How Mathematics Improves The World | Making Traffic Lights Cooperate Instead of Compete
A traffic light can do something completely sensible and still make the city worse.
Give one road more green time.
The queue disappears.
Excellent.
Except the cars arrive at the next junction in one enormous platoon.
That junction cannot store them.
The queue spills backwards.
A bus is trapped.
A pedestrian crossing is delayed.
A side road waits through another cycle.
The first traffic light won its local problem.
The network lost.
This is why traffic control is a beautiful Mathematics problem.
It is not really about red, amber and green.
It is about many competing flows sharing limited space through time.
Quick Read
Traffic signals allocate a scarce resource: right of way. At an intersection, incompatible movements cannot all proceed safely at once. The controller therefore divides time among phases. Fixed-time systems follow schedules designed from expected traffic patterns. Actuated systems respond to local detectors. Adaptive systems go further, using real-time traffic measurements and algorithms to adjust splits, cycle lengths, offsets or phase timing as conditions change.
The Mathematics ranges from elementary ratios to queueing theory, optimisation, prediction, feedback control, graph models, probability and simulation. A signal plan can minimise average delay, reduce queue length, maximise throughput, favour buses, protect pedestrians, prevent spillback or balance several objectives. There is no single universal “best” timing because different objectives can conflict.
The Federal Highway Administration describes adaptive traffic signal control as a system in which traffic is detected and algorithms predict where traffic will be, allowing downstream signals to adjust in real time. FHWA also stresses that good detection is essential: an adaptive algorithm fed bad observations can make bad decisions faster.
One-sentence answer: Mathematics improves the world by turning traffic lights from isolated timers into coordinated decision systems that measure demand, predict arrivals, allocate scarce green time and manage whole-network queues rather than merely clearing whichever approach looks busiest.
A Green Light Is a Time Allocation
At its simplest, a traffic signal divides a repeating cycle into pieces.
Suppose a cycle lasts 90 seconds.
Some seconds go to north–south traffic.
Some go to east–west traffic.
Some may be reserved for protected turns.
Pedestrian crossing intervals must fit too.
Amber and all-red clearance times consume part of the cycle.
Every extra second given to one movement is a second unavailable to another.
This is already Mathematics.
A traffic controller is allocating a finite resource under safety and fairness constraints.
The First Question: How Much Traffic Arrives?
Traffic flow is often measured in vehicles per hour.
But averages can hide important structure.
Six hundred vehicles per hour could mean ten vehicles arriving almost every minute.
Or it could mean intense bursts separated by quiet periods.
The signal experiences the second pattern very differently.
Traffic engineers therefore care about arrival distributions, peak flow rates, platoons and turning proportions—not merely daily totals.
The central question is:
How quickly are vehicles arriving compared with how quickly the intersection can serve them?
Queues Appear When Arrival Outruns Service
If a movement receives 30 seconds of useful green in a 60-second cycle, it is not serving vehicles continuously.
During red, arrivals accumulate.
During green, the queue discharges.
If the discharge rate is high enough, the queue clears.
If not, some vehicles remain when the signal turns red again.
Now the next cycle begins with inherited queue.
Repeat this and the queue grows.
This is the heart of queueing theory:
arrivals, service, waiting and capacity.
When the average arrival rate approaches the effective service capacity, delays can increase sharply. Near saturation, small disturbances produce large queues because the system has little spare capacity to absorb variation.
Saturation Flow: How Fast Can a Queue Actually Leave?
When the signal turns green, the first driver does not teleport forward.
There is reaction time.
Then the queue begins moving.
After a few vehicles, discharge settles toward a characteristic headway: the time gap between successive vehicles crossing the stop line under continuous green.
The corresponding maximum sustained discharge rate is called the saturation flow.
It depends on lane geometry, vehicle mix, grades, turning movements, driver behaviour and other local conditions.
This is why twenty seconds of green is not twenty seconds of perfect service.
Some time is lost at the beginning and end of phases.
Effective green is what matters.
Cycle Length: Too Short Wastes Time, Too Long Creates Delay
Make a cycle very short and the intersection changes phases frequently.
That may reduce maximum red time.
But every phase change consumes lost time through amber, all-red and start-up effects.
Make the cycle very long and lost time becomes a smaller percentage of the cycle.
But vehicles on low-priority movements may wait a long time.
Pedestrians may face uncomfortable delays.
Queues may grow large during long reds.
So cycle length is a trade-off.
Classical traffic engineering developed mathematical timing formulas that balance these effects for expected flows.
