How MRT Network Resilience Is Measured Using Mathematics: What Happens When a Railway Loses Part of Itself

An MRT system is easiest to admire when nothing goes wrong.

Trains arrive. Doors open. Passengers transfer. The sequence continues.

Resilience becomes visible only when part of that sequence is lost.

A train is withdrawn. A station becomes temporarily unavailable. A power or signalling problem constrains one section. A large passenger surge overwhelms the normal flow. One line cannot carry its usual service, so passengers move towards other lines, buses and routes.

At that moment, a railway is no longer solving only:

How do we run the normal timetable?

It is solving a harder problem:

How much of the city’s mobility can still be preserved, how far does the disruption spread, and how quickly can the transport system return to useful service?

This is the mathematics of MRT network resilience.

It combines graph theory, probability, passenger-flow modelling, queueing theory, optimisation, control theory, statistics and infrastructure planning. It also requires a distinctly human measure of success, because a rail graph can remain connected on paper while real passengers are stranded, delayed or unable to use the remaining path.

Singapore’s current rail work makes this a timely subject. On 13 February 2026, LTA, SMRT and SBS Transit announced that recommendations from the Rail Reliability Taskforce would be progressively implemented across the network. The work spans asset management, workforce capabilities and service recovery, with public measures involving power, signalling, trains, monitoring, training and response readiness.

Meanwhile, Circle Line Stage 6 opened for passenger service on 12 July 2026, completing the Circle Line between HarbourFront and Marina Bay. LTA explicitly described the completed 33-station, 12-interchange circle as strengthening connectivity and resilience across the rail network.

These are two sides of resilience:

make failure less likely
and
make the network better able to continue and recover when disruption occurs

This article continues the eduKateSG MRT mathematics series. Begin with How MRT Timing Works Using Mathematics, then explore braking, headway, station dwell, delay recovery, timetable construction, energy optimisation, passenger capacity, and routes and transfers.

The RFE — Why Does MRT Resilience Mathematics Exist?

A weak definition of resilience would be:

make the railway impossible to disrupt

No large physical system can honestly promise that.

Equipment ages. Components fail. Passenger demand changes. Weather and external conditions create stress. Human beings make mistakes. Construction, renewal and expansion require temporary operating changes.

The Reason for Existence is therefore not perfection.

It is continuity of public purpose.

reduce the chance of failure
+
limit the immediate loss of function
+
prevent cascading passenger harm
+
provide alternative mobility
+
restore service intelligently
+
learn and adapt afterwards

The railway does not exist merely to preserve trains and tracks.

It exists to preserve people’s ability to reach work, school, examinations, hospitals, homes and one another.

Resilience is the ability of the transport system to keep enough of the human journey alive while part of the machine is hurt.

This changes what must be measured.

Counting intact stations is not enough.

We must also count:

• passengers who can still complete journeys,
• additional travel time,
• queues created elsewhere,
• alternative-route capacity,
• accessibility of the remaining paths,
• time required to restore service,
• and total loss accumulated during recovery.

That is the RFE of the mathematics.

Prompt 1 — What Is the Difference Between Reliability, Robustness, Recovery and Resilience?

These words are related, but they are not interchangeable.

Reliability

Reliability asks:

How consistently does the system perform without failing?

A general reliability function can be written:

R(t) = P(Tfailure > t)

It is the probability that the system or component continues operating beyond time t under the stated conditions.

Another common engineering quantity is availability:

A ≈ MTBF / (MTBF + MTTR)

where MTBF is mean time between failures and MTTR is mean time to repair or restore.

This is a generic engineering relationship, not a statement of Singapore’s operational metric definitions.

Robustness

Robustness asks:

When disruption begins, how much function remains?

Let normal functionality be Q₀ and the minimum functionality after a disturbance be Q_min.

A simple robustness ratio is:

Robustness = Qmin / Q0

If the railway retains 70 per cent of its chosen functionality measure:

Robustness = 0.70

Robustness is about the depth of the fall.

Recoverability

Recoverability asks:

How effectively and how quickly can function return?

If recovery begins at t₁ and acceptable service is restored at t₂:

Trecovery = t2 − t1

A crude average recovery rate is:

vrecovery
=
(Qrestored − Qmin)/(t2 − t1)

Recoverability is about the slope and duration of the return.

Resilience

Resilience combines the full journey through disturbance:

preparedness
→ resistance
→ degradation
→ partial operation
→ recovery
→ adaptation

It includes whether the system avoids a deep collapse, whether alternatives exist, whether passenger harm remains bounded, and whether restoration is fast and intelligent.

One mathematical form is the normalised area under a functionality curve:

Resilience
=
(1/TQ0) ∫₀ᵀ Q(t)dt

A value near 1 means the system retained or rapidly restored most of its function over the study window.

A lower value means deeper or longer performance loss.

Adaptability

Adaptability asks whether the system learns or changes after the event.

Does it restore the old state exactly?

Or does it return stronger through revised maintenance, new infrastructure, better training, changed operating procedures or additional alternatives?

post-event capability
>
pre-event capability

when genuine improvement has occurred.

