K-core Percolation Analysis

• K-core percolation analysis is a framework that defines maximal subgraphs where every vertex meets a minimum degree, highlighting network resilience and vulnerability.
• It employs analytical techniques such as generating functions, message-passing, and branching processes to detail phase diagrams and capture hybrid transitions.
• The approach is applied to diverse systems, from material rigidity to infrastructure networks, with extensions to multiplex, hypergraph, and spatial configurations.

K-core percolation analysis is a central framework in network science to characterize the emergence and stability of maximally connected subgraphs subject to local degree constraints. The k-core of a network is defined as the maximal subgraph in which all vertices have degree at least k, obtained via iterative pruning of nodes failing to meet this threshold. K-core percolation thus generalizes standard connectivity percolation (the case k=1) and is fundamental for understanding resilience, rigidity, and functional breakdown in complex systems. The phenomenon exhibits rich critical behavior, including mixed-order (hybrid) transitions, tricritical points in heterogeneous systems, and universal avalanche statistics. Analytical methods, such as generating-function self-consistency, message-passing equations, and branching-process theory, enable detailed characterization of phase diagrams and scaling laws, while extensions encompass multiplex, spatial, hypergraph, and heterogeneous networks.

1. Formalism of K-core Percolation and Basic Equations

Given a graph $G=(V,E)$, its k-core is the unique maximal subgraph in which every node has degree at least k. For an ensemble of random (configuration model) networks with degree distribution $P(q)$, the k-core percolation transition is characterized analytically via recursive self-consistency relations. The standard approach introduces $u$ as the probability that a randomly chosen edge leads to a vertex not in the k-core. For homogeneous thresholds and random initial damage (fraction $1-p$ of vertices removed), the foundational equations are: $$u = 1 - p + p\, \sum_{q=k-1}^{\infty} Q(q)\sum_{m=0}^{k-2}\binom{q}{m}u^{q-m}(1-u)^m,$$ where $Q(q) = [(q+1)P(q+1)]/\langle q \rangle$, and

$$P_\infty = p\left[1 - \sum_{q=0}^{k-1}P(q)u^q\right],$$

with $P_\infty$ the fraction of nodes in the giant k-core (Xue et al., 2023, Zhu et al., 2017, Panduranga et al., 2017).

The percolation threshold $p_c$ is located by the bifurcation (tangency) condition: $$1 = p_c\,\Phi'(f_c),\quad f_c = \Phi(f_c)/\Phi'(f_c),$$ where $P(q)$0 is the corresponding edge-rooted generating function (Yuan et al., 2016).

For Erdős–Rényi (ER) graphs, these reduce to explicit formulas utilizing Poisson degree distributions. In dense-graph limits, multi-type branching-process/graphon techniques generalize the analysis to arbitrary sequences converging in the cut-metric (Bayraktar et al., 2020).

2. Mixed-Order Transitions, Scaling, and Universality

For $P(q)$1 (giant component) and $P(q)$2, the k-core percolation transition is continuous (second order), while for $P(q)$3 the transition is hybrid: a discontinuous (first-order) jump in the giant k-core size is accompanied by critical singularity above threshold, $P(q)$4 (Zhu et al., 2017, Lee et al., 2016, Morone et al., 2018).

Critical exponents for $P(q)$5 on random graphs (mean-field) are:

• Order parameter exponent: $P(q)$6;
• Susceptibility exponent: $P(q)$7;
• Correlation-size exponent: $P(q)$8;
• Avalanche-size exponent distribution: $P(q)$9, with cutoff $u$0 (Lee et al., 2016).

Finite-size scaling forms and hyperscaling relations $u$1 hold within each observable class; the two coupled exponent sets (order parameter, avalanche statistics) reflect the hybrid nature (Lee et al., 2016). The appearance of a finite k-core at $u$2 is a macroscopic instability, interpretable in applications ranging from jamming transitions in granular matter (Morone et al., 2018) to abrupt infrastructure collapse (Xue et al., 2023).

