Optimal feedback control of nonlinear systems (Q1090919)

The authors consider a finite-horizon, deterministic optimal control problem without control or endpoint constraints. They propose the following recipe to determine an optimal feedback control law \(u=u(t,x):\) Assume a solution exists, and use it to define \[ (*)\quad z(t)=(\partial /\partial u)H(t,x(t),\quad p(t),\quad u(t,x(t))), \] where H is the usual Hamiltonian. The Pontryagin maximum principle asserts that z(t)\(\equiv 0\), which obviously implies that all derivatives of z are identically zero. Compute these derivatives from (*) and eliminate p from the resulting system to obtain a set of quasi-linear partial differential equations satisfied by u(t,x). Solve it. Two illustrative examples are given.