Globally Coupled Maps (GCM)

• Globally Coupled Maps (GCMs) are discrete-time high-dimensional dynamical systems where every unit evolves by combining its own dynamics with a mean-field influence from all other units.
• They utilize various local functions—such as logistic, circle, or piecewise linear maps—to uncover phenomena like synchronization, chimera states, and phase ordering through both analytical and numerical approaches.
• The global coupling framework drives complex behaviors including turbulent chaos, invariant measure convergence, ergodicity breaking, and notable finite-size scaling effects.

Globally coupled maps (GCMs), often described as globally coupled map lattices (GCMLs) when the units are indexed as lattice sites, are discrete-time high-dimensional dynamical systems in which each element is influenced by the collective state of all other elements through a mean-field coupling term. In the finite-dimensional setting they are written as coupled recursions for $N$ maps; in the thermodynamic limit they become self-consistent dynamics on spaces of probability measures. Across these formulations, GCMs are used to study synchronization, clustering, chimera states, ergodicity breaking, invariant measures, turbulent collective chaos, and mean-field statistical behavior (Yuan, 29 Aug 2025, Velasco et al., 2021, Sélley et al., 2021, Bahsoun et al., 2022).

1. Canonical formulations

A canonical GCML with $N$ lattice points is given by

$$x_{n+1}(i) = (1-\epsilon) f(x_n(i)) + \frac{\epsilon}{N-1} \sum_{j=1, j\neq i}^N f(x_n(j)),$$

where $x_n(i)$ is the state of the $i$th node at time $n$, $\epsilon$ is the global coupling strength, and $f$ is the local map. In this form, each node’s next state is a weighted mixture of its own evolution and the mean-field contribution from all others (Yuan, 29 Aug 2025). A widely used alternative convention is

$$x_{t+1}^i=(1-\varepsilon)f(x_t^i)+\frac{\varepsilon}{N}\sum_{j=1}^N f(x_t^j),$$

which appears in globally coupled chaotic maps, logistic-map GCMs, and clustered/chimera models (Cano et al., 2017, Alvarez-Llamoza et al., 2014, Wada et al., 1 Oct 2025).

The local dynamics $f$ varies with the problem class. Explicit examples in the literature summarized here include the logistic map $N$0 on $N$1, piecewise linear expanding maps such as $N$2 and $N$3, the circle doubling map $N$4, bistable piecewise linear chaotic maps, the sine circle map, and fractional Gauss maps (Wada et al., 1 Oct 2025, Zhang et al., 19 Jul 2025, Sélley et al., 2015, Alvarez-Llamoza et al., 2014, Singha et al., 2019, Pakhare et al., 2022). The coupling is often diffusive and nonlocal; for mean-field coupled circle maps one writes

$N$5

with

$N$6

This discontinuous odd coupling function is central in several rigorous analyses of expanding circle maps (Sélley et al., 2015, Bálint et al., 2017).

In the thermodynamic limit, the state is no longer an $N$7-vector but a probability measure, or an $N$8-tuple of measures when multiple clusters are treated as subsystems. A representative self-consistent formulation is

$N$9

with

$$x_{n+1}(i) = (1-\epsilon) f(x_n(i)) + \frac{\epsilon}{N-1} \sum_{j=1, j\neq i}^N f(x_n(j)),$$0

and $$x_{n+1}(i) = (1-\epsilon) f(x_n(i)) + \frac{\epsilon}{N-1} \sum_{j=1, j\neq i}^N f(x_n(j)),$$1. In this description the dynamics acts directly on distributions, not only on microscopic trajectories (Sélley et al., 2021).

2. Synchronization and collective ordering

Synchronization is the most studied collective regime in GCM theory, but it is not the only one. In finite-dimensional formulations, a synchronized state is characterized by equality of site variables; in the measure-theoretic thermodynamic limit, a cluster is synchronized if its state is a Dirac measure, and complete synchronization occurs if all clusters are at Dirac measures (Sélley et al., 2021). This characterization is especially useful because it translates a geometric notion in phase space into an invariant set for a self-consistent transfer operator.

