Case Study of Bayesian Batch Optimization Strategies for Heterogeneous Catalysts in CO2 Hydrogenation
5 affiliations
- 1SwissCAT+ East, ETH Zurich
- 2Department of Chemistry and Applied Biosciences
- 3ETH Zurich
- 4Department of Computer Science
- 5ETH AI Center
Abstract
High-throughput and automated experimentation platforms combined with artificial intelligence are increasingly used to accelerate catalyst discovery. Batch Bayesian optimization (BO) (i.e., the simulta neous selection of multiple candidate experiments per iteration) is widely adopted for navigating large compositional spaces; however, its practical behavior under realistic high-throughput and batch con straints remains poorly understood. In this work, we performed a case study on four BO strategies to optimize the catalyst composition for the hydrogenation of CO2 to methanol. For each BO strategy, 48 catalysts were synthesized by co-precipitation, utilizing an automated synthesis platform, and evaluated in high-throughput fixed-bed reactors over seven batch iterations. The composition space comprised up to four metallic elements selected from a pool of nine. Although all BO strategies identified competitive catalyst formulations, most optimization trajectories rapidly converged toward distinct local optima that were strongly influenced by the initial random design and exhibited limited global exploration of the compositional space. This case study provides practical insights into the limitations of BO for batch catalyst optimization and highlights their strong dependence on the initial random design, parameter space definition, and batch execution strategy.
Related
- Optimization of Bimetallic Catalysts for the Hydrogenation of Liquid Organic Hydrogen Carriers using Contaminated Hydrogen Gas
- Engineering MoOₓ-based catalysts and reaction conditions to balance deoxygenation and hydrogenation in glycerol conversion for stable propylene production
- Tuning supported Fe-Ag catalysts for selective acetylene hydrogenation: A combined experimental and computational study
- Controlling Hydrogenation–Ammonolysis Competition in the Selective Reductive Amination of Biomass‐Derived Furfural Over Pt–Ni Catalysts
- A closer look into transfer hydrogenation between perhydro-dibenzyltoluene (H₁₈-DBT) and acetone for safe hydrogen release using Pt-based catalysts
BibTeX
@misc{lam2026case,
title = {{Case Study of Bayesian Batch Optimization Strategies for Heterogeneous Catalysts in CO2 Hydrogenation}},
author = {Erwin Lam and Kourosh Shariat and Leo Chevalier and Guillaume Draznieks and Scott Sussex and Riccardo De Santi and Samuel Brugger and Timon Gradinger and Yuhui Hou and Hannes Frey and Mojmir Mutny and Andreas Krause and Christophe Copéret and Paco Laveille},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15001703/v3},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15001703/v3},
note = {CC BY},
}
RIS
TY - UNPB
TI - Case Study of Bayesian Batch Optimization Strategies for Heterogeneous Catalysts in CO2 Hydrogenation
AU - Erwin Lam
AU - Kourosh Shariat
AU - Leo Chevalier
AU - Guillaume Draznieks
AU - Scott Sussex
AU - Riccardo De Santi
AU - Samuel Brugger
AU - Timon Gradinger
AU - Yuhui Hou
AU - Hannes Frey
AU - Mojmir Mutny
AU - Andreas Krause
AU - Christophe Copéret
AU - Paco Laveille
PY - 2026
DA - 2026/09/14
PB - ChemRxiv
DO - 10.26434/chemrxiv.15001703/v3
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15001703/v3
AB - High-throughput and automated experimentation platforms combined with artificial intelligence are increasingly used to accelerate catalyst discovery. Batch Bayesian optimization (BO) (i.e., the simulta neous selection of multiple candidate experiments per iteration) is widely adopted for navigating large compositional spaces; however, its practical behavior under realistic high-throughput and batch con straints remains poorly understood. In this work, we performed a case study on four BO strategies to optimize the catalyst composition for the hydrogenation of CO2 to methanol. For each BO strategy, 48 catalysts were synthesized by co-precipitation, utilizing an automated synthesis platform, and evaluated in high-throughput fixed-bed reactors over seven batch iterations. The composition space comprised up to four metallic elements selected from a pool of nine. Although all BO strategies identified competitive catalyst formulations, most optimization trajectories rapidly converged toward distinct local optima that were strongly influenced by the initial random design and exhibited limited global exploration of the compositional space. This case study provides practical insights into the limitations of BO for batch catalyst optimization and highlights their strong dependence on the initial random design, parameter space definition, and batch execution strategy.
ER -