Mutual Percolation in Interdependent Networks

• Mutual percolation is a framework defining connectivity in multilayer systems where a node is functional only if it appears in the giant component of every interdependent network.
• Analytical methods like the cavity (message-passing) technique yield self-consistency equations that quantify the size and critical transitions of the giant mutually connected component.
• The concept applies to real systems such as coupled power grids and communication networks, explaining abrupt hybrid transitions and cascading failures under interdependency.

Mutual percolation is the generalized percolation phenomenon arising in interdependent or multiplex network systems, where functionality or connectivity of nodes simultaneously depends on their percolating status across multiple structural layers or networks. This extends classical percolation theory, in which a node is considered functional if it belongs to the giant component of a single graph, by imposing the stricter requirement that a node is only functional if it is jointly connected through giant components in all participating layers. Mutual percolation models capture phenomena in systems with strong interdependencies, such as power grids coupled to communication networks, or multilayered infrastructure, and are a canonical framework for analyzing the robustness, failure cascades, and phase transitions of such composite networks.

1. Fundamental Definitions and Mathematical Formulation

A mutual percolation system comprises $M$ layers (networks) on the same or overlapping set of $N$ nodes. The mutually connected (giant) component—often called the Giant Mutually Connected Component (GMCC) or Mutually Connected Giant Component (MCGC)—is defined as the maximal set of nodes that are simultaneously present in the giant (macroscopically connected) component of every layer and satisfy all interdependency constraints. In the standard (multiplex/interdependent) model, interdependency is realized by either node-to-node one-to-one matching (typically, a node in layer $\alpha$ is dependent on a corresponding node in layer $\beta$) or random matching across layers.

A node $(i, \alpha)$ belongs to the GMCC if and only if: 1. It has at least one intra-layer neighbor in layer $\alpha$ that is also in the GMCC. 2. All nodes it is interdependent with (across other layers) are also members of the GMCC (Bianconi et al., 2014).

This generalizes the notion of the percolating cluster from single-layer networks to multilayer and interdependent systems, imposing mutuality in connectivity across prescribed interdependencies.

2. Analytical Frameworks and Self-Consistency Equations

The primary analytical tool for mutual percolation is the cavity (message-passing) method, which captures both intra-layer percolation and interdependency constraints. For infinite, locally tree-like networks, this translates to recursive equations for the probability that following an edge leads to the GMCC, subject to the mutuality condition.

For two-layer multiplexes, with degree distributions $P_k^{(a)}$ and generating functions $G_0^{(a)}$, $G_1^{(a)}$, the essential system is

$$R_A = p[1 - G_1^{(A)}(1 - R_A)][1 - G_0^{(B)}(1 - R_B)], \qquad R_B = p[1 - G_1^{(B)}(1 - R_B)][1 - G_0^{(A)}(1 - R_A)].$$

The MCGC size is then

$N$0

These equations can be generalized to multilayer and nonidentical-layer systems, arbitrary interdependency topologies (“supernetworks”), and variable interdependency probabilities (partial interdependence) (Li et al., 2021, Bianconi et al., 2014).

In the case of arbitrary finite-size layered networks with given adjacency matrices, an exact cavity-equation framework provides, for each node $N$1,

$N$2

where $N$3 and $N$4 is a cavity message recursively defined on the (directed) network (Radicchi, 2015).

3. Structure-Induced Phase Transitions: Hybrid, Continuous, and Discontinuous Behavior

A key property of mutual percolation is the emergence of nontrivial phase transitions, often of mixed (hybrid) order, as the percolation control parameter $N$5 (fraction of initially surviving nodes or edges) is varied.

• Hybrid Transitions: In many interdependent network models, e.g., two symmetrically coupled Erdős–Rényi (ER) layers, the mutual percolation transition is strongly discontinuous (“hybrid”): as $N$6 increases, the size of the GMCC jumps abruptly from zero to a finite value $N$7 at a critical $N$8, with $N$9 exhibiting a square-root singularity immediately above $\alpha$0:

$\alpha$1

For two identical ER layers with mean degree $\alpha$2, theory and simulations yield $\alpha$3, markedly larger than the single-layer value of $\alpha$4 (Li et al., 2021, Bianconi et al., 2014).

• Role of Interdependency Topology: The global structure of interdependency links (“supernetwork”) fundamentally determines phase transition nature. Tree-like supernetworks exhibit a single hybrid transition; random-regular or random supernetworks with heterogeneous interdependency (superdegree) realize a cascade of sequential hybrid transitions, where layers with higher superdegree $\alpha$5 join the GMCC at higher $\alpha$6, producing a staircase of jumps in the order parameter (Bianconi et al., 2014, Gao et al., 2013).
• Loop Effects: For systems with loops in the supernetwork (e.g., ring-of-layers), under perfect interdependency ($\alpha$7), the GMCC is destroyed outright—no mutual component survives for any $\alpha$8 (Bianconi et al., 2014). Loops stabilize the system against “deep” cascades (tree amplifier effect) but can also produce catastrophic collapse at high coupling.
• Partial Interdependence and Tricriticality: If only a fraction $\alpha$9 of interdependencies are enforced, the first-order jump can vanish at a model-dependent tricritical $\beta$0, replaced by a continuous (second-order) transition. In regimes with partial interdependence, different families of layers (e.g., with superdegree $\beta$1) may undergo either continuous or hybrid transitions depending on $\beta$2 (Bianconi et al., 2014).

