Carathéodory-Hamilton-Jacobi theory for nonsmooth problems of Lagrange and problems of optimal control (Q1067198)

This paper is devoted to the nonsmooth analogue of Carathéodory's method for the Lagrange problem with fixed end points in the calculus of variations. The normal integrand T(t,x,z) is measurable, is l.s.c. in (x,z), is convex in z and satisfies a growth condition. The necessary conditions of optimality are obtained in a Hamilton-Jacobi-type form by using convex analysis. Next, an analogue of the classical Weierstrass sufficiency theorem is proved. In the second part these results are applied to the optimal control problem of minimizing a Lagrange-type functional on trajectories of a general nonlinear system with variable domain of control and with phase restraints. This problem is reformulated by using Rockafellar's method. Conditions which guarantee the existence of measurable optimal feedback control are obtained. This method is illustrated by two examples: the linear-quadratic problem for a one-dimensional autonomous system and the problem of minimizing an integral functional on trajectories of a one- dimensional bilinear system.