Strength of Interlayer Metal-Metal Coupling as Key Active Site Configuration and Atomic Descriptor for Single-Atom...

Single-atom catalysts (SACs) are widely considered for large-scale applications due to their exceptional activity, selectivity, and near-unity atom economy, where active-site configuration is critical. Yet how the active site regulates catalytic performance remains incompletely understood. A long-standing example is layered Fe-N-C, which shows a large discrepancy between theoretical predictions and experimental oxygen reduction reaction (ORR) activity, limiting rational design and predictability. Here, guided by X-ray absorption fine structure (XAFS) observations and supported by simulations and machine learning, we identify a stacked bilayer metal-metal coupling (MMC) configuration as an important active-site motif in these catalysts. Without invoking constant-potential treatments or surface hydroxyl coverage, our model, constructed within the standard computational hydrogen electrode (CHE) framework, reproduces the experimental activity and reconciles theory with experiment, supporting MMC as a plausible and important mechanistic contributor to ORR. By constructing and comparing 15 single-layer (SL) and MMC configurations, we quantify how metal-metal coupling strength governs activity and establish a criterion for selecting appropriate modeling strategies based on coupling intensity. Using machine learning and data mining, we further identify local electronic-structure descriptors that enable quantitative structure-activity relationships to guide catalyst design. Finally, we synthesize a series of molecular catalysts featuring bilayer MMC motifs or isolated single-metal sites, thereby experimentally validating the proposed structural motif and its role in ORR characteristics. This discovery provides guidance for the rational design of SACs and highlights that physically faithful active-site modeling is a prerequisite for predictive theory, with transferable implications for electrocatalytic systems beyond ORR.