Asymptotic theory of noncentered mixing stochastic differential equations (Q2485802)

Consider the stochastic systems \[ \frac{d}{dt}x^{\varepsilon}(t,s,x)=\varepsilon^{-1}F^{\varepsilon}(t,t/\varepsilon^2,x^{\varepsilon}) + G^{\varepsilon}(t,x^{\varepsilon}) \;,\quad x^{\varepsilon}(s,s,x)=x \in\mathbb R^d \;, \] over an interval \((0,t^*)\), where \(F^\varepsilon\) is a zero mean random field and \(G^\varepsilon\) is a deterministic field. Let \(\{F_s^t:\;0\leq s\leq t\leq \infty\}\) be a filtration of \(\sigma\)-fields satisfying \[ \sup_{s\geq 0} \sup_{A\in F_{s+t}^{\infty},\,B\in F_0^s} | P(A| B)-P(A)| =\rho(t)\downarrow 0,\quad \text{as \(t\uparrow \infty\),} \] with \(\int_0^{\infty} \rho^{1/2}(s)\,ds < \infty\). Assume that \(F^\varepsilon\) is measurable, and, for each fixed \(t\) and \(x\), \(F^\varepsilon(t,t/\varepsilon^2,x)\) is \(F^{t/\varepsilon^2}_{t/\varepsilon^2}\)-measurable. Assume that the interval \((0,t^*)\) can be divided into non-overlapping intervals \(I_n\), \(n=1,\dots,m_0\), such that \(\| \nabla\otimes\cdots\otimes\nabla\otimes (F^{\varepsilon},G^{\varepsilon})\| _0\leq C_{n,\varepsilon}\) on \(I_n\) up to order 4, where the constants \(C_{n,\varepsilon}\) are uniformly bounded in \(n\) and satisfy \(\sum_{n=1}^{m_0}C_{n,\varepsilon} \sim O(\varepsilon^{-\gamma})\) for some \(\gamma<1\). Then, for each \(f\in\mathcal C^4\), there exists a constant \(C_f\) independent of \(\varepsilon\) such that \[ \sup_{0\leq t\leq t^*} \| E\{U^{\varepsilon}(0,t)f\}-A^{\varepsilon}(0,t)f\| _0 \leq \varepsilon^{1-\gamma}C_f, \] where the operator \(A^{\varepsilon}\) solves a certain final value problem, which can be interpreted as an asymptotic Kolmogorov backward equation. A corresponding forward equation for the transition density can be deduced and the solution can be obtained in closed form.