The Long Term Behaviour of Day-to-Day Traffic Assignment Models

The dynamical behaviour of deterministic process, day-to-day traffic assignment models is sometimes characterised by convergence to a variety of different fixed equilibrium points dependent upon the initial flow pattern, even though individual trajectories are unique for a given start point. This non-uniqueness is seemingly in sharp contrast to the evolution of stochastic process, day-to-day models; under certain assumptions these converge in law to a unique stationary distribution, irrespective of the start point. In this article, we show how models may be constructed which exhibit characteristics of both deterministic models and stochastic models, and illustrate the ideas by using a simple example network.