Most passengers think about an MRT timetable in minutes.
Two minutes. Three minutes. Five minutes.
Railway engineers often have to think in seconds.
Not because seconds sound more precise, but because a difference of twenty seconds repeated across an hour can change how many trains can pass, how many passengers can be moved, how crowded platforms become, how much recovery space remains in the timetable and how quickly a delay can spread.
This is the mathematics of headway: the time separation between successive trains.
A metro is not only a system for moving trains. It is a system for controlling the gaps between trains.
This article continues our Singapore MRT mathematics series. Start with the main pillar, How MRT Timing Works Using Mathematics, which explains how speed, dwell, braking, signalling, energy and delay fit together.
The RFE — Why Does Headway Mathematics Exist?
The purpose is not to make the number between trains as small as possible.
The purpose is to create a train flow that is:
- safe,
- regular,
- high-capacity,
- recoverable,
- compatible with station dwell,
- compatible with terminals and junctions,
- and useful to the passengers actually waiting on the platform.
So the real objective is not:
minimise H at all costs
It is closer to:
find a headway H that is safely achievable, operationally stable, capacity-efficient, and robust to real-world variation
That distinction is the Reason for Existence of this mathematics.
Prompt 1 — What Exactly Is Headway?
Suppose Train A departs a station at:
08:00:00
Train B departs at:
08:02:00
The departure headway is:
H = 120 s
If train i departs at time Di and the previous train at Di−1, then:
Hᵢ = Dᵢ − Dᵢ₋₁
We can measure headway at arrival, departure or at another reference point on the railway.
The important concept is separation in time.
Passengers experience headway as waiting.
Operators experience it as train flow.
Signalling systems experience it as a moving safety and capacity problem.
Those are three perspectives on the same number.
Frequency and headway are inverses
If headway H is measured in seconds, theoretical train flow q in trains per hour is:
q = 3600 / H
This simple equation is one of the most important equations in urban rail.
At a headway of 180 seconds:
q = 3600 / 180 = 20 trains/hour
At 120 seconds:
q = 30 trains/hour
At 100 seconds:
q = 36 trains/hour
At 90 seconds:
q = 40 trains/hour
The relationship is nonlinear in a useful way. Cutting ten seconds from an already short headway can be much more valuable than it first appears.
Prompt 2 — Why Are 100 Seconds and 120 Seconds So Different?
Singapore gives us a concrete example.
LTA states that the upgraded Communications-Based Train Control signalling on the North-South and East-West Lines has the capability to run trains at a 100-second headway, compared with 120 seconds before 2018.
The numerical difference is:
120 − 100 = 20 seconds
Twenty seconds sounds small.
But train throughput changes from:
3600/120 = 30 trains/hour
to:
3600/100 = 36 trains/hour
The increase is:
(36 − 30) / 30 × 100% = 20%
So a 16.7 per cent reduction in headway produces a 20 per cent increase in theoretical train paths per hour.
This is why railway engineers can care intensely about twenty seconds.
A small unit of time, repeated many times, becomes infrastructure capacity.
Ten-minute intuition
There are 600 seconds in ten minutes.
At 120-second spacing:
600/120 = 5 trains
At 100-second spacing:
600/100 = 6 trains
One extra train every ten minutes can be a large difference when each train carries many passengers.
Prompt 3 — How Does Headway Become Passenger Capacity?
Suppose each train can carry an effective passenger load P under the assumptions of our model.
Theoretical passenger throughput per hour is:
C = qP = (3600/H)P
For illustration only, suppose P = 1,000 passengers.
At 120-second headway:
C = 30 × 1000 = 30,000 passengers/hour
At 100 seconds:
C = 36 × 1000 = 36,000 passengers/hour
The train did not become larger.
The train did not necessarily become faster.
The extra capacity came from using the track more frequently.
This is a central property of metro systems.
Track is a scarce corridor. Headway determines how often that corridor can be reused.
However, the calculation above is theoretical. Real capacity depends on actual train configuration, passenger-loading standards, station demand, operating pattern, terminal capacity, signalling, power and reliability.
