pith. sign in

arxiv: 2606.30898 · v1 · pith:OOSXBEAHnew · submitted 2026-06-29 · ❄️ cond-mat.mtrl-sci · physics.chem-ph

Bridging electrode preparation and electrocatalyst performance with physics-based causal AI

Evelyna Wang , Linda Hung , Sam Witty , Michaela Burke Stevens , Kevin Tran This is my paper

Pith reviewed 2026-07-01 01:20 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci physics.chem-ph
keywords electrocatalysisstructural causal modelscausal inferenceelectrode preparationoxygen reductionmanganese-antimony oxiderotating disc electrodeVulcan carbon
0
0 comments X

The pith

Physics-based structural causal models separate the effects of support-to-catalyst ratios and electrode thickness on catalytic performance from small multi-modal datasets.

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

This paper demonstrates that physics-based structural causal models can extract quantitative causative insights into how electrode preparation parameters influence electrocatalyst performance. The models handle complex heterogeneous systems even when the dataset is small (under 10 samples) but multi-modal (over 10 modes). They separate the distinct roles of varying support-to-catalyst ratios and total material loadings. The approach further disentangles whether performance changes stem from the number of active sites or from electrode thickness. This separation matters for translating material-level observations to device-level outcomes in cases where no closed-form physical model exists.

Core claim

The proof-of-concept model applied to manganese-antimony oxide oxygen reduction electrocatalysts on Vulcan carbon supports in alkaline media using rotating disc electrode tests shows that the SCM quantitatively separates the role of support-to-catalyst ratios and total material loadings on catalytic performance and disentangles the contribution of the number of active sites from the contribution of electrode thickness.

What carries the argument

The physics-based structural causal model (SCM) that encodes domain knowledge and physical constraints to identify causal effects from n<10 samples with >10 modes.

If this is right

  • The SCM enables root-cause analysis on cyclic voltammograms of the tested manganese-antimony oxide catalysts.
  • Performance can be attributed separately to active-site count versus electrode thickness.
  • Preparation variables such as support-to-catalyst ratio and total loading can be ranked by their causal impact.
  • Insights from material studies become translatable to device configurations without requiring a known mathematical model.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same SCM structure could be reused on other oxide or carbon-supported catalysts tested under similar rotating-disc conditions.
  • Collecting one additional mode, such as direct active-site counting via a separate technique, would provide an external check on the thickness-versus-sites separation.
  • If the causal graph holds, the model predicts that changing only electrode thickness at fixed active-site density produces a measurable performance shift independent of ratio effects.

Load-bearing premise

The physics-based constraints and domain knowledge encoded in the SCM are sufficient to identify causal effects from fewer than 10 samples across more than 10 modes without additional validation data or sensitivity checks on the assumed causal graph.

What would settle it

An independent experiment that varies support-to-catalyst ratios while holding total loading fixed and measures whether the SCM-predicted change in performance matches the observed change in cyclic voltammograms.

read the original abstract

State-of-the-art artificial intelligence (AI) and Machine-Learning (ML) tools have not yet enabled rapid design of next-generation materials. Detailed physical understanding of how material properties affect device performance is required to advance materials development. For example, optimization of ink parameters for electrocatalysts has no known physical mathematical model and thus insights are difficult to translate from material studies to device studies. Herein, we demonstrate how to use the emerging AI tool, physics-based structural causal models (SCMs), to extract quantitative causative insights from complex heterogeneous electrochemical systems with small (n < 10), but multi-modal datasets (modes > 10). Our SCM quantitatively separates the role that varying the support-to-catalyst ratios and total material loadings plays on catalytic performance. The proof of concept model developed in this work enables root-cause-analysis on the cyclic voltammograms of manganese-antimony oxide oxygen reduction electrocatalysts on Vulcan carbon supports tested in alkaline media using a rotating disc electrode device configuration. Our preliminary causal analyses quantitatively disentangle how the catalyst performance is affected by the number of active sites versus the thickness of the electrode. To the best of our knowledge, this is the first demonstration of physics-based SCMs applied to electrochemical materials and their performance.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript presents a proof-of-concept for applying physics-based structural causal models (SCMs) to small (n<10), multi-modal (>10 modes) datasets from rotating-disk electrode tests of Mn-Sb oxide oxygen-reduction electrocatalysts on Vulcan carbon. It claims that the SCM quantitatively separates the causal effects of support-to-catalyst ratio and total loading on performance and disentangles the contributions of active-site number versus electrode thickness.

