SNIC Bifurcation in Dynamical Systems

• SNIC bifurcation is defined by the collision of a saddle-node equilibrium with an invariant circle, resulting in the emergence or disappearance of periodic orbits with diverging periods.
• It features square-root scaling of the oscillation period and serves as a key mechanism in neuronal excitability, Josephson junction networks, and coupled oscillator systems.
• Its well-characterized normal forms and geometric signatures enable precise diagnosis and parameter estimation in both experimental and modeling contexts.

A saddle-node on invariant circle (SNIC) bifurcation is a codimension-one global bifurcation where a saddle-node equilibrium collides with a closed invariant curve (cycle), resulting in the annihilation of both equilibria and the emergence or disappearance of a family of periodic orbits whose period diverges at onset. SNIC bifurcations constitute a standard mechanism for the transition from excitable to oscillatory dynamics in a broad class of planar and networked systems, especially in Type I neuronal models and mixed oscillator populations. The defining scaling law is the square-root divergence of the oscillation period as the bifurcation point is approached, distinguishing SNIC from local bifurcations such as Hopf and global bifurcations such as the homoclinic loop. SNIC bifurcations possess canonical normal forms, robust geometric signatures, and substantial impact on phase response, synchronization, and collective dynamics.

1. Mathematical Formulation and Normal Forms

The canonical local normal form for the SNIC in planar systems takes

$\dot x = x^2 + \mu,\qquad \mu \in \mathbb{R},$

where for $\mu<0$ a node and a saddle exist on a closed invariant set, coalescing at $\mu=0$, and for $\mu>0$ a stable periodic orbit emerges with infinite period at onset. On the circle, this becomes

$\dot\phi = \mu + (1-\cos\phi),$

or, equivalently, $\dot\theta = \mu + C\sin\theta$ after rescaling and phase-shift. Here, $\mu$ serves as the unfolding parameter controlling approach to the bifurcation.

In neuronal networks, reduced mean-field models close to the SNIC critical point take the form

$$\begin{aligned} \dot\rho_e &= -\rho_e + \Psi(\rho_e,\rho_i;\langle n\rangle), \ \dot\rho_i &= -\rho_i + \Psi(\rho_e,\rho_i;\langle n\rangle), \ \end{aligned}$$

where $\langle n\rangle$ is the bifurcation parameter, and $\Psi$ implements thresholding of total input (Lopes et al., 2017).

In coupled Josephson junctions, the reduced dynamics near SNIC also conforms to the local quadratic normal form (Hens et al., 2015), establishing the universality class of this bifurcation.

2. Bifurcation Structure and Global Geometry

The geometric structure of SNIC bifurcations is characterized by the collision of a stable node and a saddle on a closed invariant curve. For $\mu<0$0, the system exhibits three fixed points: a stable node, an intermediate saddle (the threshold), and a high-activity unstable focus (in certain neuron models). As $\mu<0$1, the stable node and the saddle approach each other and annihilate on the invariant circle at $\mu<0$2. For $\mu<0$3, no fixed points remain on the circle, and the system admits a large-amplitude, continuous limit cycle—the oscillatory regime.

This scenario underlies sharp transitions such as the resting-to-spiking switch in class I neuron models (Hesse et al., 2016), the onset of seizure-like oscillations in mean-field neural networks (Lopes et al., 2017), and the emergence of parabolic bursting in globally coupled oscillator networks (Hens et al., 2015). The bifurcation diagrams typically show:

• A region of stable equilibrium,
• Emergence of bistability or intermittent behavior (in some parameter ranges),
• A unique onset of oscillations with period divergence at threshold,
• Further transitions (e.g., Hopf, homoclinic) at higher parameter values.

There is no associated hysteresis with the SNIC itself, though bistable dynamical regimes can occur in the pre-bifurcation region due to noise or multi-network interactions (Lopes et al., 2017, Baesens et al., 2013).

3. Period Divergence and Scaling Laws

A defining feature of the SNIC bifurcation is the divergence of the period $\mu<0$4 as the control parameter approaches the bifurcation value from above: $\mu<0$5 This universal scaling arises directly from the normal form: integrating $\mu<0$6 yields the square-root law (Hens et al., 2015, Hesse et al., 2016, Baesens et al., 2013, Lopes et al., 2017). In network and neuronal models, control variables such as mean endogenous spike rate $\mu<0$7 or bias current $\mu<0$8 play the role of $\mu<0$9, and period divergence corresponds to critical slowing-down and the ability to sustain arbitrarily slow oscillations at threshold (Lopes et al., 2017, Hens et al., 2015).

For stimulated systems, thresholds for spiking and interspike-intervals follow

$\mu=0$0

as $\mu=0$1 (Lopes et al., 2017).

