You are driving along a motorway at 100 km/h. Without any accident, any roadwork or any visible reason, the brake lights ahead come on. Everyone slows to a crawl, then ten minutes later the road opens and you speed up again. You pass the place where it happened and there is nothing there. Where did the jam come from?
The answer is the first lesson of this course: a jam is not a thing, it is a wave in the traffic itself. And traffic engineering is the craft of keeping the numbers away from the point where those waves begin, and keeping people alive when they are wrong about them.
Who takes it: civil engineers (and urban planners and some industrial engineers), usually in year 3 or 4, as one course in highway and traffic engineering. Programmes shaped by the ASCE Body of Knowledge all teach it; planning schools teach the demand and land-use half.
What it's for: moving people and goods safely and efficiently through roads, junctions and transit, and understanding how travellers respond to what we build.
Comes after: Probability & Statistics (34/05), Differential Equations (34/03).
Leads to: Construction Management (34/70) to build it, Corporate Finance (34/51, engineering economics) for judging whether it is worth it, and the Systems Engineering capstone (34/22).
Honest line: you won't be able to design an interchange after this. You will see why a ring road makes phantom jams, why green time is a budget, why 30 km/h is a different world from 50, and what the evidence says about "just build more lanes".
Look at one lane from above. Three numbers describe it. Density (k) is how many cars per kilometre. Speed (v) is how fast they go. Flow (q) is how many cars pass a point per hour. They are tied by pure bookkeeping.
Now add one fact about drivers: they slow down as the road fills. At low density, cars go at the free-flow speed and the flow rises with density. But at some point the cars are too close to keep going fast, speed drops, and flow falls even though density keeps rising. The plot of flow against density is the fundamental diagram, and it rises to a peak: the lane's capacity. Past the peak the road is congested: more cars on it means fewer cars getting through. The picture below uses a simple triangle (a free-flow line up to a peak, then a congested line down to jam density) with round, illustrative numbers.
Capacity is not an absolute: a motorway lane tops out at very roughly 2,000 to 2,400 vehicles an hour in good conditions (the Highway Capacity Manual in the US publishes the standard values). Engineers grade a road by level of service, from A (free-flowing) to F (breakdown), using density or delay, and the key danger is the knee near capacity, where a small disturbance turns into a long jam.
Illustrative triangular diagram: free-flow speed 110 km/h, peak at 22 cars/km (2,420 cars/h), congestion wave moving backward at 20 km/h, jam density 143 cars/km. Real roads vary. The "state" labels are mine, not Highway Capacity Manual grades.
Drivers do not react instantly or exactly. One driver brakes slightly harder than needed; the next, with a delay, brakes a bit harder still; and a few cars back the road has come to a stop. That stop-start region moves backwards through the traffic like a wave, while the cars move forwards through it. Near the capacity peak, the traffic is unstable: small hesitations grow, not fade.
In 2008 a Japanese team (Sugiyama and colleagues) put vehicles on a circular track, asked the drivers to hold the same speed, and watched. Starting from uniform flow, a jam cluster appeared on its own and travelled backward at about the speed seen on real motorways. Their conclusion: a bottleneck is only a trigger, and the real origin of a jam is the collective behaviour of the vehicles once density passes a critical value.
You can see it in a model. The Nagel–Schreckenberg model (1992) is a ring of cells with cars that obey four rules each step: speed up (to a limit), slow down to avoid the car ahead, occasionally brake at random for no reason, and move. Without the random rule, nothing happens below a critical density. With it, jams are born, travel backward, and cost the road a fraction of its flow.
Each row is one second, the road runs left to right and wraps around (200 cells of 7.5 m). Red cars are stopped, blue are moving fast. Jams show up as red streaks leaning backward. Maximum speed is 5 cells per second.
A signal gives different streams the junction in turn. Every change of phase wastes a few seconds (the first car to start up is slow, and a safety clearance is needed), so a longer cycle wastes proportionally less. But a long cycle means long waits, with long queues. There is a best cycle length in between.
Frank Webster, at Britain's Road Research Laboratory, worked this out in 1958 and gave two classics. The best cycle length is roughly C = (1.5 L + 5) / (1 − Y), where L is the lost time per cycle and Y is the sum of the "critical flow ratios" (the flow divided by the saturation flow, for each phase). The closer Y gets to 1, the faster the best cycle grows, because the junction is nearly full. And green time is split in proportion to those flow ratios. Webster's delay formula, used below, has a bowl shape: too short a cycle wastes time in change-overs, too long makes cars wait. Near the bottom the curve is flat, so the engineer has slack.