Reliability tries to avoid the fall. Robustness limits the fall. Recovery climbs back. Resilience measures the whole passage. Adaptation changes what happens next time.

Prompt 2 — How Does Graph Theory Measure the Structural Resilience of an MRT?

Represent the rail network as a graph:

G = (V,E)

where:

• V is the set of stations,
• E is the set of rail links connecting them.

A disruption can remove or weaken a node, an edge, a section, a service pattern or a layer of the wider transport network.

Graph theory asks what happens to connectivity afterwards.

Degree

The degree of station i is the number of direct neighbouring stations or links in the chosen graph representation:

ki = number of edges incident to node i

An interchange may have higher degree than an ordinary station.

But degree alone does not prove criticality. A low-degree station on a unique branch can be vital to the communities beyond it.

Connected components

A connected component is a set of stations that remain mutually reachable.

Let S be the size of the largest connected component and N the original number of stations.

A simple connectivity ratio is:

LCC = S/N

If a failure divides the network into isolated pieces, LCC falls.

However, LCC treats all stations as equally important and says little about travel time or passenger demand.

Shortest-path length

Let d_ij be the shortest path between stations i and j.

A disruption may preserve connectivity while increasing d_ij.

Passengers can still travel, but only by a longer route.

Average path length is:

Lavg
=
1/[N(N−1)] Σi≠j dij

Lower average path length generally means stations are closer in the graph.

Global efficiency

When nodes become disconnected, their shortest-path distance is effectively infinite.

Using 1/d solves this neatly because:

1/∞ = 0

Global network efficiency is:

Eglobal(G)
=
1/[N(N−1)] Σi≠j 1/dij

It falls when paths become longer or impossible.

This makes it more informative than counting connected stations alone.

Betweenness centrality

A station can be structurally important because many shortest paths pass through it.

BC(v)
=
Σs≠v≠t σst(v)/σst

where:

• σ_st is the number of shortest paths from s to t,
• σ_st(v) is the number passing through station v.

A high-betweenness interchange may be a powerful connector.

It may also be a concentration of risk if too many journeys depend on it.

Loops and redundancy

A tree-shaped network has exactly one structural path between many station pairs.

A loop can provide another route.

For an undirected connected graph, the number of independent cycles is:

L = |E| − |V| + 1

A normalised loop measure is:

Loop density = L/N

A 2026 study of 45 global metro networks found that connectivity, redundancy, loop structure and modular organisation were important in explaining resilience across varied failure and restoration simulations.

Loops are not automatically good in every operational sense. They can introduce routing and control complexity.

But they can create alternative paths when one link is lost.

Node-disjoint paths

Two routes are node-disjoint if they do not share intermediate stations.

If an origin–destination pair has several genuinely independent paths, one station failure is less likely to destroy them all.

κod = number of internally node-disjoint paths from o to d

But a path that exists structurally may lack spare train or station capacity.

Structural redundancy is therefore only the first layer.

Circle Line Stage 6 as a real loop example

Circle Line Stage 6 opened on 12 July 2026 and closed the rail gap between HarbourFront and Marina Bay through Keppel, Cantonment and Prince Edward Road.

The completed Circle Line has 33 stations and 12 interchanges connecting to all existing MRT lines. LTA states that this strengthens network connectivity and resilience.

In graph language, CCL6 added nodes and edges that:

• completed a loop,
• created direct paths that did not previously exist,
• reduced transfer dependence for some journeys,
• and added alternative routing possibilities around the network.

LTA’s public example says Telok Blangah to Marina Bay can now be completed without the earlier transfers among the CCL, NEL and NSL, saving around ten minutes.

Closing a railway circle does more than shorten one journey. It changes the topology through which every disruption can be rerouted.

Prompt 3 — How Do Mathematicians Simulate MRT Failure?

Resilience cannot be understood from one disruption scenario.

Different failures produce different network consequences.

Node failure

A station or interchange becomes unavailable in the model.

G′ = G − v

Edge failure

A rail link or section becomes unavailable.

G′ = G − e

Capacity degradation

The link remains open but operates at reduced capacity or longer travel time.

ce′ = αce,  0 < α < 1

or

te′ > te

Service-pattern failure

The physical track exists, but normal train service does not cover it in the usual way.

This reminds us that infrastructure topology and service topology are different graphs.

Random versus targeted removal

A random-failure simulation removes stations or links in random order.

A targeted simulation removes them according to a property such as:

• degree,
• betweenness,
• passenger volume,
• or the largest marginal effect on network performance.

The 2026 global metro study compared several centrality-based, greedy and random strategies across 45 networks. Its purpose was not to reproduce ordinary outages, but to probe intrinsic structural vulnerability, fragmentation and recovery efficiency consistently across very different systems.

This distinction is important.

A mathematical stress test need not claim that a scenario is likely.

It asks what the system would reveal under controlled pressure.

Failure fraction

Let f be the fraction of stations or links removed.

Network performance becomes:

Q = Q(f)

A robust network loses performance slowly as f increases.

A fragile network shows a sharp collapse after a small number of critical losses.

A structural robustness score can be the area under the degradation curve:

Rfailure = ∫₀¹ Q(f)df

Higher area means more functionality remains across the simulated damage sequence.

