Globally Coupled Logistic Maps

• Globally Coupled Logistic Maps are nonlinear discrete systems where each element is updated via a logistic map combined with a mean-field term, leading to rich collective dynamics.
• They exhibit phenomena like synchronization, clustering, bifurcations, and chaotic itinerancy, making them pivotal for understanding complex behavior in high-dimensional systems.
• The interplay of local nonlinearity, coupling strength, and cluster formation has been rigorously analyzed and experimentally validated, revealing multistability and finite-size effects.

Globally coupled logistic maps (GCM), often also discussed under the broader label of globally coupled map lattices when the local map is logistic or quadratic-logistic, are discrete-time dynamical systems in which each degree of freedom is updated by a nonlinear logistic map together with a mean-field interaction involving all units. In canonical form one writes

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

with either the quadratic logistic form $f_a(x)=1-a x^2$ or the standard logistic form $f(x)=\alpha x(1-x)$; these parameterizations are topologically conjugate in the unimodal quadratic class and underpin much of the modern theory of synchronization, clustering, collective chaos, and chaotic itinerancy in high-dimensional nonlinear dynamics (Mierski et al., 30 Jul 2025, Shimada et al., 2011).

1. Canonical formulation and state-space organization

A standard Kaneko-type GCM uses the quadratic local map

$$f_a(x)=1-a x^2, \qquad x_{n+1}(i)=(1-\varepsilon)f_a(x_n(i))+\frac{\varepsilon}{N}\sum_{j=1}^{N}f_a(x_n(j)),$$

with $a$ controlling local nonlinearity and $\varepsilon$ the global coupling strength (Mierski et al., 30 Jul 2025, Shimada et al., 2011). A widely used alternative writes the local dynamics as

$f(x)=\alpha x(1-x),$

so that the globally coupled logistic system evolves on $[0,1]^N$ under

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

or, in a closely related variant,

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

which excludes the self-term from the mean field and replaces $f_a(x)=1-a x^2$0 by $f_a(x)=1-a x^2$1 in the normalization (Wada et al., 1 Oct 2025, Yuan, 29 Aug 2025). The distinction is structurally minor in large systems but relevant when comparing precise definitions.

The mean-field structure produces a large family of invariant subspaces corresponding to cluster partitions. In the $f_a(x)=1-a x^2$2 formulation, a permutation $f_a(x)=1-a x^2$3 defines an invariant set

$f_a(x)=1-a x^2$4

so cluster states can be treated as symmetry-restricted dynamics on lower-dimensional manifolds (Wada et al., 1 Oct 2025). A local “effective dimension”

$f_a(x)=1-a x^2$5

then provides a coarse-grained count of the number of clusters near a given state (Wada et al., 1 Oct 2025). In this sense, GCM phase space is naturally stratified by synchronization patterns rather than by geometry in the usual lattice sense.

The two-map case already contains the essential mean-field symmetry. For $f_a(x)=1-a x^2$6,

$f_a(x)=1-a x^2$7

which is exactly the globally coupled form specialized to two units and has been used as an analytically tractable prototype for symmetry breaking, invariant manifolds, and bifurcation analysis (Mareno et al., 2020, Romero et al., 2019).

2. Collective regimes, clustering, and synchronization

Classical GCM phenomenology is organized into a small number of collective regimes. In the Kaneko classification, one encounters a coherent phase, an ordered phase, a partially ordered phase, and a desynchronized phase (Ohara et al., 2020). In the quadratic-map formulation these appear as coherent chaos, two-clustered periodic states, periodic-cluster regimes, and turbulent states as the coupling is decreased at fixed local nonlinearity (Shimada et al., 2011). The large-$f_a(x)=1-a x^2$8 end is characterized by full synchronization, so the entire ensemble collapses onto the dynamics of a single map. At weaker coupling, cluster states emerge, and at still smaller coupling the system enters turbulent many-cluster dynamics.

