We consider a random growing graph model with two \emph{types}, or \emph{communities}, that belongs to the class of preferential attachment models. Starting from an initial graph $G_0$ in which each node has a type, the graph $G_{n+1}$ is generated by addition of a new node, which attaches through an edge to a node $v$ of $G_n$. The probability of attaching to $v$ depends on the communities of both the new node and $v$, and is proportional to $\operatorname{deg}_n(v)^γ$, where $γ\in[0, 1]$. We establish a convergence result for the limiting proportion of vertices of each degree and in each community in a certain regime of the parameter $γ$. To this end, we first use stochastic approximation techniques to reduce the convergence problem to the study of a non-linear evolution equation. We then use a probabilistic coupling approach to investigate the long-time behavior of the unique solution to this evolution equation. We emphasize that the techniques previously used to handle the single-community case cannot be extended directly to our setting. Indeed, the associated continuous-time embedding is not a multitype branching process in the usual sense, since the transition rates at any given time $t$ depend on the entire graph structure at that time.
🌊
