A new iterative method for discrete HJB equations (Q957940)

The goal of this paper is to propose a successive relaxation iterative algorithm for discrete Hamilton-Jacobi-Bellman equation: \((1) \max_{1\leq j\leq K} \{A^JU-F^J\}=0\) where \(A^j \in \mathbb R^{n \times n}, F^j \in \mathbb R^n, j=1,2,\dots K\). Equation (1) is a system of nonsmooth nonlinear equations. A successive iterative scheme, similar to the Jacobi iteration for a linear system is presented. Main result: Numerical tests show that the authors' iterative scheme is faster than algorithm A (a simultaneous iterative scheme). It may be a new choice for solving the subproblems of domain decomposition methods for (1). It is also a possible nonlinear smoother if a new multigrid method for (1) is constructed. The new iterative scheme and the convergence theorem (the monotone convergence for the algorithm) are proposed. Finally some theoretical discussions are also presented.