Stability Index σ(x): Measuring Attraction

• Stability Index σ(x) is a quantitative measure that encodes scaling laws to assess local and global basin structures across diverse mathematical contexts.
• It employs techniques such as Lyapunov exponents, thermodynamic formalism, and eigenvalue ratios to extract critical stability properties and inform system behavior.
• Applications include analyzing riddled basins, heteroclinic cycles, sensitivity in control systems, and the stabilization of D-finite functions.

The stability index, denoted σ(x), is a quantitative tool for assessing the local or global measure-theoretic structure of basins of attraction in a variety of mathematical contexts—including dynamical systems, stochastic processes, control theory, and symbolic differential equations. Although its exact definition and interpretation depend on the particular setting, in all cases σ(x) encodes the scaling properties of measures or the algebraic order of particular functional or geometric features near a point x. The stability index concept arises prominently in the analysis of riddled or intermingled basins of attractors, robustness of heteroclinic cycles, local regularity and tail behavior in stochastic processes, Lyapunov-based system stability, and the algebra of D-finite functions. It is fundamentally connected to Lyapunov exponents, thermodynamic formalism (notably, pressure zeros and dimensions), multifractal geometry, and the structure of invariant measures.

1. Definitions Across Core Contexts

Dynamical Systems and Basins of Attraction

In the context of invariant sets X for smooth flows φᵗ on ℝⁿ with basin N=B(X), the stability index at x∈X is defined via the Lebesgue measure μ of local neighborhoods. For ε>0, let Bₑ(x) denote a ball of radius ε around x and define

$$\Sigma_ε(x) = \frac{\mu[B_ε(x)\setminus N]}{\mu[B_ε(x)]}.$$

One-sided logarithmic exponents are

$$σ_-(x) = \lim_{ε\to0} \frac{\ln \Sigma_ε(x)}{\ln ε},\qquad σ_+(x) = \lim_{ε\to0} \frac{\ln[1-\Sigma_ε(x)]}{\ln ε},$$

and the stability index is

$σ(x) = σ_+(x) - σ_-(x).$

This captures the measure-theoretic prevalence of the basin in small neighborhoods, indicating local attraction (σ(x)>0), repulsion (σ(x)<0), or riddling (σ(x)=0) (Podvigina et al., 2010).

Skew-Product and Concave Map Attractors

For skew-product systems with a hyperbolic base map S, as in chaotically driven concave maps, the local stability index at (θ,0) is based on the scaling of the subgraph of an upper semicontinuous invariant function φ: $$\Sigma_ε(v) := \frac{1}{ε \cdot |U_ε(v)|} \int_{U_ε(v)} \min\{\varphi_∞(t), ε\}\;dt,$$ with U_ε(v) = (v–ε, v+ε), and

$$σ_-(v) := \lim_{ε\to0} \frac{\log \Sigma_ε(v)}{\log ε},\qquad σ_+(v) := \lim_{ε\to0} \frac{\log(1-\Sigma_ε(v))}{\log ε},\qquad σ(v) := σ_+(v)-σ_-(v).$$

The stability index quantifies the local dominance of upward vs. downward attraction (Keller, 2012).

Multistable Stochastic Processes

In multistable processes, the local stability index σ(t) (often denoted α(t)) at time t is the local index of a stable process tangent to the process at t. It determines local jump intensities and heavy-tail exponents. Estimation is performed using (local) moment-ratio functionals and consistency arguments on local increments (Guével, 2010).

Lyapunov-Based System Analysis

For linear dynamical systems arising in control or power systems, the stability index σ(x) at a state x is defined as the smallest real number η such that

$$-\,J(x)^T P - P J(x) + \eta I \succeq 0, \quad P\succ0,$$

where J(x) is the system Jacobian and P a Lyapunov matrix. σ(x) thus measures the "distance to instability" and the convergence rate after disturbances, and is the optimal value of a semidefinite program (Wang et al., 2023).

D-finite Functions

For a linear differential operator L acting on a D-finite function, the stability index σ(L) is the least integer m such that for all degrees i≥m, the inhomogeneous equation L^*(y)=p admits a rational solution y for some polynomial p of degree i. σ(L) quantifies when the sequence of orders of iterated integral operators stabilizes (Chen et al., 2023).

2. Explicit Formulas and Evaluation

Lyapunov and Thermodynamic Exponents

In skew-product systems, the explicit value of the stability index at regular points depends on local exponents: $$Γ(v) = \lim_{n\to\infty} \frac{1}{n} \log g_n(v), \qquad Λ(v) = \lim_{n\to\infty} \frac{1}{n} \log |(S^n)'(v)|,$$ and the global thermodynamic exponent s_*, the unique root of a pressure function,

$ψ(s) = P(-\log |S'| - s\,\log g),\qquad ψ(s_*)=0.$

The explicit regime is

• If Γ(v)+Λ(v)>0: σ(v)=\frac{Γ(v)+Λ(v)}{Λ(v)}\,s_*,
• If Γ(v)+Λ(v)<0: σ(v)=-\frac{Γ(v)+Λ(v)}{Λ(v)}.

This ratio expresses the balance between fiber multiplier contraction/growth and base-map expansion (Keller, 2012). For dynamically defined Weierstrass functions, the typical index is

$$σ_\mu(x,t) = \frac{s^*\int\log\lambda\,d\mu}{\int\log|T'|\,d\mu}$$

for t above the invariant graph (Walkden et al., 2017).

