The Braess Paradox
When More Makes Things Worse
Adding a new highway can make traffic worse. Expanding capacity can reduce throughput. This isn't a paradox - it's game theory. Understanding why individual optimization creates collective harm is essential for effective infrastructure and network policy.
SEOUL, 2003
They Tore Down a Highway
Traffic got better. Much better.
When More Makes Things Worse
In 2003, Seoul demolished the Cheonggyecheon Expressway - a six-lane elevated highway carrying 168,000 vehicles per day through the city centre.
Traffic planners expected chaos. Instead, traffic improved. Average travel times across the network decreased. The city replaced the highway with a park.
This wasn't magic. It was mathematics - specifically, a phenomenon first identified by German mathematician Dietrich Braess in 1968. He proved that adding capacity to a network can, counterintuitively, make overall performance worse.
The implications extend far beyond traffic. Power grids, internet routing, supply chains - anywhere individuals make locally optimal choices that create globally suboptimal outcomes, the Braess Paradox can emerge.
The Classic Network
Consider a simple network. 100 cars need to travel from point A to point B. There are two routes: one through the north (via N), one through the south (via S).
Each route has two segments. Some segments have fixed travel time (45 minutes regardless of traffic). Others are congestion-dependent (travel time equals the number of cars using them).
In equilibrium, traffic splits evenly. Each route takes 95 minutes. Total system travel time: 9,500 car-minutes.
Now add a shortcut - a fast road connecting N to S. Surely this helps?
[Interactive Network Diagram]
Route 100 cars from A to B. Watch as adding the shortcut makes everyone's journey longer.
The classic Braess network. Before the shortcut: 95 minutes per car. After: 100 minutes per car.
The Mechanism
Here's what happens when the shortcut opens:
- 1.Individual incentive: Each driver sees a faster route through the shortcut. Taking A-N-S-B appears to save time.
- 2.Collective shift: When everyone makes this locally optimal choice, traffic concentrates on the congestion-sensitive segments (A-N and S-B).
- 3.New equilibrium: All 100 cars now use the same route. Travel time increases to 100 minutes per car. Total system time: 10,000 car-minutes.
The shortcut didn't reduce travel time - it eliminated the only mechanism forcing traffic to distribute efficiently.
Real-World Evidence
The Braess Paradox isn't just theory. It's been documented in real infrastructure:
Demolished 6-lane elevated highway
Traffic speeds improved 3-6% citywide
Closed major road for construction
City-wide congestion decreased during closure
42nd Street closed for Earth Day
Predicted gridlock never materialized
Modeled selfish vs coordinated routing
30% efficiency loss from selfish routing identified
Embarcadero Freeway collapsed (earthquake)
Traffic improved; highway never rebuilt
Congestion pricing introduced
Traffic reduced 30% in charging zone
Nash Equilibrium vs System Optimum
The Braess Paradox illustrates a fundamental tension in game theory.
Nash equilibrium occurs when no individual can improve their outcome by changing their behavior alone. Each driver is taking their best available route given what everyone else is doing.
System optimum occurs when total travel time is minimized. This requires some drivers to take routes that are worse for them personally, but better for everyone collectively.
In networks with certain characteristics, the Nash equilibrium can be significantly worse than the system optimum. This gap is called the "Price of Anarchy" - the cost society pays for uncoordinated decision-making.
Policy Implication
When the Price of Anarchy is high, infrastructure additions can make things worse. The solution isn't always more capacity - sometimes it's better coordination mechanisms.
Beyond Traffic
The Braess Paradox appears wherever agents independently optimize their routes through a shared network:
- -Power grids: Adding transmission lines can destabilize networks if power flows shift to overload other components.
- -Internet routing: Adding bandwidth can increase latency when packets reroute to congest new links.
- -Supply chains: Opening new suppliers can reduce reliability when all buyers shift to the same source.
- -Organisations: Adding communication channels can slow decision-making when everyone routes through the new shortcut.
A Different Question
The Braess Paradox teaches us to ask a different question.
Not "Will this new capacity be used?" - it almost certainly will be. But "Will individual optimization create system-level dysfunction?"
Sometimes the answer is yes. Sometimes removing options improves outcomes. Sometimes constraints create efficiency by forcing coordination that wouldn't happen voluntarily.
Infrastructure policy isn't just engineering. It's game theory - and the optimal solution isn't always more.
Going Deeper
Technical background for those who want the formal framework.
Formal Definition
The Braess Paradox occurs when adding an edge to a network increases the cost at Nash equilibrium. Formally: let G be a network with edge costs cₑ(fₑ) depending on flow fₑ. Adding edge e' creates network G'. The paradox occurs when:
Cₙₑ(G') > Cₙₑ(G)
Where Cₙₑ is the total cost at Nash equilibrium.
Roughgarden and Tardos (2002) showed that for linear latency functions, the Price of Anarchy is bounded by 4/3 - meaning selfish routing can be at most 33% worse than optimal. For more general functions, the gap can be arbitrarily large.
Mechanism Design Solutions
If we can't rely on individual optimization to produce good outcomes, how do we achieve coordination?
- -Congestion pricing: Charge for externalities. London's congestion charge makes drivers internalize the cost they impose on others.
- -Strategic capacity removal: As Seoul demonstrated, sometimes the solution is less infrastructure, not more.
- -Information design: Control what routing information is available. Navigation apps already do this, sometimes controversially.
- -Centralized routing: Remove individual choice entirely. Autonomous vehicles could enable system-optimal routing.
Historical Context
Dietrich Braess published his paradox in 1968 in the German journal Unternehmensforschung. It remained relatively obscure until the 1990s, when researchers began finding real-world examples.
The paradox connected to broader work in game theory, particularly Wardrop's equilibrium conditions (1952) and the emerging field of algorithmic game theory. Today, the Price of Anarchy is a standard measure in network design.
Braess himself noted the philosophical implications: "For each point in time, all users of the network can agree that they should be better off... and yet they cannot achieve this improvement." It's a striking illustration of how rational individual behavior can produce irrational collective outcomes.
Further Exploration
Recommended Reading
- Algorithmic Game Theory - Nisan et al.Comprehensive technical reference
- Thinking in Systems - Donella MeadowsAccessible systems thinking primer