The scenario matrix

Failure type is only one dimension.

A useful scenario has:

location
× time of day
× duration
× passenger demand
× available alternative capacity
× recovery resources

The same closed link can be manageable late at night and extremely costly during the morning peak.

The same one-hour interruption can have different effects if it occurs at an interchange rather than near a line end.

Therefore scenario s has probability p_s and consequence L_s.

Expected loss is:

E[L] = Σs psLs

But rare, extreme scenarios may deserve attention beyond their expected value.

A resilience plan therefore considers both:

frequent moderate disturbances
and
rare severe disturbances

Resilience is not measured by asking whether one preferred scenario survives. It is measured by asking how the system behaves across a family of plausible stresses.

Prompt 4 — Why Is a Connected MRT Map Not Necessarily a Functioning MRT?

Suppose a disruption occurs, but every station remains connected to every other station through some path.

Has the network preserved full function?

No.

The remaining path may:

• take much longer,
• require several additional transfers,
• lack enough train capacity,
• create unsafe or unmanageable platform queues,
• or be difficult for passengers needing accessible transfer paths.

This is the difference between structural resilience and functional resilience.

Origin–destination demand

Let q_od be the number of passengers wishing to travel from origin o to destination d during the study period.

Normal journey time is T_od⁰.

Disrupted journey time is T_od^D.

Passenger delay loss is:

Lpassenger
=
Σod qod max(0, TodD − Tod0)

If a journey becomes impossible, the model must assign a large penalty or track unserved demand separately.

U = Σod qod I(od unreachable)

where I is 1 when the journey is unreachable and 0 otherwise.

A network can keep a large connected component while U remains high if a small disconnected branch carries substantial passenger demand.

Passenger-weighted efficiency

Ordinary global efficiency gives every station pair equal weight.

A passenger-weighted version is:

Epassenger
=
[Σod qod / Tod]
/
[Σod qod]

or, when using graph distance:

Eflow
=
[Σod qod / dod]
/
[Σod qod]

The exact form depends on the chosen unit and model.

The principle is the same:

A route used by 100,000 passengers matters more to aggregate mobility than an equally long route used by ten.

Capacity-constrained rerouting

Suppose passengers from a failed line reroute through another line.

Let x_e be passenger flow on alternative edge e and c_e its usable capacity.

xe ≤ ce

When rerouted demand exceeds available capacity, passengers queue.

So the existence of an alternative route does not prove that the alternative can absorb the disrupted flow.

A 2023 passenger-flow interruption study modelled passengers, trains and platforms as interacting agents because disruption changes network topology, train schedules and passenger path choice simultaneously.

This is functional resilience:

path exists
+
service exists
+
capacity exists
+
passenger can reach it
+
journey remains acceptable

Accessibility-weighted resilience

Suppose the only remaining transfer path uses stairs.

For some passengers, its generalised cost is finite.

For a wheelchair user:

wstairs = ∞

The network is structurally connected but personally disconnected.

For passenger group g, define reachable opportunity A_g.

An equitable resilience score might be:

Requity
=
Σg wg Agdisrupted/Agfull

where w_g gives appropriate importance to different passenger groups.

This is an educational framework, not an LTA metric.

It protects an essential idea:

A transport network is not resilient if its remaining paths work only for the passengers easiest to move.

Prompt 5 — How Does One MRT Disruption Create Cascading Overload Elsewhere?

When one route loses capacity, passenger demand does not disappear.

It moves.

Passengers may:

• wait for service to return,
• use another MRT line,
• take a bridging or regular bus,
• walk to another station,
• use a taxi or private-hire vehicle,
• change destination,
• or abandon the trip.

Every choice changes load elsewhere.

Load ratio

For network component e:

ue = xe/ce

where:

• x_e is flow,
• c_e is usable capacity.

When u_e approaches 1, spare capacity is disappearing.

If u_e exceeds 1 in the demand model, the excess becomes queueing, denied boarding, detouring or unserved demand.

Nonlinear crowding cost

Crowding cost can rise sharply near capacity.

An illustrative function is:

ceffective(x)
=
c0 + α(x/c)β

with β greater than 1.

At low load, another passenger adds little cost.

Near capacity, another passenger can increase:

• boarding time,
• denied boarding,
• dwell variability,
• platform crowding,
• and transfer delay.

Queue propagation

Let Q_k be the waiting queue before train k.

If λ passengers arrive during headway H and B can board:

Qk+1 = max(0, Qk + λH − B)

A disrupted route sends additional arrivals into the alternative station.

So λ becomes:

λnew = λnormal + λrerouted

If boardable capacity does not rise accordingly, the queue grows.

This can lengthen dwell, distort headways and reduce effective train capacity.

rail disruption
→ passenger rerouting
→ alternative-line crowding
→ longer dwell and queues
→ reduced regularity
→ capacity loss on the alternative
→ wider disruption footprint

This is a cascading failure without another physical component necessarily breaking.

The second failure is functional overload.

Cascading threshold

Suppose alternative edge e can absorb a reserve fraction r_e:

reservee = ce − xenormal

Rerouted flow Δx is safe from overload only if:

Δx ≤ reservee

If:

Δx > reservee

queues or further rerouting begin.

