Noisy FitzHugh–Nagumo Model

• The paper demonstrates how adding noise to the classic FitzHugh–Nagumo system induces limit cycles and novel bifurcations absent in the deterministic model.
• It utilizes large deviation theory and quasipotential analysis to quantify noise-induced transitions between metastable states and oscillatory regimes.
• The model provides a rigorous framework linking noise intensity and time-scale separation to dynamic behaviors like bursting and noise-dependent bifurcation thresholds.

A noisy FitzHugh–Nagumo (FHN) model is a stochastic extension of the classical two-variable slow–fast system widely used to capture spike-generation and excitability in neurons and excitable media. In this context, the addition of noise fundamentally alters the dynamics of the system, enabling behavior—including oscillations, bursting, and novel bifurcations—that are inaccessible in the purely deterministic setting. Analysis of such noise-induced effects in the FHN model, both at the level of a single cell and in networks, establishes a rigorous mathematical framework for understanding stochastic transitions, metastability, and noise-induced phenomena in neurodynamical systems.

1. Slow–Fast Structure and Stochastic Formulation

The deterministic FHN system is a two-dimensional slow–fast dynamical system:

• $\delta\,\dot{x}_t = -y_t + f(x_t)$
• $\dot{y}_t = x_t - a$,

with $f(x)$ typically a cubic polynomial, e.g., $f(x) = -x(x-\alpha)(x-\beta)$ for fixed $\alpha < 0 < \beta$. Here $x_t$ models the fast membrane potential (activator), $y_t$ is the slow recovery variable (inhibitor), $a$ is a bifurcation parameter, and $\delta \ll 1$ characterizes the time-scale separation.

The stochastic model introduces an additive white noise in the fast voltage variable:

• $$\delta\,dX_t = [-Y_t + f(X_t)]\,dt + \sqrt{\varepsilon}\,dW_t$$
• $dY_t = (X_t - a)\,dt$,

where $W_t$ is a standard Wiener process (Brownian motion), and $\varepsilon > 0$ is the noise intensity (0906.2671).

Time-scale separation ($\delta \ll 1$) ensures that the fast $X_t$-dynamics quickly equilibrates for quasi-static $Y_t$, while $Y_t$ evolves adiabatically.

2. Noise-Induced Transitions: Large Deviations and Quasipotential Theory

A central feature in the noisy FHN system is the occurrence of transitions between metastable basins in the fast subsystem due to rare, noise-induced excursions. For fixed $y$, the fast subsystem possesses two stable equilibria $x^{\ast}_{-}(y)$, $x^{\ast}_{+}(y)$ separated by an unstable $x_0^{\ast}(y)$ (since $f(x) = y$ is cubic).

Noise enables crossing of the separatrix via large deviation events, quantified by the action

$$S_{t_1}^{t_2}(\varphi) = \frac{1}{2}\int_{t_1}^{t_2} \lvert \dot{\varphi}(u) - b(\varphi(u)) \rvert^2 du$$

and the corresponding quasipotentials

$$V_{-}(y) = -2\int_{x^{\ast}_{-}(y)}^{x^{\ast}_0(y)} [-y+f(u)]du, \quad V_{+}(y) = -2\int_{x^{\ast}_{+}(y)}^{x^{\ast}_0(y)} [-y+f(u)]du$$

(0906.2671).

The mean exit time from a metastable basin scales as $\tau \sim \exp[V/(\varepsilon/\delta)]$. The ratio $c = (\varepsilon |\log \delta|)/\delta$ crucially organizes the regime of system behavior as $(\delta,\varepsilon)\to 0$.

3. Noise-Induced Oscillations: Stochastic Bifurcation and New Critical Parameters

Deterministically, oscillations (limit cycles) in the FHN model can only arise for $a\in (a_0, a_1)$—an interval defined by Hopf bifurcation points corresponding to zeros of $f'$ evaluated at equilibrium. For $a$ values outside this window, the deterministic system is quiescent.

Random perturbation fundamentally changes this picture:

• For $c \in (0, S)$ (with $S$ defined by $V_{-}(y)= V_{+}(y) = S$), and $a \in (x_-(c), x_+(c))$ (where $V_{-}(y_-(c)) = c = V_{+}(y_+(c))$), the system exhibits a noise-induced periodic orbit (“noise-induced limit cycle”).
• These are characterized by periodic functions $(\Phi_{(c)}^a, \Psi_{(c)}^a)$, with $\Psi_{(c)}^a$ satisfying

$$\dot{\Psi}_{(c)}^a(t) = \begin{cases} x^{\ast}_{+}(\Psi_{(c)}^a(t)) - a, & t \in [0, T_1^a(c)) \ x^{\ast}_{-}(\Psi_{(c)}^a(t)) - a, & t \in [T_1^a(c), T_1^a(c) + T_2^a(c)) \ \end{cases}$$

with the cycle durations

$$T_1^a(c) = \int_{y_-(c)}^{y_+(c)} \frac{dy}{x_+^*(y)-a}, \quad T_2^a(c) = \int_{y_-(c)}^{y_+(c)} \frac{dy}{|x_-^*(y)-a|}.$$

