Limit distribution for sums of inhomogeneous Markovian Bernoulli variables

Abstract: Let ${\eta_i}{i\ge 1}$ be a sequence of dependent Bernoulli random variables. While the Poisson approximation for the distribution of $\sum{i=1}^n\eta_i$ has been extensively studied in the literature, this paper establishes novel convergence regimes characterized by non-Poisson limits. Specifically, when ${\eta_i}{i\ge 1}$ exhibits a Markovian dependence structure, we show that $\sum{i=1}^n\eta_i,$ under appropriate scaling, converges almost surely or in distribution as $n\to\infty$ to a random variable with geometric or gamma distribution. As concrete applications, we derive the distribution of the number of times the population size in certain branching processes attains a given level.