The network may then cross from a stable stressed state into an unstable cascade.

Multimodal spillover

Rail resilience is therefore connected to buses, taxis, roads and walking paths.

During the September 2024 East-West Line disruption, the Ministry of Transport reported that 80 bridging buses were deployed during the morning peak on the second day, with regular buses, queue management and free chartered taxis for students and teachers travelling to PSLE and N-level examination venues.

This public example shows three levels of resilience:

• the rail system was impaired,
• the wider transport network substituted part of the lost function,
• and support was targeted towards passengers facing unusually high consequences.

A bus is not merely an emergency replacement train.

It has different capacity, road travel time, stopping pattern and boarding constraints.

So multimodal substitution is another optimisation problem:

allocate limited alternative vehicles
across disrupted passenger demand
so as to minimise total passenger harm

A resilient rail network is partly made from transport modes that are not rail.

Prompt 6 — How Is MRT Recovery Measured as a Curve?

Let Q(t) represent the chosen level of system functionality over time.

Before disruption:

Q(t) = Q0

When disruption occurs, Q falls.

During partial operation, it may remain at a reduced level.

During recovery, it rises.

normal
→ shock
→ degradation
→ minimum function
→ stabilisation
→ recovery
→ restored or adapted state

The resilience triangle

The area between normal performance and actual performance is accumulated loss:

Loss
=
∫t0t1 [Q0 − Q(t)]dt

This area incorporates both:

• how deeply service fell,
• and how long it remained below normal.

A shallow two-hour degradation may cause less total loss than a moderate two-day degradation.

A normalised resilience score is:

R
=
1 − Loss/[Q0(t1 − t0)]

which is equivalent to the normalised area under Q(t) when Q₀ is constant.

The 2026 global metro study used area under the functionality curve to compare degradation and recovery under varied intervention strategies.

A worked recovery curve

Suppose normal functionality is 1.0.

A fictional disturbance produces:

time 0: Q = 1.0
time 1: Q = 0.6
time 3: Q = 0.5
time 4: Q = 0.7
time 5: Q = 0.9
time 6: Q = 1.0

Using trapezoidal integration, area under Q(t) over six hours is:

AUC = 4.25 functionality-hours

The maximum possible area is:

6 × 1.0 = 6

So normalised resilience is:

R = 4.25/6
  ≈ 0.708

Accumulated loss is:

Loss = 6 − 4.25
     = 1.75 functionality-hours

Now compare another system that falls only to 0.7 but takes ten hours to recover.

It may have stronger immediate robustness but weaker overall resilience because the loss lasts longer.

This is why one metric cannot replace the curve.

Passenger functionality curve

Q(t) should be chosen carefully.

Possible definitions include:

• fraction of stations served,
• fraction of train-kilometres operated,
• fraction of origin–destination demand that can complete journeys,
• inverse of passenger travel-time loss,
• or a weighted combination of capacity, accessibility and waiting.

A passenger-centred functionality might be:

Qpassenger(t)
=
served passenger demand(t)
/
normal passenger demand(t)

Another might penalise additional travel time:

Qtime(t)
=
1 /
[1 + average excess journey time(t)]

The choice of Q determines what the resilience score values.

A metric based only on trains can declare success while passengers still face severe queues.

A metric based only on average passenger time can hide harm to a smaller but vulnerable group.

Before calculating resilience, decide whose function the curve is measuring.

Recovery is not the same as repair

A damaged component can be repaired while passenger service remains irregular.

Trains may still be displaced.

Platforms may still contain accumulated queues.

Alternative routes may still be overloaded.

So recovery can have stages:

fault controlled
→ limited safe movement restored
→ passenger flows stabilised
→ train sequence rebuilt
→ normal headway restored
→ residual queues cleared

The component recovery time and the passenger recovery time can differ.

Preparedness as measured recovery capability

On 24 January 2026, LTA conducted Exercise GREYHOUND 2026 at Serangoon MRT interchange with around 170 personnel from LTA, SBS Transit, SMRT, SPF and SCDF.

LTA described the exercise as validating coordinated responses to disruption and supporting swift recovery while minimising commuter impact.

Mathematically, an exercise tests whether assumed recovery parameters are realistic.

planned mobilisation time
vs
observed mobilisation time

planned passenger-routing capacity
vs
observed queue processing

planned communication sequence
vs
observed coordination

Simulation converts preparedness from confidence into evidence.

A recovery plan becomes resilience only after it survives rehearsal and measurement.

Prompt 7 — How Is Resilience Built Before an MRT Disruption Happens?

Recovery begins before failure.

The network can be made more resilient through several different mathematical strategies.

1. Reduce failure probability

Condition monitoring, preventive maintenance, predictive analytics, renewal and workforce capability aim to increase reliability.

If component hazard rate is h(t), maintenance aims to reduce the probability of failure in the operating window.

R(t) = exp[−∫₀ᵗ h(u)du]

for a general time-varying hazard model.

The 2026 Rail Reliability Taskforce recommendations include stronger monitoring, digitalisation, automation, predictive maintenance capability and scenario-based training.