This leads to the next question.
Prompt 4 — Why Can’t We Just Keep Reducing Headway?
If 100 seconds is better than 120 seconds, why not 80?
Why not 60?
Because signalling is only one part of a railway.
A useful conceptual equation is:
Hoperating ≥ max( Hsignal, Hstation, Hjunction, Hturnback, Hpower, Hfleet, Hrecovery )
Each term represents a possible bottleneck.
Signalling
Trains need safe movement authority and sufficient braking protection.
Stations
A platform needs enough time for doors, passengers and safe departure. If dwell takes too long, the following train eventually catches up.
Junctions
Conflicting train routes cannot occupy the same track path simultaneously.
Terminals
Trains must reverse direction, cross over and re-enter service. A slow turnback can constrain an entire line.
Power
More trains operating simultaneously create different traction-power demands.
Fleet
Shorter headway requires enough available trainsets to fill the timetable.
Recovery
If every gap is compressed to the theoretical minimum, there may be little room to absorb ordinary variation.
This gives us the infrastructure bottleneck principle:
The usable capacity of a railway is often governed by the slowest constraint, not the fastest component.
LTA’s renewal of the NSEWL illustrates this systems view. The signalling upgrade was accompanied by renewals to track circuits, power supply, trains and other core infrastructure rather than being treated as an isolated improvement.
Prompt 5 — How Do Passengers Change Headway?
This is where the railway becomes a human system.
Suppose passengers arrive at a platform at an average rate λ passengers per second.
If headway is H seconds, the approximate number arriving between trains is:
N ≈ λH
Suppose λ = 2 passengers per second.
At 180 seconds:
N = 2 × 180 = 360 passengers
At 120 seconds:
N = 240 passengers
At 100 seconds:
N = 200 passengers
A shorter headway can therefore reduce the number of passengers accumulating between trains.
But the relationship runs both ways.
Passenger accumulation influences dwell time.
Dwell time influences departure time.
Departure time determines the next headway.
Headway → passenger accumulation → dwell → departure time → new headway
The output of one step becomes the input of the next.
This is a feedback system.
A simple queue condition
Suppose a train can accept K of the waiting passengers before reaching its effective loading constraint.
For the platform queue to remain stable in a simplified model:
λH < K
If instead:
λH > K
then more passengers arrive between trains than the next train can absorb.
The queue grows:
Qₙ₊₁ = Qₙ + λH − K
Headway has become queueing theory.
Prompt 6 — Why Is Regularity Sometimes More Important Than Average Frequency?
This is one of the most surprising results.
Suppose two services both average one train every two minutes.
Service A is perfectly regular:
120, 120, 120, 120, 120 seconds
Service B is irregular:
60, 180, 60, 180, 120 seconds
The average may still be around 120 seconds.
But passengers do not experience averages.
They experience the actual gap in front of them.
A long gap collects more passengers.
The train after that long gap becomes more crowded and may dwell longer.
The train behind it encounters a shorter gap and may carry fewer passengers.
This can create bunching.
The mathematics of expected waiting
For passengers arriving randomly relative to a high-frequency service, a useful result from renewal theory is that expected waiting time depends not only on the mean headway but on its variability.
A standard expression is:
E[W] = E[H²] / (2E[H])
Because:
E[H²] = Var(H) + E[H]²
we can rewrite:
E[W] = E[H]/2 + Var(H)/(2E[H])
This equation is beautiful.
It says average waiting consists of:
half the mean headway + a penalty for irregularity
If headway variance is zero, the second term disappears.
If headways become irregular, expected waiting rises even when the average frequency does not change.
Passengers pay for variance.
This is one reason a good metro does not merely chase a low average headway. It tries to keep the gaps orderly.
Prompt 7 — How Does One Delay Distort the Gaps?
Let three trains be A, B and C.
Suppose the desired headway is:
H* = 120 s
Initially:
A → 120 s → B → 120 s → C
Now Train B experiences a 30-second delay at a crowded station.
