Skip to content
DraconiticResearch
Search Sources API
203,203 studies31 sourcesUpdated 43 min ago
PreprintCC BYChemRxiv

Is characterization worth the cost in data-driven catalyst discovery?

  • Ricardo Mathison1,2,3
  • Harsh R. Darji1,2,3
  • Jacob Huppin1,2,3
  • Fillipp Edvard L. Salvador1,2,4
  • Faysal Ibrahim1,2,4
  • Matias Alvear1,2,3
  • Ive Hermans1,2,3,4
4 affiliations
  1. 1Ralph O’Connor Sustainable Energy Institute
  2. 2Johns Hopkins University
  3. 3Department of Chemical and Biomolecular Engineering
  4. 4Department of Chemistry
Open at sourceDOICite

Abstract

New catalysts are often discovered by trial and error, a search in which most candidates fail. Machine learning now shortens that search by predicting how untested formulations perform from composition alone. Measuring the physicochemical properties that govern performance should shorten it further, yet this expectation remains untested in a discovery campaign. We tested it on Pt-based propane dehydrogenation catalysts, evaluating discovery speed, prediction accuracy, and physical insight on the same materials. Here we show that measured properties neither accelerate the search nor improve predictions beyond composition alone. However, they resolve deactivation into separable contributions from exposed platinum and surface acidity. These results explain why brute-force screening dominates industry and define, in catalysis and beyond, when to characterize a material rather than test the next one.

Related

  1. Implementing modular design principles for molecular homogeneous catalyst discovery.
    Francis Millward, Benjamin S Grills, Elizabeth M Dauncey +2
    Catalysis science & technologyPeer-reviewedCC BYFull textOrganic chemistryEurope PMC
  2. A Catalyst Discovery Workflow Coupling MLIP with MOBO for Nitrite-to-Hydroxylamine Electroreduction
    Zhiyi Yu, Huaiwen Zhang, Weiming Zhao
    ChemRxivPreprintCC BYFull textMaterials & polymersChemRxiv
  3. Hierarchical Multi-agent Large Language Model Reasoning for Autonomous Heterogeneous Catalyst Discovery
    Samuel Rothfarb, Megan Christina Davis, Ivana Matanović +3
    npj Computational MaterialsPeer-reviewedCC BY-NC-NDFull text14 citedMaterials & polymersOpenAlex
  4. Accelerating Catalyst Discovery with Artificial Intelligence Using Molecular Dual-Atom Site Catalysts as a Case Study
    Denis Savchenko, Bruno Ina cio da Silva Roux Leite, Timmo-Henrik Pukk +3
    ChemRxivPreprintCC BYFull textMaterials & polymersChemRxiv
  5. Artificial Intelligence- and Machine Learning-Driven Strategies for Catalyst Design and Sustainable Chemical Processes
    Amra Bratovčić, Vesna Tomašić
    ProcessesPeer-reviewedCC BY10 citedMaterials & polymersOpenAlex

Cite

BibTeXRIS
BibTeX
@misc{mathison2026characterization,
  title = {{Is characterization worth the cost in data-driven catalyst discovery?}},
  author = {Ricardo Mathison and Harsh R. Darji and Jacob Huppin and Fillipp Edvard L. Salvador and Faysal Ibrahim and Matias Alvear and Ive Hermans},
  howpublished = {ChemRxiv},
  year = {2026},
  month = oct,
  doi = {10.26434/chemrxiv.15011162/v1},
  url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15011162/v1},
  note = {CC BY},
}
RIS
TY  - UNPB
TI  - Is characterization worth the cost in data-driven catalyst discovery?
AU  - Ricardo Mathison
AU  - Harsh R. Darji
AU  - Jacob Huppin
AU  - Fillipp Edvard L. Salvador
AU  - Faysal Ibrahim
AU  - Matias Alvear
AU  - Ive Hermans
PY  - 2026
DA  - 2026/10/27
PB  - ChemRxiv
DO  - 10.26434/chemrxiv.15011162/v1
UR  - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15011162/v1
AB  - New catalysts are often discovered by trial and error, a search in which most candidates fail. Machine learning now shortens that search by predicting how untested formulations perform from composition alone. Measuring the physicochemical properties that govern performance should shorten it further, yet this expectation remains untested in a discovery campaign. We tested it on Pt-based propane dehydrogenation catalysts, evaluating discovery speed, prediction accuracy, and physical insight on the same materials. Here we show that measured properties neither accelerate the search nor improve predictions beyond composition alone. However, they resolve deactivation into separable contributions from exposed platinum and surface acidity. These results explain why brute-force screening dominates industry and define, in catalysis and beyond, when to characterize a material rather than test the next one.
ER  -