2. Add redundancy to critical systems

A redundant component provides another way to perform the same essential function.

For two independent components working in parallel with reliabilities R₁ and R₂:

Rparallel
=
1 − (1−R1)(1−R2)

Real railway systems are not always independent, so common-cause failures must be considered.

But the formula explains the value of genuine backup.

The Rail Reliability Taskforce’s public recommendations include power-supply strengthening for the North East Line and Sengkang-Punggol LRT, including additional traction power substations and exploration of a dedicated backup power intake.

The resilience value is not only extra capacity.

It is reduced dependence on one pathway.

3. Add alternative rail paths

New lines, loops and interchanges can reduce single-path dependence.

CCL6 has already completed one such structural change.

LTA states that the future Downtown Line 2 Extension will enhance network resilience in the north-west by providing alternative travel and transfer options, including a new interchange with the North-South Line.

LTA also states that the Jurong Region Line will provide alternative travel options and redistribute commuter load across the network, enhancing rail-system resilience.

The mathematical benefit of a new link can be measured as:

ΔR = R(G + enew) − R(G)

But a new edge is valuable only if it supplies useful capacity and remains sufficiently independent of the failure it is meant to bypass.

4. Preserve spare capacity

An alternative line operating permanently at 100 per cent of practical capacity cannot absorb much disrupted demand.

For component e:

Reservee = ce − xenormal

Network redundancy without reserve capacity can be structurally impressive and operationally weak.

5. Prepare multimodal substitution

Buses, walking routes, taxis and other rail lines can preserve mobility when a section is unavailable.

A bus-bridging plan can be formulated as a vehicle-allocation problem.

Let y_r be buses assigned to replacement route r and K_bus the usable hourly capacity per bus under the model.

Creplacement,r = yr Kbus

subject to:

Σr yr ≤ available buses

and road travel-time, stop-capacity and boarding constraints.

A 2014 research framework showed how local metro–bus integration can be optimised before disruption to improve metro resilience.

6. Build operational flexibility

Crossovers, turnback facilities, train rescue capability, approved reduced-service patterns and safe fallback procedures can allow partial operation rather than complete stoppage in some circumstances.

The Taskforce’s public recommendations include studying additional procedures and bypass arrangements intended to support faster restoration without compromising safety.

This article deliberately does not reproduce detailed operating procedures.

The mathematical principle is sufficient:

more safe operating states
→ fewer scenarios in which one fault forces total loss of service

7. Train the human network

Rail resilience is not contained entirely in hardware.

It depends on operators, maintainers, station staff, bus operators, emergency services, communications teams and passenger guidance.

Coordination time can be modelled as a critical path through incident response.

If tasks j have durations t_j and precedence constraints, total restoration time depends on the longest necessary task chain:

Tresponse = length of the critical path

Training can shorten tasks, reduce variance and improve coordination.

That is why exercises and simulators belong inside resilience engineering.

The network that recovers is not only the network of tracks. It is also the network of people who know what to do next.

Prompt 8 — How Should Singapore Decide Which Resilience Improvement Comes First?

No railway has unlimited money, engineering hours, staff or physical space.

Resilience therefore requires prioritisation.

Expected benefit

For intervention a and disruption scenario s, let:

ΔLas = loss avoided by intervention a under scenario s

Expected avoided loss is:

Ba = Σs ps ΔLas

If intervention cost is C_a, a simple benefit-cost ratio is:

BCRa = Ba/Ca

But expected value can undervalue rare catastrophic losses.

Worst-case and tail-risk planning

A max–min strategy seeks the intervention that performs best under the worst modelled scenario:

maximise mina Rs(a)

More commonly, the choice would be:

choose a to maximise mins Rs(a)

Tail-risk methods such as Conditional Value at Risk focus on the expected loss in the worst fraction of scenarios.

CVaRα(L)
=
expected loss given that loss lies in the worst α-tail

The precise formulation depends on policy and data.

The lesson is that resilience planning should not optimise only the average day.

Marginal resilience gain

Suppose a proposed new connection or backup system increases resilience from R to R′.

Marginal gain
=
(R′ − R)/Cost

A greedy restoration strategy can similarly select the next component that provides the largest functionality increase per unit of restoration time or cost:

choose v* = arg maxv ΔQv/cv

The 2026 global metro study compared centrality-based and greedy restoration strategies and found that restoration order materially changes recovery performance.

This does not mean public articles should identify and rank sensitive local vulnerabilities.

It means infrastructure owners should use evidence to decide where reinforcement or restoration creates the greatest public value.

Multi-objective resilience

A resilience investment can improve one objective and worsen another.

For example:

• more reserve capacity improves disruption absorption but costs more,
• more crossovers improve operational flexibility but require construction and maintenance,
• more route alternatives improve redundancy but can increase network complexity,
• faster restoration may require larger standby resources,
• and intensive engineering works may require temporary service adjustments.

A conceptual objective is:

maximise J
=
w1 reliability
+ w2 robustness
+ w3 recoverability
+ w4 passenger accessibility
+ w5 equity
− w6 cost
− w7 maintenance burden
− w8 normal-service disruption

subject to budget, land, engineering-hour and safety constraints.

