PaperlayerView exact revision on arXiv

Source-linked AI summary

Synchronization in complex networks

Alex Arenas, Albert Diaz-Guilera, Jurgen Kurths, Yamir Moreno, Changsong Zhou

arXiv:0805.2976v3physics.soc-ph

TL;DR

Synchronization in complex networks remains difficult to understand because real systems rarely use all-to-all interactions. This review synthesizes analytical and numerical approaches to network-constrained oscillators and finds that interaction topology is central to synchronizability, with the MSF formalism enabling topology-dependent predictions.

  • Problem

    The review addresses how synchronization changes when interacting oscillators are constrained by the non-all-to-all topologies found in real systems.

  • Method

    The paper reviews analytical and numerical studies of networked oscillators, including MSF-based theory and applications across several disciplines.

  • Results

    The review shows that interaction topology shapes synchronization, while the MSF formalism supports predictions of synchronized-system evolution from network structure.

  • Takeaways & Limitations

    Synchronization provides a useful abstraction for processes across biological, technological, and social contexts, with topology informing their analysis.

  • Takeaways & Limitations

    Conclusions about synchronizability robustness and fragility under node attacks require cautious interpretation because reported observations are inconsistent and need reexamination.

Abstract

from arXiv · show

Synchronization processes in populations of locally interacting elements are in the focus of intense research in physical, biological, chemical, technological and social systems. The many efforts devoted to understand synchronization phenomena in natural systems take now advantage of the recent theory of complex networks. In this review, we report the advances in the comprehension of synchronization phenomena when oscillating elements are constrained to interact in a complex network topology. We also overview the new emergent features coming out from the interplay between the structure and the function of the underlying pattern of connections. Extensive numerical work as well as analytical approaches to the problem are presented. Finally, we review several applications of synchronization in complex networks to different disciplines: biological systems and neuroscience, engineering and computer science, and economy and social sciences.

1 Introduction · 2 Complex networks in a nutshell

The supplied passage provides only the paper’s submission metadata and does not state substantive content about the introduction or complex networks.

  • 2 Complex networks in a nutshell: The preprint was submitted to Physics Reports on November 26, 2024.

1. Introduction

Synchronization is widespread across natural and artificial systems, motivating the study of oscillating units coupled through complex network topologies rather than idealized all-to-all interactions. This review examines the interplay between network structure and synchronization, including stability analysis and applications across scientific disciplines.

  • Motivation: Synchronization occurs across biology, ecology, climatology, sociology, technology, and the arts, from cellular metabolism to collective cognitive tasks.
  • Motivation: All-to-all coupling can describe spontaneous order but becomes difficult to realize in large populations because of physical constraints such as energy or cost minimization.
  • Complex-network foundations: The Watts-Strogatz model introduced a network substrate for synchronization by interpolating between regular lattices and random graphs through link rewiring.
  • Review scope: The review develops complex-network descriptors, analyzes synchronization and fully synchronized-state stability through the Master Stability Function formalism, and surveys applications across scientific fields.

2. Complex networks in a nutshell

Complex networks are represented as graphs whose structural organization is characterized by degree distributions, path lengths, clustering, and community structure. These measures distinguish network types and support comparisons between connectivity patterns and modular organization.

  • Network representation: A complex network is a graph G with N nodes and M links, represented by an adjacency matrix A, where ki is node i’s degree.The adjacency entry aij equals 1 when a directed link from j to i exists and 0 otherwise; weighted networks generalize this representation.
  • Degree distribution: The degree distribution P(k) gives the probability that a node has degree k and is the primary basis for classifying network heterogeneity.Exponentially decaying tails indicate homogeneous networks such as Erdös-Rényi graphs, whereas heavy tails indicate heterogeneous networks, including scale-free networks.
  • Path length: Average shortest-path length ℓ distinguishes network organization, scaling as ℓ∼N 1/d in d-dimensional lattices but as ℓ∼ln(N)/ln(k̄) in random networks.Thus, random networks can have small average shortest-path lengths even when they contain very large numbers of nodes.
  • Clustering: The clustering coefficient C measures local transitive connectivity: a large C implies many transitive connections and redundant paths, whereas a low C implies the opposite.It is calculated from ni, the connections between node i’s nearest neighbors, and ki, its degree.
  • Community structure: Community structure consists of densely connected groups with sparser connections between them, and modularity Q is optimized to compare alternative network partitions.Finding the best partition is difficult; larger Q indicates a more modular network, potentially revealing structure-function relationships.

