakdowns are given for the F1 and composite scores. The authors should report the per-paper results, the number of data points underlying each score, and an uncertainty estimate, or temper the claims accordingly.
minor comments (5)
  1. [Introduction and throughout] There are several typos and language issues, including 'thermondynamics' in the Introduction, 'ORC result' in the Data Mining section, 'successful rate' in the Data Mining section, and 'as Gemini 1.5-Pro' in the LLM comparison paragraph; these should be corrected.
  2. [Author contributions] The author contributions list 'A.H.' but the author list uses 'Haydn Anderson'; the initials should be made consistent.
  3. [Results and Discussion, Data Mining by LMExt] The claim that LMExt processes 'thousands of pages' by a single person in a few days would be more useful with a quantitative throughput estimate, such as pages per hour and cost per page, especially since the authors note that reasoning models are expensive; this would help readers assess practical applicability.
  4. [Method, Text data mining code] The statement that the code cannot be shown for intellectual property reasons is understandable, but the paper should clarify exactly which components are proprietary and which are available at the provided repository, so that reproducibility claims are precise.
  5. [Results and Discussion, ML-based Thermodynamic Property Prediction] The sentence 'the trained model successfully predicted ΔH°f of REE minerals within the test dataset' should be rephrased: a test-set evaluation shows correlation, not success in predicting 'minerals not included in its training set' unless a grouped split is used.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the ML prediction is trained on a literature-derived dataset with a held-out test split, and self-citations are contextual rather than load-bearing.

full rationale

The paper's central ML claim—CatBoost achieving a test MAE of 46.5 kJ/mol for standard enthalpy of formation—is a standard supervised regression result. The target values are held out in an 80/20 random split, and the input features (mineral type, chemical formula characteristics, constituent elements, atoms per formula unit) do not include the target enthalpy value, so no equation or fitted parameter is equivalent to the label by construction. The training dataset is cited to the authors' Thermochemical Database (reference 26), but that database is a literature-derived compilation and the model does not generate, define, or fit the labels; the self-citation is therefore contextual rather than load-bearing. The LMExt extraction-accuracy numbers (43.8% pre-2000, 84.2% post-2000, 100% for the IUPAC stability-constant sample) are evaluated against manually inspected ground truth, which is an external check rather than a circular reduction. The skeptic's concern that a random split with one-hot encoded mineral-type features may allow identity memorization is a legitimate generalization-risk critique, but it is not circularity: test values are not used as inputs, and the reported error is not forced by construction. No load-bearing uniqueness claim is imported from prior work, and no derivation step reduces to its own input. Score 0.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

No new physical entities are posited. The free parameters are standard ML hyperparameters and fitted scalers; the main hidden assumptions are the accuracy of the source literature and the OCR conversion, both of which are partially undermined by the reported pre-2000 extraction accuracy of 43.8%.

free parameters (1)
  • CatBoost hyperparameters = Table S1 (not visible in preprint)
    Hyperparameters were tuned by Optuna, but the specific values, search ranges, and validation protocol are not shown in the main text, making it unclear how much overfitting occurred.
assumptions (3)
  • domain assumption Literature thermodynamic values are accurate ground truths.
    The ML model is trained to reproduce these values; if the literature contains measurement errors or transcription mistakes, the model inherits them as correct labels.
  • domain assumption OCR output from Mistral at 400 DPI is a faithful text representation of the source PDFs.
    The entire extraction pipeline depends on OCR fidelity; the paper notes pre-2000 scans remain problematic, so this is load-bearing for the dataset quality.
  • domain assumption Feature engineering via Mendeleev captures chemistry relevant to formation enthalpy.
    The model uses element properties, formula counts, and mineral type; if these features do not encode the physical determinants of enthalpy, the reported accuracy may not generalize beyond the test split.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Thermodynamic Prediction Enabled by Automatic Dataset Building and Machine Learning." pith.science (2026). https://pith.science/paper/BQQOWBXT

@misc{pith2026250707293,
  author       = {Pith},
  title        = {Pith review of: Thermodynamic Prediction Enabled by Automatic Dataset Building and Machine Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BQQOWBXT}},
  note         = {Machine review of arXiv:2507.07293}
}
read the original abstract

New discoveries in chemistry and materials science, with increasingly expanding volume of requisite knowledge and experimental workload, provide unique opportunities for machine learning (ML) to take critical roles in accelerating research efficiency. Here, we demonstrate (1) the use of large language models (LLMs) for automated literature reviews, and (2) the training of an ML model to predict chemical knowledge (thermodynamic parameters). Our LLM-based literature review tool (LMExt) successfully extracted chemical information and beyond into a machine-readable structure, including stability constants for metal cation-ligand interactions, thermodynamic properties, and other broader data types (medical research papers, and financial reports), effectively overcoming the challenges inherent in each domain. Using the autonomous acquisition of thermodynamic data, an ML model was trained using the CatBoost algorithm for accurately predicting thermodynamic parameters (e.g., enthalpy of formation) of minerals. This work highlights the transformative potential of integrated ML approaches to reshape chemistry and materials science research.

