On the Bellman equation for some unbounded control problems (Q1376807)

Optimal control problems with unbounded components lead to Bellman equations which frequently do not satisfy the growth conditions which guarantee uniqueness of viscosity solutions. In addition, problems with unbounded components may not even have a continuous value function. This paper tries to address this problem for control problems which include the linear quadratic problem, which had fallen through the cracks of the uniqueness theory. The basic assumption on the dynamics for \(y'=f(y,\alpha)\) is that \[ (f(x,a)-f(y,a))\cdot(x-y) \leq L| x-y| ^2 \] and the controls \(\alpha\) take values in some closed set \(A\), possibly unbounded. The cost function is \(J(x,t,\alpha)=\int_0^t h(y,\alpha)e^{-\lambda s} ds + e^{-\lambda t}g(y_x(t))\). The functions \(h\) and \(g\) are assumed continuous and bounded from below. The value function \(V(x,t)=\inf_{\alpha}J(x,t,\alpha)\) is the object of study and the authors determine general conditions under which \(V\) is continuous, and can be characterized as the unique solution of the Bellman equation.