Transfer learning with domain knowledge adaptation for stability evaluation of dual-atom catalysts on nitrogen-doped carbon.

Dual-atom catalysts are innovative materials with exceptional activity and selectivity. Yet, their stability remains a key challenge in catalyst design. Conventional characterization and synthesis techniques struggle to precisely identify active sites due to their vast configurations, while distinguishing the metal speciation remains challenging. In a theoretical approach, density functional theory based high-throughput screening is constrained by computational cost and time. Herein, we implemented transfer learning with domain knowledge adaptation for the evaluation of the stability against metal aggregation of DACs on nitrogen-doped carbon. The transferability of the stability descriptors applied to single-atoms on doped carbon to dual-atom catalysts on nitrogen-doped carbon, demonstrates their universality to more complex systems. Valuable insights were gained for the design of stable catalysts with the identification of the optimum metal pair and coordination environment combination, and the examination of stability-synergistic effect tradeoff.