Measure solutions for semilinear and quasilinear evolution equations and their optimal control (Q1570456)

Optimal control problems for ordinary differential systems \(y'(t) = f(t, y(t), u(t))\) may lack solutions without \(u\)-convexity assumptions on \(f(t, y, u);\) the fix for this problem is to enlarge the class of controls to that of relaxed controls \(\mu(t)\) whose values are probability measures on the control set \(U.\) Trajectories corresponding to relaxed controls are uniform limits of trajectories corresponding to minimizing sequences of ordinary controls. In optimal control problems for partial differential equations, for instance those than can be modelled as the Banach space system \[ y'(t) = Ay(t) + f(t, y(t), u(t)),\tag{1} \] we find a new sort of situation, in which trajectories corresponding to a minimizing sequence are merely weakly convergent. Since nonlinearities (except in special cases) are not weakly continuous, taking limits leads to a notion of ``measure solution'' of (1) where the measure averages the nonlinearity over a subset of the state space; due to noncompactness, the measure is merely finitely additive. The author introduces a definition of measure solution along these lines and then shows how to use these solutions in infinite dimensional problems, in particular one involving optimal control of fluid flow.