3. Coupled phase oscillator models on complex networks

Early studies of synchronization in coupled oscillators were motivated largely by biological neural networks and examined phase nonlinear oscillators interacting through network structures.

  • 3. Coupled phase oscillator models on complex networks: Biological neural networks motivated the first studies of synchronization in coupled oscillators.
  • 3. Coupled phase oscillator models on complex networks: Strogatz and Mirollo, followed by Niebur et al., studied collective synchronization of phase nonlinear oscillators.
  • 3. Coupled phase oscillator models on complex networks: Their models used random intrinsic frequencies, varied coupling schemes, and 2D lattice interaction networks.

3.1. Phase oscillators

The Kuramoto model captures synchronization as a transition from incoherence to phase locking, characterized by an order parameter and critical coupling. Complex network topology changes normalization, critical behavior, and the analytical picture of synchronization onset.

  • 3.1. Phase oscillators: The model assumes natural frequencies drawn from a usually unimodal, symmetric distribution g(ω), with rotational symmetry allowing the mean frequency to be set to zero.The 1/N factor ensures appropriate behavior in the thermodynamic limit N →∞.
  • 3.1. Phase oscillators: Synchronization emerges when coupling exceeds a critical value: oscillators with |ω_i| ≤ Kr phase-lock, while the rest drift around the circle.The order parameter distinguishes coherent and incoherent states, with r ≃1 and r ≃0 respectively.
  • 3.1. Phase oscillators: Near onset, the Kuramoto order parameter follows mean-field square-root scaling, r ∼(K − K_c)^β, with β = 1/2.Numerical simulations verified the predicted transition and scaling behavior.
  • 3.1. Phase oscillators: Complex-network formulations require coupling normalization, such as σ_ij = K/k_i, to keep interactions intensive and address degree heterogeneity that can suppress synchronization.Using σ_ij = K/N in complex networks can produce a different dependence on network size than in the all-to-all model.
  • 3.1. Phase oscillators: Small-world networks synchronize at finite K after rewiring only a tiny fraction of ring links, while scale-free networks can retain a critical point even for γ ≤3.For BA networks, σ_c = 0.05(1) and β = 0.46(2), close to mean-field scaling.
  • 3.1. Phase oscillators: Analytical results for general complex networks remain unavailable, and predictions for uncorrelated scale-free networks with γ ≤3 conflict with numerical evidence.The critical coupling is analytically related to 1/g(0) and the largest adjacency eigenvalue λ_max, while the method does not explain the γ ≤3 thermodynamic-limit behavior.

3.2. Pulse-coupled models

Pulse-coupled network models, especially integrate-and-fire and neuron-like oscillators, show that topology and coupling heterogeneity can determine synchronization speed, activity patterns, and transitions between coherent and disordered states. Studies report topology-induced periodic or disordered dynamics, with average path length influencing synchronization and heterogeneous coupling replacing global synchrony with periodic firing.

  • Topology and synchronization time: Random networks synchronize more slowly than regular square lattices, especially when sparse, while Watts–Strogatz networks can further increase synchronization time.The comparison identifies synchronization time as dependent on network topology and rewiring probability.
  • Topology-induced activity patterns: In small-world rings of integrate-and-fire oscillators, low random-connection density preserves periodic patterns, whereas high density produces long transients and disordered activity.At large density, overlapping activity between excited domains causes synchronized self-sustained activity to collapse.
  • Neuron-like pulse-coupled models: Networks of nonidentical Hodgkin–Huxley elements can develop either coherent oscillations or asynchronous states under excitatory synaptic coupling across random, regular, and small-world topologies.The model neurons remain below bifurcation until incoming input forces a saddle-node bifurcation on a limit cycle.
  • Network structure and synchronization: For pulse-coupled Bonhoeffer-van der Pol-FitzHugh-Nagumo oscillators, average path length strongly affects synchronization, while clustering and loop coefficients appear less important.The authors caution that conclusions based on individual network characteristics remain inconclusive.
  • Heterogeneous coupling: In pulse-coupled networks with complex connectivity, coupling heterogeneity induces periodic firing patterns that replace global synchrony and become asynchronous beyond a critical value.The reported transition links coupling heterogeneity to qualitatively different collective firing states.

