Braess Paradox: Can a Zero-Minute Shortcut Slow Everyone Down?
More route choice can worsen a congestion-game equilibrium.
Precise claimIn some congestion games, adding a route can worsen user equilibrium because individually rational rerouting changes shared edge costs for everyone.
Applies
A static nonatomic 4,000-driver directed Braess network with two flow/100 edges, two fixed 45-minute edges and one optional zero-minute edge.
Does not prove
The page does not predict real-road projects, model queues or induced demand, claim every new road is harmful, or equate user equilibrium with system optimum.
Portable rule
If a shared system adds a locally beneficial option, then recompute the equilibrium after everyone can respond before claiming the system outcome improves.
Declare the discussion context, then copy a stable link. No account or personal data is attached.
game-model puzzle
01 · Commit before the network changes
Braess Paradox: Can a Zero-Minute Shortcut Slow Everyone Down?
Make the locally fastest route choice, then keep demand and every cost function fixed while adding only one zero-minute directed edge.
Demand 4,000 driversOld routes 65 min eachYou one negligible driver
MechanismTHE SHORTCUT TRAPBefore the change
Current assignment before the shortcut is used: 2,000 drivers on each old route. You are the next negligible driver.
Your verdictOne choice · no rerun yet
02 · Reveal the missing response
The shortcut is the rational choice—against the old traffic.
The fastest route for one driver can become the slower equilibrium for everyone.
You committed to: —
Old upper65Old lower65New path40 min
The 40-minute route is not a trick: at the initial 2,000/2,000 split it uses both 20-minute variable edges and the zero-minute link. The missing step is to let the other 3,999 drivers respond too.
In some congestion games, adding a route can worsen user equilibrium because individually rational rerouting changes shared edge costs for everyone.
The rebuilt model
You assumed:
An optional zero-cost road cannot make anyone worse off because drivers can always ignore it.
The actual model:
Available choices alter incentives; incentives alter flows; flows alter congestion costs. A Wardrop equilibrium is stable against unilateral deviation, but stability does not minimize total travel time.
The variable that failed you:
Endogenous congestion response — the shortcut is evaluated against the old flow even though making it available changes the flow on both variable-cost edges.
Reproduces the 4,000-driver, flow/100 and 45-minute Braess teaching network used by the page.
official-doc — checked 2026-07-21.
Scope: A static nonatomic 4,000-driver directed Braess network with two flow/100 edges, two fixed 45-minute edges and one optional zero-minute edge.
Does not prove: The page does not predict real-road projects, model queues or induced demand, claim every new road is harmful, or equate user equilibrium with system optimum.
If a shared system adds a locally beneficial option, then recompute the equilibrium after everyone can respond before claiming the system outcome improves.
Routing products
Test network-wide equilibrium effects rather than ranking a new edge only at yesterday's flows.
Queue design
A fast lane can attract enough demand to change every waiting time; measure the post-choice steady state.
Platform incentives
A shortcut that helps one seller or agent under fixed behavior may change participation and load once universally available.