Large-Scale Dynamic Optimization with the Directional Second Order Adjoint Method

Efficient solution of large-scale dynamic optimization problems can be achieved by exploiting the advantages of state-of-the-art integration and large-scale nonlinear programming tools. TruncatedNewton method provides an effective way to solve large-scale optimization problems by achieving savings in computation and storage. For dynamic optimization, the Hessian-vector products required by these methods can be evaluated accurately at a computational cost which is usually insensitive to the number of optimization variables using a novel directional Second Order Adjoint (dSOA) method. The case studies presented in this paper demonstrate that a dSOA powered truncated-Newton method is a promising candidate for the solution of large-scale dynamic optimization problems.