Significance. If the SCM identifiability claims hold after validation, the work would constitute the first reported use of physics-constrained causal models in electrochemical materials systems and could offer a route to root-cause analysis when conventional physical models are absent and data are scarce.

major comments (2)
  1. [Abstract and §3] Abstract and §3 (SCM construction): the central claim that the SCM 'quantitatively separates' the effects of support-to-catalyst ratios and loadings and 'quantitatively disentangles' active-site number versus thickness rests on the untested assertion that the physics-encoded graph renders these effects identifiable. No causal graph, functional forms, or identifiability proof is supplied, and with n<10 and >10 modes even modest misspecification produces large changes in estimated effects.
  2. [§4] §4 (results): no sensitivity analysis, alternative graphs, simulation-based validation, or hold-out testing is reported to confirm that the physics constraints suffice for identification at this sample size. This directly undermines the quantitative causal claims.
minor comments (2)
  1. [Abstract] Abstract: the phrase 'to the best of our knowledge, this is the first demonstration' should be supported by a brief literature comparison rather than left as a claim.
  2. [§2] Notation for the SCM variables (e.g., active-site density, thickness) is introduced without an explicit variable table or diagram, reducing readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments on the identifiability and validation of the physics-based SCM. We address each major point below.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (SCM construction): the central claim that the SCM 'quantitatively separates' the effects of support-to-catalyst ratios and loadings and 'quantitatively disentangles' active-site number versus thickness rests on the untested assertion that the physics-encoded graph renders these effects identifiable. No causal graph, functional forms, or identifiability proof is supplied, and with n<10 and >10 modes even modest misspecification produces large changes in estimated effects.

    Authors: We agree that the manuscript would be strengthened by explicitly presenting the causal graph and functional forms. In revision we will add the physics-constrained DAG, detail the functional forms based on electrochemical principles, and include a short identifiability discussion using d-separation and the encoded constraints. We will also note the potential sensitivity to misspecification at small n. revision: yes

  2. Referee: [§4] §4 (results): no sensitivity analysis, alternative graphs, simulation-based validation, or hold-out testing is reported to confirm that the physics constraints suffice for identification at this sample size. This directly undermines the quantitative causal claims.

    Authors: We acknowledge these analyses are absent from the current version. We will add sensitivity analysis over plausible graph variations, simulation-based validation on synthetic data, and hold-out testing (where feasible) to the revised §4. revision: yes

Circularity Check

0 steps flagged

No circularity detectable; no equations or derivation chain visible for inspection.

full rationale

The abstract and provided text describe application of physics-based SCMs to separate effects of support ratios, loadings, active sites, and electrode thickness from n<10 multi-modal data. No equations, functional forms, fitted parameters, causal graph construction details, or self-citations are shown. Without quotable steps that reduce a claimed quantitative separation to inputs by definition or fit, none of the enumerated circularity patterns (self-definitional, fitted-input-called-prediction, etc.) can be exhibited. The derivation is therefore self-contained against external benchmarks in the supplied material.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The approach implicitly assumes that domain knowledge can be encoded as a valid causal graph without introducing untestable assumptions.

pith-pipeline@v0.9.1-grok · 5757 in / 1065 out tokens · 17268 ms · 2026-07-01T01:20:04.367851+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

57 extracted references · 1 canonical work pages

  1. [2]

    & Wei, Z

    Nie, Y., Li, L. & Wei, Z. Recent advancements in pt and pt-free catalysts for oxygen reduction reaction.Chemical Society Reviews44, 2168–2201 (2015)

    2015

  2. [3]

    Computational Materials Science170, 109202 (2019)

    Fu, C.et al.Dft calculations: A powerful tool for better understanding of electrocatalytic oxygen reduction reactions on pt-based metallic catalysts. Computational Materials Science170, 109202 (2019)

    2019

  3. [4]

    & Lin, X

    Hua, M., Tian, X., Li, S. & Lin, X. Pdag/ag (111) surface alloys: A highly efficient catalyst of oxygen reduction reaction.Nanomaterials12, 1802 (2022)

    2022

  4. [5]

    Yarlagadda, V.et al.Preparation of pemfc electrodes from milligram-amounts of catalyst powder.Journal of The Electrochemical Society164, F845–F849 (2017)

    2017

  5. [6]

    S.et al.Best practices and testing protocols for benchmarking orr activities of fuel cell electrocatalysts using rotating disk electrode.Electrocatalysis 8, 366–374 (2017)

    Kocha, S. S.et al.Best practices and testing protocols for benchmarking orr activities of fuel cell electrocatalysts using rotating disk electrode.Electrocatalysis 8, 366–374 (2017)

