Networks of reinforced stochastic processes: asymptotics for the empirical means (Q2325374)

Consider a network of \(N\) interacting stochastic processes, positioned at the vertices of an underlying graph governing which processes interact directly. The processes at each vertex take values in \(\{0,1\}\), updating at each time \(n\) according to conditionally independent Bernoulli random variables with parameters which are linear combinations of ``inclination'' random variables \(Z_{n,j}\) at time \(n\) for those vertices \(j\) with which a given vertex interacts directly. The authors prove limiting results for the empirical means of these stochastic processes, that is, asymptotics for the proportion of time each process spends in the state \(\{1\}\). Their first result shows that these empirical means all converge almost surely to the same random variable, \(Z_\infty\). To complement this result, the authors also prove joint central limit theorems for these empirical means and the ``inclination'' variables. The covariance structures of the limiting Gaussian random variables obtained in these results depends on the asymptotic behaviour of a sequence governing the update of the ``inclination'' random variables at each time step, the underlying graph, and the random variable \(Z_\infty\). The authors also discuss some statistical applications of their results, such as the construction of asymptotic confidence intervals for statistics of interest and inference on the underlying graph structure.