Rigorous threshold results are available for specific globally or locally coupled map systems. For two-node coupled chaotic map lattices with identical piecewise linear expanding maps,

$$x_{n+1}(i) = (1-\epsilon) f(x_n(i)) + \frac{\epsilon}{N-1} \sum_{j=1, j\neq i}^N f(x_n(j)),$$2

the transition between synchronization and intermittent-synchronization is controlled by the transverse Lyapunov exponent

$$x_{n+1}(i) = (1-\epsilon) f(x_n(i)) + \frac{\epsilon}{N-1} \sum_{j=1, j\neq i}^N f(x_n(j)),$$3

For the Lorenz map $$x_{n+1}(i) = (1-\epsilon) f(x_n(i)) + \frac{\epsilon}{N-1} \sum_{j=1, j\neq i}^N f(x_n(j)),$$4, the threshold is $$x_{n+1}(i) = (1-\epsilon) f(x_n(i)) + \frac{\epsilon}{N-1} \sum_{j=1, j\neq i}^N f(x_n(j)),$$5; for the slope-3 map, the threshold is $$x_{n+1}(i) = (1-\epsilon) f(x_n(i)) + \frac{\epsilon}{N-1} \sum_{j=1, j\neq i}^N f(x_n(j)),$$6. When $$x_{n+1}(i) = (1-\epsilon) f(x_n(i)) + \frac{\epsilon}{N-1} \sum_{j=1, j\neq i}^N f(x_n(j)),$$7, transverse perturbations are suppressed and synchronization occurs; when $$x_{n+1}(i) = (1-\epsilon) f(x_n(i)) + \frac{\epsilon}{N-1} \sum_{j=1, j\neq i}^N f(x_n(j)),$$8, intermittent-synchronization occurs, meaning that almost every orbit repeatedly approaches and leaves the diagonal (Zhang et al., 19 Jul 2025). The same work gives a necessary-sufficient condition linking uniqueness of the absolutely continuous invariant measure to intermittent synchronization.

Strong-coupling synchronization also has a continuum counterpart. For globally coupled expanding circle maps with discontinuous diffusive coupling, sufficiently small coupling yields a unique absolutely continuous invariant distribution, whereas sufficiently strong coupling drives a wide class of initial measures toward a point mass with support moving chaotically on the circle. The support shrinks exponentially fast, and the limiting behavior can be interpreted as synchronization in a chaotic state (Bálint et al., 2017). In this regime, the collective population collapses onto a single chaotic trajectory of the local map.

Global coupling can also organize bistable local dynamics into phase-ordered states. In a system of globally coupled maps with bistable chaotic local dynamics, the collective properties are described by a persistence probability for spin variables and by a magnetization-like order parameter. The persistence probability saturates for all values of the coupling parameter, and a discontinuous transition from a non-ordered state to a collective phase-ordered state takes place at a critical value of the coupling. On an interval of the coupling parameter, three distinct realizations of the phase-ordered state are observed and can be discerned by the corresponding values of the saturation persistence (Alvarez-Llamoza et al., 2014). A common misconception is therefore that mean-field coupling merely homogenizes the system; the phase-ordering results show that global interaction can also generate sharply distinct ordered macrostates.

3. Cluster states, chimera states, and asymmetric collective dynamics

Cluster and chimera states are central alternatives to full synchronization. For globally coupled map lattices, a cluster state is a configuration in which a subset of sites is perfectly synchronized, while a chimera state is a configuration in which a nontrivial synchronized subset coexists with desynchronized sites (Caby et al., 2023). In measure-based network formulations, chimera states correspond to invariant sets where some cluster measures are Dirac and others are non-Dirac (Sélley et al., 2021).

The principal local stability diagnostic for synchronized blocks is the transverse Lyapunov exponent. For a cluster space $$x_{n+1}(i) = (1-\epsilon) f(x_n(i)) + \frac{\epsilon}{N-1} \sum_{j=1, j\neq i}^N f(x_n(j)),$$9 in a globally coupled map lattice,

$x_n(i)$0

Negative $x_n(i)$1 implies transverse stability, while positive $x_n(i)$2 implies transverse instability (Caby et al., 2023). Numerical and analytical studies show attracting chimeras with chaotic dynamics as well as periodic behaviors. In strong coupling regimes, there are also results ruling out the existence of absolutely continuous invariant measures supported on attracting chimera states; under the strong-coupling condition given in the paper, the non-wandering set of the reduced cluster-space dynamics has zero Lebesgue measure (Caby et al., 2023). This excludes “fat” attracting chimera measures in that regime.