The gaps can become approximately:
A → 150 s → B → 90 s → C
The mean across the two gaps is still:
(150 + 90)/2 = 120 s
Average frequency looks unchanged.
But the system is now less regular.
More passengers collect in the 150-second gap.
Fewer collect in the 90-second gap.
Train B may therefore encounter heavier boarding and become even later, while Train C encounters a lighter platform and catches up.
large gap → more passengers → longer dwell → still larger gap small gap → fewer passengers → shorter dwell → still smaller gap
This is positive feedback.
Headway error
Define desired headway H* and actual headway Hi.
Then headway error is:
eᵢ = Hᵢ − H*
A control objective is to bring:
eᵢ → 0
without compromising safety or creating a larger problem elsewhere.
The railway can potentially influence running profiles, dwell, departure timing and operational ordering within permitted rules to improve regularity.
This is control theory expressed as two trains on a platform display.
Prompt 8 — What Is the Best Headway?
There is no universal answer.
The best headway depends on demand, infrastructure, signalling, train fleet, terminals, dwell, power, maintenance needs and the amount of robustness the timetable needs.
Mathematically, we can imagine an objective:
minimise J(H)
where J may include:
passenger waiting + crowding + operating cost + energy + headway variance + delay risk + insufficient recovery margin
subject to constraints such as:
H ≥ Hsafe H ≥ Hstation H ≥ Hturnback fleet requirement ≤ available fleet power demand ≤ usable system envelope
If H is too large, passengers wait longer and platforms collect more people.
If H is pushed too close to the practical limit, the railway may have less flexibility to absorb ordinary variation.
The engineering problem is therefore not to make H mathematically tiny.
It is to make H appropriately small and operationally stable.
The best headway is not merely the shortest gap a railway can survive once. It is a gap the railway can sustain repeatedly.
A Worked Singapore-Style Headway Example
Let us build a deliberately simplified educational example.
Suppose one direction of a metro line can operate at a 120-second service interval.
Train flow:
q120 = 3600/120 = 30 trains/hour
Suppose the infrastructure is improved such that a 100-second capability is achievable under the relevant conditions.
q100 = 3600/100 = 36 trains/hour
Additional theoretical train paths:
36 − 30 = 6 trains/hour
Now suppose passenger arrival rate at one busy platform is:
λ = 3 passengers/s
Expected arrivals during a 120-second interval:
N120 = 3 × 120 = 360
During 100 seconds:
N100 = 3 × 100 = 300
Difference:
60 fewer passenger arrivals between trains
This can reduce the passenger batch confronting each arriving train, although actual crowding depends on many other factors.
Now assume both cases have the same average passenger demand per hour.
The shorter headway divides that demand into smaller, more frequent batches.
This can change:
- platform accumulation,
- doorway pressure,
- train load distribution,
- expected waiting time,
- and the probability of extreme crowding events.
Twenty seconds has entered the passenger experience through several pathways at once.
Why Average Headway Can Hide a Bad Railway
Suppose a reporting dashboard says:
Average headway = 120 seconds
That sounds good.
But compare two sequences.
Perfectly regular:
120, 120, 120, 120
Highly irregular:
30, 210, 30, 210
Both average 120 seconds.
They are not remotely the same passenger service.
This is why statistics matters.
A railway should care about:
mean headway variance maximum gap distribution shape recovery after disturbance
not merely one average.
Headway and the Terminal Problem
Every train eventually reaches a terminal or reversing point.
If the terminal needs at least Tturn seconds between successive turnbacks, then:
H ≥ Tturn
when terminal processing is the active bottleneck.
This is why physical rail infrastructure can alter a timetable equation.
The Pasir Ris rail turnback completed by LTA in 2024 is one Singapore example of infrastructure intended to allow faster crossing-over of trains and support shorter East-West Line frequencies when required.
The timetable is not floating in software.
It is anchored to steel, switches, power equipment, platforms and physical space.
Headway and Fleet Mathematics
Shorter headway also means more trains may be required.
Let total operating cycle time be C minutes.
Let headway be H minutes.
