A class of infinitely divisible distributions connected to branching processes and random walks (Q1876709)

The authors introduce and investigate a class of infinitely divisible (id) probability measures on the nonnegative integers whose probability generating functions (pgf) are solutions of the differential equations \(P'(z)=cP^2(z)+zP(z)P'(z)\) with parameter \(c>0\). For any \(0<c<1\) the solution is shown to be the pgf of the total progeny in a certain (sub)critical branching process. This connection is used to explicitly calculate the id distributions with the knowledge of a corresponding negative binomial offspring. The explicit representation extends to the case \(c>1\) but an interpretation in connection with branching processes is given so far only if \(c/(c-1)\) is an integer. In this case the offspring is binomial and the id distribution is known as Consul's distribution. The case \(c=1\) is dealt with by continuation but different id distributions arise when letting \(c\uparrow1\) on the side of the offspring respectively the original pgf. The latter is connected to Lambert's W-function. Further, complete monotonicity for all \(c>0\) is shown. As a consequence, the id distributions are special negative binomial convolutions and can be represented as Poisson mixtures.