Extragradient method for solving an optimal control problem with implicitly specified boundary conditions (Q2357118): Difference between revisions

The control system is \[ x'(t) = D(t)x(t) + B(t)u(t) + f(t) \quad (t_0 \leq t \leq t_1) \, , \quad x(t_0) = x_0 \] where the control \(u(t)\) is in \(L^2(t_0, t_1; {\mathbb R}^r)\) and the trajectory \(x(t)\) takes values in \({\mathbb R}^n.\) Here \(D(t),\) \(B(t)\) and \(f(t)\) are matrices of appropriate dimensions with piecewise continuous entries. The control problem is \[ \text{minimize} \;J(u) = g^0(x(t_1, u)) \quad \text{subject to} \;u \in U_0, \;g^j(x(t_1, u)) \leq 0, \] where \(U_0\) is convex and closed in \(L^2(t_0, t_1; {\mathbb R}^r)\) and \(g^0(x), g^1(x), \dots,g^m(x)\) are convex and continuously differentiable in \({\mathbb R}^n\) with locally Lipschitz continuous gradient. If the minimum is unique then \(x(t_1, u)\) is determined by the problem, hence the ``implicitly specified boundary condition'' in the title. The authors observe that a saddle point \((u^*, \lambda^*)\) of the Lagrangian \[ L(u, \lambda) = g^0(x(t^1, u)) + \lambda_1 g^1(x(t_1, u)) + \dots + \lambda_m g^m(x(t_1, u)) \] gives a solution \(u^*\) of the control problem and prove that an iteration method (called \textit{extragradient}) for approaching the saddle point is convergent.