A Phase Description of Mutually Coupled Chaotic Oscillators

Abstract: The synchronization of rhythms is ubiquitous in both natural and engineered systems, and the demand for data-driven analysis is growing. When rhythms arise from limit cycles, phase reduction theory shows that their dynamics are universally modeled as coupled phase oscillators under weak coupling. This simple representation enables direct inference of inter-rhythm coupling functions from measured time-series data. However, strongly rhythmic chaos can masquerade as noisy limit cycles. In such cases, standard estimators still return plausible coupling functions even though a phase-oscillator model lacks a priori justification. We therefore extend the phase description to the chaotic oscillators. Specifically, we derive a closed equation for the phase difference by defining the phase on a Poincaré section and averaging the phase dynamics over invariant measures of the induced return maps. Numerically, the derived theoretical functions are in close agreement with those inferred from time-series data. Consequently, our results justify the applicability of phase description to coupled chaotic oscillators and show that data-driven coupling functions retain clear dynamical meaning in the absence of limit cycles.

• The paper derives a closed phase-difference equation for weakly mutually coupled chaotic oscillators by combining cell-based phases, amplitude return maps, invariant measures, and averaging.
• Numerical tests on coupled Lorenz and Rössler–Sprott–N systems show close agreement between theoretical and data-inferred coupling functions, even when standard limit-cycle assumptions fail.
• The theory explains a chaos-specific deformation of the Arnold tongue caused by coupling-dependent invariant measures, while remaining limited by strong rhythmicity and near-synchronization assumptions.

Phase reduction theory provides a universal low-dimensional description of coupled rhythmic systems, but its derivation rests on the existence of limit cycles. The paper "A Phase Description of Mutually Coupled Chaotic Oscillators" (2602.17519) addresses a fundamental gap: rhythmic time series generated by chaotic attractors can be statistically indistinguishable from noisy limit-cycle oscillations, yet standard phase reduction has no a priori justification for them. Furukawa, Imai, and Aoyagi derive, for the first time, a closed equation for the phase difference between two weakly mutually coupled chaotic oscillators, and show numerically that it matches coupling functions inferred from data by methods that assume a limit cycle.

Motivation and problem statement

The starting observation is a methodological inconsistency. Given nearly periodic signals with sharp spectral peaks, a practitioner would apply a standard data-driven procedure—define phases via Poincaré sections, then infer an effective interaction function $$\Gamma_{\rm odd}(\psi) = \Gamma_{12}(-\psi) - \Gamma_{21}(\psi)$$ for the phase difference $\psi = \phi_1 - \phi_2$. The authors demonstrate that such plausible inferred functions can arise even when the underlying system is a pair of mutually coupled chaotic oscillators, for which the limit-cycle hypothesis underlying phase reduction fails entirely. Because experimental noise makes limit-cycle versus chaotic classification genuinely difficult—for example, whether EEG or cardiac rhythms are truly chaotic remains debated—the validity of data-driven coupling inference on rhythmic chaos is an open practical concern.

Chaotic oscillators do exhibit phase synchronization (as documented since Lorenz and Rössler), suggesting that some phase-like description should exist; the contribution here is to construct one rigorously at the level of the phase difference.

Construction of the theory

The central difficulty is that chaotic attractors lack isochrons, so no well-defined phase response $\partial \phi / \partial X$ exists and perturbative calculation is impossible. The authors' workaround proceeds as follows:

• Cell-based phase assignment: A transverse cross section $\mathcal{S}_a$ with small return-time variability replaces the isochron. Phases growing linearly in time are assigned only at finitely many points (cell centers) indexed by $i$, with $\omega_a(i) = 2\pi/T_a(i)$, and interpolated smoothly elsewhere.
• Amplitude return maps: The remaining $N{-}1$ degrees of freedom form an "amplitude" $R_a$, whose deviation dynamics are linearized to $O(\varepsilon)$ around uncoupled orbits. This yields discrete maps $f_a$ taking the amplitude from one crossing of $\psi = \phi_1 - \phi_2$0 to the next.
• Invariant measures: Two measures, $\psi = \phi_1 - \phi_2$1 and $\psi = \phi_1 - \phi_2$2, are defined as invariant measures of two distinct induced return maps on $\psi = \phi_1 - \phi_2$3, capturing the chaotic statistics of amplitude fluctuations.
• Averaging: The cell-dependent phase equation is averaged over these measures, producing a closed equation

$\psi = \phi_1 - \phi_2$4

A technically important subtlety is that which measure applies depends on $\psi = \phi_1 - \phi_2$5 and $\psi = \phi_1 - \phi_2$6: because crossings alternate ($\psi = \phi_1 - \phi_2$7), the measure $\psi = \phi_1 - \phi_2$8 governs part of each cycle and $\psi = \phi_1 - \phi_2$9 the remainder. The framework builds on prior work on periodically forced single chaotic oscillators (2602.17519), extending it to mutual coupling.

Numerical validation

The validation strategy is deliberately adversarial to the theory's own premise: coupling functions are inferred from time series using Gaussian process regression under the explicit assumption of a limit-cycle model, then compared against the theoretical prediction.

For the coupled Lorenz system with $\partial \phi / \partial X$0 and oscillator mismatch parameter $\partial \phi / \partial X$1, the inferred function agrees closely with the theoretical interaction function across multiple parameter sets. More strikingly, the theory reproduces a chaos-specific distortion of the Arnold tongue for $\partial \phi / \partial X$2: weak coupling perturbs the trajectories and hence the invariant measures themselves, modifying both the averaged frequency and interaction terms. This mechanism has no counterpart in standard limit-cycle phase reduction, where the Arnold tongue boundary is not deformed through measure dependence.

Applicability to heterogeneous systems is demonstrated with a Rössler–Sprott–N pair. Here simple planar sections produce excessive return-time variability, so the authors employ data-driven optimal isophases as Poincaré sections. Theory and inference remain in quantitative agreement both inside and outside the synchronization region, confirming that appropriate section choice extends the method to attractors of very different geometry.

Assumptions, limitations, and open questions

The reduction carries several substantive assumptions stated plainly by the authors. It requires cross sections with small return-time variability—a condition guaranteed only by the "strongly rhythmic" nature of the attractor—and presumes the existence of invariant measures for the induced return maps. The averaging treats the phase difference as constant within each integration window, valid only near 1:1 synchronization where a time-scale separation holds; fluctuations of $\partial \phi / \partial X$3 are handled only post hoc via a moving average. Perturbation theory is carried to leading order in $\partial \phi / \partial X$4, ignoring return-time fluctuations of that order. Extensions to networks of three or more oscillators are described as conceptually straightforward but computationally expensive, and the effective interactions may depend on phases beyond the oscillator pair in question; a systematic network-level reduction is explicitly left open. The paper also concedes that the choice of which invariant measure to use involves a subtle $\partial \phi / \partial X$5- and $\partial \phi / \partial X$6-dependent issue deferred to the Supplemental Material.

Conclusion

This work establishes that coupled chaotic oscillators admit a closed, quantitatively accurate phase-difference description despite lacking isochrons, thereby giving dynamical meaning to coupling functions inferred from data even when no limit cycle exists. The reproduction of the chaos-specific Arnold tongue distortion shows that the extension is not merely a formal unification but captures behavior inaccessible to classical phase reduction. The main open questions are the extension to larger networks with multi-oscillator phase dependence and a systematic treatment of the regime where time-scale separation between amplitude and phase-difference dynamics breaks down.