Simple Heuristics for Advanced Sampling of Reactive Species on Surfaces

Understanding interactions between reactive species and surfaces is a central task in materials science and heterogeneous catalysis. Decades of research have produced strategies to automatically detects surface active sites on computationally modeled slabs, leveraging symmetry, tessellation or relying on efficient dynamics. Regardless of the method or protocol, the quality of starting configurations (fragment-surface structures) remains pivotal. However, a universal, systematic approach for placing any molecular species on any surface remains lacking. For complex interfaces, where automatic protocols reach their limit, human curation of configurations is required, making the study of such systems slow, biased and unscalable. Machine learning approaches offer a promising alternative, but they demand large, diverse, and unbiased datasets—highlighting the need for better data generation strategies. We introduce Surrogate-SMILES (*SMILES), a representation of molecules and their fragments bound to surfaces, along with AutoAdsorbate, an automated method for generating chemically meaningful configurations of any molecule (or fragment) on any surface. Our approach enables uniform sampling and the creation of high-quality datasets, supporting both physics-based simulations and data-driven models—including one-shot learning and efficient initialization of active learning protocols for interatomic potentials. This advancement expands applications across catalysis, nanotechnology, and environmental science, where precise control over surface interactions is essential for the rational design of functional materials.