Automating the Analysis of Substrate Reactivity through Environment Interaction Mapping.

Exploring the interaction configurations between substrates and atomic or molecular systems is crucial for various scientific and technological applications, such as characterizing catalytic reactions, solvation structures, and molecular interactions. Traditional approaches for generating substrate-reactant configurations often rely on chemical intuition, symmetry operations, or random initial states, which can be inefficient and challenging for systems with low symmetry or unknown interaction mechanisms. This work introduces a systematic and automated methodology to explore the configuration space between substrates and adsorbates using symmetry-invariant features that characterize the local atomistic topology of the substrate. The approach involves three key components: (1) defining and discretizing a contact space surrounding the substrate, (2) utilizing symmetry-invariant descriptors to capture local atomic environments, and (3) employing unsupervised machine-learning techniques for clustering and hierarchical analysis of interaction sites. The method ensures comprehensive yet nonredundant sampling of the configuration space, independent of the substrate dimensionality. Applications to simple ideal substrates show that symmetry intuition and high-symmetry sites are correctly recovered. Moreover, the method is shown to translate seamlessly to less symmetric substrates.