Figures

Figures reproduced from arXiv: 2507.07293 by the authors.

Figure 4
Figure 4. Training ML models to predict (a) properties of mineral and demonstration by training a CatBoost model to predict (b) standard enthalpy formation of REE minerals [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 52 canonical work pages

  1. [1]

    A. M. Schweidtmann, E. Esche, A. Fischer, M. Kloft, J.-U. Repke, S. Sager and A. Mitsos, Chemie Ingenieur Technik, 2021, 93, 2029-2039

    work page 2021

  2. [2]

    Seifrid, R

    M. Seifrid, R. Pollice, A. Aguilar-Granda, Z. Morgan Chan, K. Hotta, C. T. Ser, J. Vestfrid, T. C. Wu and A. Aspuru-Guzik, Accounts of Chemical Research, 2022, 55, 2454-2466

    work page 2022

  3. [3]

    C. Lv, X. Zhou, L. Zhong, C. Yan, M. Srinivasan, Z. W. Seh, C. Liu, H. Pan, S. Li, Y . Wen and Q. Yan, Advanced Materials, 2022, 34, 2101474

    work page 2022

  4. [4]

    Morgan, G

    D. Morgan, G. Pilania, A. Couet, B. P. Uberuaga, C. Sun and J. Li, Current Opinion in Solid State and Materials Science, 2022, 26, 100975

    work page 2022

  5. [5]

    Pilania, Computational Materials Science, 2021, 193, 110360

    G. Pilania, Computational Materials Science, 2021, 193, 110360

    work page 2021

  6. [6]

    J. A. Keith, V . Vassilev-Galindo, B. Cheng, S. Chmiela, M. Gastegger, K.-R. Müller and A. Tkatchenko, Chemical Reviews, 2021, 121, 9816-9872

    work page 2021

  7. [7]

    C. A. Meza Ramirez, G. Michael, A. Lorna and I. u. and Rehman, Applied Spectroscopy Reviews, 2021, 56, 733-763

    work page 2021

  8. [8]

    C. Chen, Y . Zuo, W. Ye, X. Li, Z. Deng and S. P. Ong, Advanced Energy Materials, 2020, 10

    work page 2020

Show all 58 references
  1. [9]

    Fedik, R

    N. Fedik, R. Zubatyuk, M. Kulichenko, N. Lubbers, J. S. Smith, B. Nebgen, R. Messerly, Y . W. Li, A. I. Boldyrev, K. Barros, O. Isayev and S. Tretiak, Nature Reviews Chemistry, 2022, 6, 653-672

    2022

  2. [10]

    Meuwly, Chemical Reviews, 2021, 121, 10218-10239

    M. Meuwly, Chemical Reviews, 2021, 121, 10218-10239

    2021

  3. [11]

    K. Chen, C. Guo, C. Wang, S. Zhao, B. Xiong, G. Lu, J. R. Reinfelder and Z. Dang, Water Research, 2024, 256, 121580

    2024

  4. [12]

    Kanahashi, M

    K. Kanahashi, M. Urushihara and K. Yamaguchi, Scientific Reports, 2022, 12, 11159

    2022

  5. [13]

    C. Li, E. O. Adeniyi and P. Zarzycki, Scientific Reports, 2024, 14, 6603

    2024

  6. [14]

    Scharrer, G

    M. Scharrer, G. A. Agbanga, B. L. Brugman, M. Guild, G. Oliveira, J. Liu, X. Guo and A. Navrotsky, Materials Today, 2025, DOI: https://doi.org/10.1016/j.mattod.2025.05.010

    2025 doi

  7. [15]

    J. Liu, X. Li, Y . Feng, Z. Wang, K. M. Rosso, X. Guo and X. Zhang, arXiv preprint arXiv:2504.07007, 2025, DOI: https://doi.org/10.48550/arXiv.2504.07007

    2025 doi arXiv

  8. [16]

