Differential games and BV functions (Q792235)

Let V(t,x) be the value of a two-person zero-sum differential game ending at time T, with payoff \[ g(x(T))+\int^{T}_{t}h(\zeta,x(\zeta),y(\zeta),z(\zeta))d\zeta, \] where \(x(\zeta)\in {\mathbb{R}}\) is the state at time \(\zeta\) and \(y(\zeta)\) and \(z(\zeta)\) are the controls of the two players at time \(\zeta\). It is assumed that the Isaacs condition holds (so that V exists). Let H(t,x,p) be the Hamiltonian, so that \[ \frac{\partial V}{\partial t}+H(t,x,\frac{\partial V}{\partial x})=0\quad a.e.\quad for\quad O<t<T,\quad x\in {\mathbb{R}} \] is the Hamilton-Jacobi equation that V solves. The main result is that if \(H_ x\) is continuous in all variables, if \(H_ x(t,x,p)\) is Lipschitz continuous in x and p, if g'' is Hölder continuous, and if some relatively standard technical conditions hold, then \(\partial V/\partial x\) is a function of bounded variation (i.e. a BV function) on (0,T,)\(\times {\mathbb{R}}\). If the same conclusion can be proved witth slightly weaker conditions on H and without being restricted to a one-dimensional state variable, this would be a major tool in the study of singular surfaces (the set where \(\nabla_ xV\) fails to be continuous) and therefore of great help in the study of optimal controls, both in differential games and in optimal control theory.