Adaptive systems make the question dynamic:
what cycle length is appropriate now?
Split: Who Gets How Much Green?
The proportion of a cycle given to each phase is often called the split.
If north–south demand is twice east–west demand, equal green may be wasteful.
But proportional allocation is still not enough.
Pedestrian minimum times impose lower bounds.
Turning lanes have finite storage.
Bus priority may justify extra green.
A side road may need a fairness guarantee even when lightly used.
The split is therefore an optimisation variable surrounded by constraints.
Offset: The Mathematics Between Two Intersections
Now put two signals on the same road.
If both turn green at exactly the same clock time, cars leaving the first may reach the second during red.
The timing difference between signals is called the offset.
Choose the offset so a platoon departing one green arrives at the next during green and traffic can progress smoothly.
This is the familiar “green wave”.
But a green wave in one direction may disadvantage the opposite direction.
Side roads complicate the pattern.
Different vehicle speeds disperse the platoon.
Bus stops disturb it.
A parking exit injects new vehicles halfway along the corridor.
Coordination is therefore a network timing problem, not a trick of making every light green together.
The Time–Space Diagram: Draw Traffic as Geometry
Traffic engineers often use time–space diagrams.
Distance runs along one axis.
Time runs along the other.
A moving vehicle becomes a sloping line.
A green interval becomes a window at an intersection.
Signal coordination becomes a geometric problem:
Can the vehicle trajectories pass through a sequence of green windows?
A traffic problem that feels messy on the road becomes visible when represented in the right coordinate system.
This is one of Mathematics’ oldest powers.
Choose a representation where the hidden structure becomes drawable.
Platoons: Cars Are Not Independent Particles
A traffic light releases vehicles in groups.
That group is called a platoon.
As it travels, the platoon spreads.
Drivers accelerate differently.
Some change lanes.
Some turn off.
New traffic enters.
By the time the group reaches the next signal, its shape has changed.
Adaptive control can estimate these arrivals and adjust downstream green time accordingly.
FHWA describes adaptive signal control in exactly this spirit: detection upstream and downstream feeds algorithms that predict when and where traffic will arrive, then timing is adjusted in response.
Fixed-Time Control: A Plan for an Average Day
Fixed-time signal plans are designed from historical traffic patterns.
Morning peak.
Midday.
Evening peak.
Night.
If today resembles the expected pattern, fixed timing can work extremely well.
The weakness appears when today stops resembling the average.
A crash closes one lane.
A concert releases twenty thousand people.
Rain changes travel behaviour.
A school event shifts demand.
Fixed timing keeps executing yesterday’s expectation.
Adaptive control tries to measure today instead.
Actuated Control: Let the Junction Listen
Actuated traffic signals use detectors to respond to local demand.
If no vehicle is waiting on a side road, the controller may skip or shorten that phase.
If vehicles continue arriving, the green may extend up to a maximum.
This is already feedback control.
Measure.
Respond.
But local actuation still has a limitation.
The junction may respond sensibly to its own queue while ignoring what is happening downstream.
This is why network coordination matters.
Adaptive Control: Predict, Do Not Merely React
An adaptive signal system uses live traffic observations to alter timing parameters as conditions evolve.
Some systems operate with rolling horizons:
predict traffic a short distance into the future;
choose timings that perform well over that horizon;
implement part of the plan;
measure again;
recalculate.
This resembles model-predictive control used in many engineering systems.
The plan is deliberately temporary because new measurements will arrive.
Mathematics becomes a living negotiation with the road.
The Detector Is Part of the Mathematics
Adaptive algorithms sound intelligent.
But they only know traffic through sensors.
Inductive loops.
Radar.
Video analytics.
Bluetooth travel-time measurements.
Connected-vehicle data.
Each has noise and failure modes.
FHWA explicitly warns that adaptive control depends critically on accurate, maintained detection.
A broken detector can make an empty approach look full.
Or a real queue look absent.
The optimisation may then be mathematically correct relative to false input.
This is another reason Mathematics in the world always includes measurement quality.
Spillback: When a Queue Occupies the Wrong Intersection
A road segment has finite storage.
If the queue reaches the upstream intersection, it can block movements that are not even trying to use the congested downstream approach.
This is spillback.
It is one of the most dangerous network effects because capacity collapses nonlinearly.
A blocked box prevents cross traffic.
That creates queues on another road.
Those queues spill into another intersection.
A local congestion problem becomes a network gridlock problem.
FHWA’s adaptive-signal guidance treats queue management as a distinct objective and describes “gating” traffic upstream when a downstream link has insufficient storage.
