Bekki–Nozaki Equation and Chaotic Solitons

• Bekki–Nozaki equation is a forced dissipative variant of the nonlinear Schrödinger and cubic Ginzburg–Landau equations, modeling chaotic soliton dynamics with attractor behavior.
• It incorporates time-periodic forcing, dissipative terms, and nonlinearities to simulate phenomena such as soliton bouncing, limit cycle attractors, and quasi-integrable dynamics.
• The framework underpins applications in semiconductor ring lasers and polariton condensates, with physics-informed neural networks used for both forward simulation and inverse parameter analysis.

Searching arXiv for recent and relevant papers on the Bekki–Nozaki equation and closely related Nozaki–Bekki soliton literature. The Bekki–Nozaki equation denotes, in one important contemporary usage, a forced dissipative variant of the focusing nonlinear Schrödinger equation (NLS), and, in another, the cubic complex Ginzburg–Landau equation (CGLE) in the parameterization used for Nozaki–Bekki hole solutions. In the forced-dissipative NLS setting, it takes the form

$$iq_t + q_{xx} + 2|q|^2q = i\varepsilon_1 \exp(i\omega t) + i\varepsilon_2 \exp(2i\omega t) + i\gamma q_{xx},$$

with $q=q(x,t)$, small real constants $\varepsilon_1,\varepsilon_2,\gamma$, and positive forcing frequency $\omega$ (Sawado et al., 8 Jul 2025). In the CGLE setting, the same historical attribution is attached to

$$\frac{\partial E}{\partial t} = E + (1+i c_{\mathrm{D}})\,\frac{\partial^2 E}{\partial z^2} - (1+i c_{\mathrm{NL}})\,|E|^2 E,$$

which is the form used to describe Nozaki–Bekki solitons in semiconductor ring lasers (Opačak et al., 2023). Across these settings, the Bekki–Nozaki name is associated with the breakdown of integrability, the emergence of attractors or limit cycles, traveling hole-like coherent structures, and sensitivity characteristic of dissipative nonlinear wave dynamics (Sawado et al., 8 Jul 2025, Opačak et al., 2023).

1. Historical usage and model identity

The term “Bekki–Nozaki equation” is not confined to a single canonical PDE in the material considered here. In the 2025 study of PINNs and chaotic dynamics, it refers explicitly to the “forced dissipative non-linear Schrödinger equation,” introduced by Bekki and Nozaki, in which the integrable focusing NLS is perturbed by time-periodic forcing and a dissipative term proportional to $q_{xx}$ (Sawado et al., 8 Jul 2025). In the 2023 optical-soliton study, the authors explicitly employ the cubic CGLE and identify it as the “Bekki–Nozaki equation” in the context of Nozaki–Bekki hole solutions (Opačak et al., 2023).

This dual usage reflects a historical association with dissipative extensions of conservative envelope equations and with the coherent structures first derived by Nozaki and Bekki in the CGLE literature. The 2025 NLS-based work cites Bekki & Nozaki (1983, 1984) and follows their demonstration that suitably tuned forcing and dissipation can generate “solutions with an attractor or a limit cycle,” and that chaotic behavior can emerge in the time evolution (Sawado et al., 8 Jul 2025). The 2023 laser work places the theory in the classical cubic CGLE framework under which Nozaki & Bekki originally found traveling “hole” solutions (Opačak et al., 2023). A plausible implication is that the modern literature uses the name both for a specific forced-dissipative NLS perturbation and for the broader Bekki–Nozaki coherent-structure framework in dissipative nonlinear wave equations.

2. Forced dissipative nonlinear Schrödinger form

In the forced-dissipative NLS formulation, the equation studied is

$$iq_t + q_{xx} + 2|q|^2q = i\varepsilon_1 \exp(i\omega t) + i\varepsilon_2 \exp(2i\omega t) + i\gamma q_{xx},$$

where $q(x,t)$ is a complex field, $q_t$ is the first time derivative, and $q_{xx}$ is the second spatial derivative (Sawado et al., 8 Jul 2025). The term $q=q(x,t)$0 is the cubic Kerr-type nonlinearity of the NLS, while $q=q(x,t)$1 and $q=q(x,t)$2 control temporally periodic external forcing at the fundamental frequency $q=q(x,t)$3 and its second harmonic $q=q(x,t)$4, respectively; the forcing is added as pure-imaginary amplitudes $q=q(x,t)$5 and $q=q(x,t)$6 (Sawado et al., 8 Jul 2025). The parameter $q=q(x,t)$7 multiplies $q=q(x,t)$8 and is described as a linear dissipative or gain-type perturbation coupled to the second spatial derivative (Sawado et al., 8 Jul 2025).

