A parallel Frank-Wolfe/gradient projection method for optimal control

The authors propose a parametrized gradient projection algorithm for solving large-scale problems, and in particular discrete optimal control problems with linear constraints. They present numerical results concerning the sequential implementation of the algorithm both with nonadaptive and adaptive rules for the choice of the parametrizing sequence, and the parallel implementations of the various primitives used by the parametrized algorithm are discussed. It is shown that an adaptive choice of parameters results in a considerable decrease in the number of iterations required to reach a predetermined solution neighborhood, and that the algorithm can be implemented in parallel resulting in a further decrease in computational time