This is mathematically interesting because the best action can be to hold vehicles farther away from the visible congestion.
Sometimes the correct way to clear a bottleneck is to stop feeding it.
Throughput and Delay Are Not the Same Objective
Suppose one signal plan moves 5,000 vehicles per hour but gives a small side road terrible delays.
Another moves 4,900 but gives much more balanced waiting times.
Which is better?
The answer depends on the objective.
Traffic optimisation may target:
- minimum average delay;
- minimum maximum queue;
- maximum throughput;
- minimum number of stops;
- bus priority;
- pedestrian accessibility;
- emergency-vehicle priority;
- reduced emissions;
- equity among approaches;
- spillback prevention.
A mathematical optimiser does not know which value society cares about most.
Humans choose the objective.
The algorithm reveals the consequences.
Fairness: How Long Is Too Long to Make One Road Wait?
If a small side road carries only 5% of traffic, a pure throughput optimiser may starve it for a long time.
Mathematically efficient.
Socially irritating.
Potentially unsafe if drivers begin accepting risky gaps.
Signal systems therefore use minimum greens, maximum reds and service guarantees.
FHWA notes that adaptive objectives may change by time of day: peak periods may favour throughput while other periods may place more emphasis on equitable access.
This is a good lesson for students.
Fairness often enters Mathematics as a constraint.
Not because Mathematics invented fairness.
Because society decided the optimum must remain inside it.
Pedestrians Change the Optimisation
A vehicle-centred optimisation can produce a poor street.
Pedestrians need time to enter and clear crossings.
Older adults and children may walk more slowly.
People with mobility impairments need accessible timings and geometry.
Long waiting times encourage risky crossing behaviour.
So signal control is a multi-user allocation problem.
A junction is not successful merely because cars move quickly through it.
The definition of success depends on whom the street is for.
Bus Priority: One Vehicle Can Represent Fifty People
A car and a bus each occupy one vehicle slot in a detector count.
They do not carry the same number of people.
If the objective is person delay rather than vehicle delay, buses become much more important.
Transit-signal priority may extend a green or shorten a red for an approaching bus.
But priority has a cost.
Cross traffic waits longer.
A poorly designed priority rule can destabilise coordination.
Again, the question is not whether bus priority is good.
It is how to express and manage the trade-off mathematically.
Emergency Vehicles: Sometimes the Objective Changes Instantly
An ambulance changes the value of time.
Signal pre-emption can clear an approaching path by interrupting normal coordination.
But once the vehicle passes, the network must recover.
Queues may have accumulated on competing approaches.
Offsets are disrupted.
The control problem becomes:
How do we temporarily prioritise the emergency without creating a long-lived network failure afterwards?
Mathematics helps systems change objective functions when circumstances change.
Traffic Waves: Congestion Can Travel Backwards
Cars move forward.
Traffic jams can move backward.
This sounds impossible until you watch a queue form.
Vehicles enter the back of the queue upstream.
Vehicles leave the front downstream.
The boundary between moving and stopped traffic can propagate backward even though every vehicle moves forward.
Macroscopic traffic-flow theory models density, flow and speed in ways analogous to fluids.
Shock-wave analysis describes how discontinuities in traffic state propagate.
This matters to signal control because releasing too much traffic can send a congestion wave upstream into another junction.
Fundamental Diagrams: More Cars Eventually Produce Less Flow
At low density, adding more vehicles increases flow.
More cars pass a point each minute.
But only up to a point.
As density rises, vehicles interfere with one another.
Speeds fall.
Beyond critical density, adding more vehicles can reduce total flow.
This relationship between flow, density and speed is represented by a fundamental diagram.
The counterintuitive lesson is important:
more vehicles do not always mean more throughput.
Past capacity, the system begins choking on its own demand.
Ramp Metering and Gating: Sometimes You Improve Flow by Holding Cars Back
Freeway ramp meters deliberately hold vehicles at an entry ramp.
This feels unfair to the driver waiting.
But if unregulated inflow pushes the freeway beyond critical density, total throughput can fall and everyone slows dramatically.
A controlled inflow can keep the mainline near a more productive operating point.
Signal gating uses a related idea on arterial networks:
hold traffic where there is storage instead of releasing it into a saturated downstream link with nowhere to go.
This is one of the clearest examples of local frustration producing global improvement.
Prediction Is Hard Because Drivers Respond to the System
Traffic is not weather.
Drivers can observe the control system and change behaviour.
A navigation app reroutes traffic.
A new signal plan makes one road faster.