The left-hand side is the standard focusing NLS operator, whereas the right-hand side adds explicit time dependence and dissipation (Sawado et al., 8 Jul 2025). The relation to the integrable NLS,

$q=q(x,t)$9

is central: the unperturbed equation admits exact soliton solutions and infinitely many conserved quantities, and the paper quotes the bright soliton

$\varepsilon_1,\varepsilon_2,\gamma$0

as a canonical example (Sawado et al., 8 Jul 2025). When “small dissipation and external forces are applied to the integrable NLS equation, the soliton … can continue to function as the attractor of the system” (Sawado et al., 8 Jul 2025). Such perturbations destroy integrability and produce quasi-integrable behavior, attractors, and chaos (Sawado et al., 8 Jul 2025).

The initial condition used in the PINN study is

$\varepsilon_1,\varepsilon_2,\gamma$1

with typical values $\varepsilon_1,\varepsilon_2,\gamma$2 and $\varepsilon_1,\varepsilon_2,\gamma$3 (Sawado et al., 8 Jul 2025). The paper focuses on soliton-type localized initial data on a one-dimensional spatial domain, with boundary conditions enforced by the boundary-loss term in the PINN rather than through explicitly specified spatial boundaries (Sawado et al., 8 Jul 2025).

3. Attractors, collapse, and chaotic sensitivity

The reported dynamics in the forced-dissipative NLS regime include attractors, periodic behavior, collapse, and pronounced sensitivity to parameters and initial conditions (Sawado et al., 8 Jul 2025). The phase-space trajectory is defined using

$\varepsilon_1,\varepsilon_2,\gamma$4

where $\varepsilon_1,\varepsilon_2,\gamma$5 is the position where $\varepsilon_1,\varepsilon_2,\gamma$6 attains its maximum; the trajectory is plotted in the phase space $\varepsilon_1,\varepsilon_2,\gamma$7 (Sawado et al., 8 Jul 2025). Contour plots of $\varepsilon_1,\varepsilon_2,\gamma$8 together with this reduced phase-space projection are used to visualize “bouncing,” attractors, and collapse (Sawado et al., 8 Jul 2025).

For $\varepsilon_1,\varepsilon_2,\gamma$9, $\omega$0, $\omega$1, and $\omega$2, an attractor is observed for $\omega$3, whereas trajectories “collapse as increasing time” for $\omega$4 and $\omega$5 (Sawado et al., 8 Jul 2025). A representative parameter set used throughout is $\omega$6, $\omega$7, $\omega$8, again with $\omega$9 and $$\frac{\partial E}{\partial t} = E + (1+i c_{\mathrm{D}})\,\frac{\partial^2 E}{\partial z^2} - (1+i c_{\mathrm{NL}})\,|E|^2 E,$$0; for this case, contour plots and attractor trajectories show “bouncing of norm and the presence of the attractor” (Sawado et al., 8 Jul 2025). The same parameter set also exhibits long-time decay, with the norm decaying and “the attractor collapses for more than 150-seconds” (Sawado et al., 8 Jul 2025).

A second parameter regime, labeled Set (B), is given by $$\frac{\partial E}{\partial t} = E + (1+i c_{\mathrm{D}})\,\frac{\partial^2 E}{\partial z^2} - (1+i c_{\mathrm{NL}})\,|E|^2 E,$$1, $$\frac{\partial E}{\partial t} = E + (1+i c_{\mathrm{D}})\,\frac{\partial^2 E}{\partial z^2} - (1+i c_{\mathrm{NL}})\,|E|^2 E,$$2, $$\frac{\partial E}{\partial t} = E + (1+i c_{\mathrm{D}})\,\frac{\partial^2 E}{\partial z^2} - (1+i c_{\mathrm{NL}})\,|E|^2 E,$$3, $$\frac{\partial E}{\partial t} = E + (1+i c_{\mathrm{D}})\,\frac{\partial^2 E}{\partial z^2} - (1+i c_{\mathrm{NL}})\,|E|^2 E,$$4, $$\frac{\partial E}{\partial t} = E + (1+i c_{\mathrm{D}})\,\frac{\partial^2 E}{\partial z^2} - (1+i c_{\mathrm{NL}})\,|E|^2 E,$$5, and $$\frac{\partial E}{\partial t} = E + (1+i c_{\mathrm{D}})\,\frac{\partial^2 E}{\partial z^2} - (1+i c_{\mathrm{NL}})\,|E|^2 E,$$6, for which the solution “keeps bouncing for a long time” (Sawado et al., 8 Jul 2025). The study emphasizes that “Since the chaotic behavior is quite sensitive to the initial condition, even a slight variation of the perturbation parameters with a fixed initial condition might result in significant changes later on” (Sawado et al., 8 Jul 2025). It does not report Lyapunov exponents, Poincaré sections, or spectral analyses; the diagnosis of chaos is qualitative and based on sensitivity to parameters, long-time attractor behavior, and decay (Sawado et al., 8 Jul 2025).