Two phases, one lane each, saturation flow 1,800 cars/h, lost time 6 s per phase. Webster's delay formula for random arrivals. Red marker: your cycle. Green line: Webster's best. When a road is full (above about 97% of its capacity) delay is unbounded and the curve is cut off.
A road is designed around what the driver can see and do. The master idea is stopping sight distance: if an obstacle appears at the limit of vision, can the driver stop before it? It adds two distances: how far the car travels while the driver notices and reacts, and how far it needs to brake.
The usual North American design values are a reaction time of 2.5 s and a deceleration of 3.4 m/s² (gentle, so that wet or worn tyres still work). At 100 km/h, that gives about 183 m. The first term grows with speed, the second with the square of speed, so at high speed braking dominates. The same square law sets the tightest curve: the radius must satisfy R = V² / (127(e + f)), where e is the banking (superelevation) and f is the sideways friction the tyres can give while keeping passengers comfortable. Codes differ by country, and the numbers here are examples, not design values.
Braking deceleration 3.4 m/s². The curve uses side friction f = 0.12, an illustrative mid value (design tables use lower values at higher speeds). Try 4 s: it is what you might allow for a driver looking at a phone.
Roads kill. The WHO estimates about 1.16 million people die each year in road crashes, 92% of them in low- and middle-income countries, which have about 60% of the world's vehicles; more than half are pedestrians, cyclists or motorcyclists. The rate per 100,000 people differs by a factor of ten between countries. Using the WHO estimates compiled on Wikipedia (the years differ): Norway 2.1 (2022), Sweden 2.1 (2022), Japan 2.1 (2023), India 12.6 (2023), Mexico 12.0 (2021), the United States 14.2 (2021), Brazil 15.7 (2021) and Nigeria 20.7 (2019). Countries do not count the same way, and WHO adjusts for that, so national reports can differ.
The physical reason speed matters is energy: the kinetic energy goes as v², and the stopping distance as v². The WHO says each 1% rise in mean speed raises the fatal-crash risk about 4%, and the risk of a pedestrian dying when hit by a car front rises about 4.5 times from 50 km/h to 65 km/h. A 2011 AAA Foundation study (Tefft) put the average death risk for a pedestrian hit by a car at about 10% at 23 mph (37 km/h), 25% at 32 mph, 50% at 42 mph, 75% at 50 mph and 90% at 58 mph (93 km/h).
Vision Zero begins from this. Sweden's parliament adopted it in 1997: nobody should be killed or seriously injured, because "life and health can never be exchanged for other benefits". The roads should be designed for human error, with the designer sharing the responsibility, and speeds matched to what bodies survive (30 km/h where cars meet pedestrians). Sweden's road deaths fell from 541 in 1997 to 192 in 2021, though progress has stalled since about 2013. Elsewhere results are mixed: Washington, D.C. saw deaths rise in most years after adopting it in 2015, so the label alone changes nothing without road changes.
Dry road, braking at 7 m/s² after your reaction time. Death risk interpolated from the Tefft (2011) points above; at low speeds the risk is lower than 10% but not zero. An illustration, not a prediction for any one crash.
The natural response to a congested road is to widen it. Sometimes this works for a while. But the evidence shows that new capacity is largely taken up by new traffic: people drive more, at more times, to farther places, and others move out. This is induced demand. The best-known study, by Gilles Duranton and Matthew Turner (American Economic Review, 2011), found that on US interstate highways in metropolitan areas, vehicle-kilometres travelled rose proportionally with lane-kilometres: an elasticity close to 1. In this case the new lanes fill to about the old level of traffic per lane and congestion is back.
Not every study finds that much. A 2004 meta-analysis put the response to a 10% capacity increase at about 4% more driving at once and about 10% within a few years. Mokhtarian and colleagues (2002) found little effect, and critics such as Polzin argue that much extra driving is just population growth or otherwise benign. The reverse is also seen: when roads lose capacity, some trips simply disappear. The safe summary: expect much of the new capacity to fill, more over years than months, and more in growing metropolitan areas; the exact amount depends on the place.
Traffic after = (1 + added share)^ε, where ε is the demand elasticity: 0 means no new trips, 1 means traffic grows exactly with capacity (the Duranton–Turner estimate for US interstates). Traffic per lane is a rough guide to congestion, not a full model.
