In heterogeneous catalysis, density functional theory (DFT) calculations are often performed starting from multiple distinct initial adsorption configurations, many of which converge to the same final stable structure after relaxation. This redundancy wastes computational resources and complicates post-analysis. To address this challenge, we developed and benchmarked graph-theoretic approaches for systematically eliminating redundant configurations in catalyst systems. Configurations are represented as atomic graphs, and graph isomorphism algorithms are employed to identify structurally equivalent adsorption states. We compared the classical VF2 matcher, the mature Bliss canonical labeling algorithm, and a custom implementation of Luks’ 1982 bounded-valence algorithm paper. VF2 demonstrated high accuracy but proved prohibitively slow for large-scale workflows. Bliss achieved an effective balance between speed and correctness, enabling efficient identification of unique configurations across large datasets. Our custom Luks-based approach showed strong accuracy and, with further refinement, offers potential for greater scalability by incorporating orbit partitioning and quotient graph methods. Applied to CO–OH co-adsorption on Pt(111), the framework effectively identified unique binding motifs and reduced redundant configurations, thereby lowering computational overhead. Future work will focus on refining the Luks implementation with canonical labeling strategies to improve runtime efficiency further.