4. Multiplexity, Link Overlap, and Viability

The presence of overlapping (shared) links between network layers substantially alters the mutual percolation landscape.

• Overlapping vs. Nonoverlapping Links: Overlap links (present in both layers) act as “superconnectors,” forming clusters that are mutually connected by default. The rigorous cavity framework must treat clusters of overlapping links (termed "supernodes") as indivisible units for mutual connectivity calculations (Min et al., 2014).
• Self-Consistency for Overlap: Let $\beta$3 be the fraction of overlapping clusters (“supernodes”) of size $\beta$4, with corresponding distributions for the number of nonoverlap-layer-1 and layer-2 connections. The giant mutual cluster size $\beta$5 then satisfies

$\beta$6

where $\beta$7 is the cavity probability for layer $\beta$8 (Min et al., 2014).

• Transition Character: For any nonzero overlap $\beta$9, the transition remains discontinuous except in the single-layer limit. Overlaps lower the percolation threshold and furnish robustness but produce vulnerability: attacks on overlapping links destroy mutual connectivity more efficiently than attacks on nonoverlapping links (Min et al., 2014).

5. Mutual Percolation in Networks of Networks (NetONet) and Generalizations

For network-of-networks (NetONet) systems—multiple interdependent networks with arbitrary dependency topologies and inhomogeneous coupling fractions ($(i, \alpha)$0)—one obtains a hierarchy of mutual percolation models.

• General Steady-State Equations: For $(i, \alpha)$1 networks, each depending on $(i, \alpha)$2 others, the steady-state equations couple the sizes of their respective mutual clusters $(i, \alpha)$3 as

$(i, \alpha)$4

where $(i, \alpha)$5 encodes the percolation characteristics of layer $(i, \alpha)$6, and $(i, \alpha)$7 the pre-cascade survivor fractions (Gao et al., 2013).

• Thresholds and Collapse: Depending on the interdependency fraction $(i, \alpha)$8 and network topologies, transitions can be second-order (continuous) or first-order (discontinuous), with an abrupt disappearance of the mutual cluster above a maximal coupling threshold $(i, \alpha)$9. The presence of feedback (bidirectional intra-node dependencies) can eliminate discontinuous regimes and lower $\alpha$0 (Gao et al., 2013).
• Loop and Tree Topologies: In tree-like NetONet, mutual cluster sizes depend on the global system size $\alpha$1, but, in loopy random-regular NetONet, only the layer degree $\alpha$2 governs the steady-state; $\alpha$3 drops out. Loops reduce cascade depth but can cause abrupt collapse at moderately high interdependence (Gao et al., 2013, Bianconi et al., 2014).

6. Mutual Percolation in Statistical Physics Models

In statistical mechanical models, mutual percolation manifests in coupled or constrained graphical models (e.g., joint Potts models on a planar graph and its dual with non-intersecting interfaces, “spin-coupled mutual percolation” (Lis, 2019)), and in multi-type percolation problems (e.g., percolation of multiple fluids or competing innovations) (Novikov, 2019, Roca et al., 2011).

• Mutual Percolation and Height Functions: In the coupled Potts or random current model, mutual percolation is associated with global delocalization of a height function, and existence of simultaneous infinite clusters is tightly linked to the non-intersecting hard constraints in the joint system (Lis, 2019).
• Multi-percolation and Competition: In systems modeling competing innovations or opinions, two-way bootstrap percolation models exactly capture the survival dynamics: only innovations that realize their own giant percolating cluster can survive in the long run, and when both percolate, a stable coexistence phase emerges. Thresholds and transition orders (continuous vs. discontinuous) are determined by network degree and strategic payoff structure (Roca et al., 2011).

7. Implications for Information Propagation and Statistical Inference

Mutual percolation concepts also illuminate upper bounds for information-theoretic problems on graphs, specifically in establishing impossibility regimes. The “information-percolation” framework provides upper bounds on mutual information between distant nodes in graphical models, via mapping to a bond-percolation problem with per-edge opening probabilities determined by contractivity properties or $\alpha$4-information. When the percolation model does not percolate, statistical reconstruction in the original problem is impossible, and this holds even in loopy or non-tree topologies (Abbe et al., 2019, Polyanskiy et al., 2018, Abbe et al., 2018).

These methods generalize subadditivity theorems classically available only on trees to broader classes of graphs (e.g., series–parallel), with broad applications in synchronization, community detection, and inference with noisy observations.

---

References:

• (Bianconi et al., 2014) Percolation in networks of networks with random matching of nodes in different layers
• (Li et al., 2021) Percolation on complex networks: Theory and application
• (Gao et al., 2013) Percolation of a general network of networks
• (Radicchi, 2015) Percolation in real interdependent networks
• (Baxter et al., 2016) A unified approach to percolation processes on multiplex networks
• (Min et al., 2014) Link overlap, viability, and mutual percolation in multiplex networks
• (Abbe et al., 2019) Subadditivity Beyond Trees and the Chi-Squared Mutual Information
• (Abbe et al., 2018) An Information-Percolation Bound for Spin Synchronization on General Graphs
• (Polyanskiy et al., 2018) Application of information-percolation method to reconstruction problems on graphs
• (Roca et al., 2011) Percolate or die: Multi-percolation decides the struggle between competing innovations
• (Lis, 2019) Spins, percolation and height functions
• (Novikov, 2019) Percolation of three fluids on a honeycomb lattice