Optimal control of backward doubly stochastic system

摘要

An optimal control problem for backward doubly stochastic system is considered, where the control domain is not necessarily convex. By the method of classical spike variation and duality technique, one necessary condition and one sufficient condition are established for this kind of optimal control problem. The related adjoint process is characterised by the solution of a forward doubly stochastic differential equation, which formulates a forward-backward doubly stochastic differential equation coupled with the state equation. As an illustration, the authors' theoretical results are applied to study an optimal harvesting problem and a linear-quadratic optimal control problem. Moreover, the corresponding maximum principle with an initial state constraint is obtained by Ekeland's variational principle.