A Time-Space Network is a directed graph in which each node represents a location at a specific discrete time step, and each arc represents either a movement (travel arc) or a waiting action (holding arc) between consecutive time periods. A travel arc from node (A, t) to node (B, t+τ) says: "a unit can move from location A to location B in τ time steps." A holding arc from (A, t) to (A, t+1) says: "a unit can wait at location A for one time period."
The critical property is flow conservation at every node: every unit that arrives at a (location, time) node must either depart on a travel arc or depart on a holding arc. Nothing disappears. This single constraint — inherited from classical network flow theory — encodes that a resource can only be in one place at one time, without any additional scheduling constraints being written explicitly.
The payoff: a problem that was originally expressed as a scheduling or routing problem over time is now a minimum-cost flow or shortest-path problem on a static graph. That transformation imports decades of highly efficient exact algorithms — Ford-Fulkerson, successive shortest paths, network simplex — and makes problems that would otherwise require complex integer programming formulations tractable as continuous LP relaxations with integer structure arising naturally.
The problem: An airline has three airports (A, B, C) and a planning horizon of four time periods. It must assign specific aircraft (identified by tail number) to flight legs while satisfying maintenance windows and minimising expected delay propagation. Each aircraft must be in exactly one place at any time.
Time-Space Network construction: Create a node for every (airport, time) combination: A@t1, A@t2, A@t3, A@t4, B@t1, B@t2, B@t3, B@t4, C@t1, C@t2, C@t3, C@t4. Add a travel arc from A@t1 to B@t2 for every scheduled flight A→B departing at time 1. Add holding arcs (A@t1→A@t2, A@t2→A@t3, etc.) for each airport to represent aircraft waiting on the ground.
The formulation: Route a unit of aircraft flow from each aircraft's starting node to its ending node through the time-space graph, minimising total cost. Flow conservation at every node ensures no aircraft is in two places simultaneously. Constraints on travel arcs encode flight legality; constraints on holding arcs can encode maintenance requirements.
Why this matters: The Air France tail-assignment paper covered in the 3 May 2026 DO Radar issue (arXiv:2602.10866) uses exactly this structure, extending it stochastically: the pricing subproblem in their column generation is a stochastic shortest path over the time-space graph, explicitly accounting for delay propagation along travel arcs.
Confusion 1: Time-Space Network as visualisation, not formulation. The most common misread is treating a Time-Space Network diagram as a Gantt chart variant — a picture of a schedule. It is not. The graph is a mathematical object with flow-conservation semantics. Those semantics are what allow you to solve the scheduling problem using network flow solvers rather than bespoke combinatorial algorithms. A diagram without the conservation constraints is not a Time-Space Network in any operational sense; it is a picture of one.
Confusion 2: Time-Space Networks require a fixed time grid. Practitioners sometimes assume the method only applies when all events occur on a regular discrete clock (every 15 minutes, every hour). In fact, the time axis need only be discretised to the resolution of the problem's decision points — flight departures, delivery windows, maintenance check-ins. Irregular grids are common and are handled identically: each (location, event-time) pair becomes a node. The number of nodes scales with the number of events, not with calendar time.
Confusion 3: Holding arcs are optional. Practitioners sometimes omit holding arcs for "simplicity," assuming they can add resting constraints separately. This breaks the conservation structure. If a unit arrives at a node and no holding arc departs from that node, there is nowhere for the unit to "go" between events — the network becomes infeasible or requires artificial slack arcs. Holding arcs are not cosmetic; they are the mechanism by which the network encodes that resources can persist in place across time steps.
A Time-Space Network converts scheduling into flow by making time a spatial dimension: nodes are (place, time) pairs, arcs are moves or waits, and flow conservation enforces that each resource is in exactly one place at any moment.