3.3. Coupled maps

Coupled-map networks exhibit synchronization and partial synchronization through interactions between local dynamics, connectivity structure, coupling, and delays. Studies identify topology-dependent stability conditions, long-range-coupling effects, degree-dependent transitions, and distinct mechanisms of cluster formation.

  • Topology and synchronization: For logistic maps with random fixed-degree connectivity, synchronization occurs when k > 4, while synchronization time decreases with connectivity and saturates as system size grows.In a modified Watts–Strogatz model, any nonzero shortcut-addition probability guarantees synchronization in the thermodynamic limit.
  • Topology and synchronization: Long-range coupling induces synchronization in sine-circle-map systems that do not synchronize without shortcuts.The effect is measured through a parameter related to winding-number dispersion.
  • Stability conditions: Stability of synchronized states is determined by normalized-Laplacian eigenvalues and the map’s Lyapunov exponent, with random networks synchronizing for arbitrarily large size above a neighbor threshold.The framework is demonstrated for regular, globally coupled, ring, and complex connectivity patterns using quadratic maps.
  • Degree-dependent coupling: Degree-dependent coupling produces a first-order transition between coherent and noncoherent phases in BA networks, governed by mean connectivity and coupling strength.For α > 0, the transition occurs at smaller interaction values than in the usual case; deterministic pseudofractal and Apollonian networks lack coherence when α = 0 and a = 2.
  • Delays and collective states: Uniform connection delays facilitate synchronization across general topologies, while ER and SF networks are easier to synchronize than regular or small-world networks.Delays can also enable new collective phenomena, and coupled-map phase diagrams contain turbulent, partially ordered, and ordered stationary configurations.
  • Cluster formation: Partial synchronization forms through either self-organized clusters driven by intracluster coupling or driven clusters caused by intercluster coupling.In WS networks of chaotic Rössler oscillators, synchronization increases with coupling strength and mean phase differences decrease as shortcut probability increases.

4. Stability of the synchronized state in complex networks

This section examines the stability of completely synchronized states in populations of identical oscillators on complex networks. It reviews the master stability function formalism, which evaluates linear stability as a necessary but insufficient condition for synchronization.

  • Research direction: The stability of the completely synchronized state in networks of identical oscillators became a parallel research direction to synchronization studies of phase oscillators.Barahona and Pecora’s seminal work initiated this line of research.
  • Master stability function: The master stability function formalism assesses the linear stability of the completely synchronized state, providing a necessary but not sufficient condition for synchronization.The section reviews the formalism and its main results.

4.1. Master Stability Function formalism

The Master Stability Function formalism separates oscillator dynamics from network structure by assessing synchronization through coupling-matrix eigenvalues. It establishes stability conditions and shows how network topology, shortcuts, connectivity, and heterogeneity affect synchronizability.

