Rate-Induced Fractals in Dynamics

• Rate-induced fractals are fractal structures that emerge when system parameters change too rapidly, causing abrupt transitions without a bifurcation.
• They result from non-attracting invariant sets, like chaotic saddles, whose fractal basin boundaries determine critical tipping rates.
• Understanding these fractals provides insights for predicting transitions in applications such as climate modeling, ecological collapse, and engineering stability.

Rate-induced fractals are a class of fractal structures arising in dynamical systems when non-autonomous parameter variation, often at a rate exceeding a critical threshold, leads to transitions that are governed by fractal sets in phase and parameter space. Unlike bifurcation-induced transitions, @@@@1@@@@ (R-tipping) occurs without the destruction or destabilization of attractors via bifurcation; instead, the transition is linked purely to the rate at which system parameters evolve. The existence of non-attracting fractal saddles in the autonomous system underlies the formation of fractals in the space of rates and parameters, creating boundaries across which outcome prediction is fundamentally limited (Wang et al., 23 Jan 2026). Complementary phenomena are observed in stochastic systems where fragmentation rates control fractal dimension, as in stochastic dyadic Cantor sets (Rahman et al., 2020).

1. Rate-induced Tipping and Critical Transitions

Rate-induced tipping (R-tipping) refers to situations in non-autonomous dynamical systems where the time-dependent variation of a control parameter $\lambda$, moving at rate $r$, can force the system state to depart from a quasi-static attractor even though no bifurcation is traversed. In discrete and continuous time, the system has the form: $$x_{n+1} = F(x_n; \Lambda(rn)) \quad \text{or} \quad \dot x = F(x; \Lambda(rt))$$ For slow rates ($r \ll 1$), trajectories track the attractor $x^*(\Lambda(s))$. As $r$ exceeds a critical value $r_c$, the trajectory departs its basin of attraction, resulting in tipping. Unlike bifurcation-induced tipping, which requires structural instability, R-tipping is controlled solely by the rate parameter.

2. Non-attracting Fractal Sets and Edge States

In many nonlinear dissipative systems, transient chaos gives rise to non-attracting invariant sets, often chaotic saddles $\Lambda$. These sets exhibit the following properties:

• $F(\Lambda) = \Lambda$
• The stable manifold $W^s(\Lambda)$ is fractal, separating basins of distinct attractors.
• The co-dimension $\alpha = \dim(\Omega) - D(W^s)$, where $D(W^s)$ is the fractal dimension, typically satisfies $0 < \alpha < 1$.
• $W^s(\Lambda)$ forms fractal basin boundaries.

A canonical example is the open tent map repeller, where the invariant set is a middle-third Cantor set of dimension $\ln 2 / \ln 3$. More generally, such fractal saddles are defined via Lyapunov exponents with $D(W^s) = d - \kappa/\lambda_u$, where $\kappa$ is the escape rate and $\lambda_u$ the unstable exponent.

3. Mechanism: Induced Fractals from Phase to Parameter Space

The fractal structure governing outcome prediction in R-tipping is inherited from phase-space sets. Critical rates $r_c$ are defined by the condition that infinitesimal changes $\delta r$ switch the final state from “track” to “tip.” Near $r_c$, the system trajectory spends a long dwell time near the saddle $\Lambda$; small parameter changes scale the phase-space deviation as $|r - r_c|^{-\lambda_s/\lambda_u} \delta r$. Crossing the local stable manifold boundary requires that the deviation equals the basin boundary thickness $\varepsilon$, so the measure of uncertain $r$-intervals scales as $(\Delta r)^\alpha$ with $\alpha = \kappa/\lambda_u$, matching the basin boundary co-dimension.

An analogous fractalization occurs when the final parameter $\lambda_+$ is varied at infinite rate. The fractal structure of tipping sets in $(r, \lambda_+)$ space stems directly from the fractal properties of $W^s(\Lambda)$.

