Warning: following the coronavirus emergency, all the talks will be given remotely online. The link to connect to the seminar will be available in due time. A playlist of the remote talks already online is available on our YouTube channel »
Clustering clusters: unsupervised machine learning on globular cluster structural parameters
Mario Pasquato (INAF - Osservatorio Astronomico di Padova)
Tuesday 22/01/2019 @ 14:00, Sala IV piano Battiferro
I use a cluster analysis algorithm, Partitioning Around Medoids (PAM), to subdivide globular clusters (GC) in groups in the parameter space defined by their structural, chemical, and orbital parameters, based on the most recent catalogs. I find evidence that the natural number of groups is either two or three, with the latter being slightly preferred but not significantly so. The groups roughly correspond either to a disk/halo dichotomy or to a disk/inner halo/outer halo trichotomy, confirming previous results obtained qualitatively. I quantify the strength of the association of each GC with its assigned group using silhouette widths, which are also the basis for measuring the overall quality of the clustering structure.