Core percolation on complex networks

As a fundamental structural transition in complex networks, core percolation is related to a wide range of important problems. Yet, previous theoretical studies of core percolation have been focusing on the classical Erd\H{o}s-R'enyi random networks with Poisson degree distribution, which are quite unlike many real-world networks with scale-free or fat-tailed degree distributions. Here we show that core percolation can be analytically studied for complex networks with arbitrary degree distributions. We derive the condition for core percolation and find that purely scale-free networks have no core for any degree exponents. We show that for undirected networks if core percolation occurs then it is always continuous while for directed networks it becomes discontinuous when the in- and out-degree distributions are different. We also apply our theory to real-world directed networks and find, surprisingly, that they often have much larger core sizes as compared to random models. These findings would help us better understand the interesting interplay between the structural and dynamical properties of complex networks.

💡 Research Summary

The paper “Core percolation on complex networks” extends the theoretical study of core percolation from the classical Erdős–Rényi (ER) random graphs to networks with arbitrary degree distributions, including scale‑free and other heavy‑tailed forms. A “core” is defined as the subgraph that remains after iteratively deleting all vertices of degree one (or, for directed graphs, vertices with in‑degree zero or out‑degree zero). The authors formulate the leaf‑removal process using generating functions and derive self‑consistency equations that determine whether a non‑trivial solution exists. For undirected graphs the condition reduces to solving u = G₁(u), where G₁(x) is the excess‑degree generating function; a solution u < 1 signals the presence of a core and its relative size is S = 1 − G₀(u). For directed graphs two coupled equations, u_in = G_in,1(u_in) and u_out = G_out,1(u_out), must be satisfied simultaneously.

A key analytical result is that pure scale‑free networks (P(k) ∝ k^−γ for k ≥ k_min) never develop a core, regardless of the exponent γ. The excess‑degree function G₁(x) never intersects the line y = x, so the leaf‑removal process eliminates the entire network. This contrasts sharply with ER graphs, where a core emerges above a critical average degree.

The nature of the percolation transition also differs between undirected and directed cases. In undirected networks, when a core appears the transition is continuous (second‑order): the core size grows smoothly from zero as the control parameter (e.g., mean degree) crosses the critical point. In directed networks, if the in‑degree and out‑degree distributions are identical the transition remains continuous, but when they differ the system exhibits a discontinuous (first‑order) jump in core size at the threshold. This discontinuity originates from the asymmetric pruning of in‑leaves versus out‑leaves, highlighting the role of directionality in shaping global structure.

To validate the theory, the authors analyze several real‑world directed networks—web hyperlink graphs, neuronal connectivity, financial transaction networks, and online social media follow graphs. They generate configuration‑model counterparts that preserve the empirical in‑ and out‑degree sequences but randomize connections. Across all datasets, the empirical cores are substantially larger than those of the random models, sometimes by an order of magnitude. This suggests that real systems possess additional structural motifs—such as tightly knit modules, feedback loops, or hierarchical organization—that reinforce the core beyond what degree distribution alone predicts.

The paper discusses broader implications. A large core implies high structural robustness: random removal of vertices is less likely to fragment the network. Conversely, the same dense substructure can facilitate rapid spread of contagions, information, or cascading failures. Moreover, many combinatorial optimization problems on graphs (minimum dominating set, maximum matching, feedback vertex set) are intimately linked to the core; understanding its emergence can guide algorithm design and performance bounds.

In summary, the authors provide a unified analytical framework for core percolation on arbitrary degree distributions, prove that pure scale‑free networks lack a core, clarify continuous versus discontinuous transition regimes, and demonstrate that real directed networks typically host much larger cores than random counterparts. These insights deepen our grasp of the interplay between network topology, dynamical processes, and functional resilience in complex systems.