• If $a < x_-(c)$ or $a > x_+(c)$, the system instead converges to a unique equilibrium—oscillations “disappear,” though the equilibrium is now noise-modified.
• For $c > S$, the slow variable stabilizes to a new equilibrium $y^*$ defined by $V_-(y^*) = V_+(y^*) = S$. The fast variable then stochastically switches between $x_-^*(y^*)$ and $x_+^*(y^*)$, resulting in a degenerate stationary state.

The emergent bifurcation structure thus contains noise-dependent “critical parameters” $x_-(c), x_+(c)$ whose role as oscillation thresholds supersedes those in the deterministic model (0906.2671).

4. Mechanism and Timescales of Noise-Induced Oscillatory Behavior

The mechanism for these oscillations is the alternation between metastable wells in the fast variable’s effective potential landscape. Noise events supply the energy required for rare excursions; the rate of such transitions is controlled by the quasipotentials $V_{\pm}(y)$ and the effective parameter $c$.

When the exit time out of a well (set by large deviations) and the rate of change of the slow variable $y$ are commensurate, the stochastic system realizes a periodic alternation between wells—a limit cycle with period given by the sum $T_1^a(c) + T_2^a(c)$. When the noise amplitude (relative to $\delta$) is too small or too large, this balance is lost, and the system respectively reverts to a monostable stationary state or attains a different stationary regime wherein $y$ becomes pinned at $y^*$ (0906.2671).

5. New Bifurcation Structure: Stochastic Resonance, Bursting, and Metastability

The addition of noise not only initiates oscillations but qualitatively reshapes the bifurcation diagram:

• The deterministic bifurcation window in $a$ is replaced by a noise-dependent window $(x_-(c), x_+(c))$ within which noise-induced limit cycles persist.
• The critical value $S$ acts as a bifurcation threshold, separating oscillatory and stationary dynamical regimes.
• When $c \to S^-$, the period of oscillations diverges (a signature of critical slowing down near the stochastic bifurcation).
• The succession of (stochastically induced) passages between metastable states produces bursting or mixed-mode oscillatory behavior, controlled by the scaling regime $(\varepsilon, \delta)$ and the noise intensity (0906.2671).

Table: Key Noise-induced regime boundaries

Parameter interval	Dynamical Regime	Description
$a \in (x_-(c), x_+(c))$	Noise-induced limit cycle	Oscillatory, stochastically locked cycle
$a < x_-(c)$ or $a > x_+(c)$	Monostable (noise-modified) equilibrium	No periodicity, small fluctuations
$c > S$	Stationary degenerate state	$Y$ pinned, $X$ fluctuates between two values

6. Quantitative Theory: Action Functionals and Scaling of Transition Rates

The noise-induced switching rates between basins arise from large deviation theory and are determined by the action functional $S_{t_1}^{t_2}(\varphi)$, yielding exponential scaling of mean exit (or first-passage) times:

$$\tau \sim \exp\left[\frac{V(y)}{\varepsilon/\delta}\right] = \delta^{-\frac{1}{(c - V)}}$$

The relative magnitudes of $\varepsilon$ (noise amplitude) and $\delta$ (time-scale separation) thus crucially control the ability of noise to drive transitions and hence the system’s long-time behavior.

As the noise amplitude increases or $\delta$ is reduced, these rare events become more likely, the alternations between potential wells more frequent, and the probability measure over system trajectories transitions from being unimodal (quiescent) to supporting oscillatory or metastable dynamics (0906.2671).

7. Implications, Applications, and Generalizations

The analysis of the noisy FitzHugh–Nagumo system provides a rigorous, quantitative understanding of:

• How oscillations can be generated in parameter regimes where deterministic dynamics would converge to a stable fixed point.
• The emergence and tunability of limit cycles, bursting, and metastability through noise amplitude and time-scale separation.
• Shifts in bifurcation structure and the creation of new critical parameters that organize the observed phenomena.
• The applicability of action functional and quasipotential theory to more general slow–fast stochastic systems.

This framework directly applies to neurobiological settings in which channel noise, synaptic variability, or other sources of random perturbations are present, and in which neural or excitable media activity displays noise-driven rhythmicity, bursting, or transitions between firing and rest states. The structure elucidated in the noisy FHN model is also representative for a wider class of excitable systems where stochastic inputs profoundly shape the system-level dynamics.