Scaling limit of dynamical percolation on critical Erdős-Rényi random graphs (Q2227717)

Consider Erdős-Rényi random graphs \(\mathcal G(n,p)\) obtained from the complete graph with \(n \) vertices by deleting edges independently with probability \(1-p\). As \(n\to \infty\), there is a critical phase when \(p:=p(\lambda, n) =\frac 1 n + \frac \lambda {n^{4/3}}\) for some \(\lambda \in{\mathbb R}\). It was proved in [\textit{L. Addario-Berry} et al., Probab. Theory Relat. Fields 152, No. 3--4, 367--406 (2012; Zbl 1239.05165)] that the critical Erdős-Rényi random graphs with appropriate scalings convergence for the Gromov-Hausdorff topology to \(\mathcal G_\lambda\), a collection of real random graphs. In [\textit{L. Addario-Berry} et al., Ann. Probab. 45, No. 5, 3075--3144 (2017; Zbl 1407.60013)], the convergence was proved for the Gromov-Hausdorff-Prokhorov topology. In the present paper, a dynamical percolation on critical Erdős-Rényi random graphs \(\mathcal G(n,p)\) is considered. It can be constructed from two Poisson processes associated to each edges of the complete graph. One process determines the times when an edge is added (or kept if there is already one) and the second process determines the times when an edge is deleted (if there is one). A dynamical percolation process can also be constructed on the continuous limit \(\mathcal G_\lambda\). It is shown that under appropriate scalings, this process is the limit of the dynamical percolation process starting from a critical Erdős-Rényi random graph. The appropriate topology is a new intricate version of the Gromov-Hausdorff-Prokhorov distance. The proofs require a detailed analysis of the structure of a multigraph and rely on the properties of the multiplicative coalescent obtained in [\textit{D. Aldous}, Ann. Probab. 25, No. 2, 812--854 (1997; Zbl 0877.60010)]. Convergence is also established for the coalescence process and for the fragmentation process under appropriate scalings. In the first case, edges are never deleted and a Poisson process determines the times where edges are added whilst in the second case, edges are never added and a Poisson process determines the times where edges are deleted.