4. Quantitative Relations among Fractal Dimensions

There exist precise correspondences among the co-dimensions of fractal boundaries: $$\alpha_1 = \alpha_2 = \alpha_3 = \frac{\kappa}{\lambda_u}$$ where $\alpha_1$ is the co-dimension of $W^s(\Lambda)$ in phase space at $\lambda_+$, $\alpha_2$ is the co-dimension of the fractal boundary in parameter space, and $\alpha_3$ is the co-dimension of fractal sets of critical rates. The fractal dimension of the set of tipping rates is $D_r = 1 - \alpha_3 = 1 - \frac{\kappa}{\lambda_u}$. In parameter space $\mathbb{R}^m$, $\dim(\{\lambda_+:\text{tip}\}) = m - \alpha_2$ (Wang et al., 23 Jan 2026).

5. Paradigmatic Examples of Rate-Induced Fractals

Three examples illustrate rate-induced fractal boundaries:

System	Autonomous Fractal Set	Co-dimension $\alpha$	Observable Fractal Boundary
Piecewise-linear 1D map	Tent map Cantor repeller	$\approx 0.49$	Tipping rates $r$, parameter $\lambda_+$
Henon map (2D)	Chaotic saddle basin boundary	$\approx 0.34$	Track vs. tip in $(a_+, b_+)$ and in $r$
Forced pendulum (continuous)	Fractal basin boundary in Poincaré	matches basin	Critical set of ramp rates $r$

• For the piecewise-linear map (tent + stable branch), the fractal uncertainty exponent $\alpha \approx 0.49$ contains corrections from reinjection to the Cantor set value $\ln2/\ln3$.
• In the Hénon map, both the phase-space and parameter-space boundaries share co-dimension $\alpha \approx 0.34$.
• The forced pendulum exhibits fractal critical rates $r$, coinciding with the frozen-system fractal basin boundary.

6. Rate-induced Fractals in Stochastic Fragmentation Models

A related mechanism for fractal formation controlled by rate parameters occurs in stochastic binary-fragmentation models. In this context, the evolution of the segment-size density $n(x,t)$ under fragmentation with survival probability $p$ and fragmentation rate determined by shape parameter $\alpha$ yields a fractal set whose Hausdorff dimension $d_f$ is determined by the condition $(1+p)B(\alpha,d_f) = B(\alpha,1)$, where $B$ is the Beta function (Rahman et al., 2020). The dimension increases monotonically in $p$ and $\alpha$. The $d_f$-th moment $M_{d_f}$ is conserved, and the scaling form $n(x,t) = t^{\theta} \varphi(\xi)$ with exponents determined from scaling and moment hierarchy. Monte Carlo simulations confirm these analytic predictions, validating the rate-induced character of fractal formation.

7. Applications, Implications, and Open Problems

Rate-induced fractal boundaries imply extreme sensitivity to uncertainty in rates or amplitudes, fundamentally limiting predictability in affected systems. The co-dimension $\alpha$ quantifies the danger: smaller $\alpha$ (thicker fractal boundary) means more robust tipping regions and greater unpredictability. These phenomena are prevalent in climate modeling, ecological collapse, and engineering systems (such as power-grid overload), wherever non-attracting fractal saddles arise.

Open research problems involve:

• R-tipping with multi-parameter drifts in higher-dimensional parameter spaces
• Effects of stochastic noise, which interacts non-trivially with fractal boundaries to modify tipping rates
• Time-dependent protocols, such as unbounded ramps, without asymptotic limits
• Formal proofs for the occurrence and structure of rate-induced fractals in infinite-dimensional (PDE) systems (Wang et al., 23 Jan 2026)

In all cases, the existence of a fractal edge state saddle with a fractal stable manifold in the frozen system gives rise to corresponding fractal sets in parameter and rate space, establishing universal co-dimension relationships and indicating fundamental barriers to prediction and control.