Equity in prioritisation

A purely passenger-volume-based model may always favour the busiest central stations.

But a quieter branch with no alternative may suffer a larger proportional accessibility loss when disrupted.

A social priority score can include:

Priority(v)
=
passenger volume
× disruption duration
× lack of alternatives
× accessibility impact
× consequence severity

The exact weights are policy choices.

Mathematics does not remove values from the decision.

It makes the values visible.

The best resilience investment is not necessarily the one that protects the largest station. It is the one that prevents the greatest avoidable human loss across the system.

A Complete Worked Fictional MRT Resilience Model

Let us build a small fictional network with nine stations:

A — B — C — D — E — F — A
        |           |
        G — H ------
        |
        I

For clarity, the actual edge set is:

AB, BC, CD, DE, EF, FA,
CG, GH, HE,
CI

This network contains:

• a six-station main loop A–B–C–D–E–F–A,
• a secondary alternative C–G–H–E,
• and a single-path branch to I through C.

This is a teaching graph, not a model of Singapore’s actual rail topology.

Step 1 — Measure the intact network

The graph has:

N = 9 nodes
M = 10 edges

Independent loops:

L = M − N + 1
  = 10 − 9 + 1
  = 2

Normalised loop density is:

L/N = 2/9 ≈ 0.222

Using unweighted station-to-station shortest paths, global efficiency is approximately:

Eglobal ≈ 0.586

Average shortest-path length is approximately:

Lavg ≈ 2.056 edges

Station C has the highest normalised betweenness in this fictional graph:

BC(C) ≈ 0.464

because it connects the single-path I branch and lies on many efficient routes.

Step 2 — Remove Station C

After C is removed, station I becomes isolated.

The other seven surviving stations remain connected through the loops.

Largest connected component relative to the original nine-node network is:

LCC = 7/9
    ≈ 0.778

Global efficiency, calculated with the original nine-node denominator and zero contribution for disconnected or removed pairs, falls to approximately:

Eglobal,C removed ≈ 0.323

Relative efficiency retained is:

0.323/0.586
≈ 0.551

About 55 per cent of the chosen efficiency measure remains.

But the passenger impact depends on who uses station I.

If I serves 50 passengers per hour, isolation creates one scale of loss.

If it serves 5,000 passengers per hour with no nearby alternative, the same topology creates a much larger human consequence.

Step 3 — Add a resilience link before failure

Now add an alternative edge between I and B:

enew = BI

The intact graph now has three independent cycles and global efficiency rises to approximately:

Eglobal,new ≈ 0.606

Now remove C again.

All eight surviving nodes remain connected because I can travel through B.

LCC = 8/9
    ≈ 0.889

Global efficiency with the original denominator is approximately:

Eglobal,new,C removed ≈ 0.396

The new edge improves post-failure efficiency from roughly 0.323 to 0.396:

relative improvement
=
(0.396 − 0.323)/0.323
≈ 22.6%

More importantly, station I is no longer disconnected.

The new route may be longer, but it preserves reachability.

Step 4 — Add passenger demand

Suppose 1,000 passengers per hour travel between I and F.

Before failure, one shortest route is:

I → C → B → A → F

After C fails without the new BI edge:

journey impossible

With the new edge:

I → B → A → F

The passenger path remains feasible.

But if edge BI can carry only 400 additional passengers per hour, the remaining 600 still require another mode, a queue or a different journey plan.

structural resilience = path survives
functional resilience = enough passengers can use it

Step 5 — Compare two investments

Investment A adds BI at a cost of 10 units and avoids 6,000 passenger-hours of modelled disruption loss across the scenario set.

BCRA = 6000/10 = 600

Investment B reinforces a different link at a cost of 5 units and avoids 2,000 passenger-hours.

BCRB = 2000/5 = 400

Investment A has greater benefit per cost under this fictional expected-loss model.

But suppose B protects a hospital-access route for passengers with few alternatives.

A purely aggregate calculation may no longer be ethically sufficient.

The final model must include consequence severity and equity.

The graph tells us where the network bends. Passenger demand tells us where the city hurts.

The Resilience Deletion Tests

Remove one layer and observe what the model gets wrong.

Remove reliability

The model accepts frequent failures as long as recovery is fast.

Remove robustness

The network can collapse deeply and still score well if it eventually recovers.

Remove recovery time

A network that returns in one hour looks identical to one that returns in one week.

Remove passenger flows

A quiet station and a major interchange are treated as equal dots.

Remove capacity

Every structural alternative is assumed able to absorb unlimited rerouted demand.

Remove accessibility

A staircase counts as a universal alternative path.

Remove buses and other modes

The model underestimates substitution outside the rail graph.

Remove common-cause failure

Two backups sharing one vulnerable dependency are mistakenly treated as independent.

Remove adaptation

The railway returns to the old state but never becomes better prepared.

The deletion tests show why resilience is multidimensional.

A resilient network is not merely hard to break. It is difficult to isolate, difficult to overload, quick to restore, inclusive while degraded, and capable of learning afterwards.