3. Heterogeneous Thresholds and Tricriticality

When node thresholds are heterogeneous ($u$3 varying by node), the critical phenomena become even richer (Cellai et al., 2012, Cellai et al., 2011, Cellai et al., 2013):

• Binary mixtures: For a fraction $u$4 of nodes with $u$5 and $u$6 with $u$7, the phase diagram contains both continuous and hybrid transitions, which may coalesce at a tricritical point (TCP) for special $u$8. For the $u$9 mixture, the TCP occurs at $1-p$0, with order-parameter exponent interpolating from $1-p$1 (continuous) to $1-p$2 (hybrid), and $1-p$3 at the tricritical point (Cellai et al., 2011, Cellai et al., 2012).
• Ternary mixtures: Systems with three threshold types exhibit swallowtail (A₄) singularities, cusp (A₃), and hybrid (A₂) points; order-parameter exponents at these singularities are respectively $1-p$4, $1-p$5, and $1-p$6 (Cellai et al., 2013).

These multicritical phenomena directly parallel higher-order singularities in mode-coupling theory of glass transitions and facilitated spin models, underscoring a universality in the structure of constraint-percolation criticality (Cellai et al., 2012, Cellai et al., 2013).

4. Extensions: Multiplex, Hypergraph, and Spatial k-core Percolation

Multiplex Networks

In multiplex systems (multiple edge types/layers), the k-core is generalized to a vector threshold $1-p$7, and the percolation analysis requires coupled self-consistency equations for edge-rooted survival probabilities in each layer: $1-p$8 The nature of phase transitions mirrors the single-layer case: hybrid for maxima with at least one $1-p$9, continuous for $$u = 1 - p + p\, \sum_{q=k-1}^{\infty} Q(q)\sum_{m=0}^{k-2}\binom{q}{m}u^{q-m}(1-u)^m,$$0 (Azimi-Tafreshi et al., 2014). Applications include multiplex air transportation and interdependent infrastructures (Zheng et al., 2021).

Hypergraph k-core Percolation

In hypergraphs, the $$u = 1 - p + p\, \sum_{q=k-1}^{\infty} Q(q)\sum_{m=0}^{k-2}\binom{q}{m}u^{q-m}(1-u)^m,$$1-core is the maximal induced subhypergraph where all vertices have degree at least $$u = 1 - p + p\, \sum_{q=k-1}^{\infty} Q(q)\sum_{m=0}^{k-2}\binom{q}{m}u^{q-m}(1-u)^m,$$2 and hyperedges have cardinality at least $$u = 1 - p + p\, \sum_{q=k-1}^{\infty} Q(q)\sum_{m=0}^{k-2}\binom{q}{m}u^{q-m}(1-u)^m,$$3. The combinatorial structure and percolation phenomena differ sharply from factor-graph representations:

• First-neighbor and second-neighbor pruning lead to distinct threshold diagrams.
• For random Poissonian hypergraphs, $$u = 1 - p + p\, \sum_{q=k-1}^{\infty} Q(q)\sum_{m=0}^{k-2}\binom{q}{m}u^{q-m}(1-u)^m,$$4-core exhibits a tricritical point, and for $$u = 1 - p + p\, \sum_{q=k-1}^{\infty} Q(q)\sum_{m=0}^{k-2}\binom{q}{m}u^{q-m}(1-u)^m,$$5, only hybrid transitions occur (Bianconi et al., 2023).

