On large deviations for SDEs with small diffusion and averaging. (Q1877523)

A system of stochastic differential equations \[ \begin{aligned} dX^ \varepsilon _ {t} &= f(X^ \varepsilon _ {t}, Y^ \varepsilon _ {t})\,dt + \varepsilon \sigma _ 1 (X^ \varepsilon _ {t},Y^ \varepsilon _ {t})\,dw^ 1_ {t} + \varepsilon \sigma _ 3(X^ \varepsilon _ {t},Y^ \varepsilon _ {t}) \,dw^ 3_ {t}, \quad X^ \varepsilon _ 0 = x, \\ dY^ \varepsilon _ {t} &= \frac 1{\varepsilon ^ 2} B(X^ \varepsilon _ {t},X^ \varepsilon _ {t})\,dt + \frac 1{\varepsilon }C_ 1(Y^ \varepsilon _ {t})\,dw^ 1_ {t} + \frac 1{\varepsilon } C_ 2(Y^ \varepsilon _ {t}) \,dw^ 2_ {t}, \quad Y^ \varepsilon _ {0} = y, \end{aligned} \] in \(\mathbb R^ {d}\times \mathbb R^ {l}\) is studied. It is assumed that \((w^ 1,w^ 2,w^ 3)\) is a standard \(3d\)-dimensional Wiener process, \(\varepsilon >0\) a small parameter, \(f\), \(\sigma _ 1\), \(\sigma _ 3\), \(C_ 1\), \(C_ 2\) are bounded Borel functions, and \(B\) is a Borel function of linear growth. The functions \(B(\cdot ,y)\), \(\sigma _ 1(\cdot ,y)\), \(\sigma _ 3(\cdot ,y)\) are supposed to be continuous for each \(y\). Let \(f\) be globally Lipschitz in \((x,y)\), let the function \(CC^ {*} \equiv C_ 1C^ {*}_ 1 + C_ 2C^ {*}_ 2\) be continuous and \(C_ 2C^ {*}_ 2\) uniformly nondegenerate. Finally, it is assumed that either the matrix \((\sigma _ 1,\sigma _ 3)(\sigma _ 1,\sigma _ 3)^ {*}\) is nondegenerate and continuous, or all coefficients of the system are Lipschitz continuous. Under the stability hypothesis \[ \varlimsup _ {| y| \to \infty } \sup _ {| \beta | <a} \sup _ {x} \| CC^ {*}(y)\| ^ {-1}\bigl \langle B(x,y) + C_ 1(y)\sigma ^ {*}_ {1}(x,y)\beta , | y| ^ {-1}y\bigr \rangle = - \infty \] for every \(a>0\), it is proven that the slow components \(\{X^ \varepsilon \}_ {\varepsilon >0}\) obey a large deviation principle in the space \(C([0,1];\mathbb R^ {d})\) with a normalizing coefficient \(\varepsilon ^ {-2}\) and the rate function is identified.