A central structure is the two-clustered periodic regime. There the population splits into two synchronized groups, and the cluster-population ratio

$f_a(x)=1-a x^2$9

acts as an internal bifurcation parameter. In GCML phase diagrams, the balanced case $f(x)=\alpha x(1-x)$0 yields a maximally symmetric two-cluster attractor, while departures from balance produce a successive period-doubling cascade of the cluster orbit, the so-called $f(x)=\alpha x(1-x)$1-bifurcation (Shimada et al., 2011). This makes cluster population itself a dynamical control parameter, not merely a state descriptor.

Closely related globally coupled logistic-type systems confirm that synchronization can also be induced by size. In a multiplicative mean-field model for symbiotic logistic-type maps, the case $f(x)=\alpha x(1-x)$2, $f(x)=\alpha x(1-x)$3 changes qualitatively when $f(x)=\alpha x(1-x)$4 is increased from $f(x)=\alpha x(1-x)$5 to $f(x)=\alpha x(1-x)$6: the mean field develops a clear period-2 oscillation and the single-unit dynamics becomes entrained to it, an explicit example of size-induced synchronization (Lopez-Ruiz, 2019). This suggests that finite-population effects in GCMs are not limited to fluctuation amplitudes; they can reorganize the attractor itself.

3. Chaotic itinerancy and attractor-ruins

One of the most distinctive dynamical signatures of GCMs is chaotic itinerancy: trajectories wander among quasi-stable collective states, stay there for irregular residence times, and then depart through chaotic transition episodes. In an entropy-and-clustering analysis of the quadratic GCM

$f(x)=\alpha x(1-x)$7

local Shannon entropy and local permutation entropy were used to identify parameter regions likely to support itinerancy, and density-based clustering then separated dense “attractor-ruin” regions from sparse transition states (Mierski et al., 30 Jul 2025). For $f(x)=\alpha x(1-x)$8, a broad band in the $f(x)=\alpha x(1-x)$9-plane running approximately from $$f_a(x)=1-a x^2, \qquad x_{n+1}(i)=(1-\varepsilon)f_a(x_n(i))+\frac{\varepsilon}{N}\sum_{j=1}^{N}f_a(x_n(j)),$$0 to $$f_a(x)=1-a x^2, \qquad x_{n+1}(i)=(1-\varepsilon)f_a(x_n(i))+\frac{\varepsilon}{N}\sum_{j=1}^{N}f_a(x_n(j)),$$1 was identified as the region where chaotic itinerancy is most pronounced, matching Kaneko’s intermittent or partially ordered phase (Mierski et al., 30 Jul 2025).

At a representative point $$f_a(x)=1-a x^2, \qquad x_{n+1}(i)=(1-\varepsilon)f_a(x_n(i))+\frac{\varepsilon}{N}\sum_{j=1}^{N}f_a(x_n(j)),$$2, $$f_a(x)=1-a x^2, \qquad x_{n+1}(i)=(1-\varepsilon)f_a(x_n(i))+\frac{\varepsilon}{N}\sum_{j=1}^{N}f_a(x_n(j)),$$3, $$f_a(x)=1-a x^2, \qquad x_{n+1}(i)=(1-\varepsilon)f_a(x_n(i))+\frac{\varepsilon}{N}\sum_{j=1}^{N}f_a(x_n(j)),$$4, DBSCAN identifies 12 clusters that organize into three 4-cycles, and these are naturally grouped into three attractor ruins. Time series segments assigned to those ruins are ordered and low-dimensional, whereas unassigned points correspond to chaotic transition states (Mierski et al., 30 Jul 2025). Statistical tests on the sequence of visited ruins and on the residence times were then used to show that the switching process is irregular but stationary rather than simply periodic (Mierski et al., 30 Jul 2025).

A complementary description uses cluster statistics and optimal transport. In the standard logistic-map formulation,

$$f_a(x)=1-a x^2, \qquad x_{n+1}(i)=(1-\varepsilon)f_a(x_n(i))+\frac{\varepsilon}{N}\sum_{j=1}^{N}f_a(x_n(j)),$$5

an attractor-ruin is treated as the remnant of a destabilized cluster attractor and is associated with a Milnor attractor (Wada et al., 1 Oct 2025). Cluster configurations are compressed into a probability vector over cluster sizes, and an optimal-transport distance between successive cluster distributions detects when the system is stationary in cluster space. The entropy of the effective-dimension distribution at zero-transport times then measures the “strength” of attractor-ruins. This strength is maximal in the partially ordered phase, especially in the regime labeled Partially Ordered Phase II, where chaotic itinerancy is strongest (Wada et al., 1 Oct 2025).