4. Comparison with Other Bifurcation Mechanisms

SNIC bifurcations are global but of codimension one and differ sharply from Hopf and homoclinic bifurcations:

Bifurcation	Amplitude at Onset	Period Scaling	Geometry	PRC Symmetry
SNIC	Large	$\mu=0$2	Node-saddle collision	Symmetric
Hopf	Small	$\mu=0$3 finite	Local (complex pair)	Sinusoidal, Class II
Homoclinic	Large	$\mu=0$4	Saddle to saddle loop	Skewed, asymmetric

In neuronal excitability, SNIC corresponds to "class I" onset: the system can fire at arbitrarily low frequency near threshold, whereas Hopf bifurcation corresponds to "class II" onset with nonzero minimal firing frequency (Lopes et al., 2017, Hesse et al., 2016, Hens et al., 2015). Homoclinic bifurcations also yield infinite period but involve trajectory folding back on a single saddle, which leads to asymmetric spike shapes and enhanced parameter sensitivity (Lopes et al., 2017, Hesse et al., 2016).

The SNIC period divergence law and symmetric PRC ($\mu=0$5) distinguish it from other oscillatory emergences (Hesse et al., 2016).

5. Phase Response, Synchronization, and Network Dynamics

The SNIC onset fundamentally shapes the phase response curves (PRCs) and synchronization properties. The phase sensitivity function for the normal form is symmetric and—at the SNIC onset—takes a U-shape, reflecting time-reversal symmetry in the approach/exit from the bottleneck (the "ghost" of the saddle-node) (Hesse et al., 2016, Zhu, 2021). The overall shape is: $\mu=0$6 implying Type I PRC in the singular limit; however, in smooth systems, the PRC is always Type II (changes sign) (Zhu, 2021).

Bursting dynamics in networks of SNIC elements, such as Josephson junctions (Hens et al., 2015), emerges via slow–fast interactions: the effective control variable ($\mu=0$7) is modulated by network feedback, repeatedly driving the system through the SNIC transition and producing parabolic ("circle–circle") bursting, with interspike intervals peaking at threshold and following the expected $\mu=0$8 divergence.

In coupled SNIC units, weak mutualistic coupling yields regions of bistability (resting and periodic attractors), global mode-locking structures, and quasi-periodic regimes on the torus $\mu=0$9. Mode-locking tongues, tori, cantori, and Cherry flows are canonical features (Baesens et al., 2013). In large networks, Ott–Antonsen reduction reveals how SNIC-induced collective rhythmic transitions are organized by global bifurcations such as contractible and noncontractible limit cycles, connected by heteroclinic orbits on the state-cylinder (Medvedev et al., 2022).

Synchronization—particularly under noisy input—depends critically on spatial structure of phase sensitivity. The spatial phase sensitivity function (sPSF) localizes extreme phase sensitivity near the saddle-node, so only space-dependent perturbations or noise overlapping this bottleneck region effectively synchronize a population (Zhu, 2021).

6. High-Codimension Generalizations and Complex Unfoldings

Beyond the classical case, SNIC bifurcations are embedded in higher-codimension scenarios involving interactions with global connections (heteroclinic or homoclinic loops). The non-central SNICeroclinic ("SNICeroclinic," Editor's term) bifurcation is a codimension-three event organizing transitions between distinct global oscillatory and equilibrium behaviors (Nechyporenko et al., 2024).

In the planar non-central SNICeroclinic unfolding, three parameters control the local saddle-node and global connections; the SNIC arises as a codimension-two slice. The geometry involves return maps patched together from local and global segments, and parameter cross-sections display coexistence of heteroclinic, homoclinic, SNIC and purely periodic regimes—all governed by the interplay of global and local bifurcation mechanisms. Both the birth and death of oscillations can occur via such global events, allowing for richer organizing centers in network and spatially extended systems (Nechyporenko et al., 2024).

7. Applications in Neuroscience, Josephson Networks, and Network Oscillators

SNIC bifurcations govern key dynamical transitions in diverse systems:

• Neuronal models: The transition from resting to repetitive spiking in Type I neurons, and the onset of epileptiform rhythmogenesis, are prototypically organized by SNIC (Lopes et al., 2017, Hesse et al., 2016).
• Josephson junction networks: Parabolic bursting, clustering, and mixed-mode oscillations in globally coupled populations arise when single-junction dynamics is SNIC-governed (Hens et al., 2015).
• Coupled phase oscillators: Collective synchrony and the emergence of global rhythms in active-rotator networks are explained by SNIC bifurcations in the order parameter ODE after Ott–Antonsen reduction (Medvedev et al., 2022).
• General excitable media: SNIC defines the universal route to oscillation in any smooth planar system with a nondegenerate saddle-node collision on a closed invariant set.

Critical slowing down, square-root scaling of the interspike interval, and robustness of the large-amplitude periodic orbit at onset are thus characteristic signatures enabling diagnosis and parameter estimation in experimental and modeling contexts (Lopes et al., 2017, Hens et al., 2015).

---

References:

(Baesens et al., 2013, Hens et al., 2015, Hesse et al., 2016, Lopes et al., 2017, Zhu, 2021, Medvedev et al., 2022, Nechyporenko et al., 2024)