Viscosity solutions of optimal stopping problems (Q4385261)

Let \(\{X_t\}_{t\geq 0}\) be an Itô diffusion in \(I R^n\) with the corresponding infinitesimal generator \(L\). The value function \(\Phi\) of the optimal stopping problem \(\Phi (x)=\sup _{\tau}E^x\left[\int _0^{\tau}f(X_t)dt+g(X_{\tau})\right]\), where the stopping times \(\tau\) act before leaving a given region \(S\subset \mathbb{R}^n\), under some smoothness conditions can be found as the unique solution of corresponding variational inequalities, cp. \textit{B. Øksendal} [``Stochastic differential equations'' (1995; Zbl 0841.60037)]. But it seems that \(\Phi\) is not always smooth enough. Connected with stochastic control problems, \textit{M. G. Crandall} and \textit{P.-L. Lions} [Trans. Am. Math. Soc. 277, 1-42 (1983; Zbl 0599.35024)]\ have developed the so-called viscosity solution of the corresponding Hamilton-Jacobi-Bellman equations. Under mild conditions, \(\Phi\) is the unique viscosity solution of the variational inequalities \[ \min\{-L\Phi (x)-f(x),\Phi (x)-g(x)\}=0\quad \forall x\in S,\quad\Phi (x)=g(x)\quad\forall x\in\partial S. \] So the viscosity solutions got a stochastic interpretation in a very generality. An illustrative example with a non-smooth \(\Phi\) is given.