Maximum process problems in optimal control theory (Q2387499)

Given a standard Brownian motion \(B\), the author associates to the controlled stochastic differential equation \(X_{t}=x+\int_{0}^{t}v_{s}\,ds+\sqrt{2}B_{t}\), \(t\geq0\), the maximum process \(S_{t}=s\vee\max_{0\leq r\leq t}X_{r}\) and studies the value function \(V(x,s)=\sup_{v(.)}E\left[ S_{\tau}-c\tau\right] \), where \(c>0\) and the supremum is taken over all admissible control processes \(v(.)\) which are assumed to take their values in the interval \([\mu_{0} ,\mu_{1}]\)\thinspace\ (\(\mu_{0}<0<\mu_{1})\); \(\tau\) is the first moment when \(X\) leaves the interval \((\ell_{0},\ell_{1})\) (\(\ell_{0}<0<\ell_{1}\)).\ The author discusses the Hamilton-Jacobi-Bellman equation satisfied by the value function \(V\) and determines explicitly the optimal control \(v^{\ast}(.)\).\ It is shown that this control is of bang-bang type: At each time it pushes or pulls as hard as possible, i.e., \(v_{t}^{\ast}=\mu_{0}\) if \(X_{t}<g_{\ast }(S_{t})\) and \(v_{t}^{\ast}=\mu_{1}\) if \(X_{t}>g_{\ast}(S_{t})\), where the switching function \(g_{\ast}\) is derived from the Hamilton-Jacobi-Bellman equation.