A note on preservation of self-decomposability under geometric compounding (Q1096239)

The author considers compound-geometric random variables, i.e. random sums of the form \(C_ 0=\sum^{N_ 0}_{0}X_ j\) and \(C_ 1=\sum^{N_ 1}_{1}X_ j\), where the \(X_ j\) are i.i.d. and \(N_ k\) has a geometric distribution on \(\{k,k+1,...\}\) \((k=0,1)\). He especially considers \(C_ 1\)-compounds of self-decomposable distributions, and of `renewal distributions' \(\tilde F\) corresponding to F, having densities of the form \(\tilde f(x)=(1-F(x))/\mu.\) Let \(E_ n\) denote the Erlang distribution of order n. The following results are proved: 1. \(C_ 1\)-compounds of \(E_ 2\) are self-decomposable. 2. \(C_ 1\)- compounds of \(\tilde E_ 2\) are not self-decomposable. 3. \(C_ 1\)- compounds of \(E_ 3\) are not self-decomposable. The proofs are given by means of Kolmogorov's canonical measure. The paper concludes with some remarks of queueing models, where compound- geometric distributions arise naturally.