Primal-dual properties of sequential gradient-restoration algorithms for optimal control problems. II: General problem (Q1089909): Difference between revisions

These two papers are dealing with primal-dual properties of sequential gradient-restoration algorithms (SGRA) for optimal control problems. While in the former paper [see the preceding review] the minimization of a functional is considered subject to differential and final constraints (the initial state being given), here the same problem is considered subject to differential, nondifferential, initial, and final constraints: \[ \text{minimize the functional } I=\int^{1}_{0} f(x,u,v,\pi,t)dt+ [h(x,\pi)]_ 0+ [g(x,\pi)]_ 1, \] with respect to the n-state vector state x(t), the m-control vector u(t), the c-control vector v(t), and the p-vector parameter \(\pi\) which satisfy the constraints (DC) \(\dot x+\phi(x,u,v,\pi,t)=0\), \(0\leq t\leq 1\), (NDC) \(S(x,u,v,\pi,t)=0\), \(0\leq t\leq 1\), (IC) \([\omega (x,\pi)]_ 0=0\), (FC) \([\psi (x,\pi)]_ 1=0\) (here f,g,h are scalars, \(\phi\) an n-vector, S a c-vector, \(\omega\) an a- vector, \(a\leq n\), and \(\psi\) a b-vector, \(b\leq n\); f, g, h, \(\phi\), S, \(\omega\), \(\psi\) are \(C^ 2\) with respect to x, u, v, \(\pi\), \(rank[\omega_ x]_ 0=a\), rank\([\psi_ b]_ 1=b\), rank\(S_ v=c\) everywhere). The authors present both the primal and the dual formulation of the problem and hence obtain primal and dual SGRA. The control vector is split into an independent control vector and a dependent control vector which enhances the computational efficiency of both formulations. Twelve numerical examples were solved using both formulations and the comparison of primal and dual SGRA is given.