Why a new road can make every trip longer
Build a new road into a busy network, let every driver pick the quickest route for themselves, and everyone can arrive later than before. Nobody is foolish and nothing is broken. It happens because what is quickest for one driver is not what is quickest for the crowd.
A town with two routes
Here is a version with round numbers we chose. Drivers leave Start for End. There are two routes. Each has one narrow road, where the trip takes one minute for every 100 drivers on it, and one wide road, which always takes 45 minutes however many use it. The upper route goes narrow then wide. The lower route goes wide then narrow.
Send 4,000 drivers. If they split evenly, each narrow road carries 2,000 and takes 20 minutes, so every trip is 20 + 45 = 65 minutes. Nobody gains by switching: the other route would then be the more crowded one.
Why the shortcut makes it worse
Start from the old even split. A driver who switches to the zigzag drives two narrow roads at about 20 minutes each and skips both 45-minute wide roads. That looks like 40 minutes instead of 65, so they switch. So does the next driver, and the next. Each one who switches adds traffic to both narrow roads, which slows everyone on them.
The switching stops only when no driver can do better by changing alone. With all 4,000 on the zigzag, each narrow road takes 40 minutes and the trip is 80. A driver who tries the old upper route now gets 40 on the full narrow road plus 45 on the wide one: 85. The lower route is the same 85. So all of them stay at 80, and none of them would choose to stay if any other route were quicker.
That resting state is called a Nash equilibrium, or in traffic, Wardrop's principle: every route in use takes the same time, and no unused route is quicker. Youn, Gastner and Jeong put the point plainly: the best choice for each driver is not always the best outcome for all of them. The gap between the two has a name, the price of anarchy.
Interactive Open or close the shortcut, and change how many drivers set out. Drag the slider, or drag across the chart below it. Each route's time is shown, so you can check that no driver could do better by switching alone. The button and slider work with Tab, Space and the arrow keys.
| route | drivers | trip time |
|---|---|---|
| upper: S→A→E | ||
| lower: S→B→E | ||
| zigzag: S→A→B→E |
It depends on how many drive
Move the driver count in the figure. With fewer than 3,000 drivers the shortcut helps. At exactly 3,000 it makes no difference. Between 3,000 and 9,000 it makes every trip longer, and from 9,000 up nobody takes it at all. Verbree and Cherukuri prove a general version of this: a road that causes the paradox at some level of traffic must have been a genuine help at some lower level. That is part of why planners cannot just spot a bad road and remove it. The same road can help in light traffic and hurt in heavy traffic. They also note that deciding whether a network suffers from the paradox at all is, in general, a computationally hard problem.
Is this only a puzzle?
Dietrich Braess described it in 1968, in a paper on road planning. Youn, Gastner and Jeong simulated drivers on the main roads of Boston, London and New York. On Boston's network, with 10,000 vehicles an hour from Harvard Square to Boston Common, the drivers' total time came out about 30% higher than a coordinated routing would need, and closing any one of six particular streets made the equilibrium trip shorter.
The same thing happens without anyone choosing. In a power grid, current divides by physical law, not by choice. Schäfer and colleagues built a small test grid with real generators and upgraded one line so it could carry more. The load fell on the line next to it but rose on a line farther away. In a 2022 study of planned extensions to the German high-voltage grid, they found two cases in which planned projects would push extra flow onto other lines: in one case onto lines with no upgrade planned, in the other enough to overload a line unless a separate upgrade of it was finished first.
In short
When each user of a network picks the best route for themselves, adding a route can make everyone slower: it tempts every driver onto roads they then crowd for one another. Whether it does depends on how much traffic there is. The same mechanism shows up in power grids, where current takes its path by physics.
Where this comes from
- The Price of Anarchy in Transportation Networks: Efficiency and Optimality Control linked only, not reproduced
arxiv.org/abs/0712.1598 - Understanding Braess' Paradox in power grids reuse permitted with attribution
arxiv.org/abs/2209.13278 - Wardrop equilibrium and Braess's paradox for varying demand reuse permitted with attribution
arxiv.org/abs/2310.04256 - Über ein Paradoxon aus der Verkehrsplanung (Unternehmensforschung 12, 258–268) linked only, not reproduced
doi.org/10.1007/BF01918335