Chemical segregation analysed with unsupervised clustering

Context. Molecular emission is a powerful tool for studying the physical and chemical structures in cold and dense cores. The distribution and abundance of different molecular species provide information on the chemical composition and physical properties in these cores. Aims. We study the chemical segregation of three molecules – c-C3H2, CH3OH, and CH3CCH – in the two starless cores B68 and L1521E, and the prestellar core L1544. Methods. We applied the density-based clustering algorithms DBSCAN and HDBSCAN to identify chemical and physical structures within these cores. To enable cross-core comparisons, the clustering input samples were characterised based on their physical environment, discarding the two-dimensional spatial information. Results. Clustering analysis showed significant chemical differentiation across the cores. The clustering successfully reproduces the known molecular segregation of c-C3H2 and CH3OH in all three cores. Furthermore, it identifies a segregation between c-C3H2 and CH3CCH, which is not apparent from the emission maps. Key features driving the clustering are integrated intensity, velocity offset, H2 column density, and H2 column density gradient. Different environmental conditions are reflected in the variations in the feature relevance across the cores. Conclusions. This study shows that density-based clustering provides valuable insights into chemical and physical structures of starless cores. It demonstrates that already small datasets covering only two or three molecules can yield meaningful results. In fact, this new approach revealed similarities in the clustering patterns of CH3 OH and CH3CCH relative to c-C3H2, suggesting that c-C3H2 traces more outer layers or lower-density regions than to the other two molecules. This allowed for insight into the CH3CCH peak in L1544, which appears to trace a landing point of chemically fresh gas that is accreted to the core, highlighting the impact of accretion processes on molecular distributions.