Research Projects

This project addresses a key challenge in sustainable energy: the discovery of efficient, cost-effective electrocatalysts for the oxygen reduction reaction (ORR), which is central to fuel cells and metal–air batteries. While platinum-based catalysts offer high activity, their prohibitive cost, scarcity, and susceptibility to degradation drive the search for alternatives that combine high catalytic efficiency with greater stability and lower cost. Sub-nanometer transition metal clusters (3–30 atoms) offer a promising solution due to their high surface-to-volume ratios and unique quantum-confined electronic properties. These clusters operate in the non-scalable regime, where catalytic behavior is highly sensitive to size, structure, and composition. Even minor changes can dramatically affect performance. However, the same fluxionality and dense landscape of metastable states that make these clusters attractive also complicate their rational design. Traditional computational strategies, which emphasize static global minima and scaling relationships, cannot capture the dynamic nature of catalysis at this scale. This research will establish a machine learning (ML)-driven framework to accelerate the discovery and understanding of sub-nanocluster ORR catalysts. Integrating global structural optimization, grand canonical density functional theory (GC-DFT), and solvation models, the project will generate a robust dataset of adsorption energetics and reaction intermediates under realistic electrochemical conditions. Advanced descriptors capturing geometric, electronic, and environmental factors will be coupled with ML models such as graph neural networks and kernel-based regressors to learn complex structure–activity relationships. The central hypothesis is that catalytic activity in sub-nanoclusters arises from dynamic ensembles of metastable isomers, whose populations are sensitive to applied potential and solvent effects. ML models trained on this data will enable rapid screening of the vast chemical space, identifying catalysts that transcend traditional volcano plots and scaling relations. Augmented volcano plots incorporating fluxionality, solvation, and potential-dependent restructuring will serve as predictive tools for catalyst discovery. The methodology combines global optimization (genetic algorithms, basin hopping) with DFT calculations of ORR intermediates (*O, *OH, *OOH, O₂, H₂O), electrochemical modeling (GC-DFT, implicit solvation), and feature engineering (SOAP, MBTR, ACSF, d-band metrics, Bader charges). Trained ML models will efficiently predict adsorption energies and overpotentials for thousands of clusters. Promising candidates will undergo DFT validation and free energy analyses (CHE model). Machine-learned interatomic potentials (MLIPs) will enable AIMD simulations with explicit solvation to capture catalyst–electrolyte dynamics under operando conditions. Outcomes will include new insights into the role of fluxionality and electronic structure in ORR catalysis, along with practical guidelines for designing platinum-free catalysts. The tools and datasets developed will broadly support AI-driven materials discovery for sustainable energy.