The effect of network structure on phase transitions in queuing networks

Recently, De Martino et al have presented a general framework for the study of transportation phenomena on complex networks. One of their most significant achievements was a deeper understanding of the phase transition from the uncongested to the congested phase at a critical traffic load. In this paper, we also study phase transition in transportation networks using a discrete time random walk model. Our aim is to establish a direct connection between the structure of the graph and the value of the critical traffic load. Applying spectral graph theory, we show that the original results of De Martino et al showing that the critical loading depends only on the degree sequence of the graph – suggesting that different graphs with the same degree sequence have the same critical loading if all other circumstances are fixed – is valid only if the graph is dense enough. For sparse graphs, higher order corrections, related to the local structure of the network, appear.

💡 Research Summary

The paper investigates the onset of congestion in transportation networks by employing a discrete‑time random‑walk queueing model. Building on the framework introduced by De Martino et al., which claimed that the critical traffic load (λ_c) depends solely on the degree sequence of the underlying graph, the authors aim to clarify the conditions under which this claim holds and to identify additional structural factors that influence λ_c in sparse networks.

First, the authors formalize the model: each node hosts an infinite‑capacity queue, and at each time step a packet at node i selects one of its k_i neighbors uniformly at random for forwarding. New packets are injected into the network at a global rate λ, which is assumed to be homogeneous across nodes. The system is said to be in the uncongested phase when the average queue length remains bounded; otherwise, it enters the congested phase, characterized by diverging queue lengths.

To connect λ_c with network topology, the authors apply spectral graph theory. They express the steady‑state queue dynamics in terms of the adjacency matrix A and its eigenvalue spectrum. For dense graphs—where the average degree ⟨k⟩ grows faster than log N—the largest eigenvalue λ_max of A is tightly concentrated around ⟨k⟩, and the classical result λ_c ≈ ⟨k⟩/⟨k²⟩ (or equivalently a function of the degree distribution alone) emerges naturally. This reproduces the De Martino et al. finding and confirms its validity in the dense regime.

The novelty of the present work lies in the analysis of sparse graphs, where ⟨k⟩ is O(1) and the spectrum of A deviates significantly from the dense‑graph approximation. The authors demonstrate that local motifs—such as triangles, short cycles, and the overall clustering coefficient—perturb λ_max and consequently shift λ_c. By performing a first‑order perturbation expansion of the eigenvalues, they derive a corrected critical load formula:

λ_c ≈ (⟨k⟩/⟨k²⟩) ·