Closing a road can make everyone's journey shorter
In a congested network where each traveller chooses the route that seems best, adding a link can increase everyone’s equilibrium travel time, and removing one can reduce it. The mathematical result is exact under its assumptions; real road removals are suggestive cases rather than controlled proofs.
Dietrich Braess demonstrated the effect in 1968. A new shortcut changes everyone’s incentives, and individually sensible rerouting can move the network to a worse equilibrium. The system optimum and the self-selected equilibrium are different objects. Cities including Seoul have removed major roads without the predicted traffic collapse, but such interventions also change transit, signals, land use and behaviour, so they do not isolate the paradox by themselves.
It is a clean example of individually rational choices creating a collectively worse result—and of subtraction being a legitimate systems intervention. More capacity is not automatically more usable capacity.
It is bounded. For networks with linear congestion costs, the price of anarchy is at most 4/3: selfish routing is never worse than a third off the optimum. That bound is oddly reassuring and also the reason the paradox stays a curiosity rather than a crisis — it is real, it is not usually catastrophic.
The same structure appears in electrical grids, in packet networks, and — most vividly — in a mechanical demonstration where two springs and two ropes hold a weight, and cutting one of the ropes makes the weight rise. Nothing about the paradox is specific to traffic. It is a property of shared-cost networks with individual routing.
Seoul was not a controlled experiment.
The expressway removal coincided with new bus rapid transit, changed signal timing and a large public campaign. The defensible claim is that removing major capacity did not produce the predicted collapse, which is still surprising, rather than that removal improved flow by itself.
A well-instrumented removal with transit and pricing held constant, showing travel times rise as classical capacity models predict.
What in your work would get faster if you removed a path rather than adding one?
Return to todayA sealed edition.
Reading an old edition does not reopen today's door. The archive is a shelf, not a feed.