    J. Liu, Z. Zhang, X. Li, M. Zong, Y . Wang, S. Wang, P. Chen, Z. Wan, L. Liu, Y . Liang, W. Wang, S. Wang, X. Guo, E. G. Saldanha, K. M. Rosso and X. Zhang, Chemical Engineering Journal, 2023, 473, 145216

    2023

  9. [17]

    Zheng, O

    Z. Zheng, O. Zhang, C. Borgs, J. T. Chayes and O. M. Yaghi, Journal of the American Chemical Society, 2023, 145, 18048-18062

    2023

  10. [18]

    Dagdelen, A

    J. Dagdelen, A. Dunn, S. Lee, N. Walker, A. S. Rosen, G. Ceder, K. A. Persson and A. Jain, Nature Communications, 2024, 15, 1418

    2024

  11. [19]

    M. P. Polak and D. Morgan, Nature Communications, 2024, 15, 1569

    2024

  12. [20]

    M. Wan, T. Safavi, S. K. Jauhar, Y . Kim, S. Counts, J. Neville, S. Suri, C. Shah, R. W. White and L. Yang, arXiv preprint arXiv:2403.12173, 2024, DOI: https://doi.org/10.48550/arXiv.2403.12173

    2024 doi arXiv

  13. [21]

    Mistral OCR, https://mistral.ai/news/mistral-ocr, (accessed 05/21/2025, 2025)

    2025

  14. [22]

    BerriAI, Journal, 2025

    2025

  15. [23]

    M. Reid, N. Savinov, D. Teplyashin, D. Lepikhin, T. Lillicrap, J.-b. Alayrac, R. Soricut, A. Lazaridou, O. Firat and J. Schrittwieser, arXiv preprint arXiv:2403.05530, 2024, DOI: https://doi.org/10.48550/arXiv.2403.05530

    2024 doi arXiv

  16. [24]

    C. R. Harris, K. J. Millman, S. J. van der Walt, R. Gommers, P. Virtanen, D. Cournapeau, E. Wieser, J. Taylor, S. Berg, N. J. Smith, R. Kern, M. Picus, S. Hoyer, M. H. van Kerkwijk, M. Brett, A. Haldane, J. F. Del Rio, M. Wiebe, P. Peterson, P. Gerard- Marchant, K. Sheppard, T...

    2020

  17. [25]

    Reback, W

    J. Reback, W. McKinney, jbrockmendel, J. V . d. Bossche, T. Augspurger, P. Cloud, gfyoung, S. Hawkins, Sinhrks, M. Roeschke, A. Klein, T. Petersen, J. Tratne, C. She, W. Ayd, S. Naveh, M. Garcia, J. Schendel, patrick, A. Hayden, D. Saxton, V . Jancauskas, A. McMaster, M. Gorel...

    2021 doi

  18. [26]

    Thermochemical Database, https://theguolab.us/Thermochemical-database/index.html, (accessed 05/25/2025, 2025)

    2025

  19. [27]

    Mentel, Journal, 2024

    L. Mentel, Journal, 2024

    2024

  20. [28]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss and V . Dubourg, the Journal of machine Learning research, 2011, 12, 2825-2830

    2011

  21. [29]

    N. L. Manakov, A. V . Meremianin and A. F. Starace, Physical Review A, 1998, 57, 3233- 3244

    1998

  22. [30]

    A. V . Dorogush, V . Ershov and A. Gulin, arXiv preprint, 2018, DOI: https://doi.org/10.48550/arXiv.1810.11363

    2018 doi arXiv

  23. [31]

    Akiba, S

    T. Akiba, S. Sano, T. Yanase, T. Ohta and M. Koyama, 2019

    2019

  24. [32]

    G. Team, R. Anil, S. Borgeaud, J.-B. Alayrac, J. Yu, R. Soricut, J. Schalkwyk, A. M. Dai, A. Hauth and K. Millican, arXiv preprint arXiv:2312.11805, 2023, DOI: https://doi.org/10.48550/arXiv.2312.11805

    2023 doi arXiv

  25. [33]

    D. Rau, S. Wang, H. Déjean and S. Clinchant, arXiv preprint arXiv:2407.09252, 2024, DOI: https://doi.org/10.48550/arXiv.2407.09252

    2024 doi arXiv

  26. [34]

    SC-Database, https://equilibriumdata.github.io/sc-database.html, (accessed 05/23/2025)

    2025

  27. [35]