Heteroclinic Cycles

For robust heteroclinic cycles in ℝ⁴, the stability index on a connection is computed in terms of eigenvalue ratios: $$a_j = \frac{c_j}{e_j}, \qquad b_j = -\frac{t_j}{e_j},$$ with explicit formulas classifying regimes: σ_j=+∞ (strong attraction), σ_j finite (power-law scaling), or σ_j=–∞ (repulsion), based on the sign and size of eigenvalue products (Podvigina et al., 2010).

Stochastic Processes

In multistable processes, the time-varying stability index α(t₀) is estimated via minimization of a divergence between empirical and theoretical p-norm ratios on increments, exploiting properties of stable distributions (Guével, 2010). Consistency results require careful regularity and moment assumptions.

Control and Sensitivity

In power systems, σ(x) arises as the minimal η in a Lyapunov matrix inequality. Sensitivity ∂σ/∂J_{ij} can be computed analytically via dual semidefinite programming as

$$\frac{\partial\sigma}{\partial J_{ij}} = 2\,[P^* Y_1^*]_{ij}$$

where (P^*, Y_1^*) are primal/dual optimizers (Wang et al., 2023). This method outperforms numerical finite-difference approaches both in accuracy and efficiency.

D-finite Functions

General upper and exact bounds are provided: $$σ(L) \leq \max\left\{0,\, \max\{s \geq 0:\, {}^{L^*}(s)=0\} + 1 + \sigma^{L^*} + \deg(d)\right\},$$ with special cases:

• Constant-coefficient L: σ(L)=0
• Katz-type (L=p(D)+q(x)): σ(L)=\deg q
• Hyperexponential: σ(y)=\deg(\text{denominator}(f)) for y'=f(x)y (Chen et al., 2023)

3. Interpretation and Geometric Significance

The stability index rigorously encodes the local scaling law of the basin of attraction in measure—the decay rate with which either the complement (if σ(x)>0) or the basin itself (if σ(x)<0) occupies vanishing ε-balls around x. Positive index indicates predominance of attraction, negative index signifies dominance of escape/instability, and zero indicates a riddled or balanced regime. In dynamical systems with intermingled basins (Milnor attractors), σ(x) is critical for distinguishing between essential, predominant, and weak attraction properties (Podvigina et al., 2010, Keller, 2012).

In skew-product or random dynamical systems, σ(x) mediates the interplay between local multipliers and base dynamics, and is intimately connected with the thermodynamic formalism—particularly the pressure root s_*, which also appears in queueing theory as Loynes's exponent (Keller, 2012, Walkden et al., 2017).

In operator/algebraic contexts, σ(L) quantifies at which order the iterative integration process "stabilizes," governing the structure of differential ideals and rational solutions (Chen et al., 2023).

4. Connections to Thermodynamic Formalism and Multifractal Analysis

The calculation of σ(x) in many cases relies on thermodynamic formalism, specifically the topological pressure associated with certain potentials, yielding an exponent s_* as the unique zero of

$P(\text{potential} + s\,\text{observable})=0.$

This exponent governs both the global distribution of small values of the invariant graph (i.e., frequency of rare events) and the multifractal spectra of stability exponents.

In dynamically defined Weierstrass functions, the multifractal spectrum of σ(x,t) is elucidated via Legendre transforms of the pressure function, yielding level sets K_μ(σ) with explicitly computable Hausdorff dimension: $\dim_H K_\mu(\sigma) = S(q)-q\,\sigma(q),$ where S(q) solves $P(-S(q)\log|T'| + q s^* \log\lambda) = 0$ (Walkden et al., 2017).

5. Relationships to Broader Notions of Stability

The stability index is deeply related to several traditional and measure-theoretic notions of stability:

• Asymptotic stability is characterized by σ(x)>0 on the invariant set.
• Milnor attraction is implied if σ(x)>–∞ for some x.
• Essential and predominant asymptotic stability are encoded in σ(x)—with p.a.s. corresponding to uniform positivity (Podvigina et al., 2010).
• In control and power systems, σ(x)<0 ensures exponential decay, while the sensitivity of σ(x) reveals how system perturbations impact stability margins (Wang et al., 2023).

6. Illustrative Examples and Key Theorems

Chaotically Driven Concave Maps

Keller (Keller, 2012) provides explicit local scaling formulas for σ(θ,0), connecting it to Lyapunov exponents and the solution s_* of a pressure equation, with rigorous analysis via Markov preimages and distortion estimates.

Heteroclinic Cycles

Podvigina & Ashwin (Podvigina et al., 2010) extend classical stability analysis of heteroclinic cycles in ℝ⁴, showing that σ_j can be computed via explicit recursions in terms of eigenvalue ratios, encompassing previous results of Krupa and Melbourne.

D-finite Functions

Chen et al. (Chen et al., 2023) classify stability for hyperexponential functions and provide both exact and effective upper bounds for general monic differential operators, settling positivity and growth of the stability index in various operator families.

Lyapunov/SDP Sensitivity

Wang et al. (Wang et al., 2023) introduce a semidefinite-program-based σ(x), its dual formulation, and an analytic sensitivity formula, showing its computational and theoretical advantages in power system stability analysis.

---

These results establish σ(x)—in its diverse formulations—as a central object for quantifying local and global stability properties across mathematical, probabilistic, and applied domains.