Spatial and Lattice k-core Percolation

Introducing spatial embedding or lattice structure (e.g., Euclidean/hyperbolic lattices) modifies critical thresholds and universality. In spatial two-dimensional networks with a tunable link-length parameter $$u = 1 - p + p\, \sum_{q=k-1}^{\infty} Q(q)\sum_{m=0}^{k-2}\binom{q}{m}u^{q-m}(1-u)^m,$$6, one observes:

• Four transition regimes: continuous (lattice-like), metastable nucleation-driven first-order, boundary mixed-order, and mean-field hybrid for $$u = 1 - p + p\, \sum_{q=k-1}^{\infty} Q(q)\sum_{m=0}^{k-2}\binom{q}{m}u^{q-m}(1-u)^m,$$7;
• Emergence of a metastable window with extreme vulnerability: local (finite) seed perturbations induce global collapse, unlike in standard percolation (Xue et al., 2023);
• On hyperbolic lattices (e.g., $$u = 1 - p + p\, \sum_{q=k-1}^{\infty} Q(q)\sum_{m=0}^{k-2}\binom{q}{m}u^{q-m}(1-u)^m,$$8), k-core percolation is continuous (no hybrid jump) even at $$u = 1 - p + p\, \sum_{q=k-1}^{\infty} Q(q)\sum_{m=0}^{k-2}\binom{q}{m}u^{q-m}(1-u)^m,$$9, attributed to the abundance of loops at all scales (Lopez et al., 2015).

5. Applications, Algorithmics, and Generalizations

K-core percolation analysis underpins models of:

• Mechanical and hydraulic rigidity in granular materials and suspensions, where the k-core emergence corresponds to the onset of force or stress-bearing backbone (k=3 in jamming, shear-thickening transitions) (Sedes et al., 2021, Morone et al., 2018);
• Network resilience under targeted, localized, or random attacks, with mappings to equivalent random-attack ensembles enabling unified analytical treatment (Yuan et al., 2016);
• Real-world complex networks (power grids, transportation, brain networks), in which detailed k-core statistics provide enhanced prediction of robustness compared to degree-based models (Hébert-Dufresne et al., 2013).

Algorithms for k-core extraction employ recursive degree-threshold pruning and generalize naturally to multiplex and hypergraph cases (Azimi-Tafreshi et al., 2014, Bianconi et al., 2023). Hard-core random network (HRN) models using coreness-degree matrices provide a maximally random ensemble conditioned on prescribed k-core structure (Hébert-Dufresne et al., 2013).

6. Critical Fluctuations, Avalanches, and Relation to Bootstrap Percolation

At hybrid transitions, the order-parameter fluctuations and finite avalanche statistics decouple but manifest universal scaling:

• Mean finite avalanche size diverges as $Q(q) = [(q+1)P(q+1)]/\langle q \rangle$0, with size distribution exponent $Q(q) = [(q+1)P(q+1)]/\langle q \rangle$1.
• Scaling relations link avalanche and order-parameter exponents: e.g., $Q(q) = [(q+1)P(q+1)]/\langle q \rangle$2 with $Q(q) = [(q+1)P(q+1)]/\langle q \rangle$3 for $Q(q) = [(q+1)P(q+1)]/\langle q \rangle$4 (Lee et al., 2016).
• K-core percolation can be exactly mapped to complementary heterogeneous bootstrap percolation via threshold duality ($Q(q) = [(q+1)P(q+1)]/\langle q \rangle$5), but the sizes of their giant components are not complementary due to distinct activation/deactivation dynamics (Muro et al., 2018).

7. Open Problems and Research Directions

Major open directions include:

• Classification of universality classes beyond the mean-field random graph, particularly in networks with strong clustering, modularity, or spatial constraints.
• Systematic computation of diagrammatic corrections (e.g., $Q(q) = [(q+1)P(q+1)]/\langle q \rangle$6 expansions) for finite-dimensional lattices.
• Quantitative characterization of hybrid transitions, avalanches, and critical scaling on arbitrary topologies (graphons, hypergraphs, multiplexes).
• Practical algorithms for real-time monitoring and control of k-core percolation in dynamic and interdependent systems.

The continued generalization and unification of k-core percolation theory across topological, spatial, and functional dimensions highlight its fundamental role in the statistical mechanics of complex networks (Zhu et al., 2017, Hébert-Dufresne et al., 2013, Azimi-Tafreshi et al., 2014, Bianconi et al., 2023).