    Lewis, E

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V . Karpukhin, N. Goyal, H. Küttler, M. Lewis, W.-t. Yih, T. Rocktäschel, S. Riedel and D. Kiela, presented in part at the Proceedings of the 34th International Conference on Neural Information Processing Systems, Vancouver, BC, Canada, 2020

    2020

  28. [36]

    Bratsch and H

    S. Bratsch and H. B. Silber, Polyhedron, 1982, 1, 219-223

    1982

  29. [37]

    K. A. Gschneidner, Journal of the Less Common Metals, 1969, 17, 1-12

    1969

  30. [38]

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, B. Ichter, F. Xia, H. Chi, Q. V . Le and D. Zhou, arXiv preprint arXiv:2201.11903, 2022, DOI: https://doi.org/10.48550/arXiv.2201.11903

    2022 doi arXiv

  31. [39]

    Reynolds and K

    L. Reynolds and K. McDonell, presented in part at the Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems, Yokohama, Japan, 2021

    2021

  32. [40]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser and I. Polosukhin, Advances in neural information processing systems, 2017, 30

    2017

  33. [41]

    F. B. Baker and C. E. Holley, Journal of Chemical & Engineering Data, 1968, 13, 405- 407

    1968

  34. [42]

    Dash and N

    S. Dash and N. K. Shukla, Journal of Thermal Analysis and Calorimetry, 2013, 112, 193- 200

    2013

  35. [43]

    S. A. Gramsch and L. R. Morss, The Journal of Chemical Thermodynamics, 1995, 27, 551-560

    1995

  36. [44]

    L. A. Reznitskii, Inorganic Materials, 2001, 37, 491-495

    2001

  37. [45]

    N. D. Stout, C. L. Hoenig and P. C. Nordine, Journal of the American Ceramic Society, 1969, 52, 145-151

    1969

  38. [46]

    S. W. Kieffer, Reviews of Geophysics, 1982, 20, 827-849

    1982

  39. [47]

    S. K. Papadopoulou, A. de Souza Braga Neto, I. Billard, C. Cousin, V . Briois, A. Beauvois, L. Michot, G. Mériguet, A.-L. Rollet and J. Sirieix-Plénet, Separation and Purification Technology, 2025, 355, 129686

    2025

  40. [48]

    J. Wei, X. Chu, X.-Y . Sun, K. Xu, H.-X. Deng, J. Chen, Z. Wei and M. Lei, InfoMat, 2019, 1, 338-358

    2019

  41. [49]

    M. Ge, F. Su, Z. Zhao and D. Su, Materials Today Nano, 2020, 11, 100087

    2020

  42. [50]

    Oviedo, J

    F. Oviedo, J. L. Ferres, T. Buonassisi and K. T. Butler, Accounts of Materials Research, 2022, 3, 597-607

    2022

  43. [51]

    A. S. Anker, K. T. Butler, R. Selvan and K. M. O. Jensen, Chem Sci, 2023, 14, 14003- 14019

    2023

  44. [52]

    Bender, N

    A. Bender, N. Schneider, M. Segler, W. Patrick Walters, O. Engkvist and T. Rodrigues, Nature Reviews Chemistry, 2022, 6, 428-442

    2022

  45. [53]

    Debus, H

    B. Debus, H. Parastar, P. Harrington and D. Kirsanov, TrAC Trends in Analytical Chemistry, 2021, 145, 116459

    2021

  46. [54]

    Houhou and T

    R. Houhou and T. Bocklitz, Analytical Science Advances, 2021, 2, 128-141

    2021

  47. [55]

    J. Liu, X. Zhao, K. Zhao, V . G. Goncharov, J. Delhommelle, J. Lin and X. Guo, Scientific Reports, 2023, 13, 5919

    2023

  48. [56]

    X. Zhao, Y . Luo, J. Liu, W. Liu, K. M. Rosso, X. Guo, T. Geng, A. Li and X. Zhang, The Journal of Physical Chemistry C, 2023, 127, 14830-14838

    2023

  49. [57]

    C. J. Taylor, A. Pomberger, K. C. Felton, R. Grainger, M. Barecka, T. W. Chamberlain, R. A. Bourne, C. N. Johnson and A. A. Lapkin, Chemical Reviews, 2023, 123, 3089-3126

    2023

  50. [58]

    Ahmadi, M

    M. Ahmadi, M. Ziatdinov, Y . Zhou, E. A. Lass and S. V . Kalinin, Joule, 2021, 5, 2797- 2822

    2021

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.