How Mathematics Improves The World | Stopping One Late Train From Making the Whole Railway Late

How Mathematics Improves The World | Stopping One Late Train From Making the Whole Railway Late

A train is five minutes late.

That sounds like one train’s problem.

Then it reaches a junction where another train was supposed to pass first.

One train waits.

A platform is now occupied at the wrong time.

A connecting service is held for transferring passengers.

The held train misses its slot through a single-track section.

A crew arrives late for its next working.

A trainset is not where tomorrow’s timetable expected it to be.

Five minutes becomes fifty trains.

This is delay propagation.

The railway is not a collection of independent trains.

It is a tightly coupled network in which time itself is shared infrastructure.

Mathematics helps keep that coupling from turning one disturbance into system-wide disorder.

Quick Read

Railway timetables allocate scarce track, junction, platform and rolling-stock capacity through time. When a train is delayed, it may conflict with trains that were scheduled later. Dispatchers then have to decide whether to retime trains, change their order, reroute them, hold connections, skip stops, short-turn services or cancel some trains so the network can recover.

Operations-research models formulate this as a constrained optimisation problem. Binary and continuous variables represent train order, route choice and event times. Constraints enforce minimum headways, track occupancy, platform compatibility, rolling-stock continuity and other operational rules. Objectives can minimise train delay, passenger delay, missed connections, cancellations or a weighted combination.

Recent railway research continues to focus on real-time optimisation. A 2025 study in the Journal of Rail Transport Planning & Management develops lightweight mixed-integer methods for rerouting, reordering and rescheduling with the aim of reducing delay propagation. A 2026 study integrates dynamic timetable rescheduling with rolling-stock reassignment under uncertain disruption duration. Other work shows why passenger-centred objectives matter: the timetable with the fewest train-minutes of delay is not always the timetable that saves passengers the most time.

One-sentence answer: Mathematics improves the world by treating railway disruption as a network scheduling problem, allowing dispatchers to choose the least damaging sequence of retiming, rerouting and connection decisions before one late train consumes the time slots needed by many others.

A Timetable Is a Reservation System for Time

A railway track can carry many trains.

Not at the same place at the same time.

The timetable reserves time windows.

Train A occupies Block 1 now.

Train B enters after a safe headway.

Train C crosses a junction at 09:14.

Train D uses Platform 3 from 09:16 to 09:19.

Capacity is therefore not only kilometres of rail.

It is railway geometry multiplied by time.

A delay consumes future reservations.

Primary Delay and Secondary Delay

A primary delay comes from the original disturbance.

door fault;
signal failure;
late crew;
weather;
track obstruction;
longer-than-planned dwell.

A secondary delay is created because another train interacts with the first.

Train B was on time.

It waits for delayed Train A to clear the junction.

B now becomes late without suffering the original fault.

Disruption management tries to prevent primary delay becoming large secondary delay.

Headway: Safety Creates a Minimum Time Gap

Trains cannot follow one another arbitrarily closely.

Signalling divides track into protected movement authorities or blocks depending on system design.

A following train must maintain enough separation to stop safely and satisfy signalling rules.

In timetable models this becomes a minimum headway constraint.

departure of B ≥ departure of A + minimum headway

If A is late, B’s earliest feasible time may move too.

Safety turns delay into an inequality.

Junction Conflicts: Two Perfectly Good Routes Can Be Incompatible

At a flat junction, two routes may physically cross.

Train A wants the junction eastbound.

Train B wants it southbound.

Both cannot occupy the conflicting route simultaneously.

The dispatcher chooses order.

A first delays B.

B first delays A.

But the correct choice depends on what happens after the junction.

One train may carry 800 passengers and connect to three later services.

The other may have recovery margin immediately downstream.

Order is a network decision, not a local coin toss.

Single-Track Sections: Meeting Points Become Scheduling Decisions

On a single-track railway, trains travelling opposite directions need passing loops or stations.

Timetable design decides where they meet.

If one train is late, the planned crossing point may no longer be best.

Hold the opposing train at Loop A?

Let it continue to Loop B?

The choice changes delay for both directions and later trains.

One-dimensional infrastructure creates combinatorial timing.

Platform Capacity: A Station Is Another Scheduling Machine

A major station may have many platforms.

Not every platform fits every train or route.

Some platforms lead only to certain tracks.

Length matters.

Electrification may differ.

Passenger interchange matters.

