A law of large numbers for stochastic difference equations (Q1185780)

The author proves a strong law of large numbers for a class of random variables and explains its scope by means of some examples. Let \(F_ 0\subset F_ 1\subset\dots\) be an increasing sequence of \(\sigma\)-fields on some probability space. Let \(X_ 0,X_ 1,\dots\) be a sequence of random variables such that \(X_ n\) is \(F_ n\)-measurable and such that the following condition is satisfied: (A) There are random variables \(\xi_ 1,\xi_ 2,\dots\) such that a.s. \(E(\xi_{n+1}\mid F_ n)=0\), and there is a positive bounded function \(h(x)\), \(x\geq 1\), such that for \(X_ n\geq 1\) and \(n\geq 0\), \(X_{n+1}/X_ n=1+h(X_ n)+\xi_{n+1}\). Let \(C_ \infty=\{X_ n\to\infty\), as \(n\to\infty\}\). Define the sequence of real numbers \((a_ n)\) by the recursion \(a_{n+1}=a_ n(1+h(a_ n))\), \(a_ 0=1\). The main result is the following theorem: Let (A) be satisfied. Let \(\phi(x)\), \(x\geq 1\), be a positive function such that \(\phi(x)/x\) is decreasing and \(\int_ 1^ \infty\phi(x)dx/x<\infty\). If for \(1\leq x\leq y\), \(h(y)\leq h(x)(1+\phi(x))\), and if there is a number \(1\leq p\leq2\) such that \(E(|\xi_{n+1}|^ p\mid F_ n)\leq h(X_ n)\phi(X_ n)\) on the event \(\{X_ n\geq1\}\), then \(X_ n/a_ n\) converges to a strictly positive limit a.s. on \(C_ \infty\). If \(h(x)=o(1)\), then \(X_ n/a_ n\to 1\) almost surely on \(C_ \infty\).