Instantaneous support propagation for \(\Lambda\)-Fleming-Viot processes (Q2105084)

This work is concerned with Fleming-Viot processes, which are measure-valued Markov processes on \(R^d\) subject to two competing mechanisms: mutation and resampling. Thinking of the process as the empirical measure of a particle system (or the limit of such a system as the number of particles goes to infinity), the mutation mechanism, which tends to spread the measure, describes the individual motion of particles independently one from the other, and is here given by a Levy process, while the resampling mechanism, which tends to contract the measure, describes how blocks of particles jump at some rate to the location of one of them. The resampling mechanism is said to come down from infinity if, in the absence of mutations, starting at time 0 from an infinite countable set of particles with different values, for all positive times, there is only a finite number of values in the system (this is for instance the case for binary resampling, namely when a resampling event is simply a single particle jumping to the location of another one). It was known that, when the mutations are given by Brownian motions and the resampling mechanism comes down from infinity, then almost surely the Fleming-Viot process has compact support for all positive time (i.e. coalescence wins over mutations, in a sense). The main result of this work is that, on the contrary, when mutations have a non-trivial Levy measure then, independently from the resampling mechanism, for all positive times, the support of the process is stable by convolution with this Levy measure (so that, in particular, it is the whole space in many cases of interest, like alpha stable processes). An ingredient of the proof is the so-called lookdown construction, which represents the process as the limit as N goes to infinity of the empirical measure of a particular system of N particles.