Disturbance neutralization principle for optimized dynamical systems (Q498280)

The dynamics of the system are given by the vector differential equation \[ x'(t) = f(t, x(t), u(t)) + w(t) \, , \quad x(t_0) = x_0 \, , \quad x(t_1) = x_1 \, , \quad u(t) \in U \] in \(n\)-dimensional space, where \(u(t)\) is an \(m\)-dimensional control and \(w(t)\) is a disturbance. The problem is to maximize the functional \[ \int_{t_0}^{t_1} f(t, x(t), u(t)) dt \, . \] Under smoothness and convexity conditions on the right side of the equation and the functional, it is known that an optimal control exists and can be expressed directly from the trajectories by a feedback law. In absence of convexity, optimal controls may be relaxed rather than ordinary, but suboptimal controls can still be constructed by feedback (a suboptimal control misses the maximum by \(\epsilon > 0,\) where \(\epsilon\) is fixed in advance). The main results start with such a feedback law for the undisturbed case \(w(t) = 0\) and end with a similar feedback law under perturbation.