These analyses correct a common oversimplification of chaotic itinerancy as a merely visual phenomenon. In GCMs it can be formalized through state-space clustering, entropy variance, residence-time statistics, and transport-based cluster diagnostics (Mierski et al., 30 Jul 2025, Wada et al., 1 Oct 2025).

4. Rigorous results, invariant structures, and finite-size scaling

Fully rigorous results for standard logistic GCMs remain limited, but two complementary lines of work are established. First, the two-site symmetrically coupled logistic map admits a detailed invariant-set analysis. For

$$f_a(x)=1-a x^2, \qquad x_{n+1}(i)=(1-\varepsilon)f_a(x_n(i))+\frac{\varepsilon}{N}\sum_{j=1}^{N}f_a(x_n(j)),$$6

the diagonal $$f_a(x)=1-a x^2, \qquad x_{n+1}(i)=(1-\varepsilon)f_a(x_n(i))+\frac{\varepsilon}{N}\sum_{j=1}^{N}f_a(x_n(j)),$$7 is invariant and carries the usual logistic dynamics (Romero et al., 2019). In explicit parameter ranges there exists a positively invariant Jordan curve $$f_a(x)=1-a x^2, \qquad x_{n+1}(i)=(1-\varepsilon)f_a(x_n(i))+\frac{\varepsilon}{N}\sum_{j=1}^{N}f_a(x_n(j)),$$8 containing $$f_a(x)=1-a x^2, \qquad x_{n+1}(i)=(1-\varepsilon)f_a(x_n(i))+\frac{\varepsilon}{N}\sum_{j=1}^{N}f_a(x_n(j)),$$9 that bounds the immediate basin of infinity and organizes the bounded set. For small coupling and suitable $a$0, every bounded orbit synchronizes to the diagonal; in other regimes off-diagonal fixed points, Cantor sets, annular basin components, and numerically observed fat attractors appear (Romero et al., 2019). This two-dimensional prototype already exhibits the synchronization manifold, symmetry breaking, and fractal basin geometry that recur in larger GCMs.

Second, rigorous continuum-limit results are known for globally coupled expanding circle maps. In that setting the state is a probability density, the weak-coupling regime admits a unique absolutely continuous invariant distribution that attracts all admissible initial densities exponentially fast in the $a$1 norm, and the invariant density depends Lipschitz-continuously on $a$2; for sufficiently strong coupling, a broad class of initial measures converges in Wasserstein distance to a moving Dirac mass, interpreted as synchronization in a chaotic state (Bálint et al., 2017). These theorems apply to uniformly expanding circle maps rather than logistic maps, but they provide the most complete rigorous mean-field picture currently available within the coupled-map framework.

At the finite-size level, the largest Lyapunov exponent in turbulent GCMs is not universal. For globally coupled maps with positive multipliers, the asymptotic law

$a$3

replaces the previously proposed universal logarithmic correction, and the exponent $a$4 depends on the multiplier statistics (Velasco et al., 2021). For strongly dissimilar multipliers, the convergence becomes even slower and no universal law is supported numerically (Velasco et al., 2021). Since standard logistic maps have derivatives of both signs, the positive-multiplier analysis does not directly resolve the classical logistic case, but it establishes that universal finite-size scaling of $a$5 cannot be assumed even in paradigmatic GCMs (Velasco et al., 2021).

5. Delays, adaptive couplings, and low-dimensional bifurcation theory

Several important generalizations preserve the GCM mean-field spirit while changing the effective collective dynamics. A delayed, distance-dependent logistic-map lattice introduces a coupling range parameter $a$6 and a delay parameter $a$7, with the global-coupling limit at $a$8. In the completely synchronized state,

$a$9

so the synchronized manifold reduces to a two-dimensional non-delayed map independent of $\varepsilon$0 (Anteneodo et al., 2018). Moderate delay regularizes chaotic orbits and makes synchronization possible even for short-range coupling, while stronger delay introduces genuinely two-dimensional phenomena such as Neimark–Sacker bifurcations and quasiperiodic synchronized states (Anteneodo et al., 2018). The globally coupled case is therefore not just a stronger version of the undelayed model; delay changes the internal dimensionality of collective motion.

