Classical and Relaxed Progressively Refining Discretization-Optimization Methods for Optimal Control Problems Defined by Ordinary Differential Equations

An optimal control problem is considered, for systems defined by nonlinear ordinary differential equations, with control and pointwise state constraints. Since the problem may have no classical solutions, it is also formulated in the relaxed form. Various necessary/sufficient conditions for optimality are first given for both formulations. In order to solve these problems numerically, we then propose a discrete penalized gradient projection method generating classical controls, and a discrete penalised conditional descent method generating relaxed controls. In both methods, the discretization procedure is progressively refining in order to achieve efficiency with reduced computational cost. Results are given concerning the behaviour in the limit of these methods. Finally, numerical examples are provided.