Nonlinear α-Stable Lévy Processes

• Nonlinear symmetric α-stable Lévy process is a stochastic process defined by stationary, independent increments that follow a heavy-tailed, symmetrical α-stable law, serving as driving noise in nonlinear SDEs and SPDEs.
• The analysis employs tools like fractional Laplacians, maximal tail inequalities, and paracontrolled calculus to handle irregular coefficients and singular drifts.
• The framework underpins applications in anomalous diffusion, kinetic systems, and non-Gaussian limit theorems, revealing threshold phenomena in existence and uniqueness.

A nonlinear symmetric α-stable Lévy process is a stochastic process whose increments are stationary, independent, and follow a symmetric α-stable law, and which appears as the driving noise in various nonlinear stochastic (partial) differential equations (SDEs/SPDEs). Nonlinearity typically enters via coefficients or drift terms that depend on the state, spatial variable, or even distributions. This class forms a central object in the analysis of heavy-tailed fluctuations, anomalous diffusion, and singular stochastic models, and has motivated significant advances in stochastic analysis, PDE theory, and statistical mechanics.

1. Construction and Properties of Symmetric α-Stable Lévy Processes

A symmetric α-stable Lévy process $L_t$ on $\mathbb{R}^d$ ($\alpha\in(0,2)$, $\alpha\ne1$) is characterized by:

• Stationary, independent increments, $L_0=0$, and $L\overset{d}{=}-L$.
• The characteristic function $$E\left[e^{i \langle \xi, L_t\rangle}\right] = \exp(-t\psi(\xi))$$, with

$$\psi(\xi) = \int_{S^{d-1}} |\langle\xi,\theta\rangle|^\alpha v(d\theta)$$

for a finite symmetric measure $v$ on the unit sphere.

• The Lévy measure $\nu(dz) = C_\alpha |z|^{-d-\alpha} dz$ gives the intensity of jumps.

The infinitesimal generator $\mathbb{R}^d$0 on smooth $\mathbb{R}^d$1 has the nonlocal form:

$\mathbb{R}^d$2

In the isotropic, rotationally invariant case,

$\mathbb{R}^d$3

where $\mathbb{R}^d$4 is the fractional Laplacian.

The finite-dimensional marginals are symmetric α-stable laws, with characteristic function

$\mathbb{R}^d$5

and distribution $\mathbb{R}^d$6 for bounded Borel $\mathbb{R}^d$7 (Balan, 2013, Kremp et al., 2020).

2. Stochastic Integration and Maximal Inequalities

Stochastic integration with respect to symmetric α-stable Lévy noise extends the Itô calculus for continuous martingales and Gaussian processes. The Giné–Marcus method, generalized to higher dimensions and non-symmetric tails, provides a construction of the integral with respect to the Lévy sheet $\mathbb{R}^d$8:

• For step processes $\mathbb{R}^d$9,

$\alpha\in(0,2)$0

• The integral is extended by density in $\alpha\in(0,2)$1.

A central tool is the maximal tail inequality

$\alpha\in(0,2)$2

This replaces the Burkholder–Davis–Gundy inequality for martingales, and provides control on the probability of large excursions of the stochastic integral (Balan, 2013).

"Truncation" schemes, where the driving noise $\alpha\in(0,2)$3 is replaced by $\alpha\in(0,2)$4 (jumps larger than $\alpha\in(0,2)$5 removed), yield convergent approximations and moment bounds:

$\alpha\in(0,2)$6

for $\alpha\in(0,2)$7 when $\alpha\in(0,2)$8, or $\alpha\in(0,2)$9 when $\alpha\ne1$0 (Balan, 2013).

3. Nonlinear SPDEs and SDEs Driven by Symmetric α-Stable Lévy Noise

The central nonlinear SPDEs and SDEs driven by symmetric α-stable noise admit both spatially extended and finite-dimensional forms.

SPDEs: Consider

$\alpha\ne1$1

where $\alpha\ne1$2 is a second-order pseudo-differential operator, $\alpha\ne1$3 is Lipschitz, and $\alpha\ne1$4 is an α-stable Lévy sheet.

The mild (integral) formulation is

$\alpha\ne1$5

with $\alpha\ne1$6 the Green kernel.

Existence and uniqueness of a predictable solution $\alpha\ne1$7 in $\alpha\ne1$8 for $\alpha\ne1$9 ($L_0=0$0) or $L_0=0$1 ($L_0=0$2) is guaranteed provided $L_0=0$3 satisfies suitable integrability and regularity (Balan, 2013).

SDEs: For

$L_0=0$4

with $L_0=0$5 a drift that may be merely measurable or even distributional, regularity and uniqueness theory depend delicately on $L_0=0$6:

• Weak solutions via the martingale problem are constructed for drifts $L_0=0$7 in Besov space $L_0=0$8 with $L_0=0$9.
• Pathwise uniqueness holds if $L\overset{d}{=}-L$0 is locally in $L\overset{d}{=}-L$1, $L\overset{d}{=}-L$2, $L\overset{d}{=}-L$3 ($L\overset{d}{=}-L$4) (Zhang, 2010, Kremp et al., 2020).

