Unsupervised Classification of Local Clathrate Hydrate Structures

Quantifying and differentiating the structural characteristics of clathrate hydrates at the molecular level is crucial for understanding the properties that underpin hydrate-based technologies. While useful, current approaches lack sufficient resolution to discern, e.g., interfacial and dynamical structures. In this study, we present an algorithm based on Density-Based Spatial Clustering of Applications with Noise (DBSCAN) that accurately identifies different water states coexisting within clathrate hydrates. A key novel component is an effective cavity-finder algorithm, which provides input to the clustering framework. The new algorithm detects hydrate cavities by analyzing the number and type of constituent molecular rings around voids. Integrating the new algorithm with widely used order parameters (e.g., F3, F4, and F4t) provides a powerful and accurate tool for analyzing hydrate structures at interfaces and phase transitions. The performance of the new algorithm is assessed for structure I (sI) CO₂ hydrates and for structure II (sII) mixed CH₄/Dioxane hydrates, demonstrating its robustness and adaptability across different clathrates. Crucially, the proposed algorithm enables us to identify partially ordered structures characteristic of the quasi-liquid layer, thereby capturing interfacial dynamics and molecular-scale details essential for understanding hydrates under realistic conditions.