A differential game with constrained dynamics and viscosity solutions of a related HJB equation (Q1849023): Difference between revisions

A first-order partial differential equation with constraints \[ \max\{-V-H(DV); V-g\} =0,\quad (DV,\gamma_i)=0, i\in\{i: x_i=0\}, \] is considered. Here \(g(t,x)\) and \(\gamma_i(t,x)\) are known bounded and continuous functions of \(t\), \(t\geq 0\), and \(x\in \mathbb{R}^d\); \(D\) denotes the gradient with respect to \(x\). The Hamiltonian \(H\) is equal to the pointwise minimum of a finite family of convex functions \(H_j\). Initial and boundary conditions have the following form: \[ V(t,0)=0, t\geq 0;\quad V(1,x)=f(x), x\in\mathbb{R}^d\setminus\{0\}. \] Such an initial and boundary-value problem is related to the value function of a differential game coming from robust queueing network optimization problems. The main result of the paper is a comparison theorem, which formulates conditions guaranteeing that any viscosity subsolution of the problem mentioned above does not exceed (pointwise) any viscosity supersolution. In particular, from this theorem, one immediately obtains uniqueness of the viscosity solution. The authors formulate finite-time and infinite-time horizon differential games whose value functions are the viscosity solutions to initial and boundary-value problems of the type mentioned above. Three examples of optimal queueing control are presented. The cost criteria and interpretation of these examples do not exactly fit into the theoretical results obtained in the reviewed paper. More adequate examples are planned to be considered in future publications.