    2017

  6. [7]

    Toyota CRDL 49, 33–46 (2018)

    Shinozaki, K.et al.Oxygen reduction reaction activity measurement for fuel cell catalysts using the rotating disk electrode technique.R&D Rev. Toyota CRDL 49, 33–46 (2018)

    2018

  7. [8]

    Electrocatalysis7, 305–316 (2016)

    Chen, G.et al.Evaluation of oxygen reduction activity by the thin-film rotat- ing disk electrode methodology: the effects of potentiodynamic parameters. Electrocatalysis7, 305–316 (2016)

    2016

  8. [9]

    & Zhang, J.Rotating Electrode Methods and Oxygen Reduction Electrocatalysts, Ch

    Du, C., Tan, Q., Yin, G. & Zhang, J.Rotating Electrode Methods and Oxygen Reduction Electrocatalysts, Ch. 5, 171–198 (Elsevier, 2014)

    2014

  9. [10]

    & Zhang, J.Applications of RDE and RRDE methods in oxygen reduction reaction, Ch

    Du, C., Sun, Y., Shen, T., Yin, G. & Zhang, J.Applications of RDE and RRDE methods in oxygen reduction reaction, Ch. 7, 231–277 (Elsevier, 2014)

    2014

  10. [11]

    C.et al.High-performance ionomerless cathode anion-exchange mem- brane fuel cells with ultra-low-loading ag–pd alloy electrocatalysts.Nature Energy 31 8, 1262–1272 (2023)

    Douglin, J. C.et al.High-performance ionomerless cathode anion-exchange mem- brane fuel cells with ultra-low-loading ag–pd alloy electrocatalysts.Nature Energy 31 8, 1262–1272 (2023)

    2023

  11. [12]

    R.et al.Improved oxygen reduction activity on pt3ni (111) via increased surface site availability.Science315, 493–497 (2007)

    Stamenkovic, V. R.et al.Improved oxygen reduction activity on pt3ni (111) via increased surface site availability.Science315, 493–497 (2007)

    2007

  12. [13]

    & Compton, R

    Masa, J., Batchelor-McAuley, C., Schuhmann, W. & Compton, R. G. Koutecky- levich analysis applied to nanoparticle modified rotating disk electrodes: Electro- catalysis or misinterpretation.Nano Research7, 71–78 (2014)

    2014

  13. [14]

    & Daza, L

    Mart´ ın, A., Chaparro, A., Folgado, M., Rubio, J. & Daza, L. Rotating disk elec- trode analysis of oxygen reduction at platinum particles under limiting diffusion conditions.Electrochimica acta54, 2209–2217 (2009)

    2009

  14. [15]

    E.et al.Strategies for modulating the catalytic activity and selectiv- ity of manganese antimonates for the oxygen reduction reaction.ACS Catalysis 12, 10826–10840 (2022)

    Kreider, M. E.et al.Strategies for modulating the catalytic activity and selectiv- ity of manganese antimonates for the oxygen reduction reaction.ACS Catalysis 12, 10826–10840 (2022)

    2022

  15. [16]

    W., Pylypenko, S., Pivovar, B

    Shinozaki, K., Zack, J. W., Pylypenko, S., Pivovar, B. S. & Kocha, S. S. Oxygen reduction reaction measurements on platinum electrocatalysts utilizing rotating disk electrode technique: Ii. influence of ink formulation, catalyst layer uniformity and thickness.Journal of The Electrochemical Society162, F1384–F1396 (2015)

    2015

  16. [17]

    L., Mahy, J

    Deschamps, F. L., Mahy, J. G., Leonard, A. F. & Job, N. Rotating disk elec- trode measurements on low and high loading catalyst layers: Diffusion limitations and application to pt catalysts supported on porous micrometric carbon xero- gel particles designed for proton exchange membrane fuel cells.Journal of Electroanalytical Chemistry933, 117279 (2023)

    2023

  17. [18]

    & Swider-Lyons, K

    Garsany, Y., Ge, J., St-Pierre, J., Rocheleau, R. & Swider-Lyons, K. E. Analytical procedure for accurate comparison of rotating disk electrode results for the oxygen reduction activity of pt/c.Journal of The Electrochemical Society161, F628– F640 (2014)

    2014

  18. [19]

    Kokhanov, A. A.et al.Improving the reproducibility of oxygen reduction reaction activity assessment for pt-based electrocatalysts on a rotating disk electrode via catalytic layer optimization.Catalysts15, 1140 (2025)

