Random walks on dynamical percolation: mixing times, mean squared displacement and hitting times (Q495551)

The research is motivated by network models with structures evolving over time. The corresponding probabilistic set-up reads as follows. A given graph \(G=(V,E)\) evolves randomly in such a way that an edge \(e\in E\) performs \(0\to 1\) and \(1\to 0\) switches, independently of other edges, with probabilities \(p\mu\) and \((1-p)\mu\), respectively, where \(0< p\), \(\mu\leq 1\) are parameters of the induced Markov process \(\eta_t\) \((t\geq 0)\) on \(\{0,1\}^E\) having the stationary reversible product measure \(\pi_p\). This type of evolution of a random graph is known as a dynamical percolation. Next, a random walk \(X_t\) on the random graph \(\eta_t\), \(t\geq 0\), at rate \(1\) is defined, which chooses randomly a neighbor and then moves there if and only if the connecting edge is open (= at state 1) at that moment of time. As a result, the Markov process \(M(t):=\{(X_t,\eta_t): t\geq 0\}\) arises. The paper studies the case when \(G\) is either the \(d\)-dimensional discrete torus with vertex set \([0,1,\ldots,n)^d\) or the lattice \({\mathcal Z}^d\). In the framework of the above setting, the authors establish bounds on the mixing time for the whole system \(M_t\) and for the random walk \(X_t\). It is also shown that the model considered on the lattice \({\mathcal Z}^d\) exhibits the usual recurrence/transience dichotomy in \(d\).