Pattern · emergence · show the check

The Jam That Isn't There

Here is a perfectly circular road. Every driver is the same — careful, patient, identical. There is no crash, no merge, no slow truck, nothing on the road at all. Add enough cars and a traffic jam appears anyway, out of nowhere, and crawls backward against the flow. Drag the density below a threshold and the very same road refuses to jam. The jam was never on the road. It was in the crowd.

position around ring →time ↓ · dark = stopped

Cars on the road — 45 (50/km)

Density

50 /km

Mean speed

— km/h

Cars stopped

Jam wave

— km/h

Settling…

The Nudge button stops at least three adjacent drivers. Use the slider to change the density, then try another nudge. Watch whether the disturbance fades or persists at each density.

For decades this was a roadside mystery: you crawl for ten minutes, then the traffic frees up and there's nothing there — no wreck, no police, no reason. The reason is that you were never going to find a reason on the road. A dense stream of cars is an unstable medium. Above a critical packing, the smooth-flowing state can't hold: any tiny fluctuation — one driver lifting off the accelerator a fraction — forces the car behind to brake a little harder, and the one behind that harder still, because each driver reacts with a small delay and a margin of safety. The disturbance amplifies as it travels back through the line until cars far upstream are stopped dead. The cluster of stopped cars is a real, persistent object — a wave — and like the bump in a Slinky it moves opposite to the things that make it up. Cars roll forward through the jam and out the front; the jam itself slides backward.

It happens with real people, too

In 2008 a team led by Yuki Sugiyama reported the cleanest possible experiment. They put cars — driven by ordinary people — on a circular track 230 metres around, and asked them to do one thing: drive steadily at a fixed speed, keeping a safe distance. Models predict that with few enough cars, small disturbances fade and flow stays smooth. But once the track held about 22 cars or more, a jam cluster appeared on its own within a minute or two and propagated backward around the loop — no obstacle, no instruction to stop, nothing but the drivers and the density.¹ The model you are operating above reproduces exactly that transition. It is the Intelligent Driver Model² — a standard, empirically calibrated rule for how one car follows another — and nothing in it knows what a jam is. The jam is what the rule does when you crowd it.

The check — every number here is recomputed, not asserted

The simulator runs the real Intelligent Driver Model. Each car's acceleration is

v̇ = a · [ 1 − (v / v₀)⁴ − (s* / s)² ] s* = s₀ + max(0, v·T + v·Δv / (2·√(a·b)) )

where s is the bumper-to-bumper gap to the car ahead and Δv is how fast you're closing on it. The parameters are the standard highway set,² used identically here and in the offline verifier:

v₀	desired speed	30 m/s · 108 km/h
T	safe time headway	1.5 s
a	max acceleration	0.73 m/s²
b	comfortable braking	1.67 m/s²
s₀	standstill gap	2 m
—	vehicle length	5 m

The jam is in the dynamics, not the road. On a 1 km loop, a one-car velocity dip of 0.5 m/s decays below a critical density and grows into a permanent stop-and-go cluster above it. The road is byte-for-byte identical in both runs — only the car count changes. Onset of instability: ≈ 28–32 cars/km; full stop-and-go by ≈ 40 cars/km.
The jam travels backward at a near-constant speed. The verifier tracks the stopped cluster and fits its drift: ≈ −12.7 km/h, essentially the same from 36 to 60 cars/km. Negative means upstream — against traffic.
Honest gap. Real highway jam waves are measured travelling backward at about −15 to −20 km/h, famously near-constant across wildly different conditions. This model reproduces the sign, the near-constancy, and the rough magnitude, landing a little slower than the highway figure. The exact number depends on the parameters (mainly the headway T) and on how you measure the cluster; we report what the model gives rather than tuning it to the textbook value. The live "jam wave" readout is measured the same crude way and will wobble — read its sign and order of magnitude, not its last digit.

node research/phantom-traffic-jam/verify.mjs prints the density sweep, the first sampled count meeting its near-stop jam criterion, and the backward wave speed. Shared model: research/phantom-traffic-jam/idm.mjs.

Why this matters past the curiosity

If the jam needs no cause, then you cannot clear it by clearing the road — there's nothing to clear. What you can do is attack the instability itself. A single car that refuses to brake-and-surge — holding a steady speed and a generous gap, absorbing the wave instead of passing it on — can damp a phantom jam for everyone behind it; experiments with one autonomous vehicle in a ring of human drivers have shown exactly this. The continuum theory that makes the waves precise calls them jamitons:³ self-sustaining travelling waves of the traffic equations, mathematically the same kind of object as a detonation front in a gas — a shock dragging a zone of frantic readjustment behind it. The traffic on your commute is, sometimes, literally detonating in slow motion.

Corrections

Corrected 2026-10-14: The Nudge hint said one driver eased off and the wobble died below about 32–40 cars/km. The button sets at least three adjacent drivers' speeds to zero, three at the default 45 cars. Replaying Reset then Nudge at 26 and 28 cars on the 900 m ring gives speed spreads of about 8.64 and 15.74 m/s at 6,000 simulated seconds; found by the assay line, 2026-10-14.

Corrected 2026-10-14: The page said the ring experiment was done in 2008. It was reported in 2008: the paper was received on 14 November 2007 and published on 4 March 2008; found by the assay line, 2026-10-14.

Corrected 2026-10-14: The page said the low-density experimental flow stayed smooth forever. Low-density stability is a model prediction in the paper; the reported jam trials use 22 and 23 vehicles, and the paper leaves the instability range for further work; found by the assay line, 2026-10-14.

Corrected 2026-10-14: The live status said a cluster was drifting backward whenever speeds differed sharply. A speed spread does not establish direction; the status now reports the measured drift or says that a measurement is pending; found by the assay line, 2026-10-14.

Corrected 2026-10-14: The live status said uniform flow meant the road was below critical density and any disturbance would die out. Reset makes speeds equal at the same density; uniform speeds alone do not establish stability; found by the assay line, 2026-10-14.

Corrected 2026-10-14: The verifier said the nudge decayed below 40 cars/km and labelled those runs smooth. At 28 and 32 cars/km the 0.5 m/s seed instead produces settled speed spreads of about 8.82 and 19.68 m/s, without meeting its near-stop jam criterion; found by the assay line, 2026-10-14.

Corrected 2026-10-14: The verifier called 40 the smallest car count that jammed on its 1 km ring. It is the first sampled count meeting the verifier's near-stop criterion; the same simulation at 35 cars also meets it; found by the assay line, 2026-10-14.