On Convergence of Attainability Sets for Controlled Two-Scale Stochastic Linear Systems (Q3128451)

The authors consider the two-scale stochastic linear system described by \[ \begin{aligned} dx_t= \bigl(A_1(t)x_t +A_2(t) y_t+B_1(t)u_t\bigr) dt+dw^x_t, \quad & x_0=0 \tag{1} \\ \varepsilon dy_t= \bigl(A_3(t) x_t+A_4(t) y_t+B_2(t) u_t\bigr) dt+ \sqrt \varepsilon dw^y_t, \quad & y_0=0 \tag{2} \end{aligned} \] where \(x\in R^k\), \(y\in R^n\) and \(u\in {\mathcal U}\in R^d\) are the state (slow and fast) and control vector, respectively, \(w_x\in R^k\) and \(w_y\in R^d\) are standard independent Wiener processes, \(t\in[0,T]\), \(\varepsilon \in[0,1]\), \(A_1\), \(A_2\), \(A_3\), \(A_4\), \(B_1\) and \(B_2\) are matrices of appropriate dimensions. The class of admissible controls \({\mathcal U}\) is defined as the set of all predictable processes \(u=(u_t)_{t\in [0,T]}\) with values in \({\mathcal U}\). The performance index is defined as in the Mayer problem \[ I=\inf_{\mu\in {\mathcal K}_\varepsilon (T)}\int g(x,y) \mu(dx dy) \tag{3} \] where \(g\) is a function on \(R^m= R^k\times R^n\) which is integrable with respect to the measure \(\mu\) of the attainability set \({\mathcal K}_\varepsilon (T)\). The set \({\mathcal K}_\varepsilon (T)\) is defined as the set of distributions of attainable terminal values of the solutions to system (1), (2). The authors showed under suitable assumptions that \[ \lim_{\varepsilon\to 0} {\mathcal K}_\varepsilon(T) ={\mathcal K}_0(T) \tag{4} \] in the Hausdorff metric in the space of compact subsets of the corresponding spaces of probabilistic measures.