Congestion on highways often occurs by sudden transitions from free flow traffic to synchronized traffic. Associated with such phase transitions are large decreases in speed. Such phase transitions seem to be often caused by temporal distortions in traffic flows, such as a short peak in the flow of traffic entering a highway from a ramp. If the traffic flow on the ramp has decreased after the peak, congestion on the highway may persist for a longer period. The paper develops a model of congestion for such situations that is based on assumptions about driver behavior and has these characteristics. Critical properties of driver behavior are that, for obvious reasons, sudden large changes in speed are avoided and that reactions to other drivers occur with some delay. In a model where speed can be changed immediately from to any desired value (and vice versa) and drivers react immediately to each other some of the typical characteristics of congestion on highways do not occur. The model suggests that control of the traffic flow, and especially of variations in traffic density, is of major importance in dealing with congestion. Measures for reducing congestion that intend to decrease the average flow of traffic (such as road pricing) should be supplemented by measures reducing the variation in that flow.
