Traffic Jam Simulator

Most drivers assume a jam has a cause: a crash, roadworks, a merge, a red light. But a large fraction of motorway congestion has no cause at all. You crawl for five minutes, reach the front of the queue, and find nothing there. That is a phantom traffic jam — congestion that emerges purely from the interaction of drivers with each other, with no bottleneck anywhere.

Traffic physicists call the stable travelling structure a jamiton, by analogy with a soliton: a self-sustaining wave that keeps its shape as it moves. Cars pour into its upstream edge, sit in it, and eventually accelerate out of its downstream edge. The individual cars are constantly being replaced, but the jam itself persists and can survive for hours and travel tens of kilometres.

The ring road above removes every possible excuse. There is one lane, no junctions, no lights, no lane changes, no obstacles, and every driver wants to go 108 km/h. The jam still forms, and it still refuses to go away.

Every car above obeys the Intelligent Driver Model (IDM), published by Martin Treiber, Ansgar Hennecke and Dirk Helbing in 2000. It is a deterministic car-following model: each vehicle's acceleration depends only on its own speed, the gap to the car in front, and the speed difference to that car.

The bracket holds two competing pressures. The free-road term 1 − (v/v₀)^δ pushes the car towards its desired speed and vanishes as v reaches v₀. The interaction term −(s*/s)² is the braking force: it is negligible when the actual gap s is much larger than the desired gap s*, and it explodes quadratically as the gap closes.

A jam is not a group of cars. It is a boundary — the line where free-flowing traffic meets stopped traffic. Watch what happens at that boundary. At the back of the queue, a new car arrives and stops; the boundary jumps one car-length upstream. At the front, a car accelerates away; the boundary also moves upstream by one car-length. The queue gains at the back and loses at the front, so the whole structure slides backwards even though every car in it only ever moves forwards.

The rate is set almost entirely by how quickly cars leave the front. Each departing driver takes roughly the headway time T to pull away and open a gap of s₀ + ℓ (standstill gap plus vehicle length). The upstream front therefore retreats at approximately:

Notice what is not in that formula: the desired speed v₀. The wave speed depends only on how tightly cars pack when stopped and how long each takes to react and pull away — both of which are properties of drivers and vehicles, not of the speed limit. That is why measured jam waves on motorways in Germany, the Netherlands, Japan and the United States all cluster around 15 to 20 km/h backwards, whether the road is posted at 80 or 130 km/h. It is one of the most reproducible numbers in traffic science.

Measuring the simulation above directly — timing when each car first drops below 1.5 m/s and fitting a line through the jam front — gives about −14 km/h, a little shallower than the idealised formula because real IDM cars in the queue are not all at a dead stop. In the space-time strip below the road, that is the slope of the dark bands: they lean left as they descend, which is precisely what a backward-moving feature looks like when time runs downwards.

For decades this was a claim about mathematical models. In 2008, Yuki Sugiyama and colleagues published Traffic jams without bottlenecks — experimental evidence for the physical mechanism of the formation of a jam in the New Journal of Physics, and simply did it with real cars and real drivers.

They marked out a circular track roughly 230 m in circumference, placed 22 vehicles on it evenly spaced, and gave the drivers a single instruction: drive at about 30 km/h, keeping a safe distance, and do not overtake. There was no bottleneck of any kind — the track was a perfect loop.

Follow-up experiments made the same point in the other direction. In 2017 a team including Benjamin Seibold, Benedetto Piccoli and Daniel Work ran a similar ring-road experiment in Arizona, but placed a single autonomous vehicle among about 20 human-driven cars. Programmed to hold a steady, smooth gap rather than mirror the car ahead, that one vehicle damped the stop-and-go waves out of the entire ring — cutting the fleet's fuel consumption by roughly 40% and its hard braking events by around 90%.

The most important property of this system is metastability. Over a wide band of densities the smooth flow is genuinely stable against tiny disturbances — press Start perfectly smooth above and it will cruise at 49 km/h indefinitely. But it is only locally stable. One driver braking for 2.5 seconds is a large enough kick to knock the system out of that basin, and it collapses to about 22 km/h with 19 of 42 cars stopped, permanently. There is no path back to free flow without removing cars.

That asymmetry is the whole practical lesson. A jam costs almost nothing to create and is nearly impossible to undo. And because the disturbance amplifies as it passes backwards through the queue, a driver ten cars behind brakes harder than you did, and a driver thirty cars behind stops dead.

• Keep a steady gap, not a small one. The IDM shows braking force scaling as (s*/s)² — the penalty for tailgating is quadratic. Leaving a larger, constant buffer gives you room to absorb the car ahead's fluctuations instead of passing them on amplified.
• Brake early and gently. A long, mild deceleration transmits a far weaker pulse backwards than a late, hard one. Ease off the accelerator before you need the brake pedal.
• Accelerate decisively out of a queue. The slider above makes this vivid: raising a past about 1.5 m/s² dissolves the jam entirely, while dropping it to 0.2 m/s² deepens it. Hesitating at the front of a queue starves nobody — it feeds the jam behind you.
• Adaptive cruise control can help — or hurt. ACC holds a constant time headway automatically and never gets distracted, which is exactly the smoothing behaviour the 2017 experiment exploited. But ACC tuned for a short headway and aggressive catch-up is string unstable and amplifies disturbances just like a human tailgater. Settings matter more than the badge.
• Density is the real control. No driving technique beats the threshold. Drop the ring to 20 cars above and the jam dissolves on its own. This is why ramp metering and variable speed limits work: they manage how many vehicles are on the road at once, rather than trying to fix the driving.