The numerical reference solutions are obtained by fourth-order Runge–Kutta with time steps $$\frac{\partial E}{\partial t} = E + (1+i c_{\mathrm{D}})\,\frac{\partial^2 E}{\partial z^2} - (1+i c_{\mathrm{NL}})\,|E|^2 E,$$7 and $$\frac{\partial E}{\partial t} = E + (1+i c_{\mathrm{D}})\,\frac{\partial^2 E}{\partial z^2} - (1+i c_{\mathrm{NL}})\,|E|^2 E,$$8, and long-time behavior is tracked up to $$\frac{\partial E}{\partial t} = E + (1+i c_{\mathrm{D}})\,\frac{\partial^2 E}{\partial z^2} - (1+i c_{\mathrm{NL}})\,|E|^2 E,$$9 (Sawado et al., 8 Jul 2025). The long-time behavior changes with the chosen time step, including norm decay and attractor collapse, which the authors treat as motivation for mesh-free PINNs in chaotic regimes (Sawado et al., 8 Jul 2025).

4. Bekki–Nozaki equation in the CGLE and Nozaki–Bekki holes

A distinct but historically connected formulation appears in the CGLE literature, where the “Bekki–Nozaki equation” is written as

$q_{xx}$0

(Opačak et al., 2023). In this setting, the entire parameter space reduces to the plane $q_{xx}$1, and the continuous-wave linear stability boundary is given by $q_{xx}$2, so that CW is linearly stable for $q_{xx}$3 (Opačak et al., 2023).

The 2023 work derives this CGLE from a laser master equation for a unidirectional semiconductor ring laser with fast gain recovery and giant resonant Kerr nonlinearity, with $q_{xx}$4 and $q_{xx}$5 collecting dispersion, gain curvature, linewidth enhancement factor effects, and nonlinear phase modulation (Opačak et al., 2023). The canonical cubic CGLE often used in pattern-formation literature,

$q_{xx}$6

maps to the authors’ parameters as $q_{xx}$7 and $q_{xx}$8 (Opačak et al., 2023).

Within this framework, Nozaki–Bekki holes are traveling localized depressions of amplitude on a finite-amplitude plane-wave background, with a localized phase kink and finite phase slip (Opačak et al., 2023). A representative explicit form given in the paper is

$q_{xx}$9

with

$$iq_t + q_{xx} + 2|q|^2q = i\varepsilon_1 \exp(i\omega t) + i\varepsilon_2 \exp(2i\omega t) + i\gamma q_{xx},$$0

and the hole existing in the CW linearly stable region, notably for $$iq_t + q_{xx} + 2|q|^2q = i\varepsilon_1 \exp(i\omega t) + i\varepsilon_2 \exp(2i\omega t) + i\gamma q_{xx},$$1 (Opačak et al., 2023). In the authors’ parameterization, $$iq_t + q_{xx} + 2|q|^2q = i\varepsilon_1 \exp(i\omega t) + i\varepsilon_2 \exp(2i\omega t) + i\gamma q_{xx},$$2 and $$iq_t + q_{xx} + 2|q|^2q = i\varepsilon_1 \exp(i\omega t) + i\varepsilon_2 \exp(2i\omega t) + i\gamma q_{xx},$$3 (Opačak et al., 2023).

