On a stable torus in a 3D system with a saddle-focus

This paper proposes a conceptual model for the onset of a stable torus near a saddle-focus equilibrium. This bifurcation scenario is typical of slow-fast systems that generate elliptic bursting in a variety of neuronal models in mathematical neuroscience. Variants of the model also capture other dynamical regimes recurring in a neighborhood of the saddle-focus. We also discuss homoclinic bifurcations for which the model assumptions are feasible.

💡 Research Summary

The paper investigates the emergence of a stable two‑dimensional torus in a three‑dimensional slow‑fast dynamical system that contains a saddle‑focus equilibrium. The motivation comes from neuronal models that display elliptic bursting, a phenomenon where a slow variable modulates a fast relaxation oscillator, producing alternating periods of quiescence and rapid spiking. The authors begin by presenting a canonical three‑dimensional extension of the Van der Pol (or FitzHugh‑Nagumo) oscillator, written as system (1). In this model the slow variable y evolves on a timescale µ≪1, while the fast subsystem (v,w) exhibits a cubic nullcline that generates hysteresis and a subcritical Andronov‑Hopf bifurcation. By varying the parameter c, the slow nullcline (the plane v = y – c) moves relative to the slow‑motion manifold MLC, which consists of stable and unstable periodic orbits of the fast subsystem. When the nullcline intersects the inner cone of MLC, the periodic orbit loses stability and two invariant circles appear in a transverse cross‑section, signalling the birth of a torus.

To analyze the dynamics near the saddle‑focus, the authors linearize the system around the equilibrium O, obtaining the normal form (3)–(4) with eigenvalues ρ±iω (unstable spiral) and –λ (stable direction). They introduce two cross‑sections S₀ (z=1, r≤1) and S₁ (r=1, 0≤z≤1) and derive the local return map T₀:S₀→S₁, which in polar coordinates yields r(t)=r₀e^{ρt}, φ(t)=ωt+φ₀, z(t)=e^{-λt}. The transition time τ satisfies r₀e^{ρτ}=1, giving τ=−(1/ρ)ln r₀, and the map is expressed as z₁=r₀^{ν}, φ₁=φ₀+ωρln(1/r₀) (mod 2π) with ν=λ/ρ>1. The condition ν>1 is crucial: it guarantees that the stable direction contracts faster than the unstable spiral expands, a prerequisite for Shilnikov‑type complex dynamics.

The global dynamics are captured by a second map T₁ that connects the exit from the neighbourhood of O back to the entry section S₀. T₁ is written in a parametric form involving a small bifurcation parameter μ and smooth periodic functions α(φ) and β(φ). The composition T=T₀∘T₁ yields the combined map (6): (\bar z = \alpha(\phi)^{\nu} + \dots), (\bar\phi = \phi + \omega\rho\ln\frac{1}{\alpha(\phi)} + \Omega(\mu) + \dots) (mod 2π), where Ω(μ)→∞ as μ→0. After a scaling of the z‑coordinate, the map reduces to a skew‑product where the angular dynamics are governed by a circle map (\bar\phi = \phi + \omega\rho\ln\frac{1}{\alpha(\phi)} + \Omega(\mu)).

Three propositions are proved under mild smoothness assumptions:

• Proposition I (existence of a stable torus): If the derivative of the logarithmic term satisfies (\omega\rho,\alpha’(\phi)/\alpha(\phi) < 1) for all φ, then for sufficiently small μ the map possesses an attracting invariant closed curve (z = h(\phi;\mu)). This curve lifts to a smooth invariant torus in the original flow, providing a rigorous mechanism for torus creation near a saddle‑focus.

• Proposition II (hyperbolic chaotic sets): If on some interval I the function α(φ) varies sufficiently strongly so that the logarithmic stretch exceeds a multiple of 2π, then T contains a hyperbolic set conjugate to a full shift on m symbols (m≥2). This describes the torus‑breakdown scenario where the invariant circle fragments into a Cantor‑like chaotic attractor.

• Proposition III (stable periodic orbit): In the case where the global map does not wrap the unstable manifold around the origin (n=0), the fixed point of the reduced one‑dimensional map (\bar\phi = \phi + \omega\rho\ln\frac{1}{\alpha(\phi)} + \Omega(\mu)) is attracting provided (|\omega\rho,\alpha’(\phi)/\alpha(\phi)|<1). Consequently the full system admits a stable periodic orbit, reminiscent of a blue‑sky catastrophe.

The paper situates these results within the broader “Shilnikov funnel” framework, where the two‑dimensional unstable manifold of the saddle‑focus creates a funnel‑shaped region that can host regular, quasiperiodic, or chaotic dynamics depending on the geometry of the return maps. By linking the funnel construction to annulus maps studied by Afraimovich, Shilnikov, and others, the authors provide a unified description of several bifurcation phenomena observed in neuronal bursting, fluid turbulence, and other applications. The analysis also highlights how the presence of rotational symmetry (as in the example α(φ)=1+a sin φ) leads to Arnold tongues, resonant zones, and torus breakdown via period‑doubling cascades or homoclinic tangencies.

In conclusion, the work offers a mathematically rigorous yet conceptually accessible model for the birth of a stable torus near a saddle‑focus, delineates precise conditions for its persistence, and explains how slight parameter variations can trigger transitions to chaos or to a simple periodic orbit. These insights deepen our understanding of elliptic bursting and related oscillatory phenomena in neuroscience and provide a template for exploring similar mechanisms in higher‑dimensional slow‑fast systems.