A related line of work studies asymmetric cluster and chimera dynamics in globally coupled systems. In an asymmetric chimera state, the synchronized subset follows a stationary or periodic trajectory while the desynchronized subset is chaotic; in an asymmetric cluster state, the periods of trajectories in different clusters are different (Cano et al., 2017). The key analytical device is the analogy between the effective local dynamics in the GCM and a single map subject to a constant drive,

$x_n(i)$3

together with the self-consistency condition

$x_n(i)$4

This permits prediction of parameter values and subset partitions for the formation of asymmetric cluster and chimera states (Cano et al., 2017).

Chimera dynamics in globally coupled sine circle map lattices makes the same point from a different direction. In a two-group globally coupled sine circle map lattice with intra-group coupling $x_n(i)$5 and inter-group coupling $x_n(i)$6, random initial conditions can evolve to chimera states in which one group is phase synchronized and the other is phase desynchronized, with spatiotemporal intermittency-like structures in the incoherent group. The complex order parameters $x_n(i)$7 and $x_n(i)$8 identify chimera, globally synchronized, two-clustered, and fully desynchronized states. In the chimera regime the two largest Lyapunov exponents are positive, so the state is hyperchaotic, and the distributions of laminar and burst lengths show exponential behavior (Singha et al., 2019). Global coupling therefore supports coexistence of coherence and high-dimensional chaos rather than eliminating one in favor of the other.

4. Invariant measures, ergodicity, and rigorous statistical structure

The statistical theory of GCMs relies heavily on transfer operators and invariant measures. For a class of globally coupled expanding circle maps in the continuum limit, there exist regularity parameters and a threshold $x_n(i)$9 such that for $i$0 the nonlinear transfer operator admits a unique invariant density $i$1. This density exponentially attracts all initial densities in the admissible regularity class in the $i$2 norm, and the map $i$3 is Lipschitz continuous in $i$4 (Bálint et al., 2017). These results show that weak mean-field coupling need not destroy the stable statistical structure inherited from the uncoupled expanding map.

For infinite systems of globally coupled Anosov diffeomorphisms with weak coupling strength, the state space is intrinsically infinite-dimensional and the transfer operator is self-consistent and nonlinear. Using transfer operators acting on anisotropic Banach spaces, one proves that the coupled system admits a unique physical invariant state $i$5, exponential convergence to equilibrium for a suitable class of distributions, and Lipschitz continuity of $i$6 in a strong Banach norm (Bahsoun et al., 2022). The underlying functional-analytic machinery includes Lasota–Yorke inequalities and contraction estimates on anisotropic spaces $i$7. This extends rigorous statistical analysis from uniformly expanding mean-field models to globally coupled hyperbolic dynamics.

Finite systems display a different but related phenomenon: ergodicity can break through the emergence of multiple invariant components. For four globally coupled doubling maps,

$i$8

the system has a unique acim for weak interaction, but there is a critical value

$i$9

at which multiple acims appear. At least six asymmetric invariant sets emerge, and the geometry of these sets is described through symmetry reduction, polyhedral constructions, and a centrally symmetric Lorenz map on special subspaces (Sélley, 2016). This is a rigorous instance of symmetry breaking in a globally coupled map.

The finite/infinite distinction is itself structurally important. For mean-field coupling of identical expanding circle maps, the distribution-based description permits treatment of infinitely many sites and reveals continuum behavior analogous to the limit states of the contracting regime of the three-site system, while finite-site ergodicity breaking associated with labeled site order does not directly survive in the continuum description (Sélley et al., 2015). A plausible implication is that some bifurcations of finite GCMs are fundamentally combinatorial, whereas the continuum limit selects the statistically stable aspects of the dynamics.