More drivers begin using it.
A bus lane improves bus travel time.
Some travellers switch mode.
A road becomes less congested.
Demand returns.
The intervention changes the system being modelled.
This is why transport Mathematics sits partly in social science.
People respond to incentives and information.
Braess’s Paradox: More Road Can Make Traffic Worse
In some transport networks, adding a new road can make everyone’s equilibrium travel time worse.
This is known as Braess’s paradox.
Why?
Because individual drivers choose routes that are good for themselves.
The combination of individually rational choices can create a worse network equilibrium.
The paradox is not a claim that every new road worsens traffic.
It is a warning that network performance cannot be inferred by inspecting capacity additions one at a time.
Whole-network response matters.
Traffic Simulation: Build a City That Can Fail Cheaply
Before changing thousands of signals, engineers can simulate proposed timing plans.
Microsimulation represents individual vehicles and driver rules.
Mesoscopic models aggregate some behaviour.
Macroscopic models treat traffic more like continuous flow.
Each scale has different strengths and computational costs.
A model can test:
- a new signal offset;
- bus priority;
- road closure;
- pedestrian timing;
- incident response;
- event traffic;
- new development demand.
But the model is only as good as its behavioural assumptions, calibration and input demand.
A simulated queue is not a real queue until field data supports the model.
Calibration: Make the Virtual Road Behave Like the Real One
Traffic models contain parameters for:
- acceleration;
- car-following;
- lane changing;
- desired speed;
- gap acceptance;
- route choice;
- reaction time.
Use default values blindly and the model may produce a beautifully animated wrong city.
Calibration adjusts parameters so model outputs reproduce observed flows, speeds, queues and travel times reasonably well.
Validation then tests the calibrated model against different data.
This is the same verification-and-validation discipline seen in structural simulation.
Mathematics becomes trustworthy through receipts from reality.
Machine Learning Can Predict Traffic, But Prediction Is Not Control
Machine learning can estimate short-term traffic arrivals, detect incidents and forecast congestion.
Useful.
But knowing congestion is coming does not tell us automatically what signal action is safe or fair.
Control still has to respect:
- minimum pedestrian times;
- clearance intervals;
- phase conflicts;
- queue-storage limits;
- public-transport priorities;
- network coordination.
A prediction model asks:
What will probably happen?
A control system asks:
Given what may happen, what should we do under these constraints?
Different Mathematics.
Reinforcement Learning: Let the Signal Learn a Policy
Research systems increasingly explore reinforcement learning for traffic signal control.
An agent observes traffic state.
Chooses a signal action.
Receives a reward based on delay, queue or other objectives.
Then learns a policy over many simulated episodes.
The promise is adaptive decision-making in complex networks.
The challenge is safety and generalisation.
A policy trained on ordinary demand may behave badly during unusual incidents.
A reward that minimises delay may learn socially undesirable behaviour.
A simulation gap may make the learned policy brittle in the real road network.
The smarter the algorithm, the more important the boundary conditions become.
Connected Vehicles Change What the Signal Can Know
Traditional detectors observe traffic only at fixed locations.
Connected vehicles can report speed, position and trajectory information more directly, subject to privacy, standards and penetration rates.
This gives controllers a richer picture of approaching traffic.
But partial adoption creates sampling bias.
If only 20% of vehicles report data, are they representative?
Are buses overrepresented?
Do older vehicles disappear from the data?
Better data can create new statistical problems as well as solve old ones.
Privacy: A City Does Not Need to Know Who You Are to Know a Queue Exists
Traffic optimisation benefits from detailed movement data.
Citizens also deserve privacy.
The system objective is usually aggregate:
how many vehicles are arriving?
what is the queue length?
how fast is the corridor moving?
It does not necessarily require identifying the person in every car.
Good transport systems therefore separate the information needed for control from information that would merely make surveillance easier.
Mathematics can help through aggregation, anonymisation and privacy-preserving statistics.
Primary Mathematics: The Traffic-Light Game
Put two queues of counters on a table.
One queue has eight counters.
The other has four.
Each green interval lets two counters leave.
Every round, add new counters to represent arriving vehicles.
Ask the child:
- Which queue should go first?
- How long should it stay green?
- What happens if you always serve the longest queue?
- What happens to the smaller road?
- What if the next junction can store only six cars?
The child has entered queueing and control theory without knowing the names.
Secondary Mathematics: Traffic Becomes Functions and Networks
Secondary students can model:
- flow as vehicles per hour;
- queue growth as arrival minus departure;
- distance–time graphs;
- speed–flow relationships;
- signal cycles with ratios;
- corridors as networks;
- probability of arrivals;
- optimisation of simple timing plans.