At the smallest nontrivial scale, strong mean-field coupling in two globally coupled logistic maps produces explicitly computable local bifurcations. The asymmetric fixed points $\varepsilon$1 of the symmetric two-map system undergo a reverse flip bifurcation and a Neimark–Sacker bifurcation as the coupling $\varepsilon$2 is varied (Mareno et al., 2020). In the reverse flip, unstable period-2 orbits collide with the symmetry-broken fixed point and a stable 1-cycle is recovered as coupling increases; on the Neimark–Sacker curve, the asymmetric fixed point loses stability to an invariant closed curve, giving quasiperiodic motion and then more complicated attractors (Mareno et al., 2020). These results show that coupling strength can act as a primary bifurcation parameter, not merely as a synchronizing perturbation.

Global coupling can also be generalized from a static complete graph to adaptive mean-field networks. In a delayed Hebbian extension of coupled logistic maps,

$\varepsilon$3

uniform weights $\varepsilon$4 recover a standard GCM, while adaptive weights generate ten distinct network states, including pacemaker networks, paired layers, loop networks, hidden modular states, and hidden randomly connected states (Ohara et al., 2020). The familiar coherent, ordered, partially ordered, and desynchronized dynamical phases remain visible, but the coupling architecture becomes part of the emergent state (Ohara et al., 2020). A different logistic-type extension uses a multiplicative global mean field that modulates the effective growth parameter rather than adding an additive mean-field term, again producing synchronization and size-induced collective order (Lopez-Ruiz, 2019).

6. Experimental realizations and empirical hysteresis

Electronic realizations have made GCM-like dynamics experimentally accessible with high precision. A modular analog implementation of the logistic map based on an AD633 analog multiplier plus a sample-and-hold block realizes

$\varepsilon$5

with parameter control at $\varepsilon$6 resolution and reproduces the single-map bifurcation diagram and Lyapunov exponent with strong agreement between experiment and simulation (L'Her et al., 2015). A coupling block then implements

$\varepsilon$7

so the same architecture extends directly to arbitrary networks and, in the complete-graph case, to finite GCMs (L'Her et al., 2015).

A six-map experimental system with configurable diffusive coupling realized 52 distinct network topologies, including the complete graph, and operated with a signal-to-noise ratio of approximately $\varepsilon$8 (Gutiérrez et al., 2020). The isolated circuits reproduced Feigenbaum scaling with less than 8% deviation in $\varepsilon$9, while the coupled system showed robust multistable regions and hysteresis even in the presence of measured parameter mismatch and unavoidable electronic noise (Gutiérrez et al., 2020). In complete and ring-like configurations, hysteresis was observed under slow sweeps of the coupling strength $f(x)=\alpha x(1-x),$0, and across all 52 configurations a strong-coupling hysteresis around the period-2 to fixed-point transition occurred near $f(x)=\alpha x(1-x),$1 (Gutiérrez et al., 2020). The experiment therefore confirms that multistability in coupled logistic maps is not a fine-tuned numerical artifact; it survives realistic imperfections.

Experimental work also sharpens the distinction between synchronization and identicality. In the six-map setup, the effective parameters $f(x)=\alpha x(1-x),$2 differed slightly across circuits, so the exact synchronization manifold was no longer strictly invariant; nonetheless, the average pairwise variance

$f(x)=\alpha x(1-x),$3

still approached very small values over substantial coupling intervals, providing an operational notion of approximate collective synchronization in finite, imperfect GCM realizations (Gutiérrez et al., 2020).

Globally coupled logistic maps thus occupy a distinctive position in nonlinear science: simple enough to admit explicit mean-field formulations, cluster manifolds, and in some cases exact bifurcation calculations, yet rich enough to support synchronization, symmetry breaking, chaotic itinerancy, multistability, attractor-ruins, and unresolved finite-size questions. Their continuing relevance stems precisely from that combination of minimal local dynamics and nontrivial collective organization.