Researchers from the Montreal Institute for Learning Algorithms (MILA), McGill University, and the University of Toronto have published a study that challenges the current approach to computational catalyst discovery, which is crucial for advancing green chemical synthesis and electrochemical energy storage technologies. The team, led by Shahana Chatterjee and including prominent figures like Yoshua Bengio and David Rolnick, has investigated the limitations of using thermodynamic overpotential as a primary metric for identifying promising catalyst candidates. Their findings were published in the journal Nature Communications.
The study focuses on the use of adsorption energies to calculate thermodynamic overpotentials, which are commonly employed in computational and machine learning-driven catalyst discovery processes. The researchers utilized datasets from the Open Catalyst Project (OC20 and OC22) to assess the practicality of these estimates. They began by quantifying the uncertainty in predicting adsorption energies using ab initio methods, determining that a conservative estimate for the uncertainty in a single adsorption energy prediction is around 0.3-0.5 eV.
The team then computed the overpotential for all materials in the OC20 and OC22 datasets for the hydrogen and oxygen evolution reactions. While they found that the overpotential could identify known good catalysts like platinum and iridium oxides, they also discovered that the inherent uncertainty in these predictions is large enough to misclassify a significant portion of the dataset as “good” catalysts. This limitation reduces the value of overpotential as a reliable screening criterion.
The researchers conclude that relying solely on overpotential estimation is insufficient for effectively sorting through catalyst candidates. They advocate for a shift in focus within the computational catalysis and machine learning communities towards other important metrics, such as synthesizability, stability, lifetime, and affordability. This shift could lead to more practical and efficient catalyst discovery processes, ultimately benefiting the energy sector by accelerating the development of green technologies.
The study highlights the need for a more comprehensive approach to catalyst discovery, one that considers multiple factors beyond just thermodynamic overpotential. By doing so, researchers can better identify promising candidates that are not only theoretically sound but also practical and economically viable. This research was published in Nature Communications, a highly respected journal in the field of scientific research.
This article is based on research available at arXiv.