5. Turbulence, finite-size scaling, and chaotic itinerancy

Weakly coupled GCMs generically exhibit a hyperchaotic turbulent state, and one basic question is how the largest Lyapunov exponent depends on the system size $n$0. For turbulent GCMs with positive multipliers, analytical and numerical results show that

$n$1

rather than obeying a universal logarithmic law. The exponent $n$2 is parameter-dependent, with three regimes controlled by a tail index $n$3: $n$4 for $n$5, $n$6 for $n$7, and an even slower regime when $n$8 (Velasco et al., 2021). The explicit conclusion is that a universal convergence law for the largest Lyapunov exponent cannot be taken for granted in general GCMs.

Another hallmark of high-dimensional GCM dynamics is chaotic itinerancy among attractor-ruins. In a globally coupled logistic map model,

$n$9

the phase space is organized into coherent, ordered, partially ordered, and turbulent phases. The partially ordered phase is the regime where chaotic itinerancy occurs: clusterings are unstable, trajectories make long visits near clustering invariant sets, and then transition to others (Wada et al., 1 Oct 2025). The instability of orbits is analyzed through clustering encoded as probability vectors of cluster sizes, and changes in cluster structure are quantified by the optimal transport distance.

The same study evaluates the strength of attractor-ruins numerically by recording the effective dimension, defined as the number of clusters, at time steps where the optimal transport distance is zero, and then computing the Shannon entropy of the resulting distribution. The strength of various attractor-ruins is high in the partially ordered phase, where the time-averaged optimal transport distance is high and the system repeatedly approaches many different clustering invariant sets (Wada et al., 1 Oct 2025). This suggests that partially ordered mean-field chaos is characterized not merely by desynchronization, but by repeated reorganization of cluster structure across multiple metastable remnants.

6. Generalizations, analogues, and cross-domain uses

The GCM framework extends beyond memoryless maps. In coupled fractional Gauss maps, the discrete update contains a long-memory kernel,

$\epsilon$0

and the globally coupled fractional system is defined through the corresponding mean-field term. Numerical studies show that synchronization is observed over a large parameter region, synchronized periodic states with period-3 or period-6 occur even for a large lattice, and the standard deviation decays as a power law in time with the power same as fractional-order (Pakhare et al., 2022). This indicates that long-term memory changes both the speed and the type of collective ordering.

There is also a continuous-time analogue. A globally coupled lattice of Duffing flows (GCFL) is constructed as a natural extension of the globally coupled logistic map lattice. Its phase diagrams are essentially the same as those of GCML, including coherent chaos, two-clustered behavior, and turbulence. Similar to the two-clustered periodic attractor of GCML, the GCFL two-clustered attractor exhibits a successive period-doubling with an increase of population imbalance between the clusters, although the symmetry of the Duffing equation produces a non-trivial distinction between GCML and GCFL attractors (Shimada et al., 2011). The recurring appearance of the same phase structure across maps and flows supports the view that mean-field coupling organizes collective dynamics through mechanisms broader than any single local model.

The conceptual reach of GCML has also extended into optimization. In globally coupled particle swarm optimization, GCML is integrated into PSO by modifying the velocity update so that each particle is influenced by the positions of all birds, with the strength of the impact distinguished by the size of the weight. When $\epsilon$1, the method reduces to standard PSO; when $\epsilon$2, it uses full swarm information in a way structurally paralleling GCML (Yuan, 29 Aug 2025). This transfer does not redefine GCM theory, but it does show that the mean-field idea of global coupling can be used as a design principle outside dynamical-systems analysis.

Taken together, these developments establish GCMs as a common framework for collective chaos, synchronization, cluster formation, chimera dynamics, invariant-measure theory, and high-dimensional statistical behavior. The same mean-field architecture supports unique invariant densities and exponential convergence in weakly coupled regimes, multiple invariant components and symmetry breaking at stronger coupling, intermittent synchronization and chaotic synchronization, as well as turbulent and partially ordered phases with nontrivial finite-size effects (Bálint et al., 2017, Sélley, 2016, Zhang et al., 19 Jul 2025, Velasco et al., 2021, Wada et al., 1 Oct 2025).