Phase transition of the contact process on random regular graphs (Q287739)

The contact process on a graph is a Markov process on subsets of vertices, where vertices neighboring the current set are added with rate \(\lambda\), while vertices included anywhere in the current set are removed with rate \(1\). Since the empty set is an absorbing state, a natural question is to investigate whether extinction occurs when the graph is infinite. More generally, one can look at local extinction: does the process started from some set \(A\) eventually does not contain any vertice of \(A\) at some later time? If for any finite set \(A\) local extinction occurs with probability \(1\) even though global extinction does not, we say the process survives weakly. Given an infinite graph, there exists \(\lambda_1 \leq \lambda_2\) such that global extinction occurs for \(\lambda< \lambda_1\), local extinction occurs for \(\lambda \in (\lambda_1, \lambda_2)\), and local extinction has a probability strictly smaller than one for any finite set if \(\lambda>\lambda_2\). While for \(\mathbb{Z}^d\) we have \(\lambda_1 = \lambda_2\), the two values are different for infinite regular trees. The present work is interested in the asymptotics of the global extinction time \(\tau_{G_n}\) for growing finite graphs \(G_n\) (for which extinction always eventually occurs) that converge to some infinite graph \(G\). A phase transition has been shown to hold for certain classes of graphs: the extinction time grows logarithmically in the size of \(G_n\) when \(\lambda<\lambda_2(G)\), and exponentially for \(\lambda>\lambda_2(G)\). Such results have been previously obtained for \(\mathbb{Z}^d\) (see [\textit{T. M. Liggett}, Stochastic interacting systems: contact, voter and exclusion processes. Berlin: Springer (1999; Zbl 0949.60006)] and references therein) and for infinite regular trees [\textit{M. Cranston} et al., ALEA, Lat. Am. J. Probab. Math. Stat. 11, No. 2, 385--408 (2014; Zbl 1300.82019)]. In this work, the authors study the case of random regular graphs. The main result is that a similar phase transition holds, but unlike previous results it occurs at \(\lambda_1(\mathbb{T}^d)\) instead of \(\lambda_2\) (where \(\mathbb{T}^d\) is the infinite \(d\)-ary tree). More precisely, they show that: (i) For any \(\lambda<\lambda_1(\mathbb{T}^d)\) there exists \(C> 0\) such that \[ \lim \mathbb{P}(\tau_{G_n}<C\log n) = 1; \] (ii) For any \(\lambda> \lambda_1(\mathbb{T}^d)\) there exists \(c>0\) such that \[ \lim \mathbb{P}(\tau_{G_n}> e^{cn}) = 1; \] where \(G_n\) is a random \((d+1)\)-regular graph with \(n\) vertices. Part (i) is obtained by comparison with the situation for regular trees, while part (ii) relies on geometric properties of the random graph.