The mitigation of climate change and the stability of the future chemical industry depend on a shift to renewable energy sources and sustainable raw materials. To address this grand challenge, societal changes as well as scientific innovations are needed. The vision of Power-to-X, where renewable energy can be stored and utilized when and where it is needed, can only be facilitated by new catalyst materials. These must be made from earth-abundant elements and exhibit high catalytic activity for the technologies to mature and scale up to fulfill the goals of a global green transition.
High-entropy alloys are a novel material that could prove a powerful discovery platform for addressing this requirement. As solid solutions with typically five or more principal elements, they possess a high degree of tunability towards many di↵erent applications.
However, both the number of unique materials and the heterogeneous structure of these alloys make it an arduous task to experimentally search these high-dimensional composition spaces for catalyst materials. Therefore, a high-throughput predictive model is needed for an exhaustive screening.
This work strives to develop a computational framework for predicting the electrocatalytic oxygen reduction reaction activity of a high-entropy alloy surface by adaptation of the canonical descriptor-based approach to catalyst discovery. This framework employs machine-learning algorithms to infer the adsorption energies of *OH and *O. These energies are used to establish the characteristic gross and net adsorption energy distributions of a surrogate surface model, from which the catalytic activity can be predicted. As this process only takes a few seconds, it can be implemented in an exploration/exploitation search strategy to identify promising catalyst compositions marked for synthesis and electrochemical testing.
Generally, we find good agreement between predicted trends and experimental measurements. By using the measured values to fit our predictive model, we enhance the accuracy of subsequent predictions and bring theory into the active learning loop. This not only enables extrapolation to other composition spaces, but allows us to extract information about the descriptor-to-activity mapping and ultimately gain insight into the underlying reaction mechanisms.
There are still obstacles to overcome before we can e↵ectively optimize the catalytic activity of high-entropy alloys. For example, assertion of catalyst stability and elucidation of the compositional changes that the surfaces undergo during reaction conditions. However, with this work, we have taken important steps towards modeling catalysis on high-entropy materials and it will be exciting to see what new discoveries will be made in this vast composition space.