An operator theoretic approach to nonparametric mixture models (Q2328057)

This research considers the distribution-free estimation of a normal mixture. In detail, in absence of any information on the distribution for each mixture component, it provides a quantification on the number of observations needed -- as a function of the number of the mixture components -- so that the mixture is identifiable. Additionally, this article quantifies the identifiability of mixtures with linearly independent and jointly irreducible components; it should be noted that the provided results are definite in the sense that they cannot be improved any further. An application, based on multinomial mixture models concludes this work.