  • 4.1. Master Stability Function formalism: The formalism considers time-continuous systems with identical vector output functions H(x), while also applying to time-discrete maps.The output function generates the signal sent between oscillators and may select only components of the oscillator state.
  • 4.1.1. Linear Stability and Master Stability Function: For complex coupling-matrix eigenvalues, the MSF must be evaluated over the complex plane, where stable regions may be bounded or semi-bounded.This case is mathematically more intricate, and fewer results are available than for real eigenvalues.
  • 4.1.2. Measures of synchronizability: Synchronization is impossible when R > α2/α1; when R < α2/α1, stability requires σmin < σ < σmax.The thresholds are σmin = α1/λ2 and σmax = α2/λN.
  • 4.1.2. Measures of synchronizability: Smaller eigenratio R indicates greater synchronizability, while larger λ2 lowers the synchronization threshold and influences the time needed for complete synchronization.R and λ2 depend only on the network structure, allowing oscillator-independent characterization for bounded MSFs.
  • 4.1.2. Measures of synchronizability: For general directed networks, complex spectra make it unclear how to define simple oscillator-independent synchronizability measures.This limitation helps explain the focus on undirected and unweighted networks with real spectra.
  • 4.1.3. Synchronizability of typical network models: Regular networks have poor synchronizability because λ2 ∼1/N^2, whereas adding random shortcuts to small-world networks reduces R and improves synchronizability.In small-world networks, λ2 ≈2S = 2kp, so shortcut density is a principal determinant of synchronizability.
  • 4.1.3. Synchronizability of typical network models: Random networks are synchronizable only when f ≳2 ln N/(N + 2 ln N), while semirandom small-world backbones remain connected and synchronizable for k ≥1.For small f ≲ln N/N, purely random networks are almost surely disconnected and non-synchronizable.
  • 4.1.4. Synchronizability and structural characteristics of networks: In rewired small-world networks, the small-world transition begins at pSW = 1/Nk, but synchronizability improves only when shortcut density is independent of N and beyond Ssync ∼k.At p = pSW, S ∼1/N and λ2 ∼1/N, so rewiring does not yet enhance synchronizability.

4.2. Design of synchronizable networks

The section reviews approaches for designing synchronizable networks by reweighting couplings, adapting weights, perturbing topology, and searching for optimal network structures. These schemes generally improve synchronizability by homogenizing coupling intensities, directing information flow, or aligning connectivity with oscillator properties, while their applicability and theoretical basis remain limited.

  • Weighted coupling: Weighted couplings can enhance synchronizability in degree-heterogeneous networks by balancing degree heterogeneity toward a homogeneous intensity distribution.Betweenness-dependent weights achieve a minimum eigenratio at 0 < α ≲ 1, slightly improving synchronizability over the optimal degree-based weighting; for larger networks, further gains become negligible.
  • Weighted coupling: For the two-parameter weighting scheme, synchronizability is optimal at β = 1, where coupling intensity is fully uniform, with an optimal α for each fixed β.When β ≠ 1, intensity becomes more heterogeneous as |1 − β| increases.
  • Weighted coupling: Asymmetric couplings based on node age enhance synchronizability in scale-free networks when older, typically higher-degree nodes drive younger nodes, corresponding to θ < 0.The review notes that the mechanism is not fully understood and that changes may primarily reflect reduced heterogeneity in node intensity distributions.
  • Adaptive coupling: Adaptive local synchronization can tune degree–weight correlations with θ ≈ −0.5 and significantly enhance synchronizability relative to unweighted networks.The adaptation increases connection strengths among each node and its direct neighbors, producing generally asymmetric input and output weights.
  • Topology optimization: Topology optimization can slightly improve synchronizability while transforming scale-free networks toward disassortativity, whereas imposing directed spanning trees, no directed loops, and normalized input strengths yields feedforward optimal networks.For phase oscillators, rewiring toward connections between oscillators with similar average frequencies also enhances synchronization and can produce many cliques and large average distance.

4.3. Beyond the Master Stability Function formalism

Beyond complete synchronization, effective synchronization patterns can still be analyzed using mean-field approximations and linear criteria related to the Master Stability Function. In heterogeneous networks, hubs synchronize more closely with the mean field and form effective clusters, while global stability requires additional oscillator and network constraints.

  • Effective synchronization patterns: Mean-field analysis examined effective synchronization in unweighted scale-free networks under subthreshold coupling, oscillator mismatches, and noise perturbations.The oscillators used chaotic dynamics with coupling function H(x) = x.
  • Effective synchronization patterns: In heterogeneous networks, hubs with ki > kth synchronize more closely with the mean field and form effective synchronization clusters.The clusters satisfy |xi −xj| < ∆th, but no unique threshold defines them.
  • Beyond complete synchronization: For H(x) = x, the largest Lyapunov exponent becomes negative when σki > λF 1, and for large k satisfying σk ≫λF 1, λmax(k) ≈−σk.These linear criteria extend MSF-based reasoning to degree-dependent stability away from complete synchronization.
  • Beyond complete synchronization: The asymptotic synchronization difference scales as ∆X(k) = c/|λmax(k)|, explaining the numerically observed dependence for perturbed or noisy systems.Here c > 0 denotes the perturbation level and depends on either noise level D or coupling strength σ; the same scaling was found for Kuramoto recovery times.
  • Global stability: Global synchronization stability requires additional constraints on individual oscillator dynamics and, in general networks, conditions involving link path lengths and betweenness-related quantities.The cited result for undirected networks also applies to directed networks when input and output degrees are equal at every node.

