SCAIM Seminar - Jan Blechschmidt

Hamilton-Jacobi-Bellman (HJB) equations are nonlinear partial differential equation that arise as optimality conditions of stochastic control problems. HJB equations often possess a variety of difficulties, e.g., discontinuous coefficients, vanishing viscosity, unknown boundary conditions, etc. One particular issue is the handling of non-existing second-order derivatives. In this presentation we focus on the discretization of HJB equations with a fully implicit timestepping scheme based on a semi-Lagrangian approach. We discuss the algorithmic idea in the context of a finite difference approximation and present numerical examples.