When a delayed train occupies Platform 4 beyond its scheduled departure, an arriving train may need another platform or wait outside.

Platform assignment and train rescheduling become coupled.

Dwell Time: Passengers Can Create Delay Feedback

A crowded platform takes longer to board.

A late train accumulates more waiting passengers.

More passengers increase dwell time.

Longer dwell makes the train later.

This positive feedback can destabilise frequent metro services.

late train → more passengers → longer dwell → later train

Behind it, the following train finds fewer passengers and catches up.

Trains bunch.

Headway control can therefore be more important than strict timetable adherence in high-frequency urban rail.

Slack: Timetables Need Space to Breathe

A timetable could schedule every movement at the theoretical minimum running time and headway.

It would be wonderfully fast.

It would also be fragile.

A thirty-second disturbance would propagate because no recovery margin exists.

Timetables therefore include buffer and recovery time.

Too little buffer creates instability.

Too much makes journeys unnecessarily slow and reduces capacity.

Robust timetable design optimises speed against resilience.

Recovery Time: Hide Small Delays Before Passengers Notice Them Later

Suppose a train is scheduled for 31 minutes between A and B but normally needs 29.

The two-minute margin can absorb minor variation.

But where should recovery time be placed?

Spread evenly?

Concentrated before important junctions?

Before major interchange?

Placement matters because one recovered minute before a conflict can prevent several secondary minutes downstream.

Delay Propagation Is a Graph Problem

Represent timetable events as nodes.

Departure from Station A.

Arrival at B.

Departure from B.

Add edges for dependencies.

running time;
dwell time;
headway;
connection;
rolling-stock continuation;
crew continuation.

Delay travels along dependency edges.

The timetable becomes a directed graph whose topology determines how disruption spreads.

A highly connected event can be a delay amplifier.

Event-Activity Networks: Timetable as Difference Constraints

Let π_i be time of event i.

An activity from i to j with minimum duration L gives:

π_j − π_i ≥ L

A maximum duration U can add:

π_j − π_i ≤ U

Running, dwell and transfer requirements become inequalities.

Timetable feasibility becomes a system of temporal constraints.

Binary Decisions: Who Goes First?

At a conflict point, either A precedes B or B precedes A.

That is a binary choice.

Mixed-integer programming uses binary variables for such decisions.

If x=1, A goes first.

If x=0, B goes first.

Big-M or stronger disjunctive formulations enforce the corresponding headway constraints.

Thousands of pairwise ordering choices produce a combinatorial problem.

That is why real-time rescheduling is computationally hard.

Retiming: Keep the Route, Change the Clock

The simplest recovery action is retiming.

A train keeps its route and stops but departure or arrival times shift.

Retiming can resolve small conflicts.

Under large disruptions, it may not be enough.

The network may need structural changes.

Reordering: Let Another Train Go First

Express Train A is late.

Local Train B is on time ahead of it.

Should B be held so A can pass?

Maybe.

But if A is still ten minutes away, holding B wastes capacity and delays B’s passengers.

Reordering decisions compare future consequences.

A nominal priority rule can be suboptimal after disruption.

Rerouting: Use a Different Path Through the Network

A route is blocked.

If alternative track exists, trains can be diverted.

Rerouting consumes capacity elsewhere.

A freight bypass may have lower speed.

An alternative platform may require conflicting crossings.

A diverted passenger train may skip an important station.

Network topology creates options and externalities simultaneously.

Short-Turning: Do Not Send Every Train to the End

A disruption blocks the final five stations of a line.

Instead of cancelling the whole service, turn trains back at an earlier station with suitable track.

Passengers beyond the turn point need alternatives.

But the remaining network retains more frequency.

Research on disruption rescheduling explicitly includes flexible short-turning because partial service can reduce total passenger delay.

Skipping Stops: Save Time by Creating Local Inconvenience

A delayed train can recover time by skipping a low-demand stop.

Passengers at that station wait longer.

Passengers already on board reach downstream stations sooner.

The decision is not morally neutral.

A passenger-centred optimisation weights who gains and who loses.

Flexible stopping research shows how such choices can be integrated with retiming, reordering and short-turning.

Cancellation: Sometimes Removing a Train Saves the Railway

Cancellation sounds like failure.

Under severe disruption, keeping every scheduled train can produce gridlock.

A cancelled service releases track and platform capacity.