5. Applications

This section reviews applications of synchronization in complex networks across biology and neuroscience, engineering and computer science, and economy and social sciences.

  • Applications: The review applies synchronization concepts and techniques to specific problems across several scientific fields.These applications follow the review’s theoretical and computational treatment of synchronization processes in complex networks.
  • Biology and neuroscience: Biology and neuroscience are among the fields covered by the review’s applications.
  • Engineering, computer science, economy and social sciences: Engineering and computer science, as well as economy and social sciences, are also application areas.

5.1. Biological systems and neuroscience

Synchronization in complex networks provides a framework for analyzing dynamical interactions across biological scales, from genetic and cellular systems to populations and neural networks. Applications show how coupling structure shapes biological rhythms, population synchrony, and brain activity.

  • Biological systems and neuroscience: Synthetic gene networks such as the three-gene repressilator model cyclic repression and provide a system for studying biological rhythms and synchronization.Its protein products repress transcription of one another, while modular additions can introduce diffusion-based coupling between yeast cells.
  • Biological systems and neuroscience: Plant circadian oscillators synchronize through vein-network connections that combine nearest-neighbor coupling with long-range material transport across tissues.The plant vein system serves as the network substrate for experimentally investigated synchronization.
  • Biological systems and neuroscience: Population synchrony requires integrating spatial and trophic couplings, while complex food-web topology and conservation corridors complicate metacommunity dynamics.The Moran effect offers one explanation through synchronous environmental forcing, but other mechanisms also motivate explicit spatial and trophic modeling.
  • Biological systems and neuroscience: Synchronization is especially relevant in neural systems, whose billions of neurons are coupled through hierarchical complex-network connectivity.Research spans detailed cellular circuits and larger-scale brain networks.
  • Biological systems and neuroscience: The synchrony-to-wiring-length ratio is optimized in the small-world regime, where most connections are local and only a small fraction of neurons have long-range connectivity.The long-range neurons have large global impact despite being rare, matching observed neural organization.
  • Biological systems and neuroscience: Brain simulations show that long-term functional connectivity is shaped by structural connectivity, while short-term functional connectivity changes over time.These findings support viewing the brain as an active network capable of generating spontaneous activity without external signals.

5.2. Computer science and engineering

Synchronization in computer science and engineering supports distributed simulation, consensus, data mining, wireless timing, logistics, and power-grid operation. Complex network structure shapes synchronization feasibility, convergence speed, robustness to delays, and recovery after perturbations.

  • Parallel distributed simulations: Parallel distributed simulations use synchronization to coordinate local state variables across processors in large interacting systems.Applications include financial markets, epidemic spreading, traffic, and physical-system dynamics.
  • Parallel distributed simulations: In conservative PDES, the simulated-time horizon behaves like a nonequilibrium surface governed in one dimension by the Edwards-Wilkinson Hamiltonian.The analogy connects synchronization performance to the steady-state roughness of a macroscopic landscape.
  • Parallel distributed simulations: On regular lattices, synchronization-landscape width diverges as w ∼ N^1/d, while adding a few random links produces a small-world structure that suppresses large fluctuations.The corresponding lateral correlation length also diverges, and small-world links decorrelate it.
  • Consensus: Consensus is formulated as asymptotically stable agreement on a unique common value, with applications including cooperative control, formation control, and distributed sensor networks.For fixed strongly connected digraphs, average consensus holds if every node has equal indegree and outdegree.
  • Consensus and wireless networks: Consensus convergence speed is linked to λ2, can increase by orders of magnitude after rewiring a regular lattice, and wireless synchronization supports location, proximity, energy efficiency, and mobility.For fixed undirected connected networks with equal delay τ > 0, average consensus holds when τ ∈(0, π/2λN).

