Towards real-world complexity: an introduction to multiplex networks

Kyu-Min Lee; Byungjoon Min; Kwang-Il Goh

Source-linked AI summary

Towards real-world complexity: an introduction to multiplex networks

Kyu-Min Lee, Byungjoon Min, Kwang-Il Goh

arXiv:1502.03909v1Feb 13, 2015physics.soc-phcond-mat.stat-mechcs.SI

TL;DR

Multiplex networks require analysis that accounts for structural and functional coupling among layers, beyond single-network frameworks. This Colloquium organizes and reviews statistical-physics research by layer-coupling context, highlighting cooperative and complementary systems and their distinctive phenomena. It also identifies scope boundaries and open directions for extending the review’s framework.

Problem
Single-network theory does not fully capture multiplex systems, whose layer coupling can produce non-additive and nonlinear effects in structure and function.
Method
The paper provides an organized review of multiplex-network research, categorizing problems broadly by the functional context of layer coupling.
Results
The review distinguishes cooperative and complementary layer-coupling problems and surveys major concepts, methods, and phenomena associated with them.
Takeaways & Limitations
Properly identifying multiplexity context is important for formulating and solving multiplex-network problems.
Takeaways & Limitations
The review focuses mainly on strict multiplex networks and multiplex random networks, leaving more heterogeneous layers such as scale-free networks less covered.

Abstract

from arXiv · show

Many real-world complex systems are best modeled by multiplex networks of interacting network layers. The multiplex network study is one of the newest and hottest themes in the statistical physics of complex networks. Pioneering studies have proven that the multiplexity has broad impact on the system's structure and function. In this Colloquium paper, we present an organized review of the growing body of current literature on multiplex networks by categorizing existing studies broadly according to the type of layer coupling in the problem. Major recent advances in the field are surveyed and some outstanding open challenges and future perspectives will be proposed.

1 Introduction

Multiplex networks model systems in which the same nodes interact through multiple link types, but their behavior depends on both structural and functional coupling between layers. The paper reviews this literature by organizing problems according to functional layer coupling.

Multiplex networks represent systems whose same nodes are connected through multiple types of links, each forming a network layer.
Because functional coupling can produce non-additive and nonlinear effects, multiplex analysis requires concepts and computations beyond single-network theory.
Multiplexity has structural and functional facets: layers may be coupled in their pattern and in how one layer’s function affects another.
Functional layer coupling can be complementary/connective, as in transportation, or cooperative/dependent, as in critical infrastructure.
The Colloquium organizes current multiplex-network research by type of multiplexity and surveys cooperative, complementary, and other layer-coupling problems.

2 Characterization and modeling of the multiplex structure

The paper characterizes multiplex structures through empirical data, structural measures, and random-graph models. These tools support systematic and analytic study of multiplex systems.

The characterization program begins with empirical multiplex-network data from social, man-made, and biological systems.
It introduces basic representations and structural measures for multiplex-network ensembles.
Random-graph models are used as theoretical tools for systematic and analytic understanding of multiplex systems.

2.1 Empirical studies and multiplex network data

Multiplex-network data arise across social, transportation, biological, and other real-world systems. Increasingly large-scale measurements provide empirical material for studying correlated multilayer structure.

Empirical network data motivate newer network concepts and the analytic tools needed to study them.
Social multiplex data include surveys, multiplayer online games, and communication-collaboration networks among scientists.
Transportation multiplexes span multiple modes, airline routes operated by different companies, and coupled airport–seaport systems.
Biological multiplex networks occur across cellular, physiological, and ecological levels of organization.
In social-network usage, multiplexity has often referred specifically to overlap in links across different layers.

2.2 Mathematical formulations and measures

Multiplex structure requires measures that capture layer participation, interlayer correlations, overlap, and cross-layer paths rather than simply adding a layer index to single-network definitions. The section develops representations and measures whose appropriateness depends on the coupling context.

2.2 Mathematical formulations and measures: Simple generalization of single-layer concepts can miss multiplex context, so the discussion restricts basic development to simple undirected unweighted layers.
2.2 Mathematical formulations and measures: A simple multiplex requires each multiplex node to participate in all layers; systems with every node participating are fully multiplexed, otherwise partially multiplexed.
2.2.1 Matrix representation: The supra-adjacency matrix represents an ℓ-layer multiplex as an ℓ×ℓ matrix of N×N blocks containing layer-to-layer adjacency information.
2.2.2 Multiplex degree: Multiplex degree lists node degrees across layers, while suitable centrality definitions depend on whether layers are complementary or contain overlapping links.
2.2.3 Correlations between layers: Joint multiplex-degree distributions encode structural information, and interlayer degree correlations compare degrees of the same nodes across layers.
2.2.2 Multiplex degree: Interlayer degree coupling can be compared through maximally positive, uncorrelated, and maximally negative arrangements based on matching or reversing degree ranks.
2.2.3 Correlations between layers: Link overlap signals non-random layer coupling and has distinctive functional roles, while cross-layer clustering extends transitivity beyond triangles confined to one layer.
2.2.4 Shortest path and other distance-based measures: Shortest paths, distances, and centralities must also be adapted to multiplex coupling, including optimal paths across complementary or connective layers.

