Region-Resolved Interface Descriptors in III-nitride-nanocluster photocatalysts
Understanding how nanocluster cocatalysts modify the electronic structure of III-nitride surfaces is central to the rational design of efficient photocatalytic interfaces. Here, we introduce a region-resolved first-principles framework for analyzing GaN-, InGaN-, and ScGaN-based slabs decorated with six-atom elemental nanoclusters. Using a region-resolved projected local density of states (PLDOS) analysis, we separate electronic response of nanocluster-covered regions from that of uncovered semiconductor surface regions, revealing that semiconductor--nanocluster interfaces behave as laterally heterogeneous systems. In this picture, nanocluster-covered regions are associated with local band bending, charge injection, and nanocluster-induced gap states, whereas uncovered nitride regions retain surface-localized states and in-plane dipoles that may influence adsorbate interactions and water-activation pathways. These distinct regional responses motivate the definition of region-resolved interface descriptors, which are further integrated into an Interface Score: a physics-informed comparative metric that combines injection, activation, and interfacial-coupling contributions to rank heterogeneous model interfaces on a common descriptor basis. Machine-learning models trained on global and region-resolved descriptors are used primarily as interpretability tools to assess their correlation with single-H adsorption energetics. The results show that spatially averaged descriptors capture broad adsorption-energy trends, while region-resolved descriptors expose complementary local information related to interfacial electrostatics and charge separation. Together, this work establishes a descriptor framework for moving beyond spatially averaged electronic-structure analysis and for bridging first-principles calculations with experimental interpretation of heterogeneous photocatalytic interfaces.