This paper examines the convergence properties of four popular traffic assignmet algorithms: frank-wolfe decomposition for fixed-demand equilibrium assignment, an ad hoc variation of the evans algorithm for elastic-demand equilibrium assignment, fixed-demand incremental assignment, and elastic-demand incremental assignment. The algorithms were evaluated according to errors associated with insufficient iterations, arbitrary selection of starting point, inexact theory, and small variations in data. Each of the four algorithms reached its intended solution, but did so very slowly. Elastic-demand incremental assignment emerged as the preferred technique, principally because of its more accurate response to small variations in data and its adaptability to various models of travel demand. This paper appears in transportation research record no. 1220, Forecasting.
Abstract