Evaluation of the classification of pre-solar silicon carbide grains using consensus clustering with resampling methods: An assessment of the confidence of grain assignments

Abstract

ABSTRACT We report the use of several cluster analysis techniques to evaluate the classification of pre-solar silicon carbide (SiC) grains. The stability of clusters and the confidence of individual cluster assignments of grains are assessed using consensus clustering with resampling methods. Our analysis shows that pre-solar SiC grains can be divided into seven groups that are found to be highly stable with most of the grains being assigned to the same cluster for at least 90 per cent of the time over multiple aggregated clustering. Among the seven groups, two groups are dominated by AB grains, three groups by MS grains, one group by Z grains, and one group by X grains. The further division of X grains into two groups is highly dependent on the chosen algorithm and is therefore uncertain. Z and Y grains are clustered jointly with MS grains, with one group dominated by Z grains, pointing to their common origins from low-mass asymptotic giant branch stars. The most stable N grain-containing clusters are dominated by 15N-rich AB grains. However, some methods assign N grains with X grains, but in less stable clusters. The suggested genetic relationship among 15N-rich AB, N, and X grains is in line with the recent proposal that all three types of pre-solar SiC grains came from core collapse supernovae. We discuss the results from different clustering techniques based on our assessment of the cluster stabilities and the extent to which the cluster assignments overlap across the different methods.