2.3 Multiplex network models

Multiplex network models extend single-network theory by representing multiple interaction layers and their structural or dynamical coupling. Random graph and growing-network approaches show how layer coupling, including coevolution, shapes multiplex structure.

Random graph models: Random graph ensembles generalize degree-constrained network models using multiplex degrees or their distributions, enabling systematic analysis of multiplex structure and percolation.For overlapping links, the multiplex degree must separately constrain degrees for overlapping and non-overlapping links.
Growing network models: Growing-network models specify generative rules for adding nodes and links, connecting microscopic linkage rules to macroscopic network properties.These models include preferential attachment and can reveal structural instability in growing multiplex networks.
Coevolving multiplex models: Coevolving layers make attachment in one layer depend on node degrees in both that layer and the other layers.In the two-layer model, the coevolution parameter ϵ controls relative dependence on the other layer.
Coevolving multiplex models: The coupled preferential-attachment kernels mix native- and cross-layer degrees according to ϵ, while a determines the layers’ native degree exponent.The two kernels exchange the roles of layers A and B symmetrically.

3 Problems with cooperative layer-coupling

Cooperatively coupled multiplex layers can produce synergistic effects that are non-additive and nonlinear across individual layer contributions. Such interactions motivate studying structural and dynamical properties beyond conventional single-layer frameworks.

Cooperative layer coupling: Cooperative activity across multiple layers can create non-additive, nonlinear synergistic effects that single-layer network frameworks cannot accurately address.The section concerns structural and dynamical properties arising from cooperative layer coupling.

3.1 Mutual percolation

Mutual percolation requires simultaneous connectivity across every layer, producing giant mutual components whose transitions can differ sharply from ordinary percolation. The review covers how partial participation, interlayer correlations, overlap, network-of-networks structure, and spatial constraints alter these transitions.

A mutual component contains node pairs connected within every layer simultaneously, and its extensive form is the giant mutual component.Mutual percolation generalizes connectivity by requiring simultaneous cross-layer connectivity.
For duplex Erdős–Rényi networks, the transition occurs at zc(2) = 2.455407 . . . with a giant-component jump Mc(2) = 0.511699 . . . .The threshold is higher than ordinary single-layer percolation, where zc(1) = 1.
Mutual percolation is a hybrid transition: the order parameter jumps, then follows critical scaling above the transition, with β = 1/2 for duplex ER networks.Unlike typical first-order transitions, it does not exhibit hysteresis, and finite mutual-component susceptibility does not diverge at the transition.
Partial multiplexity: With partial multiplexity, the transition can change from discontinuous to continuous as the multiplex fraction q and layer mean degrees vary, producing a tricritical-like point.Generalized mutual-component sizes can differ between layers when their mean degrees differ and q ≠ 1.
Extensions: Related extensions show that one-to-one interdependency can make a network of networks share the mutual component of a simple multiplex, while spatial embedding introduces correlation-driven theoretical challenges.The review also treats interdependent lattice networks, where coupling distance interpolates between local and random dependency patterns.
Correlated multiplexity: Positive interlayer degree correlation lowers the mutual-percolation transition point, making the multiplex more robust to random failure.Link overlap facilitates mutual percolation, but in multiplex ER networks the transition remains discontinuous for r < 1 and its jump vanishes gradually as r → 1.

3.2 Robustness against attacks

Multiplex robustness depends on how nodes are removed and how network layers are structurally correlated. These dependencies can produce protocol-specific and non-monotonic responses, complicating robust-system design.

The giant mutual component is the primary quantity for assessing robustness in multiplex networks with cooperative layer coupling.
Removal strategies are encoded by φ(k), including uniform random removal and intentional attacks based on total degree.For random removal, φ is constant; degree-based attacks use a cutoff or degree-dependent rule.
Positive interlayer degree correlation enhances robustness against random failures, whereas negative correlation decreases it.
Under total-degree attacks, vulnerability depends on correlated coupling and initial network density, producing non-monotonic and protocol-specific responses.
The optimal multiplex structure can depend strongly on the type of damage considered, complicating robust coupled-system design.

3.3 Cascades and complex contagion

Multiplex cascade models extend contagion and failure dynamics by combining influences across network layers. The reviewed results show that layer-coupling rules and interlayer correlations can determine whether cascades are facilitated, impeded, continuous, or discontinuous.