They can calculate whether a queue clears in one cycle.
Plot two arrival patterns with the same hourly average and see why variability matters.
Explore how one intersection’s output becomes another’s input.
The chapters join.
Advanced Mathematics: The City as a Dynamic Network
Modern traffic control draws on:
- queueing theory;
- probability and stochastic processes;
- graph theory;
- optimisation;
- control theory;
- differential equations;
- partial differential equations;
- game theory;
- simulation;
- machine learning;
- operations research.
Microscopic models simulate individuals.
Macroscopic models treat traffic as flow.
Network optimisation coordinates signals.
Game theory studies route choice.
Model-predictive control repeatedly re-optimises as conditions change.
Traffic is an entire Mathematics curriculum disguised as a commute.
Why This Improves the World
1. It reduces wasted waiting time
Better timing can reduce delay when signals respond to actual demand rather than stale schedules.
2. It prevents local queues becoming network gridlock
Spillback-aware control protects downstream storage and keeps intersections from blocking one another.
3. It gives public transport priority intelligently
Signals can optimise person movement rather than blindly treating every vehicle as equivalent.
4. It makes streets safer and more accessible
Pedestrian clearance times, maximum waits and protected phases turn safety requirements into explicit timing constraints.
5. It lets the network adapt to incidents
Adaptive systems can respond when traffic patterns differ sharply from ordinary daily plans.
6. It reveals when the wrong local improvement hurts the whole city
Network models expose interactions that are invisible when each junction is evaluated separately.
What Mathematics Does Not Do
Mathematics does not create road capacity that does not exist.
It does not make every traveller happy.
It does not decide whether cities should prioritise cars, buses, bicycles or pedestrians.
It does not make detector faults disappear.
It does not guarantee a simulation represents human driving accurately.
It does not eliminate incidents, weather or unpredictable behaviour.
And it cannot optimise away a transport system whose demand fundamentally exceeds its physical capacity.
Sometimes the mathematical answer is not a better traffic light.
It is more public transport, different land use, road pricing, active mobility or reduced demand.
Frequently Asked Questions
What is adaptive traffic signal control?
Adaptive traffic signal control uses live traffic detection and algorithms to adjust signal timing as demand changes. Depending on the system, it may modify green splits, cycle lengths, offsets or phase timing in real time.
What is a traffic-signal cycle?
A cycle is the time required for a signal controller to complete a sequence of phases before repeating. The cycle contains green, amber, all-red and other intervals allocated to different movements.
What is a signal offset?
An offset is the timing relationship between coordinated signals. Proper offsets can help vehicle platoons released from one intersection arrive during green at downstream intersections.
What is queue spillback?
Spillback occurs when a queue fills the available road segment and extends into an upstream intersection or blocks other movements. It can cause network-wide capacity loss and gridlock.
Do adaptive signals always reduce congestion?
No. Performance depends on traffic patterns, detection quality, network geometry, controller objectives, maintenance and system design. Where demand exceeds available physical capacity, signal optimisation can manage congestion but cannot eliminate the underlying shortage.
Why not give the busiest road green all the time?
Because competing road users still need service, pedestrian and safety constraints must be met, and excessive release can overload downstream links. The network optimum may require holding back the busiest movement temporarily.
Sources and Further Reading
- Federal Highway Administration, Traffic Signal Timing Manual — Adaptive Traffic Signal Control.
- Federal Highway Administration, Adaptive Signal Control Technology.
- Federal Highway Administration, Measures of Effectiveness and Validation Guidance for Adaptive Signal Control Technologies, including queue-management and variable-objective guidance.
Continue Through eduKateSG
Continue with How Mathematics Works. Then move through Batch 003 with Finding Useful Pages in a Web Too Large to Read, Designing an Evacuation Before the Emergency, and Letting Many Wireless Systems Share the Same Air.
Final Thought: The Best Signal Is Not the One That Wins Its Intersection
A traffic light is easy to judge locally.
Did this queue shrink?
Did these cars move?
Did the green arrive quickly?
The harder question is whether the movement helped the city.
Did it overload the next block?
Did it trap a bus?
Did it starve pedestrians?
Did it create a queue that blocks an unrelated movement?
Did it preserve enough space for what arrives next?
This is what Mathematics improves.
It lets us stop asking whether one component is doing well and start asking whether the system is doing well.
That is a much harder question.
It is also the question cities actually need answered.