    2025

  19. [20]

    L., Hettige, C., Zook-Gerdau, L

    Knoche, K. L., Hettige, C., Zook-Gerdau, L. A. & Leddy, J. Film permeation by rotating disk voltammetry at electrodes modified with electrochemically inert 32 layers and heterogeneous composites.Journal of The Electrochemical Society 163, H3173–H3181 (2016)

    2016

  20. [21]

    Xu, S.et al.Building upon the koutecky-levich equation for evaluation of next- generation oxygen reduction reaction catalysts.Electrochimica Acta255, 99–108 (2017)

    2017

  21. [22]

    T., Mano, N., Nam, B

    Bonnecaze, R. T., Mano, N., Nam, B. & Heller, A. On the behavior of the porous rotating disk electrode.Journal of the Electrochemical Society154, F44–F47 (2007)

    2007

  22. [23]

    Deo, S.et al.Interpretable machine learning models for practical antimonate electrocatalyst performance.ChemPhysChem25, e202400010 (2024)

    2024

  23. [24]

    M., Gasteiger, H

    Lazaridis, T., St¨ uhmeier, B. M., Gasteiger, H. A. & El-Sayed, H. A. Capabilities and limitations of rotating disk electrodes versus membrane electrode assemblies in the investigation of electrocatalysts.Nature Catalysis5, 363–373 (2022)

    2022

  24. [25]

    M.et al.Alloyed pt–zn oxygen reduction catalysts for proton exchange membrane fuel cells.ACS Applied Energy Materials5, 8282–8291 (2022)

    Dull, S. M.et al.Alloyed pt–zn oxygen reduction catalysts for proton exchange membrane fuel cells.ACS Applied Energy Materials5, 8282–8291 (2022)

    2022

  25. [26]

    Chen, W.et al.Reconsidering the benchmarking evaluation of catalytic activity in oxygen reduction reaction.IScience23(2020)

    2020

  26. [27]

    Nature Energy6, 475–486 (2021)

    Fan, J.et al.Bridging the gap between highly active oxygen reduction reaction catalysts and effective catalyst layers for proton exchange membrane fuel cells. Nature Energy6, 475–486 (2021)

    2021

  27. [28]

    & Xia, B

    Zaman, S., Tian, X. & Xia, B. Y. Bridging oxygen reduction performance gaps in half and full cells: Challenges and perspectives.Materials Chemistry Frontiers 7, 4605–4612 (2023)

    2023

  28. [29]

    Pearl, J.Causality(Cambridge university press, 2009)

    2009

  29. [30]

    Y.Actual causality(MiT Press, 2016)

    Halpern, J. Y.Actual causality(MiT Press, 2016)

    2016

  30. [31]

    & Traag, V

    Klebel, T. & Traag, V. Introduction to structural causal models in science studies. Quantitative Science Studies7, 159–178 (2026). 33

    2026

  31. [32]

    Causal modelling and brain connectivity in functional magnetic resonance imaging.PLoS biology7, e1000033 (2009)

    Friston, K. Causal modelling and brain connectivity in functional magnetic resonance imaging.PLoS biology7, e1000033 (2009)

    2009

  32. [33]

    & Barnard, A

    Motevalli, B., Sun, B. & Barnard, A. S. Understanding and predicting the cause of defects in graphene oxide nanostructures using machine learning.The Journal of Physical Chemistry C124, 7404–7413 (2020)

    2020

  33. [34]

    & Ghosh, S

    Ghosh, A. & Ghosh, S. Mapping causal pathways with structural modes finger- print for perovskite oxides.Machine Learning: Science and Technology5, 045014 (2024)

    2024

  34. [35]

    Barakati, K.et al.Mapping causal patterns in crystalline solid.APL Machine Learning3(2025)

    2025

  35. [36]

    V., Ghosh, A., Vasudevan, R

    Kalinin, S. V., Ghosh, A., Vasudevan, R. & Ziatdinov, M. From atomically resolved imaging to generative and causal models.Nature Physics18, 1152–1160 (2022)

    2022

  36. [37]

    A.Bayesian Structural Causal Inference with Probabilistic Program- ming

    Witty, S. A.Bayesian Structural Causal Inference with Probabilistic Program- ming. Ph.D. thesis, Ph. D. Dissertation. University of Massachusetts Amherst (2023)

    2023

  37. [38]

    Aykol, M.et al.Perspective—combining physics and machine learning to predict battery lifetime.Journal of The Electrochemical Society168, 030525 (2021)

    2021

  38. [39]

