A geometric characterization of viable sets for controlled degenerate diffusions (Q1611757)

Let \(X^{x,\alpha}\) denote the unique adapted solution to the equation \(dX_t= \sigma(\alpha_t, X_t)dB_t+ f(\alpha_t, X_t)dt\), \(X_0=x\), where \(B\) is an \(m\)-dimensional Brownian motion and \(\alpha\) belongs to the space of admissible controls \({\mathcal A}\) taking their values in some set \(A\). Assume that the set \(\{\text{tr} (\sigma\sigma^* (u,x)),\;f(u,x), u\in A\}\) is convex, for all \(x\in\mathbb{R}^n\). A closed set \(K\subset\mathbb{R}^n\) is viable w.r.t. the above SDE if, for every \(x\in K\), there is some \(\alpha\in {\mathcal A}\) such that \(P\{X_t^{x, \alpha}\in K\), for all \(t\geq 0\}=1\). In the deterministic case, \(\sigma=0\), and without control, this problem has first been studied by \textit{M. Nagumo} [Proc. Phys.-Math. Soc. Japan, III. Ser. 24, 551-559 (1942; Zbl 0061.17204)]. Since then a lot of authors have worked on this subject. For the stochastic case one should refer to the works by \textit{J.-P. Aubin} and \textit{G. Da Prato} [Stochastic Anal. Appl. 13, 1-11 (1995; Zbl 0816.60053); 16, 1-15 (1998; Zbl 0931.60059)] who have characterized the viability property of \(K\) with the help of a stochastic version of the Bouligand-Severi contingent cone. Further, \textit{R. Buckdahn}, \textit{S. Peng}, \textit{M. Quincampoix} and \textit{C. Rainer} [C. R. Acad. Sci., Paris, Sér. I., Math. 327, 17-22 (1998; Zbl 1036.49026)] have proved that \(K\) is viable if and only if the square of the distance from \(K\) is a viscosity supersolution of the Hamilton-Jacobi-Bellman equation associated to the control problem \(V(x)=\inf_{\alpha\in {\mathcal A}}E [\int^\infty_0 e^{-Ct} d^2_K (X_t^{x, \alpha})dt]\). The authors of the present paper use the second-order analogue of the positive polar cone \({\mathcal N}^2_K(x)\) of generalized interior normal vectors [cf. \textit{M. Bardi} and \textit{P. Goatin}, in: W. M. McEneaney (ed.) et al., Stochastic analysis, control, optimization and applications. Boston: Birkhäuser, 191-208 (1999; Zbl 0928.93067)]. They show that \(K\) is viable if and only if for every \(x\in\partial K\), \((p,Y) \in {\mathcal N}^2_K (x)\) there is some \(u\in A\) such that \(\tfrac 12\text{tr} (\sigma \sigma^* (u,x)Y)+ f(u,x)p\geq 0\). Their proof uses analytical tools of the theory of viscosity solutions of Hamilton-Jacobi-Bellman equations.