The focal point

An introduction to the bidirectional mandatory core

Published: 16 Aug 2026

Have you ever been to the airport? You probably arrived there by cab or public transport. In any case, here is an observation.

Say three people are trying to catch the same flight. One arrives by taxi. The taxi leaves him outside the departures hall, where he joins a queue, passes through security and walks toward the concourse. The second arrives by train. He comes up from a station beneath the terminal, enters from a different level and reaches the concourse without ever seeing the departures hall. The third is already inside the airport. Her first flight has just landed, and this airport is only a connection. She follows the transfer signs through the airside corridors, never passing through the train station, the taxi rank or the main security checkpoint.

Their routes are almost entirely different. They begin in different places, cross different parts of the building and are governed by different procedures. For most of the journey, there may be no location visited by all three.

Then the routes converge. All three passengers are booked on the same departing flight. However they reached the airport, each must eventually present a boarding pass at the same gate and pass from the terminal to the aircraft.

The gate occupies a special position in the structure of their journeys. It may not be the busiest place in the airport. It may not be near the geographical centre of the terminal. Another checkpoint may process more people during the day. None of that changes the role the gate plays for these passengers: every successful route from their starting regions to their aircraft passes through it.

Imagine drawing the airport as a network. Rooms, entrances and checkpoints become points. A directed connection between two points means that a passenger is permitted to move from one to the other. The details of the floor plan begin to fade, leaving only the possible routes.

taxi drop-off ───────┐
                     │
train station ───────┼──▶ boarding gate ──▶ aircraft
                     │
transfer concourse ──┘

If the gate becomes inaccessible, none of the three can board. Blocking a point on the train passenger's route stops only the passenger arriving by train. Blocking a point on the transfer route affects only the connecting passenger. The gate is different: it is shared by every route that completes the journey.

The airport gives us a simple form of structural necessity. A point becomes important because of where it sits among the possible routes.

A network that never stops #

Now consider a system with no final boarding call and no last passenger through the door: the human cardiovascular system.

At a broad level, circulation consists of two linked circuits. The systemic circuit carries oxygen-rich blood from the heart to the tissues of the body and returns oxygen-poor blood. The pulmonary circuit carries that returning blood from the heart to the lungs and brings oxygen-rich blood back. These circuits are connected through the heart, which continually drives the movement around them.^[1]

The organs of the body receive blood through a vast branching network. The brain, kidneys, muscles and digestive organs do not form one simple queue. They are served through many branches operating alongside one another. At a high level, the organs are arranged in parallel, while the systemic and pulmonary circulations are joined in series through the heart.^[2]

A deliberately simplified picture looks like this:

body tissues ──▶ heart ──▶ lungs
body tissues ◀── heart ◀── lungs

Blood returning from the body passes through the heart on its way to the lungs. Blood returning from the lungs passes through the heart on its way back to the body. The two journeys have opposite directions, yet the same organ is unavoidable in both.

This example holds richer information than the airport one. The passengers have a destination and mostly move toward it. Circulation is a continuing cycle. The significant point must support movement in both directions.

The word point deserves attention. A boarding gate can be treated as one location on an airport map. The heart is an organ with internal structure: four chambers, valves and distinct right and left sides. Blood returning from the body travels through the right side of the heart toward the lungs; blood returning from the lungs travels through the left side toward the body.^[1]

If our network represents whole organs, the heart appears as one point and occupies both directions of the cycle. If we redraw the network at the level of individual chambers, that one point unfolds into several points with different roles.

Both pictures are useful. They answer questions at different scales. The organ-level picture asks which major component connects pulmonary and systemic circulation. The chamber-level picture asks how blood moves inside that component.

The same choice was hidden in the airport example. We treated the boarding gate as one point. A more detailed model could separate the desk, scanner, door, jet bridge and aircraft door. A network never arrives without interpretation. Someone must decide which distinctions matter.

The common pattern #

An airport and a cardiovascular system look like very different things. One transports people according to schedules and permissions. The other transports blood continuously through vessels and organs. One is engineered and frequently reconfigured. The other is biological and self-regulating.

Their shared pattern becomes visible when we ignore most of those details and look only at possible movement. Both systems contain many starting places and many alternative routes. Variety dominates the outer parts of each network. Taxi, train and transfer passengers approach through different airport corridors. Blood travels through branching vessels to tissues distributed throughout the body.

Yet variety does not imply that every part of a journey is replaceable. Alternative routes can remain constrained by a common passage. In the airport, the relevant journeys converge at the boarding gate. In the organ-level circulation model, movement between the body and lungs passes through the heart.

This gives us a particular meaning of important. We often call a place important because it is busy, central, large or close to many other places. Each quality can be measured, and each answers a useful question. Here we want a different meaning:

A point may be structurally indispensable for a particular movement.

That role always depends on a question. The boarding gate is indispensable for reaching one particular flight. The heart is indispensable in our simplified organ-level model of movement between systemic tissues and the lungs. Change the origin, destination, permitted routes or level of detail, and the answer may change.

