Sensitivity, controllability, and necessary conditions of optimal control problems governed by integral equations (Q1895795)

The author considers an optimal control problem with standard cost and target structure, but in which the dynamics are governed by the controlled integral equation \[ x(t)= x(0)+ \int^t_0 \phi(t, s, x(s), u(s)) ds,\quad u(t)\in U,\quad 0< t\leq 1, \] and required to satisfy the unilateral constraint \[ g(t, x(t))\leq 0,\quad 0\leq t\leq 1. \] There are no convexity hypotheses; instead, the author works explicitly with Warga's relaxed (convexified) reformulation of the problem over a space of Radon probability measures. Through additive perturbations of the dynamic, endpoint, and state constraints (some of these in infinite-dimensional Hilbert spaces), the author establishes a relationship between (i) the multipliers appearing in the first-order optimality conditions in this problem and (ii) the generalized gradients of the problem's minimum value as a function of the perturbation vectors. The guiding principle is `proximal analysis', a technique whose usefulness for proving this kind of result is well established.