Moment conditions for a sequence with negative drift to be uniformly bounded in \(L^r\) (Q1613608)

Let \(\{X_n\}\) be a sequence of real-valued random variables. The authors study under what condition we get \(\sup _n {\mathbf E}X_n <\infty \). Since the motivation comes from stability of queueing networks they would like to avoid the assumptions including the independence, Markov property, \dots{} of increments. The main statement is as follows: Suppose that there are constants \(a>0\), \(J\), \(V\), and \(p>2\) such that \(X_0\leq J\), \({\mathbf E}(X_{n+1}-X_n \mid X_0,\dots ,X_n) \leq -a\) on the event \(\{X_n>J\}\), and \({\mathbf E}(|X_{n+1}-X_x|^p \mid X_0,\dots ,X_n)\leq V\). Then for any \(r\in (0,p-1)\) there is a \(c=c(p,a,V,J,r)\) such that \({\mathbf E}(X^+_n)^r <c\) for all \(n\). Examples show that all assumptions are essential and that the bound \(r<p-1\) is sharp.