Exponential-supermartingale escape bound
= Exponential-supermartingale escape bound
If $e^{-\theta N_{t\wedge\sigma_K}}$ is a nonnegative <supermartingale>, with $\sigma_K$ the first time $N_t\leq K$, the <optional stopping theorem> yields $\mathbb P_N(\sigma_K<\infty)\leq e^{-\theta(N-K)}$. Such a bound can prove escape of a backlog independently of heuristic <fluid models>.