Rolling stock can be repositioned.

Following trains may run more reliably.

The cost is concentrated on cancelled-train passengers.

The optimiser must compare one severe inconvenience with many smaller delays.

Passenger Delay Is Not Train Delay

Train A is ten minutes late with 50 passengers.

Train B is five minutes late with 900.

Which disruption is worse?

Train-delay minutes say A=10, B=5.

Passenger-delay minutes say A=500, B=4,500 before considering connections.

Research increasingly minimises passenger inconvenience rather than raw train delay because operational punctuality and human impact can diverge sharply.

Missed Connections Create Discrete Jumps in Cost

A passenger arrives two minutes before the connecting train leaves.

Fine.

Arrive two minutes after.

The passenger may wait thirty minutes for the next service.

A four-minute change in incoming delay creates a thirty-minute journey penalty.

Connection cost is discontinuous.

That is why holding a connecting train can be rational even if it worsens that train’s punctuality.

Hold or Go? The Connection Decision

Train B is ready to depart.

Train A will arrive four minutes late with 120 connecting passengers.

Hold B four minutes?

Benefits:

120 passengers keep connection.

Costs:

everyone on B waits;
B may lose a path at the next junction;
B may cause another connection failure later.

The right answer depends on downstream network state.

Connection holding is a decision tree through future timetable consequences.

Rolling Stock: Trains Need to Be in the Right Place Tomorrow Too

A passenger service is operated by a physical trainset.

Cancel or short-turn one trip and the vehicle may end the evening at the wrong depot.

Tomorrow morning’s first train now has no train.

Recovery therefore couples timetable rescheduling with rolling-stock circulation.

A 2026 optimisation study integrates these explicitly because a timetable that looks recovered operationally can create hidden vehicle infeasibility later.

Recovery has memory beyond the current hour.

Crew: People Have Rosters Too

Drivers and onboard staff have legal and contractual limits.

A late train can make a crew miss the next assignment.

A diversion may require route knowledge the current driver does not have.

A cancelled service may free one crew and strand another.

Full disruption management therefore becomes multi-resource scheduling:

track;
platform;
train;
crew;
passengers.

The railway is only feasible when all five layers agree.

Mixed-Integer Linear Programming: Put Operational Choices Into Variables

A typical rescheduling MILP contains:

continuous variables for event times;
binary variables for train ordering;
binary route choices;
binary cancellation decisions;
possibly passenger-flow variables.

Constraints encode:

minimum running times;
minimum dwell;
headways;
track conflicts;
platform assignment;
route continuity;
rolling-stock feasibility.

The objective minimises weighted disruption cost.

Operations becomes algebra.

Why Exact Optimisation Can Be Too Slow

A network with hundreds of trains can create millions of possible orderings.

Mixed-integer problems are combinatorial.

The global optimum may take too long to prove.

A dispatcher needs an answer now.

Research therefore uses:

decomposition;
heuristics;
rolling horizons;
conflict graph reduction;
warm starts;
approximate optimisation.

The best mathematical answer after the network has collapsed is less valuable than a very good answer in thirty seconds.

Rolling Horizon: Solve the Near Future, Then Solve Again

A disruption duration is uncertain.

Do not optimise the whole day once.

Optimise the next twenty or thirty minutes.

Implement the first decisions.

New information arrives.

Move the horizon forward.

Optimise again.

This is model-predictive control for a railway timetable.

A 2026 study uses a rolling-horizon framework precisely to adapt timetable and rolling-stock recovery to uncertain disruption duration.

Robust Rescheduling: Do Not Bet Everything on One Repair Time

Infrastructure control says the fault will clear in 20 minutes.

Maybe it takes 40.

A plan optimised for exactly 20 can fail badly.

Robust optimisation evaluates several plausible durations or uncertainty sets.

The chosen plan may be slightly worse under the best case and much better if repair takes longer.

Recovery should survive forecast error.

Fairness: Which Train Is Always Sacrificed?

An efficiency-maximising dispatcher may repeatedly delay the low-priority regional train to protect high-speed services.

Total delay falls.

One operator or passenger group carries disproportionate cost.

A 2023 study of railway timetable rescheduling explicitly analyses trade-offs between efficiency and fairness among competing train operators.

Optimisation does not automatically create fairness.

Fairness must be measured or constrained.

Passenger Fairness: The Last Train Matters Differently

Delay a commuter service at noon by ten minutes.

