An examination of convergence error in equilibrium traffic assignment models

This research focuses on the convergence error inherent in the predictions of network equilibrium flows obtained from the Frank-Wolfe algorithm. Factors which determine the magnitude of the convergence error are identified. The effects of selection of a stopping criterion and an initial solution are examined in an empirical study. The results indicate that it is difficult to obtain precise flow predictions in some contexts, specifically where networks contain only a few O-D pairs which are connected by a number of alternative overlapping routes. In addition to having important implications for research studies, the empirical results highlight the need for both researchers and practitioners to consider the issue of convergence error when selecting stopping criteria and initial solutions for flow prediction algorithms.