(521ea) Streamlining Semiautonomous Workflows through Deep Learning for Materials Discovery in the Oxygen Evolution Reaction (OER)

This work introduces WhereWulff, a semi-autonomous workflow for modeling the reactivity of catalyst surfaces. The workflow begins with a bulk optimization task that takes an initial bulk structure and returns the optimized bulk geometry and magnetic state, including stability under reaction conditions. The stable bulk structure is the input to a surface chemistry task that enumerates surfaces up to a user-specified maximum Miller index, computes relaxed surface energies for those surfaces, and then prioritizes those for subsequent adsorption energy calculations based on their contribution to the Wulff construction shape. While the workflow handles computational resource constraints such as limited wall-time as well as automated job submission and analysis, it still suffers from the computational complexity of DFT. In order to streamline the magnetic characterization part of WhereWulff, we have devised a scheme that can predict both the magnetic ordering and moment of atoms in a crystal. We first train a classifier to generate three types of embeddings, partitioning our dataset into three spin types. A regressor is then applied to the spin types that show appreciable magnetic behavior, circumventing the difficulties faced when standard deep learning solutions are deployed on a skewed continuous outcome with many zeros. Results are promising with 93 % F1 score for the spin type classification and ~0.2 Bohr-magneton MAE on the regressor.