Inconvenient.

Cancel the final rural connection at midnight.

Passengers may be stranded overnight.

Equal delay minutes are not equal human consequences.

Passenger-centred objective functions can weight missed final connections, accessibility needs and crowding differently from ordinary minutes.

The objective function carries social priorities.

Information to Passengers Can Reduce the Cost of the Same Delay

A train will be twenty minutes late.

Tell passengers immediately and accurately.

Some choose another route.

Some postpone travel.

Some catch a bus.

The physical train delay remains twenty minutes.

Passenger inconvenience falls.

Operational control and information control are complementary.

Crowding: Rescheduling Changes Passenger Distribution

Cancel one train.

Its passengers board the next.

The next train becomes crowded.

Dwell time rises.

Some passengers cannot board.

Passenger assignment therefore changes train performance.

Integrated models combine timetable rescheduling with passenger rerouting rather than assuming passengers remain attached to their originally planned train.

Integrated Passenger Reassignment: Let Travellers Choose New Paths

A disrupted passenger can:

wait;
take another line;
transfer somewhere else;
abandon the journey.

A 2020 open-access study integrates train rescheduling and passenger reassignment using a MILP, minimising generalised travel time including waiting, in-vehicle time and transfers.

The network should be optimised for where passengers can actually go after disruption, not for a timetable that no longer exists.

Delay Trees: Which Primary Incident Caused This Late Arrival?

A train reaches its terminus 18 minutes late.

Where did the delay come from?

Five minutes from door fault.

Four waiting for another train.

Three from crowding.

Six from losing a junction slot.

Delay attribution builds causal chains.

This helps infrastructure managers identify which failures create the largest secondary consequences, not merely which incidents last longest.

Critical Nodes: Some Junctions Amplify Delay More Than Others

A remote branch terminus affects one line.

A central flat junction connects five corridors.

A five-minute blockage at the junction can propagate far more widely.

Graph centrality and simulation identify infrastructure whose disruption has disproportionate network impact.

Investment can then target flyovers, additional platforms, turnback sidings or signalling improvements where they buy the most resilience.

Capacity and Robustness Trade Off

Pack more trains into the timetable.

Capacity rises.

Recovery margins shrink.

A small disturbance propagates further.

Leave more empty paths.

Robustness rises.

Fewer passengers can travel at peak.

Railway planning therefore lives on a Pareto frontier between utilisation and resilience.

100% theoretical capacity is often poor operational capacity.

Punctuality Metrics Can Change the Timetable

If performance is measured only by “arrived within five minutes”, operators may optimise around that threshold.

A train four minutes late looks perfect in the metric.

A train six minutes late looks like failure.

The human difference is two minutes.

Threshold metrics create discontinuities and incentives.

Passenger-weighted minutes and distributional measures can provide richer performance views.

Measurement changes management.

Reinforcement Learning: Learn Dispatch Policies From Simulated Disruptions

Railway rescheduling can be represented as a sequential decision process.

State:

train positions;
delays;
track occupancy;
remaining disruption;
passenger demand.

Action:

hold;
release;
reroute;
reorder;
short-turn.

Reward:

negative passenger delay;
negative cancellations;
penalties for conflicts.

A 2026 study models train timetable rescheduling with deep reinforcement learning under perturbations and disruptions.

The promise is fast policy decisions.

The challenge is guaranteeing safety and generalisation under rare disruptions.

Simulation: Create a Thousand Bad Days Before One Happens

Block a junction for ten minutes.

Then twenty.

Remove one platform.

Add 30% passenger demand.

Delay the first train of the morning.

Rail simulation can test timetable resilience before deployment.

The design question becomes:

Under which disturbances does this timetable stop recovering?

Robustness is learned from failure scenarios, not from one perfect day.

A Classroom Thought Experiment: One Late Train

Draw a single-track line with passing loops at B and D.

Schedule Train East and Train West to meet at B at 10:10.

Delay East by eight minutes.

Should West wait at B?

Or continue to D?

Add a third train behind East.

Add twenty connecting passengers at D.

The best decision changes as network consequences are added.

Students discover secondary delay.

A Second Thought Experiment: Train Delay Versus Passenger Delay

Option A delays one train by 10 minutes carrying 50 people.

Option B delays two trains by 4 minutes each carrying 600 people.

Train-delay minutes:

A=10, B=8.