Structural importance therefore belongs to a system together with a task. We choose where movement begins, where it must end and what counts as an available route. Only then can we ask what the structure forces every successful solution to share.

Empty route #

The gate and the organ-level heart are the easiest kind of bottleneck: one visible point lies on every relevant route. Real networks are often less obliging.

Consider a small network joining regions A and B:

             ┌──▶ h ──────────▶─┐
A ───────────┤                  ├──▶ B
             └──▶ x ──▶ y ─────▶┘

The network offers two independent routes. The upper route passes through h. The lower route passes through x and then y.

Suppose we want to stop all movement from A to B by making as few internal points unavailable as possible. Closing h stops the upper route but leaves the lower route open. Closing x or y stops the lower route but leaves the upper route open. No single closure is enough.

Two closures are sufficient, and there are two smallest choices:

{h, x} or {h, y}.

The choices disagree about the lower route. Either x or y will do. They agree about h, because the upper route offers no alternative place to intervene.

This agreement is easy to miss if we inspect only the paths. The point h does not occur on the lower path, so it is not a place through which everything travels. Nevertheless, it occurs in every smallest set that blocks everything.

A mandatory point need not appear on every route. It can instead appear in every optimal way of blocking all routes.

This is the conceptual step behind the mandatory core. Rather than intersecting the routes, we intersect the smallest complete separators.

From routes to cuts #

In graph language, the locations, components or states are vertices. Directed edges describe permitted movement. We choose two disjoint regions, A and B.

An A-to-B vertex cut is a collection of internal vertices whose removal destroys every directed path from A to B. The vertices inside A and B are protected: the question is how to separate the regions by intervening somewhere between them.

A cut can contain unnecessary additions. If {h, x} already blocks every route, then {h, x, y} also blocks every route, but the extra y tells us nothing about what the structure requires. For that reason, we focus on cuts of the smallest possible size.

The forward mandatory core M(A, B) is the intersection of all minimum A-to-B vertex cuts. In the two-route example,

M(A, B) = {h}.

The definition does not ask which minimum cut an algorithm happens to return. It asks what survives disagreement among all equally small cuts.

Direction matters #

An escalator may carry passengers toward a concourse without carrying them back. A controlled door may allow exit but forbid entry. A request and its response may traverse different services. In circulation, the right and left sides of the heart play different roles in the two directions.

We must therefore compute two cores:

M(A, B) from A to B;
M(B, A) from B to A.

Their intersection is the bidirectional mandatory core:

M↔(A, B) = M(A, B) ∩ M(B, A).

It contains only the vertices present in every minimum separator in both directions. If the intersection is {h}, then the graph has singled out one vertex without using distances, traffic volumes or a centrality ranking. Every optimal separator in either direction agrees on h.

The result can also be empty or contain several vertices. An empty bidirectional core means only that no vertex belongs to every minimum separator in both directions. It does not by itself imply resilience or substitutable optimal cuts: each direction can have a unique minimum cut while the two directional cuts are disjoint, and a direction with no path has an empty core by convention. A larger core identifies a mandatory subsystem rather than a unique point.

Precise structure #

A singleton bidirectional core is a precise structural certificate. It says:

One vertex belongs to every minimum separator in both directions.

It does not say that the vertex is the busiest point, the geographical centre, the highest-scoring centrality vertex or necessarily a single point of failure. If the smallest cut contains three vertices, removing one mandatory member may still leave routes available. That member is forced into every smallest complete disruption, but it may not stop the network alone.

Nor does mandatory membership prove that every route passes through the vertex. The point h in our two-route network is the counterexample. If a security gateway or safety interlock is meant to inspect every operation, route intersection—not mandatory-cut membership—is the property that must be verified.

The two questions are complementary:

Which points does every route traverse?
Which points does every smallest complete separator contain?

The first concerns compulsory passage. The second concerns agreement among optimal interventions.

Returning to the gate #

The airport gate represents the simplest case. Separate approaches converge on one final passage to the aircraft. The same point lies on every successful route and forms the unique smallest separator in the simplified graph.

The cardiovascular example adds direction and scale. At organ level, the heart is the common component joining systemic and pulmonary circulation in both directions. At chamber level, the vertex unfolds and the singleton conclusion may change.

The network with h, x and y adds the decisive distinction. The point h is absent from one complete route yet present in every smallest complete separator. It shows why the mandatory core cannot be replaced by the simpler question, “What lies on every path?”

Across all three pictures, the method takes the same shape:

Choose the regions whose connection matters.
Decide which movements the graph permits.
Find the smallest complete separators.
Ask what all those optimal separators share.
Reverse the direction and ask again.

Many networks have no unique centre, and many legitimate notions of importance can compete. The bidirectional mandatory core makes a narrower promise. When every minimum separator in both directions agrees on one vertex, the network has made one structural choice unavoidable.