Highly singular drifts (e.g., spatial white noise in Brox diffusion) require paracontrolled calculus for well-posedness beyond the Young regime (Kremp et al., 2020).

4. Existence, Uniqueness, and Threshold Phenomena

Distinct threshold phenomena separate the regimes of solvability:

• In finite variation regime ($L\overset{d}{=}-L$5), pathwise uniqueness in SDEs with symmetric α-stable drivers requires $L\overset{d}{=}-L$6, $L\overset{d}{=}-L$7 one-sided Lipschitz (Fournier, 2010).
• For infinite variation ($L\overset{d}{=}-L$8), pathwise uniqueness holds if $L\overset{d}{=}-L$9 and $$E\left[e^{i \langle \xi, L_t\rangle}\right] = \exp(-t\psi(\xi))$$0 is one-sided Lipschitz, which is sharp (Fournier, 2010).
• For distributional drifts, transition from classical Schauder/PDE theory ($$E\left[e^{i \langle \xi, L_t\rangle}\right] = \exp(-t\psi(\xi))$$1) to regimes handled by paracontrolled distributions ($$E\left[e^{i \langle \xi, L_t\rangle}\right] = \exp(-t\psi(\xi))$$2) (Kremp et al., 2020).

In SPDEs, uniqueness and maximality of mild solutions are shown by a combination of Picard iteration on equations with truncated noise, moment inequalities, and consistency via stopping times and limit-pasting arguments (Balan, 2013).

5. Ergodicity, Linear Response, and Limit Theorems

Stochastic dynamics driven by symmetric α-stable Lévy noise exhibit rich ergodic and response properties.

• Ergodicity: The semigroup of the kinetic Fokker–Planck equation with a degenerate stochastic Hamiltonian system and α-stable driving noise converges exponentially in $$E\left[e^{i \langle \xi, L_t\rangle}\right] = \exp(-t\psi(\xi))$$3 to a unique invariant Gibbs measure, with explicit rate computed via hypocoercivity techniques beyond the classic DMS framework (Jianhai et al., 2024):

$$E\left[e^{i \langle \xi, L_t\rangle}\right] = \exp(-t\psi(\xi))$$4

This requires Poincaré inequalities for stable-like Dirichlet forms.

• Linear Response: Nonlinear SDEs with symmetric α-stable noise admit fluctuation-dissipation-type formulas. The response function to small perturbations is computed via Markov semigroup techniques and a nonlocal stationary forward equation, generalizing classical FDT to the non-Gaussian, heavy-tailed setting (Zhang et al., 2020).
• Non-Gaussian Limit Theorems: For mechanical systems with nonlinear friction and symmetric α-stable noise, macroscopic (rescaled) behavior can converge to a Lévy process with modified index and jump structure determined by the interplay between nonlinearity and jump intensity. The fractal index of the input is reshaped by the nonlinearity, leading to non-Gaussian limit processes (Kulik et al., 2017).

6. Analytic and Probabilistic Techniques

The analysis of nonlinear symmetric α-stable Lévy processes employs an array of modern methods:

• Paracontrolled distributions handle ill-posed products in the Kolmogorov backward equations for distributional drifts (Kremp et al., 2020).
• Krylov-type estimates for discontinuous semimartingales and sophisticated PDE theory (Bessel-potential spaces, fractional Sobolev and Schauder estimates) underpin pathwise uniqueness and existence proofs for SDEs with irregular coefficients (Zhang, 2010).
• Stochastic calculus for Lévy sheets, including maximal inequalities and tail estimates, is essential for random-field solutions of SPDEs (Balan, 2013).
• Hypocoercivity via non-symmetric Dirichlet forms, potential-theoretic control of nonlocal drifts, and commutator estimates provide $$E\left[e^{i \langle \xi, L_t\rangle}\right] = \exp(-t\psi(\xi))$$5-ergodic rates in degenerate kinetic systems (Jianhai et al., 2024).

7. Applications and Fundamental Examples

This framework encapsulates a range of physically and mathematically significant models:

• Brox Diffusion with Lévy Noise: Brownian motion in a random environment generalized to α-stable jumps and singular environments, solved via paracontrolled calculus (Kremp et al., 2020).
• Fractional Kinetic Equations: Ergodic properties for Hamiltonian systems with nonlocal dissipation, critical for modeling transport with anomalous diffusion (Jianhai et al., 2024).
• Nonlinear Filters and Macroscopic Limits: Mechanical systems with jump-driven inputs and nonlinear dissipation, leading to limit theorems of filtering/jump convolution type (Kulik et al., 2017).
• Linear Response for Non-Gaussian Systems: Nonlinear dynamical systems under heavy-tailed fluctuations, including nonequilibrium statistical mechanics beyond Gaussian approximations (Zhang et al., 2020).

These results delineate the mathematical landscape for nonlinear symmetric α-stable Lévy processes and provide the analytic underpinnings for their application in stochastic dynamics, statistical mechanics, and random media.