Moderate deviations of inhomogeneous functionals of Markov processes and application to averaging. (Q1879536)

Let \((\xi ,\mathbf P_ {x})\) be a continuous time Markov process on a Polish space \(E\) and \(P\) its transition function. Suppose that \(\xi \) is exponentially ergodic: There exist an invariant measure \(\pi \) for \(P\), a finite function \(M\in L^ 1(\pi )\) and \(\rho <1\) such that \(\| P_ {t}(x,\cdot )-\pi \| \leq M(x)\rho ^ {t}\), \(t\geq 0\), \(\| \cdot \| \) denoting the total variation norm. Let \(f: [0,1]\times E\to \mathbb R^ {d}\) be a bounded measurable function, \(\int _ {E} f(t,\cdot )d\pi =0\) for all \(t\geq 0\). Denote by \(\omega (\varepsilon ) = \sup _{| t-s| \leq \varepsilon ,\; x\in E} | f(t,x)-f(s,x)| \) the modulus of continuity. Assume that \(\varepsilon ^ {-1/2}\omega (\varepsilon )\to 0\) as \(\varepsilon \downarrow 0\) and set \[ M_ \varepsilon (t) = \frac 1{\sqrt {\varepsilon }h(\varepsilon )} \int ^ {t}_ 0 f(s,\xi _ {s/\varepsilon })\, ds, \] where the function \(h\) satisfies \(h(\varepsilon )\to +\infty \), \(\sqrt {\varepsilon }h(\varepsilon )\to 0\) as \(\varepsilon \downarrow 0\). Then, whenever \(\mu \) is a probability measure on \(E\) such that \(M\in L^1(\mu )\), \(\mathbf P_ \mu \circ M_ \varepsilon (\cdot )^ {-1}\) satisfies a moderate deviation principle in the space \(C([0,1];\mathbb R^ {d})\) endowed with the uniform topology, with speed \(h^ 2(\varepsilon )\) and some good rate function. This result may be applied to the averaging problem. Consider a differential equation with a random right hand side \(\dot z^ \varepsilon (t) = \varepsilon b(z^ \varepsilon (t),\xi _ {t})\), \(z^ \varepsilon (0) = x\) in \(\mathbb R^ {d}\). Let \(\bar b(x) = \int _ {E} b(x,y)d\pi (y)\), let \(y\) be a solution to \(\dot y(t) = \bar b(y(t))\), \(y(0)=x\). Under suitable assumptions on the function \(b\) the process \(\varepsilon ^ {-1/2}h^ {-1}(\varepsilon )(z^ {\varepsilon }(t/ \varepsilon ) - y(t))\), \(0\leq t\leq 1\), satisfies a moderate deviation principle in \(C([0,1];\mathbb R^ {d})\).