Passenger-delay minutes:

A=500, B=4,800.

Which objective should a railway optimise?

The exercise exposes the power of objective definition.

Primary Mathematics: Railways Begin With Timetables

Primary students already learn:

elapsed time;
distance;
speed;
tables;
ordering;
graphs.

A timetable is a table of times.

A delay is addition.

A connection is an inequality:

arrival + transfer time ≤ departure.

Operations research begins with arithmetic that has consequences.

Secondary Mathematics: The Timetable Becomes a Network of Constraints

Secondary students can add:

inequalities;
graphs;
networks;
probability;
optimisation;
statistics;
simulation.

Headways become inequalities.

Routes become graph paths.

Passenger delay becomes weighted sums.

Uncertain repair time becomes probability.

Advanced Mathematics: Railway Recovery as Real-Time Optimisation

Modern railway rescheduling draws on:

mixed-integer programming;
graph theory;
queueing theory;
stochastic optimisation;
dynamic programming;
decomposition;
simulation;
reinforcement learning;
multi-objective optimisation.

The railway is a moving resource-allocation problem.

The timetable provides the plan.

Real-time optimisation repairs the plan when reality refuses to follow it.

Why This Improves the World

1. It stops local delay becoming network delay

Conflict-aware rescheduling prevents a late train from automatically pushing every later train out of its time slot.

2. It protects passenger journeys, not only train statistics

Passenger-weighted models account for occupancy, transfers and alternative routes.

3. It creates robust timetables

Slack and recovery time can be placed where they prevent the greatest downstream propagation.

4. It makes disruption recovery faster

Real-time optimisation searches many reordering and rerouting possibilities faster than manual reasoning alone.

5. It reveals critical infrastructure

Simulation identifies junctions, platforms and single-track sections whose failure creates disproportionate delay.

6. It lets recovery remain adaptive

Rolling-horizon methods revise plans as repair times, passenger loads and train positions change.

What Mathematics Does Not Do

Optimisation does not repair broken signalling.

It does not create track capacity that does not exist.

It does not make uncertain repair times certain.

It does not automatically value one passenger group fairly.

It does not guarantee a globally optimal solution can be found fast enough for real-time use.

It does not replace dispatchers, signallers or safety rules.

And the least train delay is not necessarily the least human disruption.

Frequently Asked Questions

What is delay propagation?

Delay propagation occurs when one late train causes other trains to become late because they share tracks, junctions, platforms, passengers, rolling stock or crews.

Why can’t trains simply run in timetable order after a disruption?

Because the original timetable assumes trains are at planned positions. Once one is late, following the original order may create larger conflicts or wasted capacity. Reordering can reduce total network disruption.

What is timetable slack?

Slack is deliberately unused time or capacity inserted so small disturbances can be absorbed without causing secondary delay. Too little makes the timetable fragile; too much reduces speed or capacity.

Why would a railway cancel a train to improve service?

Under severe disruption, a cancellation can free scarce track, platform and rolling-stock capacity, preventing much wider delay. The decision is justified only if the network-wide passenger benefit outweighs the concentrated cost to cancelled-train passengers.

What is rolling-horizon rescheduling?

It repeatedly optimises a limited future window, implements near-term decisions, incorporates new information and then optimises again. This is useful when disruption duration and traffic state are uncertain.

Sources and Further Reading

Lövétei, Lindenmaier and Aradi, Efficient Real-Time Rail Traffic Optimization: Decomposition of Rerouting, Reordering, and Rescheduling Problems, Journal of Rail Transport Planning & Management, 2025.
Research on Railway Timetable Rescheduling With Flexible Stopping and Short-Turning, integrating retiming, reordering and cancellation to reduce passenger delay.
Reynolds, Ehrgott and Wang, Fairness of Railway Timetable Rescheduling, 2023.
Integrated Timetable Rescheduling and Passenger Reassignment During Railway Disruptions, 2020.
From Railway Disruptions to Recovery, 2026, integrating dynamic timetable rescheduling and rolling-stock reassignment under uncertain disruption durations.

Continue Through eduKateSG

Continue with How Mathematics Works. Compare this railway problem with Making Traffic Lights Cooperate Instead of Compete, where local timing decisions can damage a wider network, and with Keeping Aircraft Safely Apart in a Crowded Sky, where routes and time slots again become scarce shared resources.

Final Thought: The Railway Is Trying to Recover a Future