A dual ascent algorithm for traffic assignment problems

A dual decomposition algorithm is developed for large-scale traffic assignment problems. In contrast to standard methods, this algorithm does not require that the system or user optimal objective function be differentiable and it allows bounds on the arc flows. Iterates alternate between dual ascent steps and calculations of shortest paths as in the Frank-Wolfe method. Although a dual method, it produces feasible flow patterns at each iteration. Convergence of the method is proven and a computational example is given.