Multiplex cascade models integrate social influences from different layers non-additively and nonlinearly.Analytical treatments for locally tree-like uncorrelated layers follow logic related to mutual-percolation analyses.
The response function F̄(m,k) gives the probability that a node activates based on its active neighbors among its multiplex degree.
Under the OR rule, nodes activate when a threshold is met in at least one layer; under the AND rule, thresholds must be met in all social layers.The OR rule facilitates cascades, whereas the AND rule impedes them.
Load-based cascading failures differ from threshold cascades because redistribution of shortest-path loads induces failures nonlocally.Each node’s capacity is set to Ci = (1 + α)Li, where α is tolerance and Li is its load.
For E = 0.2, the transition to a global cascade is discontinuous, whereas it is continuous for E = 0.5 and E = 1.0.Here E is the fraction of OR nodes, with ρ0 = 0.001 and R = 0.18.
Positive interlayer degree correlation is beneficial in mitigating load-based cascades compared with other coupling patterns.

4 Problems with complementary layer-coupling

Complementary layers support interconnected-network analyses of percolation, diffusion, transport, and spreading. Their coupling can produce behavior that depends on layer structure, interlayer diffusion, and cross-layer transmission mechanisms.

4.1 Percolation: Interconnected-layer percolation uses generating functions to determine the giant component, with the giant bicomponent defined by two disjoint paths.The existence conditions for the giant component and giant bicomponent coincide through the largest Jacobian eigenvalue at the trivial solution.
4.1 Percolation: MP coupling lowers the duplex ER percolation threshold to zc = 0, whereas MN coupling delays it to zc = 0.838587 . . . versus zc = 1/2 for uncorrelated coupling.For MP coupling, the giant bicomponent size equals the giant component size; generally, the bicomponent grows more slowly.
4.2 Diffusion: Above the interlayer-diffusion threshold, multiplex diffusion is faster than diffusion in the slower layer, and for DAB ≫1 it can exceed diffusion in either isolated layer.This latter enhancement is termed superdiffusive, although the paper notes it does not change the dynamic exponent.
4 Problems with complementary layer-coupling: Multiplex transport also includes non-diffusive routing processes, where logical and physical layers represent distinct aspects of traffic movement.These layered formulations were applied to real traffic problems involving routing and congestion control.
4.3 Epidemic spreading: For multiplex contagion, the epidemic threshold is governed by the layer with the largest contact-probability eigenvalue, so aggregating layers can be inaccurate.Layer-crossing overhead can make the epidemic threshold depend non-monotonically on switching cost through path-dependent transmissibility.

5 Other types of layer coupling

Beyond cooperative and complementary coupling, multiplex layers may compete or influence one another directionally. Competitive coupling produces distinctive phase behavior, while directional coupling remains comparatively understudied.

Competitive coupling: Competitive coupling treats antagonistically interacting layers as functionally interlocking while emphasizing their distinctive effects.The paper discusses this coupling separately from cooperative coupling because the layers oppose one another.
Competitive coupling: Antagonistic percolation assigns a node to one layer’s percolating cluster only when it connects within that layer and lacks a corresponding connection in the other.This exclusivity defines the percolating clusters for the two-layer model.
Competitive coupling: Dense competitive layers can produce bistability with discontinuity and hysteresis, and this phase persists with only a small fraction of antagonistic nodes.The reported phase diagram includes stable solutions with these features when both layer densities are sufficiently large.
Directional coupling: Directional or hierarchical layer influence has received little attention in multiplex-network research.Temporal networks are identified as a related field where directionally coupled layers arise naturally from temporal causality.

6 Conclusion and outlook

The paper reviews multiplex networks by emphasizing layer-coupling context and distinguishing cooperative from complementary problems. It argues that multiplex coupling can generate phenomena unavailable to equivalent single-layer descriptions while leaving several foundational questions open.

Conclusion: The review organizes structural and dynamical multiplex-network problems by coupling context, with cooperative and complementary layers treated as major categories.It introduces multiplex examples and structural measures alongside statistical-physics problems.
Scope and limitations: The review is deliberately focused on the strict multiplex definition and mostly on multiplex random networks, omitting or limiting related multilayer systems and several dynamics.More heterogeneous layers, including scale-free networks, are identified as a direction for further study.
Conclusion: The paper argues that multiplex systems are not reducible to equivalent single-layer systems because layer coupling can produce novel phenomena, especially under cooperative or competitive coupling.This frames multiplexity as important for understanding emergent behavior in complex systems.
Outlook: Open problems include context-dependent network measures, multiplex controllability, and systematic classification of universality classes and critical phenomena.The paper specifically calls for identifying minimal couplings relevant to discontinuous multiplex transitions.