The MRT Resilience Paradoxes

Paradox 1 — A more connected network can spread disruption further

Interchanges create alternatives, but they also allow passenger overload and delay to propagate across lines.

Paradox 2 — Spare capacity can look inefficient until failure occurs

Unused train or bus capacity has an everyday cost, but it becomes valuable absorption capacity during disruption.

Paradox 3 — Restoring the busiest station first is not always optimal

A lower-volume bridge station may reconnect two large network components and create more immediate functionality.

Paradox 4 — Repair completion may occur before passenger recovery

Queues, displaced trains and overloaded alternatives can persist after the physical fault is cleared.

Paradox 5 — A network can be connected but not accessible

The remaining route may be too slow, crowded or physically unsuitable for some passengers.

Paradox 6 — More backup systems can share one hidden dependency

Nominal redundancy is weaker when alternatives fail for the same reason.

Paradox 7 — A slower partial service can accelerate total recovery

Safe reduced operation may preserve passenger movement and prevent queues from growing while full service is restored.

Paradox 8 — The most resilient network may not have the shortest normal journeys

Redundant paths, reserve time and backup capacity can impose everyday cost while reducing disruption loss.

How MRT Resilience Mathematics Grows from School to Research

Primary Mathematics

• fractions of service remaining,
• percentages and ratios,
• time to recover,
• simple route alternatives.

Secondary Mathematics

• graphs and networks,
• rates and recurrence relations,
• probability,
• statistics and variance,
• areas under piecewise graphs.

Junior College

• integration of performance loss,
• probability distributions,
• optimisation,
• differential and difference equations,
• expected value and conditional risk.

University and Research

• complex network science,
• percolation and graph dismantling,
• passenger assignment,
• agent-based simulation,
• stochastic programming,
• recovery optimisation,
• multilayer transport networks,
• reliability engineering,
• equitable accessibility analysis.

The railway does not experience these as separate school chapters.

It experiences one disturbance.

All the mathematics has to arrive together.

The World Return — When a Real Disruption Answers the Resilience Model

Before an incident, the model predicts:

• which stations are critical,
• how passengers will reroute,
• how quickly queues will grow,
• how many buses are required,
• which restoration order is best,
• and how long recovery will take.

Then reality returns observations.

predicted passenger diversion
vs
observed passenger diversion

predicted queue length
vs
observed queue length

planned mobilisation time
vs
actual mobilisation time

predicted recovery curve
vs
actual recovery curve

Let predicted functionality be Q̂(t) and observed functionality Q(t).

Model error is:

e(t) = Q(t) − Q̂(t)

If the model repeatedly overestimates alternative-route capacity, the capacity assumptions need revision.

If passengers ignore the recommended route, behavioural assumptions need revision.

If mobilisation is slower than planned, training or resource placement may need revision.

If one unforeseen dependency repeatedly appears, the system map is incomplete.

model
→ exercise or real event
→ observation
→ mismatch
→ diagnosis
→ revised infrastructure, procedure or training
→ new model

This is why the 2026 Taskforce process included technical audits, workshops, site visits, ground observations and staff interviews, and why Exercise GREYHOUND used simulation to validate response readiness.

The world does not merely test the railway.

It teaches the railway what its model forgot.

A resilience model that cannot learn from disruption is only a diagram of yesterday’s confidence.

RFE Return — What Does a Resilient MRT Owe the Passenger?

It does not owe a promise that nothing will ever fail.

That promise would be dishonest.

It owes something more practical:

serious prevention
+
early detection
+
safe containment
+
honest information
+
usable alternative mobility
+
protection for vulnerable and high-consequence journeys
+
competent recovery
+
learning afterwards

A resilience score should therefore not celebrate the network merely because most stations remained open.

It should ask:

• Could people still reach where they needed to go?
• How much longer did it take?
• Who could not use the alternatives?
• Where did queues and crowding move?
• How rapidly was useful service restored?
• What changed so that the next event would cause less harm?

The railway’s output is not trains.

Its output is reachable life.

The RFE of MRT resilience is to keep Singapore connected enough to continue being Singapore when part of the normal route disappears.

Conclusion — Resilience Is the Shape of What Remains

When an MRT line is working normally, the network is easy to describe.

Stations connect.

Trains run.

Passengers move.

Resilience begins when those statements stop being completely true.

Then mathematics asks:

• How much of the graph remains connected?
• How much longer have the paths become?
• Which passenger flows have no alternative?
• Can the alternative routes absorb the displaced demand?
• Will queues remain stable or cascade?
• How deep is the loss of functionality?
• How fast is the recovery curve rising?
• Which restoration action creates the largest public gain?
• Did the system return stronger than before?

The structural engineer sees components.

The graph theorist sees nodes, edges and loops.

The operations researcher sees capacity-constrained rerouting.

The statistician sees distributions and tail risk.

The control engineer sees a functionality curve returning towards its reference state.

The passenger sees whether they can still get home.

All of those views are necessary.

