Phantom Jams: How Congestion Appears from Nothing

A phantom jam is a region of stop-and-go traffic that forms spontaneously on a highway without any accident, merge, bottleneck, or on-ramp. It propagates backward --- upstream against the direction of traffic flow --- at approximately 15 km/h, while vehicles themselves move forward through it. Drivers enter the jam from the front, decelerate to a stop, wait, and eventually accelerate out the back. The jam persists as a coherent density wave even though no physical obstruction exists.

The Phenomenon

Every highway driver has experienced a phantom jam: traffic slows to a crawl or a stop, then clears, with no visible cause. No accident, no construction, no merge. The jam is simply there and then it is gone.

The phenomenon is not rare or marginal. Traffic sensor data from highways worldwide shows that phantom jams account for a substantial fraction of congestion events on high-density roads. Kerner and Rehborn (1996), analyzing loop detector data from German autobahns, documented spontaneous stop-and-go waves that formed in the absence of any identifiable bottleneck. The waves had consistent properties: propagation speed of approximately 15 km/h upstream (against traffic flow), a period of roughly 5 to 15 minutes between successive stop-and-go cycles, and persistence for hours after forming.

The propagation speed of 15 km/h is remarkably consistent. It has been measured independently on German autobahns (Kerner and Rehborn, 1996), Japanese expressways (Sugiyama et al., 2008), Dutch motorways (Helbing, 2001), and U.S. interstate highways. Lighthill and Whitham (1955) predicted this speed theoretically in their kinematic wave model of traffic flow, decades before detailed sensor data confirmed it. The speed depends on the fundamental diagram’s slope at the critical density but is insensitive to the specific characteristics of the road, the vehicle mix, or the driver population.

The Instability Mechanism

Free-flowing traffic at high density is unstable. The instability mechanism is asymmetric amplification of braking perturbations.

Consider a line of vehicles traveling at speed v with uniform headway (gap between vehicles). One driver brakes slightly --- perhaps a momentary distraction, a small adjustment to the gap. The perturbation is small: a reduction in speed of 2 km/h for one second.

The driver behind has less time to react. The gap has closed slightly, and the driver behind must brake somewhat more aggressively to avoid collision. The reduction in speed is now 3 km/h. The next driver in the chain must brake still harder. The perturbation amplitude grows with each vehicle it passes through.

The formal condition for instability in car-following models was derived by Herman, Montroll, Potts, and Rothery (1959). In the optimal velocity model (Bando et al., 1995), each driver adjusts speed toward a desired velocity that depends on the gap to the lead vehicle. The system is stable if the sensitivity parameter (how strongly a driver responds to a gap change) is below a critical value that depends on the desired speed function. Above that value, small perturbations grow exponentially. The critical condition relates reaction time, following distance, and speed: when vehicles are close enough that the response time exceeds the headway divided by the characteristic wave speed, the system is unstable.

The instability is specific to high density. At low density, gaps are large. A driver who brakes slightly opens a gap that is still well above the following distance threshold. The driver behind does not need to brake at all, or brakes minimally. The perturbation dies out. At high density, gaps are small. A slight braking event closes the gap to a level that triggers a strong braking response. The perturbation amplifies.

The critical density at which the instability sets in depends on driver characteristics (reaction time, braking aggressiveness) and road characteristics (speed limit, lane width), but the qualitative transition --- stable flow below critical density, unstable above --- is universal across traffic models.

The Sugiyama Ring Road Experiment (2008)

Yuki Sugiyama and colleagues at Nagoya University conducted the definitive empirical demonstration. Twenty-two vehicles were placed on a circular track 230 meters in circumference. Drivers were instructed to maintain 30 km/h and a comfortable following distance. The track was a circle --- no intersections, no on-ramps, no exits, no grade changes. Every confounding cause was eliminated.

Within two minutes, a phantom jam formed. The video (published with the paper in New Journal of Physics) shows the jam clearly: a region of closely spaced, stopped vehicles that rotates around the track in the direction opposite to the traffic flow. Vehicles enter the jam, stop, and eventually exit the back as new vehicles enter the front. The jam is a stable, self-sustaining wave.

The measured propagation speed was approximately 20 km/h backward, consistent with the 15 to 20 km/h range observed on real highways. The jam was not caused by any driver’s mistake or any external event. It emerged from the collective dynamics of 22 drivers each trying to drive smoothly.

The experiment was repeated with different numbers of vehicles and different densities. Below a critical density (approximately 0.5 vehicles per meter of road, corresponding to a following distance of about 2 meters), the traffic flowed smoothly. Above the critical density, phantom jams formed consistently. The transition was sharp: a small increase in density produced a qualitative change in behavior.

Stern et al. (2018) repeated the experiment with one crucial modification: one of the 22 vehicles was equipped with adaptive cruise control, following a simple algorithm that smoothed speed variations. The single automated vehicle suppressed phantom jam formation entirely. Total fuel consumption across all 22 vehicles decreased by 40 percent. One vehicle out of 22, following a more consistent version of the same local rules, changed the system’s macroscopic behavior.

Why Jams Propagate Backward

The backward propagation direction follows from causality in car-following dynamics. Each driver responds to the vehicle ahead. Information about braking travels backward through the vehicle chain: driver A brakes, causing driver B (behind A) to brake, causing driver C (behind B) to brake. The braking signal propagates upstream.

In the continuum limit, the jam is a kinematic wave --- a wave in the density field of the traffic flow. The Lighthill-Whitham-Richards (LWR) model describes traffic as a one-dimensional compressible fluid with a flow-density relationship (the fundamental diagram). The speed of a density disturbance in this model is determined by the slope of the fundamental diagram at the local density. At the critical density where flow is maximized, the slope is zero --- disturbances are stationary. At densities above the critical point, the slope is negative --- disturbances propagate backward. The characteristic speed of backward-propagating jams is the slope of the fundamental diagram in the congested regime, which is approximately -15 to -20 km/h for typical road parameters.

The jam does not move with the traffic because it is not a property of any particular vehicle. It is a density wave --- a region of high density that persists because vehicles entering at the front decelerate (adding density) at the same rate that vehicles exiting at the back accelerate (removing density). The balance of inflow and outflow sustains the wave at a fixed position (or, on a ring, at a fixed angular position that rotates backward).

This is analogous to a standing wave in a river: the wave is stationary (or moves relative to the bank), but the water flows through it. The jam is stationary in the road frame (or moves backward) while the vehicles flow through it in the forward direction.

Further Reading