Viscosity Solutions of Second-Order Equations, Stochastic Control and Stochastic Differential Games

Abstract

In this note we review, explain and detail some recent results concerning the possible relations between various value functions of general optimal stochastic control and stochastic differential games and the viscosity solutions of the associated Hamilton-Jacobi-Bellman HJB and Bellman-Isaacs BI equations. It is well-known that the derivation of these equations is heuristic and it is justified only when the value functions are smooth enough (W.H. Fleming and R. Richel [15]). On the other hand, the equations are fully nonlinear, second-order, elliptic but possibly degenerate. Smooth solutions do not exist in general and nonsmooth solutions (like Lipschitz continuous solutions in the deterministic case) are highly nonunique. (For some simple examples we refer to P.-L. Lions [24]). As far as the first-order Hamilton-Jacobi equations are concerned, to overcome these typical difficulties and related ones like numerical approximations, asymptotic problems etc. M.G. Crandall and P.-L. Lions [8] introduced the notion of viscosity solutions and proved general uniqueness results. A systematic exploration of several equivalent formulations of this notion and an easy and readable account of the typical uniqueness results may be found in M.G. Crandall, L.C. Evans and P.-L. Lions [6]. It was also observed in P.-L. Lions [24] that the classical derivation of the Bellman equation for deterministic control problems can be easily adapted to yield the following general fact: Value functions of deterministic control problems are always viscosity solutions of the associated Hamilton-Jacobi-Bellman equations. The uniqueness of viscosity solutions and the above fact imply then a complete characterization of the value functions. This observation was then extended to differential games by E.N. Barron, L.C. Evans and R. Jensen [3], P.E. Souganidis [36] and L.C. Evans and P.E. Souganidis [14].