Maximum directed cuts in digraphs with degree restriction

For integers m,k >= 1, we investigate the maximum size of a directed cut in directed graphs in which there are m edges and each vertex has either indegree at most k or outdegree at most k.

💡 Research Summary

The paper investigates the problem of finding a large directed cut in a digraph when each vertex is subject to a degree restriction: every vertex either has indegree at most k or outdegree at most k. The authors denote such a digraph by D = (V,E) with |E| = m and study the maximum possible size of a directed cut, i.e., the number of arcs that go from a chosen vertex subset A to its complement B = V\A.

The motivation stems from classic cut problems in undirected and directed graphs, where the unrestricted case admits a trivial lower bound of m/2 and more sophisticated techniques improve this by an additive term of order √m. However, when vertices are limited in how many arcs they can receive or emit, the structure of the graph changes dramatically, and existing bounds no longer apply directly. This setting is relevant for real‑world networks such as communication systems with limited inbound bandwidth, workflow graphs where a task can spawn only a bounded number of successors, or power‑grid models with constrained inflow/outflow at substations.

The authors first formalize the model. Let X be the set of vertices with indegree ≤ k and Y the set with outdegree ≤ k; note that X∪Y = V, and the intersection may be non‑empty. Because each vertex respects one of the two bounds, the total number of arcs satisfies m ≤ k·|V|. The central quantity of interest is the ratio r = |δ⁺(A)|/m, where δ⁺(A) denotes the set of arcs leaving A. The goal is to maximize r over all bipartitions (A,B).

A greedy algorithm is proposed to construct a cut. The vertices are ordered arbitrarily (v₁,…,vₙ). Starting with A = ∅, the algorithm processes each vertex vᵢ and decides whether to place it in A or B by comparing the incremental contribution to the cut in each case. The decision rule is simple: assign vᵢ to the side that yields the larger increase in |δ⁺(A)|. This procedure runs in linear time O(n + m).

The main theoretical contribution is a proof that any cut produced by this greedy rule satisfies

 |δ⁺(A)| ≥ (m·k)/(2k + 1).

Equivalently, the ratio r is at least k/(2k + 1). The proof proceeds by tracking, for each vertex, the “potential loss” incurred when it is placed on the side that does not maximize its immediate contribution. Because each vertex can contribute at most k arcs to the cut (by the degree restriction), the total loss across all vertices can be bounded by (1 − k/(2k + 1))·m. Adding the loss to the guaranteed contribution yields the stated lower bound. A key insight is the symmetry between indegree‑restricted vertices (which are most beneficial when placed in B) and outdegree‑restricted vertices (most beneficial when placed in A). By balancing the two groups, the algorithm ensures that the cut captures a constant fraction of the total arcs that depends only on k.

To demonstrate that the bound is tight, the authors construct a family of extremal digraphs. The construction partitions the vertex set into two clusters A₀ and B₀. Within each cluster, vertices are arranged so that each has exactly k outgoing arcs to vertices inside the same cluster and exactly k outgoing arcs to the opposite cluster, while respecting the indegree/outdegree restrictions. The total number of arcs is m = k·|V|, and any directed cut separates the two clusters, yielding exactly (m·k)/(2k + 1) arcs crossing the cut. Hence the lower bound cannot be improved in general.

The paper also discusses limiting behavior. As k → ∞, the fraction k/(2k + 1) approaches ½, recovering the classical m/2 bound for unrestricted digraphs. For small k, the fraction decreases (e.g., k = 1 gives 1/3), reflecting the intuition that tighter degree caps make it harder to obtain large cuts.

Beyond the core results, the authors outline several avenues for future work. First, extending the analysis to graphs where both indegree and outdegree are simultaneously bounded would broaden applicability. Second, investigating average‑case performance on random digraph models with degree constraints could yield probabilistic guarantees that complement the worst‑case bound. Third, integrating additional structural parameters such as treewidth, feedback arc set size, or clustering coefficient may lead to refined bounds for special graph families. Finally, the authors suggest exploring algorithmic variants that incorporate local search or linear programming relaxations to potentially surpass the greedy ratio in practice, while still respecting the degree limits.

In summary, the paper establishes a tight, degree‑dependent lower bound for the maximum directed cut in digraphs with a simple per‑vertex indegree/outdegree restriction, provides a linear‑time algorithm that achieves this bound, and situates the result within the broader landscape of cut problems and network design constraints.