Reliability
→ fewer disruptions

Robustness
→ shallower loss

Redundancy
→ alternative paths

Reserve capacity
→ alternatives that can actually absorb demand

Recovery
→ functionality restored

Adaptation
→ the next disruption hurts less

Circle Line Stage 6 changed the shape of Singapore’s graph by completing a loop and adding new direct and alternative journeys. Future Downtown and Jurong Region Line extensions will add more options. The Rail Reliability Taskforce is strengthening systems, capabilities and recovery. Exercises test whether response plans work when people, trains and agencies must coordinate under pressure.

None of these guarantees that a disruption will never occur.

Together, they change what happens after one does.

A resilient MRT is not a railway that never loses anything. It is a railway designed so that one loss does not automatically become the loss of the whole journey.

That is how MRT network resilience is measured using mathematics.

Not by looking only at what broke.

But by measuring the shape of what remained, the cost carried by passengers, the speed of the return and the intelligence of what changed afterwards.

Key Equations

R(t) = P(Tfailure > t)
Reliability function

A ≈ MTBF/(MTBF + MTTR)
Generic steady-state availability

Robustness = Qmin/Q0
Immediate retained functionality

Trecovery = t2 − t1
Recovery duration

R = (1/TQ0)∫Q(t)dt
Normalised resilience over time

Loss = ∫[Q0 − Q(t)]dt
Accumulated performance loss

G = (V,E)
Rail network graph

L = |E| − |V| + 1
Independent cycles in a connected undirected graph

LCC = S/N
Largest connected component ratio

Eglobal = 1/[N(N−1)]Σi≠j 1/dij
Global network efficiency

BC(v) = Σs≠v≠t σst(v)/σst
Betweenness centrality

κod = number of internally node-disjoint paths
Structural path redundancy

Rfailure = ∫₀¹Q(f)df
Area under a degradation curve

E[L] = Σs psLs
Expected scenario loss

Lpassenger = Σod qod max(0,TodD − Tod0)
Passenger travel-time loss

U = Σod qod I(od unreachable)
Unserved passenger demand

ue = xe/ce
Network-component load ratio

Qk+1 = max(0,Qk + λH − B)
Passenger queue evolution

Reservee = ce − xenormal
Spare capacity for rerouting

Rparallel = 1 − (1−R1)(1−R2)
Idealised reliability of independent parallel components

ΔR = R(G + enew) − R(G)
Resilience gain from a new connection

BCRa = [Σs psΔLas]/Ca
Expected avoided-loss benefit per cost

choose v* = arg maxv ΔQv/cv
Greedy restoration priority

Requity = Σg wg Agdisrupted/Agfull
Illustrative equitable accessibility resilience

e(t) = Q(t) − Q̂(t)
Observed minus predicted functionality

Reader-safety note: The fictional graph, numerical calculations and general equations in this article are educational abstractions. They do not identify sensitive Singapore MRT vulnerabilities, reproduce internal restoration priorities, reveal restricted control procedures or model security-critical operating details. Actual resilience planning uses confidential engineering data, validated system models and authorised operational protocols.

Continue the MRT Mathematics Series

• How MRT Timing Works Using Mathematics — the system-wide pillar.
• How MRT Braking Works Using Mathematics — stopping distance and regenerative braking.
• How MRT Headway Works Using Mathematics — train spacing and throughput.
• How MRT Station Dwell Time Works Using Mathematics — queues, doors and platform flow.
• How MRT Delays Propagate and Recover Using Mathematics — instability, regulation and recovery.
• How an MRT Timetable Is Built Using Mathematics — demand, train cycles and robustness.
• How MRT Energy Use Is Optimised Using Mathematics — traction, coasting and regenerative energy.
• How MRT Passenger Capacity Is Calculated Using Mathematics — vehicle, station and delivered capacity.
• How MRT Routes and Transfers Are Optimised Using Mathematics — graph theory, transfer cost and route choice.

The next natural leg is How MRT Predictive Maintenance Works Using Mathematics: failure probability, condition monitoring, anomaly detection, remaining useful life, inspection intervals, false alarms and how maintenance tries to act before reliability becomes a disruption.

Sources and Further Reading

• LTA, SMRT and SBS Transit — Rail Reliability Taskforce recommendations, 13 February 2026
• LTA — Exercise GREYHOUND 2026 and rail-incident operational readiness
• LTA — Circle Line Stage 6, opened 12 July 2026
• LTA — CCL6 connectivity, resilience and travel-time benefits
• LTA — Downtown Line extensions and alternative travel options
• LTA — Jurong Region Line, passenger redistribution and network resilience
• Ministry of Transport — September 2024 EWL disruption response and alternative transport
• npj Sustainable Mobility and Transport — Resilience of urban metro rail networks globally, 2026
• Communications Physics — Topological assessment of recoverability in public transport networks
• Nature Communications — Interconnectedness and multimodal public-transport resilience
• Urban Rail Transit — Passenger-flow distribution under metro interruption
• Reliability Engineering & System Safety — Dynamic passenger-level resilience assessment of metro networks
• International Journal of Disaster Risk Reduction — Passenger-flow and travel-efficiency resilience modelling
• Transportation Research Interdisciplinary Perspectives — Bus–metro multimodal resilience assessment
• Transportation Research Part E — Enhancing metro resilience through localised bus integration
• Frontiers of Engineering Management — Urban rail transit disruption management