Limit distributions for multitype branching processes of \(m\)-ary search trees (Q2451114)

This paper considers the Markov process that corresponds to random \(m\)-ary search trees. This is a multitype branching process with \(m-1\) types. Each particle of type \(j\) dies with an exponential rate \(j\) and gives rise to a particle of type \(j+1\), except when \(j=m-1\), where the dying particle produces \(m\) particles of type 1. This induces a continuous process on \(\mathbb{R}^{m-1}\) which is denoted by \(X(t)\): here, the \(k\)th coordinate of this vector is the number of particles of type \(k\). It has been known that for \(m \leq 26\), this vector converges (for \(t\rightarrow \infty\)) to a Gaussian. The authors of this paper consider the case where \(m \geq 27\) and give the limiting distribution, which in this case is a linear combination of functions of two Gamma-distributed random variables. They also consider the limiting random variable for the special case where the initial state is a single particle of type 1. Here, they give the asymptotic decay of its characteristic function as well as its Laplace transform close to the origin of the complex plane. To this end, the authors show that this random variable is the solution to a distributional fixed-point equation. They derive theorems for this solution, regarding the asymptotics of its Fourier transform and its relative density with respect to the Lebesgue measure on \(\mathbb{C}\).