These structures are realized experimentally in a ring semiconductor laser as traveling localized dark pulses on a finite-amplitude CW background (Opačak et al., 2023). The observed signatures include a localized intensity dip, a temporal phase ramp of magnitude $$iq_t + q_{xx} + 2|q|^2q = i\varepsilon_1 \exp(i\omega t) + i\varepsilon_2 \exp(2i\omega t) + i\gamma q_{xx},$$4 within the pulse width, and anti-phase synchronization in intermode phases, with two $$iq_t + q_{xx} + 2|q|^2q = i\varepsilon_1 \exp(i\omega t) + i\varepsilon_2 \exp(2i\omega t) + i\gamma q_{xx},$$5 jumps around the primary CW mode (Opačak et al., 2023). The work further reports multisoliton states and probable soliton-crystal behavior, including a probable fifth-harmonic comb with mode spacing $$iq_t + q_{xx} + 2|q|^2q = i\varepsilon_1 \exp(i\omega t) + i\varepsilon_2 \exp(2i\omega t) + i\gamma q_{xx},$$6 FSR $$iq_t + q_{xx} + 2|q|^2q = i\varepsilon_1 \exp(i\omega t) + i\varepsilon_2 \exp(2i\omega t) + i\gamma q_{xx},$$7 (Opačak et al., 2023).

5. Related dissipative structures, instability, and generalizations

The Bekki–Nozaki framework is also used to interpret long-lived quasi-stationary dark solitons in nonresonantly pumped exciton–polariton condensates (Xue et al., 2014). In that context, the open-dissipative Gross–Pitaevskii equation coupled to a reservoir can be reduced near threshold to a cubic–quintic CGLE-type amplitude equation, and the observed localized structure is described as a close analog of a Bekki–Nozaki heteroclinic hole rather than an exact solution of the idealized cubic CGLE (Xue et al., 2014). The paper states that the profile and dynamics closely resemble the Nozaki–Bekki hole solutions of the CGLE: a tanh-shaped density dip embedded in a plane-wave background, with a $$iq_t + q_{xx} + 2|q|^2q = i\varepsilon_1 \exp(i\omega t) + i\varepsilon_2 \exp(2i\omega t) + i\gamma q_{xx},$$8 phase jump and outgoing waves (Xue et al., 2014).

A key point in that work is the “acceleration instability” of NB holes due to higher-order structural perturbations (Xue et al., 2014). The effective quintic structural perturbation is

$$iq_t + q_{xx} + 2|q|^2q = i\varepsilon_1 \exp(i\omega t) + i\varepsilon_2 \exp(2i\omega t) + i\gamma q_{xx},$$9

and the sign $q(x,t)$0 is identified with abrupt decay after a long quasi-stationary stage (Xue et al., 2014). The observed phenomena of small oscillations, “soliton explosions,” and abrupt collapse are interpreted through this CGLE hole phenomenology (Xue et al., 2014). This suggests that the Bekki–Nozaki concept extends beyond exact coherent structures to a broader instability scenario for dissipative dark localized states.

A different generalization appears in coupled CGLEs, where the Bekki–Nozaki modified Hirota operator is used to derive exact domain-wall solutions (Yee et al., 2011). The modified operator is defined by

$q(x,t)$1

with the ordinary Hirota operator recovered at $q(x,t)$2 (Yee et al., 2011). The paper applies this BN operator, together with a factorization procedure, to a system of two coupled CGLEs and derives exact mutually locked front pairs with explicit amplitude, frequency, and velocity relations (Yee et al., 2011). In that usage, the Bekki–Nozaki contribution is methodological as well as dynamical: it provides a bilinearization framework for dissipative PDEs that are not integrable in the ordinary Hirota sense (Yee et al., 2011).

6. Physics-informed neural networks and inverse analysis

The 2025 study develops a PINN formulation for the forced-dissipative NLS form of the Bekki–Nozaki equation by splitting the complex field as $q(x,t)$3 and solving for $q(x,t)$4 and $q(x,t)$5 (Sawado et al., 8 Jul 2025). The real and imaginary operators are written explicitly as

$q(x,t)$6

$q(x,t)$7

with forward residuals

$q(x,t)$8

(Sawado et al., 8 Jul 2025). The physical residual entering the loss is

$q(x,t)$9

(Sawado et al., 8 Jul 2025).

The forward PINN loss is

$q_t$0

while the inverse loss for parameter discovery is

$q_t$1

(Sawado et al., 8 Jul 2025).

The forward network uses 4 hidden layers, 128 nodes per layer, and tanh activation, with $q_t$2, $q_t$3, and $q_t$4; Adam is used for early training and L-BFGS for high-precision convergence, with TensorFlow 2.18.0 and SciPy 1.15.1 (Sawado et al., 8 Jul 2025). Long-time integration is handled by dividing the time domain into 2-second blocks and solving sequentially to reach $q_t$5 (Sawado et al., 8 Jul 2025). The inverse PINN uses 6 hidden layers, 40 nodes per layer, tanh activation, and $q_t$6, with training data taken from the forward-PINN solution over $q_t$7 (Sawado et al., 8 Jul 2025).

