The averaging principle for stochastic difference equations (Q5900007)

Let for \(n=1,2,..\). the following objects be given: a flow (\({\mathcal F}_{ni}\), \(i=\overline{0,m_ n})\) of \(\sigma\)-algebras on a probability space \((\Omega_ n,{\mathcal F}_ n,P_ n)\), a partition \(O=t_{n0}<t_{n1}<...<t_{nm_ n+1}=1\) of the segment [0,1], an \({\mathcal F}_{n0}\)-measurable random variable \(X_{n0}\in R^ d\), \({\mathcal F}_{ni+1}\)-measurable random variables \(\beta_{ni}\in R^ d\), \({\mathcal B}(R^ d)\times {\mathcal F}_{ni}\)-measurable functions \(a_{ni}(x)=a_{ni}(x,\omega_ n)\in R^ d\), \(\omega_ n\in \Omega_ n\), \(i=\overline{O,m_ n-1}\). We define \(X_{ni}\), \(i=\overline{1,m_ n}\) by the difference relation \[ \Delta X_{ni}=a_{ni}(X_{ni})\Delta t_{ni}+\beta_{ni} \] and study their deviation from some nonrandom numbers \(\bar X_{ni}\) which usually are obtained from the difference relation with coefficients averaged by i and \(\omega_ n\). With this end we investigate the limiting behaviour of the processes \[ \xi_ n(t)=B_ n(X_{ni}-\bar X_{ni}),\quad t\in [t_{ni},t_{ni+1}), \] where \(B_ n\) are normalizing constants. Theorem 1 gives rather general and therefore hardly verifiable conditions for convergence of \(X_ n(t)\) to the solution to the equation \[ (1)\quad \xi (t)=\xi_ 0+\int^{t}_{0}f(u,\xi (u))du+\gamma (t) \] with nonrandom function f and process \(\gamma\) such that \(E| \gamma (t)-\gamma (s)|^ 2\leq L| t-s|\). Theorem 2 deals with the case when \(a_{ni}\) are nonrandom of x and \(z_{ni}\) where \((z_{ni})\) is a sequence of random variables such that \(\beta_{ni}\) are conditionally independent given \((z_{ni})\). Conditions are found for convergence of \(\xi_ n(t)\) to the solution to equation (1) with \(\gamma (t)=\psi (t)+\zeta (t)\) where \(\psi\) and \(\gamma\) are independent martingale and process with independent increments, respectively.