Orateur
Description
In this talk, we introduce a new model called "decoupling random walks" (DRW), as an intermediate model to the Binomial Splitting process. It consists of a system of $k = N^\gamma$ continuous-time random walks on the complete graph with $N$ vertices. Particles starting from distinct vertices evolve independently, whereas particles starting from the same vertex do not: they share the same clock and the same chosen neighbour, and the decisions of whether to jump are sampled jointly. These dependencies progressively disappear as particles separate, and eventually, all particles evolve independently.
Although DRW differs from independent random walks only by this temporary dependence, its mixing behavior turns out to be significantly more complex. Indeed, we show that DRW exhibits a cutoff phenomenon at time $\beta(\gamma) \log N$, where the mapping $\gamma \mapsto \beta(\gamma)$ is non-differentiable at four transition points, leading to five distinct regimes. This is a joint work with P. Caputo, M. Quattropani, and F. Sau.