It isn’t an easy task to design efficient new catalysts. In the case of multiple element mixtures, for example – researchers must take into account all combinations and then add other variables such as particle size or surface structure; not only does this lead them towards a massive number of potential candidates, but it becomes increasingly difficult with every change that needs consideration.
Scientists employ computational design techniques to screen material components and alloy composition, optimizing a catalyst’s activity for a given reaction. This reduces the number of prospective structures that would need testing to be developed–a combinatorial approach with theory calculations must also occur. But such methods require combinatorial approaches coupled with theory calculations, and this can be complex and time-consuming.
Carnegie Mellon University (CMU) researchers introduce a deep reinforcement learning (DRL) environment called ‘CatGym.’ CatGym is a revolutionary approach to designing metastable catalysts that could be used under reaction conditions. It iteratively changes the positions of atoms on the surface of a catalyst to find the best configurations from a given starting configuration.
The researchers use their method to predict the surface reconstruction pathways of a ternary Ni3Pd3Au2(111) alloy catalyst. Therefore, they showed that The DRL program is more than just about exploring new surfaces; it can also generate pathways based on the energetically favorable ones.
The CatGym agent (trained with TRPO algorithm and a ternary Ni3Pd3Au2(111)) not only explores more diverse local and global minima configurations compared to the baseline MH (minima hopping) method but also generates kinetic pathways that lead there.
The researchers also show that their agent consistently returns to a global minimum energy path and its associated transition state, which is in good agreement with the minimum energy path predicted by nudged elastic band (NEB) calculations.