Hamilton-Jacobi theory and parametric analysis in fully convex problems of optimal control (Q2574774)

Continuing the work in [\textit{R. T. Rockafellar} and \textit{P. R. Wolenski}, SIAM J. Control Optim. 39, No. 5, 1323--1350 (2000; Zbl 0998.49018); ibid., 1351--1371 (2000; Zbl 0998.49019)], the author proves a large number of regularity properties of the \textit{value function} \[ p(\pi,\tau):=\inf_{x(.)\in {\mathcal A}_n^1} \biggl[g(x(0))+ \int_0^\tau L(x(t), x'(t))dt+h(\pi,\tau,x(\tau))\biggr] \] in the ``fully convex case'', in which all the data, \(g(.)\), \(L(.,.)\), \(h(\pi,\tau,.)\), are extended, convex, proper and l.s.c. functions with certain growth properties. Using tools from Convex Analysis, Nonsmooth Analysis and Hamilton-Jacobi Theory, the author proves in Theorem 1 the existence and certain characterizations of the (parametric) optimal trajectories, in Theorem 2, the Lipschitz continuity and some characterizations of the subgradients of the value function, in Theorem 3, the semidifferentiability of the value function and, in Theorem 4, the (Fréchet) differentiability of the value function under some additional hypotheses on the data.