Families and clusters of diffuse interstellar bands: A data-driven correlation analysis

More than 500 diffuse interstellar bands (DIBs) have been observed in astronomical spectra, and their signatures and correlations in different environments have been studied over the past decades to reveal clues about the nature of the carriers. We compare the equivalent widths of the DIBs, normalized to the amount of reddening, EB-V, to search for anticorrelated DIB pairs using a data sample containing 54 DIBs measured in 25 sightlines. This data sample covers most of the strong and commonly detected DIBs in the optical region, and the sightlines probe a variety of interstellar medium conditions. We find that 12.9 per cent of the DIB pairs are anticorrelated, and the lowest Pearson correlation coefficient is rnorm ∼-0.7. We revisit correlation-based DIB families and are able to reproduce the assignments of such families for the well-studied DIBs by applying hierarchical agglomerative and k-means clustering algorithms. We visualize the dissimilarities between DIBs, represented by 1 - rnorm, using multidimensional scaling (MDS). With this representation, we find that the DIBs form a rather continuous sequence, which implies that some properties of the DIB carriers are changing gradually following this sequence. We also find that at that least two factors are needed to properly explain the dissimilarities between DIBs. While the first factor may be interpreted as related to the ionization properties of the DIB carriers, a physical interpretation of the second factor is less clear and may be related to how DIB carriers interact with surrounding interstellar material.