Self-decomposable laws from continuous branching processes (Q2297328)

A continuous-time and space branching process (CBP) is a Markov process \((Z_t : t \geq 0)\) taking value in \([0,\infty]\) that satisfies the branching property, i.e., that the process starting from \(z_1 + z_2\) can be constructed as the sum of two independent copies of the same process, starting from \(z_1\) and \(z_2\) respectively. The CBP can be constructed as the Lamperti transform of a spectrally positive Lévy process. Assuming that \(m := \log E(Z_1) \in (0,\infty)\), the process \((Z_t e^{-mt}, t \geq 0)\) is a non-negative martingale. In that situation, there exists a norming function \(\eta_t\) such that \(\eta_t Z_t\) converges a.s. to \(M\) as \(t \to \infty\). The law of \(M\) either is compound Poisson or self-decomposable. This paper explores some of the general properties of this law in the latter case. A fundamental question is whether the cumulant function of the martingale is a Thorin-Bernstein function. The article makes some progress in this direction by showing this cumulant is special Bernstein if the cumulant function of the subordinator generating \(Z\) also is special Bernstein. A specific one-parameter family of martingale limit cumulant functions are shown to be Thorin-Bernstein. The principal Lambert \(W\)-function being a boundary case of this family, this gives a new proof that it too is Thorin-Bernstein. Additionally, tail estimates of the distribution functions for this family are derived along with the right-hand tail and integral representation of their Lévy densities.