The forward PINN reproduces the “bouncing of norm and the presence of the attractor,” and the phase-space trajectory shows the same qualitative features as RK4; the authors conclude that “This implies that the chaos in the Bekki-Nozaki equation is not a numerical artifact, but rather an inherent property of the system itself” (Sawado et al., 8 Jul 2025). In inverse mode, the perturbative parameters are accurately identified from chaotic data. For ground truth $q_t$8, $q_t$9, the reported identified values include $q_{xx}$0 with loss $q_{xx}$1, $q_{xx}$2 with loss $q_{xx}$3, and $q_{xx}$4 with loss $q_{xx}$5 (Sawado et al., 8 Jul 2025).

The same inverse framework is also used to estimate the NLS dispersion and nonlinearity coefficients $q_{xx}$6 with perturbations fixed. For Set A,

$q_{xx}$7

and for Set B,

$q_{xx}$8

with true values $q_{xx}$9, $q=q(x,t)$00 (Sawado et al., 8 Jul 2025). Using 10-second segments in Set A, $q=q(x,t)$01 and $q=q(x,t)$02 are “generally accurately estimated both before and after the solution dissipates,” whereas with finer 5-second segments discrepancies fluctuate when “the solution's norm decays and the attractor collapsed” (Sawado et al., 8 Jul 2025). For Set B, with sustained bouncing, 5-second segments “correctly estimated” the coefficients (Sawado et al., 8 Jul 2025).

7. Interpretation, misconceptions, and significance

A recurring misconception is that the dynamics attributed to the Bekki–Nozaki equation might be reducible to numerical artifacts. The 2025 PINN study addresses this directly by showing that a mesh-free, physics-informed method reproduces the same qualitative chaotic features—bouncing norms and attractor trajectories—as RK4, leading the authors to state that the chaos is an inherent property of the system rather than a discretization artifact (Sawado et al., 8 Jul 2025). At the same time, the study does not claim a full quantitative chaos characterization by standard dynamical-systems metrics; it relies on sensitivity to perturbation parameters, long-time attractor behavior, and decay, and explicitly does not report Lyapunov exponents, Poincaré sections, or spectral analyses (Sawado et al., 8 Jul 2025).

A second potential misconception is to treat all “Bekki–Nozaki” objects as the same PDE. The sources instead support a layered interpretation. In one strand, the name designates a forced dissipative NLS whose small forcing and dissipation destroy integrability and create attractors, bouncing, and chaotic sensitivity (Sawado et al., 8 Jul 2025). In another, it denotes the cubic CGLE parameterization in which Nozaki–Bekki hole solutions exist, including stable dark-pulse states in semiconductor ring lasers (Opačak et al., 2023). In yet another, it refers to a bilinear-operator formalism for dissipative PDEs, exemplified by the BN modified Hirota operator and its use in coupled CGLE domain walls (Yee et al., 2011).

The inverse-analysis results in the PINN study motivate a further interpretive point. The authors state that “The results of the inverse analysis indicate a correlation between the governing equation's predictability and its chaotic nature of the solution” (Sawado et al., 8 Jul 2025). More specifically, when the solution continues bouncing and the soliton acts as an attractor, the original NLS terms dominate and the inverse PINN reliably recovers them, whereas after soliton decay the external forcing dominates and shorter-than-$q=q(x,t)$03 time segments can “fail to correctly interpret the system's dynamics” (Sawado et al., 8 Jul 2025). This suggests that, in this formulation, “predictability” is not simply monotone with reduced complexity; it depends on which physical terms dominate the observed dynamics.

Taken together, the literature portrays the Bekki–Nozaki equation as a paradigmatic dissipative nonlinear-wave framework at the interface of soliton theory, attractor dynamics, and coherent-structure analysis. In the forced-dissipative NLS form it provides a setting for soliton-based attractors, parameter-sensitive collapse, and PINN-based inverse inference under chaos (Sawado et al., 8 Jul 2025). In the CGLE form it underlies structurally stable traveling dark pulses in semiconductor lasers and the classical Nozaki–Bekki hole family (Opačak et al., 2023). Related developments in polariton condensates and coupled CGLEs show that the Bekki–Nozaki label continues to organize a broader class of dissipative-wave phenomena, including acceleration instability, domain walls, and bilinear solution methods (Xue et al., 2014, Yee et al., 2011).