Generalization to the sufficient conditions for a discrete random variable to be infinitely divisible (Q1105264): Difference between revisions

It is known that a discrete probability distribution \(\{p^ i\}\), \(i=0,1,2,..\). is infinitely divisible if it is log-convex. The author obtains a more general condition for infinite divisibility by showing that it is sufficient to know that the sequence \(\{p^ i\}\), \(i=m, m+1,..\). is log-convex for some \(m\geq 0\) together with finitely many explicit inequalities that can be checked recursively. The utility of these generalized sufficiency conditions is illustrated with the univariate generalized Waring distribution, which is not log-convex but has a log-convex tail for certain choices of its three parameters.