Bulirsch R. 10.1007/978-3-0348-7539-4_20 doi theory reliability necessary condition estimates control problem procedure false glider direct method solution sensitive dependence upwind extension https://doi.org/10.1007/978-3-0348-7539-4_20 initial guess step switching structure adjoint relationship multipoint boundary value problem direct collocation 1993-01-01 low accuracy difficulties control theory 1993 domain numerical solution structure large domains adjoint variables competition way convergence iterative solution control trajectory optimization problem maximization multipliers accuracy direct collocation method iteration shooting method hang glider possibility Combining Direct and Indirect Methods in Optimal Control: Range Maximization of a Hang Glider optimal control optimal control problem chapters inequality constraints optimization problem indirect method When solving optimal control problems, indirect methods such as multiple shooting suffer from difficulties in finding an appropriate initial guess for the adjoint variables. For, this initial estimate must be provided for the iterative solution of the multipoint boundary-value problems arising from the necessary conditions of optimal control theory. Direct methods such as direct collocation do not suffer from this problem, but they generally yield results of lower accuracy and their iteration may even terminate with a non-optimal solution. Therefore, both methods are combined in such a way that the direct collocation method is at first applied to a simplified optimal control problem where all inequality constraints are neglected as long as the resulting problem is still well-defined. Because of the larger domain of convergence of the direct method, an approximation of the optimal solution of this problem can be obtained easier. The fusion between direct and indirect methods is then based on a relationship between the Lagrange multipliers of the underlying nonlinear programming problem to be solved by the direct method and the adjoint variables appearing in the necessary conditions which form the boundary-value problem to be solved by the indirect method. Hence, the adjoint variables, too, can be estimated from the approximation obtained by the direct method. This first step then facilitates the subsequent extension and competition of the model by homotopy techniques and the solution of the arising boundary-value problems by the indirect multiple shooting method. Proceeding in this way, the high accuracy and reliability of the multiple shooting method, especially the precise computation of the switching structure and the possibility to verify many necessary conditions, is preserved while disadvantages caused by the sensitive dependence on an appropriate estimate of the solution are considerably cut down. This procedure is described in detail for the numerical solution of the maximum-range trajectory optimization problem of a hang glider in an upwind which provides an example for a control problem where appropriate initial estimates for the adjoint variables are hard to find. computation disadvantages appropriate initial estimates problem model variables 273-288 https://scigraph.springernature.com/explorer/license/ multiple shooting nonlinear programming problem multiple shooting method conditions detail collocation method first step optimal control theory Lagrange multipliers collocation indirect multiple shooting method example homotopy technique method high accuracy chapter approximation guess non-optimal solutions appropriate estimates optimal solution boundary value problem constraints range maximization subsequent extension 2022-08-04T17:19 shooting results programming problem technique fusion initial estimates dependence appropriate initial guess precise computation 978-3-0348-7541-7 978-3-0348-7539-4 Optimal Control Miele A. von Stryk Oskar Edda Nerz Mathematical Sciences Hans Josef Pesch Applied Mathematics Roland Bulirsch Numerical and Computational Mathematics J. Stoer Springer Nature pub.1004585554 dimensions_id Well K. Mathematisches Institut, Technische Universität München, Postfach 20 24 20, D-8000, München 2, Germany Mathematisches Institut, Technische Universität München, Postfach 20 24 20, D-8000, München 2, Germany Springer Nature - SN SciGraph project