Study of a one-dimensional optimal control problem with a purely state-dependent cost (Q2172963)

The objective is to maximize the functional \[ J(x(t)) = \int_0^T e^{-rt} \Phi(x(t)) dt \] among solutions of the system \[ \begin{aligned} x'(t) &= f(x(t)) + u(t)g(x(t)) \, ,\quad |u(t)| \le 1 \quad (0 \le t \le T),\\ x(0) &= x_0 \, , \qquad x(T) = x_T, \end{aligned} \] where both the state $x(t)$ and the control $u(t)$ are scalar functions. The function $\Phi(t)$ is assumed to have a unique maximum $x^*$ and to be increasing for $x < x^*$, decreasing for $x > x^*$. These conditions imply that the best value of the state is $x(t) = x^*$ at least in a subinterval, which gives $u(t) = - f(x^*)/g(x^*)$ in the subinterval. In view of the bound on $u(t)$, this is only possible if $|f(x^*)/g(x^*)| \le 1$. The authors call this the standard case. Both the standard and the general case yield to Pontryagin's maximum principle, but can also be solved by classical analysis and arguments based on Chyaplygin's comparison theorem for solutions of ordinary differential equations. The authors also deal with the case where the right boundary condition is omitted and a salvage term $S(x(T))$ is added to the cost functional. Finally, the results are compared with those coming from the maximum principle and applied to several concrete problems.