    Tran, K.et al.Methods for comparing uncertainty quantifications for mate- rial property predictions.Machine Learning: Science and Technology1, 025006 (2020)

    2020

  39. [40]

    Kutner, M. H.et al. Applied linear statistical models(India McGraw-Hill Education, 1984)

    1984

  40. [41]

    & Gabry, J

    Vehtari, A., Simpson, D., Gelman, A., Yao, Y. & Gabry, J. Pareto smoothed importance sampling.Journal of Machine Learning Research25, 1–58 (2024)

    2024

  41. [42]

    E.et al.A causal roadmap for generating high-quality real-world evidence.Journal of Clinical and Translational Science7, e212 (2023)

    Dang, L. E.et al.A causal roadmap for generating high-quality real-world evidence.Journal of Clinical and Translational Science7, e212 (2023)

    2023

  42. [43]

    Gelman, A.et al.Bayesian workflow.arXiv preprint arXiv:2011.01808(2020). 34

    work page arXiv 2011

  43. [44]

    Lind, M.Multilevel flow modelling of process plant for diagnosis and control (Risø National Laboratory, 1982)

    1982

  44. [45]

    Exploiting bond graph causality in physical system models

    Rosenberg, R. Exploiting bond graph causality in physical system models. Journal of Dynamic Systems, Measurement, and Control(1987)

    1987

  45. [46]

    Gridin, V.et al.Gde vs rde: Impact of operation conditions on intrinsic catalytic parameters of fenc catalyst for the oxygen reduction reaction.Electrochimica Acta444, 142012 (2023)

    2023

  46. [47]

    O., Lindbergh, G., Thomassen, M

    Darab, M., Barnett, A. O., Lindbergh, G., Thomassen, M. S. & Sunde, S. The influence of catalyst layer thickness on the performance and degradation of pem fuel cell cathodes with constant catalyst loading.Electrochimica acta232, 505– 516 (2017)

    2017

  47. [48]

    Rao, V.et al.The influence of carbon support porosity on the activity of ptru/sibunit anode catalysts for methanol oxidation.Journal of Power Sources 145, 178–187 (2005)

    2005

  48. [49]

    Gong, Q.et al.Effects of ink formulation on construction of catalyst layers for high-performance polymer electrolyte membrane fuel cells.ACS applied materials & interfaces13, 37004–37013 (2021)

    2021

  49. [50]

    Li, C.et al.Effects of ink formulation on the structure and performance of pgm-free catalyst layer in pemfcs.Electrochemical Society Transactions 240104, 327–333 (2021)

    2021

  50. [51]

    Su, Y.et al.Bridging ink formulation and proton exchange membrane fuel cell performance: Deciphering the role of catalyst ink composition in regulating ionomer behaviors and aggregates/agglomerates structures for high performance cell design.Renewable and Sustainable Energy Reviews225, 116163 (2026)

    2026

  51. [52]

    K., Devivaraprasad, R., Kar, T., Chakraborty, A

    Singh, R. K., Devivaraprasad, R., Kar, T., Chakraborty, A. & Neergat, M. Elec- trochemical impedance spectroscopy of oxygen reduction reaction (orr) in a rotat- ing disk electrode configuration: effect of ionomer content and carbon-support. Journal of The Electrochemical Society162, F489–F498 (2015)

    2015

  52. [53]

    inDirect and indirect effects(eds Geffner, H., Dechter, R

    Pearl, J. inDirect and indirect effects(eds Geffner, H., Dechter, R. & Halpern, J. Y.)Probabilistic and causal inference: the works of Judea Pearl373–392 35 (Association for Computing Machinery, New York, NY, USA, 2022)

    2022

  53. [54]

    D., Blei, D

    Hoffman, M. D., Blei, D. M., Wang, C. & Paisley, J. Stochastic variational inference.Journal of machine learning research(2013)

    2013

  54. [55]

    Institute, B. R. Chirho. https://github.com/BasisResearch/chirho (2025). V0.3.0

    2025

  55. [56]

    Xu, X.et al.Toward enhanced pyro-catalysis performance: mechanisms, strategies and challenges.Journal of Materials Chemistry A13, 14465–14490 (2025)

    2025

  56. [57]

    Paszke, A.et al.Pytorch: An imperative style, high-performance deep learning library.Advances in neural information processing systems32(2019)

    2019

  57. [58]

    NIPS 2017 Autodiff Workshop (2017)

    Paszke, A.et al.Automatic differentiation in pytorch. NIPS 2017 Autodiff Workshop (2017). 36

    2017