5.3. Social sciences and economy

Applications in social sciences and economics use synchronization-related models to study opinion consensus, market comovement, and economic-cycle correlations, while noting that some cases are weak formulations because correlated activity is interpreted as synchronization. Opinion models exhibit coupling-dependent transitions between incoherence and synchrony, whereas financial-network analyses reveal strengthened correlations and tighter asset structures during crises.

  • Opinion formation: Opinion-formation models ask whether initially different opinions can produce complete or partial consensus through agents’ mutual influence.The social objective is to determine whether consensus can emerge, regardless of the time required.
  • Opinion formation: A modified Kuramoto model treats opinion groups as synchronization, with complete synchronization corresponding to a unique opinion state and partial synchronization to persistent differences.The model uses individuals’ opinions x_i and a network, while the synchronization interpretation is specific to this formulation.
  • Opinion formation: When σ < σc, simulations show incoherent isolated opinion groups; when σ ≫σc, the society fully synchronizes, while bipolarity is possible only if σ ∼σc.The regimes are interpreted respectively as non-interacting cultures, a single way of thinking, and bipolarity near the critical coupling.
  • Economy and financial markets: Economic synchronization is commonly measured by correlation coefficients linking correlated business cycles to correlated returns, with common disturbances and interactions both contributing to market comovement.Examples of common disturbances include world interest rates, oil prices, and political uncertainty.
  • Economy and financial markets: Analyses of N = 477 New York Stock Exchange stocks from Jan 02, 1980 to Dec 31, 1999 construct time-varying asset trees from equal-time correlations.The procedure uses smoothed daily closure prices, logarithmic returns, and correlation-derived asset distances.
  • Economy and financial markets: During crises, markets become very strongly correlated, asset trees contract and tighten, and the smallest tree converges to Black Monday when the time window is reduced.The correlation structure is interpreted as an indirect measure of strongly connected financial agents, with interactions strengthened during crashes.

6. Perspectives

The review identifies limits of master stability function analyses and outlines research directions needed to understand structure–synchronization relations in realistic complex systems.

  • Perspectives: Master stability function analysis mainly addresses linear stability of complete synchronization among identical oscillators, which is often unrealistic and may relate to pathological activity.The passage cites epileptic seizure as an example of potentially pathological activity associated with very strong synchronization.
  • Spectral properties and synchronization processes: Detailed Laplacian or adjacency spectral properties, including eigenvectors, are needed to study synchronization processes and perturbation-induced dynamical patterns.Existing studies of these detailed spectral properties remain mainly restricted to random networks.
  • Directed networks and synchronization: Directed and weighted networks require analysis of their generally complex spectra and of how directionality affects synchronization in realistic complex systems.The review presents these topics as key to understanding dynamical organization.
  • Co-evolution of structure and synchronization: Future work must examine how synchronization dynamics reshape network structures through adaptation, co-evolution, and self-organization.The review highlights neural synaptic plasticity as an example of feedback from dynamics to structure and notes increasing interest in adaptation due to synchronization.

7. Conclusions

The review advances understanding of synchronization in complex networks but concludes that a general predictive theory remains incomplete. It identifies the MSF formalism and integration of dynamics across the synchronization process as important foundations and priorities for future research.

  • 7. Conclusions: The review deepens understanding of synchronization in complex networks but does not yet establish a general theory capable of making actual predictions.Topological characterization alone may not yield predictions that can be tested against observations.
  • 7. Conclusions: Synchronization is a paradigmatic phenomenon underlying biological processes and serving as a plausible abstraction across diverse applications.Even the simple Kuramoto-model approximation raises intricate questions concerning equation uniformity.
  • 7. Conclusions: The MSF formalism provides dynamics-independent theoretical predictions and links synchronized-system evolution to network topology near the synchronization state.It applies to linear systems or nonlinear systems close to synchronization and is among the few available predictive mechanisms.
  • 7. Conclusions: Future research should develop mathematical objects that combine specific dynamics with the full synchronization process and finely describe dynamics near the synchronization manifold.This integration is presented as necessary for a general theory of synchronization processes in complex networks.
Loading 0805.2976v3…