Abstract flows over time: A first step towards solving dynamic packing problems
Flows over time generalize classical network flows by introducing a notion of time. Each arc is equipped with a transit time that specifies how long flow takes to traverse it, while flow rates may vary over time within the given edge capacities. In this paper, we extend this concept of a dynamic optimization problem to the more general setting of abstract flows. In this model, the underlying network is replaced by an abstract system of linearly ordered sets, called “paths” satisfying a simple switching property: Whenever two paths P and Q intersect, there must be another path that is contained in the beginning of P and the end of Q. We show that a maximum abstract flow over time can be obtained by solving a weighted abstract flow problem and constructing a temporally repeated flow from its solution. In the course of the proof, we also show that the relatively modest switching property of abstract networks already captures many essential properties of classical networks.
💡 Research Summary
The paper extends the well‑studied “flows over time” paradigm, which augments classical network flow models with a temporal dimension, to the much broader setting of abstract flows. In an abstract flow model the underlying graph is replaced by a family of linearly ordered sets called “paths”. The only structural requirement is the switching property: whenever two paths P and Q intersect at an element e, there must exist a third path that consists of the initial segment of P up to e followed by the terminal segment of Q after e. This property captures the essence of path recombination in ordinary networks without referring to vertices or arcs.
Each element of a path is assigned a transit time τ(e) > 0, and a flow is described by a time‑dependent function f_P(t) that specifies how much flow enters path P at time t. The usual capacity constraints (the instantaneous inflow on any element cannot exceed its capacity c(e)) and flow‑conservation constraints (the net flow entering and leaving any element at any time must balance) are imposed in the temporal domain. The objective is to maximize the total amount of flow that can be sent from a designated source to a sink over a given time horizon.
The authors’ main contribution is a two‑step reduction that shows how to obtain an optimal dynamic abstract flow from a static weighted abstract flow. First, they define a static problem in which each element e receives a weight w(e)=1/τ(e). Solving this weighted abstract flow yields a set of path flows x_P that satisfy the static capacity constraints and maximize the weighted sum Σ_e w(e)·flow_on_e. Second, they construct a temporally repeated flow: the static solution x_P is dispatched repeatedly every τ(P) time units (where τ(P) is the sum of transit times along P). Because the weight of an element is the reciprocal of its transit time, the repeated schedule respects the original temporal capacities, and the total amount of flow delivered over the horizon equals the value of the static weighted solution.
The paper proves two key theorems. The first shows that any optimal solution of the weighted static problem can be transformed into a feasible temporally repeated flow for the dynamic problem. The second establishes that this temporally repeated flow is in fact optimal for the dynamic abstract flow, i.e., no other time‑varying schedule can achieve a larger total throughput. The proofs rely heavily on the switching property: it guarantees that whenever two dynamic paths intersect, one can reroute flow along a newly formed path without violating capacities, thereby mimicking the cut‑flow duality that underlies the classic max‑flow/min‑cut theorem. The authors also construct a corresponding abstract cut (a collection of elements whose total capacity bounds any feasible flow) and show that its capacity equals the value of the temporally repeated flow, completing the duality argument.
Beyond the theoretical result, the authors argue that the modest switching property already encapsulates many structural features of ordinary networks—conservation, capacity, and cut concepts—without needing an explicit graph representation. This observation suggests that a wide class of dynamic packing problems (e.g., time‑dependent knapsack, scheduling with release times, or perishable inventory routing) can be modeled within this abstract framework.
In the concluding discussion, the paper outlines several avenues for future work: (i) enriching the abstract model with additional properties such as hierarchy or precedence constraints, (ii) incorporating stochastic or uncertain transit times, (iii) developing combinatorial algorithms that exploit the structure of the switching property for faster computation, and (iv) applying the theory to real‑world logistics and communication systems where dynamic packing decisions are critical. By bridging the gap between dynamic network flows and abstract flow theory, the work provides a foundational step toward solving complex time‑sensitive packing problems in a unified, mathematically rigorous manner.