A simplified active-fleet estimate is:
N ≈ C/H
If the cycle takes 120 minutes and headway is 2 minutes:
N ≈ 120/2 = 60 trains
At 100 seconds:
100 s = 1.667 min N ≈ 120/1.667 ≈ 72 trains
This is a simplified illustration rather than an estimate of any actual Singapore line fleet requirement.
It shows why higher frequency is not free.
It can require:
- more trains in service,
- more maintenance capacity,
- more depot capacity,
- more traction-power support,
- more precise timetable management,
- and more resilience if one train leaves the pattern.
The Headway Paradox
The shortest possible headway is not always the best operational headway.
Suppose a railway can theoretically achieve Hmin.
If it schedules every train exactly at that boundary, then any small disturbance may violate the intended spacing or cause following trains to slow.
So the timetable may deliberately operate with some margin:
Hscheduled = Hminimum + recovery allowance
The railway sacrifices a little theoretical capacity in exchange for stability.
This resembles engineering across many fields.
Aircraft do not normally fly at structural failure limits.
Bridges are not designed so that ordinary operation sits exactly at collapse.
Computer systems keep spare capacity.
Railways keep operating margin.
Unused margin can look inefficient right up until the moment reality deviates from the plan.
RFE Return — What Is a Good MRT Frequency?
A passenger might define good frequency as:
the next train comes quickly
That is reasonable.
But the railway must add:
and the train after that comes regularly and the next one does too and the platform remains manageable and the terminal can reverse them and the signalling protects them and the power system supports them and a small delay does not destroy the pattern
So a good frequency is not merely high.
It is sustainable.
The railway is successful when the sequence:
Train A → H → Train B → H → Train C → H → Train D
remains close to the intended rhythm despite passengers, variations and ordinary disturbances.
Conclusion — A Railway Is Made from the Empty Space Between Trains
We usually notice the trains.
But one of the most valuable things on an MRT line is the invisible interval between them.
That interval contains:
- safety margin,
- braking space,
- passenger waiting,
- platform accumulation,
- timetable capacity,
- recovery opportunity,
- and the possibility of fitting another train into the hour.
At 120 seconds, one hour contains room for 30 theoretical train intervals.
At 100 seconds, it contains 36.
But the real achievement is not simply squeezing trains closer.
It is preserving the pattern.
Because a metro is not a collection of independent trains.
It is a sequence.
And the sequence works only if the spaces remain useful.
The passenger sees the next train. The mathematician sees the gap before it.
Key Equations
Hᵢ = Dᵢ − Dᵢ₋₁ Headway
q = 3600/H Theoretical trains/hour
C = (3600/H)P Simplified passenger throughput
N ≈ λH Passenger arrivals between trains
Qₙ₊₁ = Qₙ + λH − K Simplified queue evolution
eᵢ = Hᵢ − H* Headway error
E[W] = E[H²]/(2E[H]) Expected random-arrival wait
E[W] = E[H]/2 + Var(H)/(2E[H])
Mean wait + irregularity penalty
Nfleet ≈ Ccycle/H Simplified active-fleet relationship
Important: These are educational models. Actual Singapore MRT operating rules, line capacities and control algorithms involve train-specific, line-specific and safety-validated engineering parameters beyond these classroom equations.
Continue the MRT Mathematics Series
- How MRT Timing Works Using Mathematics — the full system pillar.
- How MRT Braking Works Using Mathematics — stopping distance, braking curves and regenerative energy.
The next layer is the station itself: how passenger arrivals, doorway flow and dwell-time variation become a mathematical bottleneck for the entire railway.
Sources and Further Reading
- Land Transport Authority — NSEWL CBTC and 100-second headway capability
- Land Transport Authority — Upgrading the North-South and East-West Lines
- Land Transport Authority — Completion of NSEWL Six Core Systems Renewal
- Land Transport Master Plan — train headway and service-level context
- LTA Annual Report 2023/24 — Pasir Ris rail turnback and rail renewal
Return to the complete MRT Mathematics hub: How MRT Works | It’s Mathematics.