Recognizing the Versatility of Heterogeneous Catalysts Can Provide New Opportunities
Abstract Heterogeneous catalysts are most often designed to facilitate a specific reaction. We propose that evaluating one catalyst across several reactions is advantageous to make more efficient use of resources and increasing the data generated for machine learning. Screening a single material for potential catalytic activity across multiple reactions and operating conditions samples a substantially wider region of the reaction condition parameter space than a single optimization. This can produce a multitude of performance metrics for an identically prepared and characterized material. Current literature reporting practices are biased toward reporting only optimal experimental outcomes. In practice, results obtained from low-conversion and low-selectivity catalysts would likely constrain a regression model to a similar degree as would data obtained from a high-performing catalyst, but these are seldom published. Systematically recording the full distribution of outcomes produces data sets that are better suited at modeling structure–property relationships, and support multitask learning, in which data from several reactions inform a shared representation of the catalyst. We showcase this by taking Pd/ZnO and Cu/ZnO that are effective for CO2 hydrogenation and applying them to other diverse reactions such as preferential oxidation of CO and photocatalytic ibuprofen decomposition in water. We find that Cu/ZnO and CuO/ZnO were more effective than Pd/ZnO in facilitating both the CO preferential oxidation reaction and ibuprofen decomposition. The versatility of the Cu-based catalyst demonstrates the potential for future studies that involve a broader approach whereby heterogeneous catalysts can be specifically repurposed for multiple processes.
Authors
- Layal Halawy
- Tshepo Molefe (ORCID: https://orcid.org/0000-0003-3039-4728)
- Graham John Hutchings (ORCID: https://orcid.org/0000-0001-8885-1560)
- Isla E. Gow (ORCID: https://orcid.org/0009-0008-8667-2575)
- Theopolina Amakali
- Veikko Uahengo (ORCID: https://orcid.org/0000-0002-7937-7286)
- Thulani M. Nyathi (ORCID: https://orcid.org/0000-0002-7684-1258)
- Roy P. Forbes (ORCID: https://orcid.org/0000-0001-6737-7775)
- Kieran J. Aggett (ORCID: https://orcid.org/0000-0002-9552-0396)
- Mohamed Fadlalla (ORCID: https://orcid.org/0000-0002-3820-1372)
- Louise R. Smith (ORCID: https://orcid.org/0000-0003-1665-3033)
- Nicholas F. Dummer (ORCID: https://orcid.org/0000-0002-0946-6304)
- Michael Claeys (ORCID: https://orcid.org/0000-0002-5797-5023)
- Dean H. Barrett (ORCID: https://orcid.org/0000-0001-5360-9336)
- Oskar Wielgosz
- Tumelo N. Phaahlamohlaka (ORCID: https://orcid.org/0000-0003-1174-5078)
- Johannes Naimhwaka
- Ateeq Rahman
- Jacobina T. Nangombe
- Katie G. Roberts
Institutions
- University of Cape Town (ZA)
- University of the Witwatersrand (ZA)
- University of Namibia (NA)
- Cardiff University (GB)
Publication Details
- Journal
- ACS Applied Catalysis
- Published
- 2026-11-14
- DOI
- https://doi.org/10.1021/acsappcat.6c00087
- Primary Topic
- Machine Learning in Materials Science
- Type
- article
- Field-Weighted Citation Impact
- 0.00