Self-organized criticality in a discrete model for Smoluchowski's equation
We study a discrete model of coagulation, involving a large number $N$ of particles. Pairs of particles are given i.i.d exponential clocks with parameter $1/N$. When a clock rings, a link between the corresponding pair of particles is created only if its two ends belong to small clusters, i.e. of size less than $α(N)$, with $1 \ll α(N) \ll N$. The concentrations of clusters of size $m$ in this model are known to converge as $N \to \infty$ to the solution to Smoluchowski's equation with a multiplicative kernel. Under the additional assumption $N^{2/3} \log^γ(N) \le α(N)$, for some $γ> 1/3$, we study finer asymptotic properties of this model, namely the combinatorial structure of the graph consisting of small clusters. We prove that this graph is essentially an Erdos-Renyi random graph, which is subcritical before time 1, and remains critical after time 1. In particular, we show that our model exhibits self-organized criticality at a microscopic level: the limiting distribution of a typical finite cluster is that of a critical Galton-Watson tree. Our approach allows in particular to verify, under our additional assumption, a conjecture of Aldous.