Nonlocal improvement of controls in dynamical systems that are quadratic with respect to the state (Q1405712)

The paper is concerned with the optimal control problem \[ \Phi(u)= \varphi(x(t_1))+ \int_T F(x(t),u(t),t)\,dt \to \min_{u \in V}, \] \[ {\dot x}(t)= f(x(t),u(t),t), \quad x(t_0)=x^{0}, \quad u(t) \in U, \quad t \in T=[t_0,t_1], \] where \(x(t)=(x_1(t),\dots,x_n(t))\), \(u(t)=(u_1(t),\dots,u_m(t))\), and \(V\) is the set of all piecewise continuous functions with values in a compact subset \(U \subset R^{m}\). The functions \(f(x,u,t)\) and \(F(x,u,t)\) are quadratic with respect to \(x\); the function \(\varphi(x)\) is quadratic on \(R^n\). With the use of exact (i.e., without remainder terms) formulas for the increment of \(\Phi(u)\), the author suggests two algorithms of nonlocal improvement of a current control. In the first algorithm, the equation for the conjugate variable \(p\) is linear with respect to \(x\) and \(p\); in the second algorithm, the equation for the state variable \(x\) is smooth in \(x\) and the equation for \(p\) is linear in \(x\) while boundary value problems of the maximum principle involve in general discontinuous right hand sides.