A Feasible Directions Algorithm for Optimal Control Problems with State and Control Constraints: Convergence Analysis (Q4210198)

The paper concerns a new first-order feasible directions algorithm for the solution of the optimal control problem with pathwise state inequality constraints: (P) \(\min_u \phi(x(1))\) s.t. \(\dot x(t)=f(t,x(t),u(t))\), a.e. on \([0,1]\), \(x(0)=x_0\), \(u(t)\in \Omega\) a.e. on \([0,1]\), \(h_i^1(x(1))=0\) \(\forall i\in E\), \(h_j^2(x(1))\leq 0\) \(\forall j\in I\), \(q(t,x(t))\leq 0\) \(\forall t\in [0,1]\) with finite sets of index values \(E\), \(I\), functions \(f:R^m\times R^n\to R^n\), \(\phi:R^n\to R\), \(h^1_i:R^n\to R\), \(h^2_j:R^n\to R\), \(q:[0,1]\times R^n\to R\), \(x_0\in R^n\) and \(\Omega=\{u\in R^m\): \(b_-^i\leq u_i\leq b_+^i\), \(i=1,2,\dots,m\}.\) The control problem (P) is expressed as an optimization problem over the set of control functions \(\mathcal U=\{u:T\to R^m: u\) is measurable and \(u(t)\in \Omega\) a.e. on \(T=[0,1]\}\) in a form \(\min_{u\in \mathcal U} \widetilde F_0(u)\) s.t. \(\widetilde h_i^1(u)=0\) \(\forall i\in E\), \(\widetilde h^2_j(u)\leq 0\) \(\forall j\in I\), \(\widetilde q(u)(t)\leq 0\) \(\forall t\in T\). The proposed algorithm aims to solve a penalized problem \((\text{P}_c)\) \(\min_{u\in \mathcal U} \widetilde F_c(u)\) s.t. \( \widetilde h^2_j(u)\leq 0\) \(\forall j\in I\), \(\widetilde q(u)(t)\leq 0\) \(\forall t\in T\) with \(\widetilde F_c(u)= \widetilde F_0(u)/c+\max_{i\in E}| \widetilde h_i^1(u)|\). The penalized problem is formulated also in the relaxed form with the set of relaxed controls \(\tilde \mathcal U\) containing measurable functions \(\mu:T\to rpm(\Omega)\), where \(rpm(\Omega)\) is the space of Radon probability measures on the Borel sets of \(\Omega\). A general convergence property of the proposed algorithm makes no reference to the existence of accumulation points. The relaxed accumulation points generated by the algorithm satisfy standard first-order necessary conditions of optimality.