Braess's paradox is a different effect, and often misreported. Dietrich Braess showed in 1968 that adding a road to a network can make everyone's trip longer, if each driver picks the quickest route for themselves. In the textbook example below, 4,000 drivers go from Start to End, two roads take T/100 minutes (T is the number of cars on them) and two take a fixed 45 minutes. Without the shortcut, 2,000 take each route and the trip is 65 minutes. Add a free, instant link and everyone takes the new route: 80 minutes. Nobody can do better by switching on their own, yet everyone is worse off. The paradox is a maths result under particular assumptions: selfish route choice, delays that depend on flow, a certain network shape. Whether real cities contain such links is a separate question. There are reported cases (the Cheonggyecheon expressway in Seoul, Stuttgart in the 1960s, 42nd Street in New York in 1990), but they are mostly anecdotes with other things changing at once. Youn, Gastner and Jeong (2008) found roads in Boston, New York and London where models suggest the effect could occur, and laboratory experiments with real people exist (Rapoport and colleagues, 2009). The safest reading: the paradox is real in theory, plausible in particular places, and no reason to think closing any given road will help.
Equilibrium: no driver can save time by switching. Start to A and B to End take T/100 minutes; Start to B and A to End take 45; the shortcut A to B takes 0. With few drivers (under 3,000) the shortcut helps; above that it hurts, until about 9,000, when traffic is so heavy it stops mattering.
A lane of cars cannot move many people: at roughly 2,000 vehicles an hour and about 1.2 people per car, it is about 2,400 people an hour. A bus running every two minutes with 80 passengers is similar. A bi-articulated bus every minute with 160 passengers carries about 9,600 an hour, and a metro train of 1,200 passengers every two and a half minutes about 28,800. (All round, illustrative figures: capacity is vehicle size times frequency.) That is the geometric advantage of transit, and it only shows up if the vehicles are frequent and well filled. An empty bus loses to a car on every measure.
Filling them is a land-use question. Transit works when many people live and work near stops, walk to them safely, and can reach many places by it; low-density, separated land uses make it expensive. Land use and transport shape each other over decades: roads create sprawl, and sprawl creates traffic. Planning research has examined how density, mix of uses, street design, distance to destinations and distance to transit each affect travel; individual effects are modest, and combinations matter more (Ewing and Cervero's 2010 review is the standard citation).
Traffic is a fluid with a peak: flow rises with density to a capacity, then falls, and near that peak jams form from nothing as waves that travel backward. Signals trade lost time against queues, and Webster's best cycle lengthens as a junction fills. Roads are designed around stopping distance, which grows with the square of speed, and the same square law decides who survives a collision. Vision Zero designs for human error. More lanes tend to fill with new trips, and Braess's paradox shows that adding a link can hurt everyone, though real examples are less clear than the maths.
Try this: on your next trip, notice when traffic slows with no visible cause, and count how far back the stop-start region reaches. Then on a street near you, pace out the distance a car at 50 km/h needs to stop (about 35 m with a 1.5 s reaction time) and see how much of the street that is.
Sources. Flow, density and speed relations, Webster's cycle and delay formulas, stopping sight distance and curve radius are standard textbook material (Mannering and Washburn; Garber and Hoel; AASHTO Green Book). Stopping distance with t = 2.5 s and a = 3.4 m/s² (about 183 m at 100 km/h): Wikipedia, Stopping sight distance. Nagel–Schreckenberg model (1992): Wikipedia. Phantom jams: Sugiyama et al., New Journal of Physics 10, 033001 (2008). Road deaths worldwide, speed and pedestrian risk: WHO fact sheet Road traffic injuries (page dated 20 July 2026); rates by country: WHO estimates compiled on Wikipedia, List of countries by traffic-related death rate (years vary). Pedestrian death risk by speed: B. C. Tefft, Impact Speed and a Pedestrian's Risk of Severe Injury or Death, AAA Foundation for Traffic Safety (2011). Vision Zero, Sweden 541 deaths (1997) and 192 (2021) and the critiques: Wikipedia, Vision Zero. Induced demand: Duranton and Turner, American Economic Review 101(6), 2011; other studies as summarised in Wikipedia, Induced demand. Braess (1968) example, Seoul/Stuttgart/42nd Street reports, Youn et al. (2008) and Rapoport et al. (2009): Wikipedia, Braess's paradox. Not re-checked for this piece: the Highway Capacity Manual capacity range, Ewing and Cervero (2010), Newman and Kenworthy (1989) and the transit capacity round numbers (labelled illustrative). All interactive numbers were computed in node.