PaperlayerView exact revision on arXiv

Source-linked AI summary

Levy Flight Superdiffusion: An Introduction

A. A. Dubkov, B. Spagnolo, V. V. Uchaikin

arXiv:0810.1492v1cond-mat.stat-mech

TL;DR

The paper addresses Lévy-flight superdiffusion as a non-Gaussian generalization of Brownian motion and develops a unified mathematical treatment. It introduces Lévy motion through self-similarity and stable distributions, derives fractional evolution equations, and analyzes stationary distributions and barrier crossing. For Cauchy noise in symmetric smooth monostable potentials, the stationary distributions are bimodal, while arrival and residence times are identified as appropriate characteristics for Lévy flights.

  • Problem

    The paper addresses anomalous diffusion and barrier-crossing behavior arising from Lévy flights, whose discontinuous heavy-tailed motion differs from ordinary Brownian diffusion.

  • Method

    It introduces Lévy flights as self-similar Lévy processes and uses functional analysis to derive a generalized Kolmogorov equation and fractional Fokker–Planck equation from Langevin dynamics with symmetric α-stable noise.

  • Results

    For Cauchy driving noise in symmetric smooth monostable potentials, the paper derives stationary probability distributions and reports that they remain bimodal as β increases, γ decreases, or D increases.

  • Takeaways & Limitations

    Arrival and residence times are identified as appropriate characteristics for investigating Lévy flights across different potential profiles.

Abstract

from arXiv · show

After a short excursion from discovery of Brownian motion to the Richardson "law of four thirds" in turbulent diffusion, the article introduces the Lévy flight superdiffusion as a self-similar Lévy process. The condition of self-similarity converts the infinitely divisible characteristic function of the Lévy process into a stable characteristic function of the Lévy motion. The Lévy motion generalizes the Brownian motion on the base of the $α$-stable distributions theory and fractional order derivatives. The further development of the idea lies on the generalization of the Langevin equation with a non-Gaussian white noise source and the use of functional approach. This leads to the Kolmogorov's equation for arbitrary Markovian processes. As particular case we obtain the fractional Fokker-Planck equation for Lévy flights. Some results concerning stationary probability distributions of Lévy motion in symmetric smooth monostable potentials, and a general expression to calculate the nonlinear relaxation time in barrier crossing problems are derived. Finally we discuss results on the same characteristics and barrier crossing problems with Lévy flights, recently obtained with different approaches.

I. INTRODUCTION

The introduction traces superdiffusion from Brownian motion and Richardson’s turbulent-diffusion law to Lévy flights, emphasizing their heavy-tailed jumps and anomalous statistics. It then outlines the paper’s self-similar-process framework, fractional equations, stationary distributions, and barrier-crossing analysis.

  • Origins of superdiffusion: Brownian motion models irregular continuous trajectories, whereas Richardson observed turbulent-cloud widening proportional to t^3/2 rather than the normal t^1/2.This established an early example of superdiffusion, defined by growth ∆∝t^γ with γ > 1/2.
  • Lévy-flight phenomenology: Lévy flights are discontinuous stochastic processes with extremely long jumps, power-law-tailed Lévy-stable lengths, and divergent second moments.This distinguishes them from Brownian motion, whose coordinate moments are finite.
  • Paper scope: The paper introduces Lévy motion as a self-similar Lévy process that generalizes Brownian diffusion within stable-distribution theory.The introduction places Lévy motion between Brownian motion and general infinitely divisible Lévy processes.
  • Paper scope: A functional approach derives a generalized Kolmogorov equation and, for Lévy-stable noise, a fractional Fokker–Planck equation from a Langevin equation.The Lévy formulation replaces white Gaussian noise with Lévy-stable noise.
  • Paper scope: For Cauchy noise in symmetric smooth monostable potentials, the paper derives stationary probability distributions and studies barrier crossing and first-passage behavior.The reported stationary distributions remain bimodal as β increases, as γ decreases, or as noise intensity D increases.

II. L´EVY PROCESSES

This section defines the process classes used in the paper, moving from random and Markovian processes to processes with stationary independent increments. These properties lead to the Lévy-process framework and its Lévy–Khintchine characteristic representation.

  • Basic process definitions: A random process is a time-indexed set of random variables defined on a common probability space.The section restricts attention to one-dimensional processes with X, x ∈ (−∞, ∞) and t ≥ 0.
  • Basic process definitions: A Markovian process makes the future independent of the past conditional on the present.The paper summarizes this as the past influencing the future only through the present.
  • Independent increments: Processes with independent increments have mutually independent initial state and successive non-overlapping increments.Such processes belong to the Markovian class.
  • Stationary increments: A process with independent increments is homogeneous or stationary when increment distributions depend on interval duration rather than starting time.The increment law is written as P{X(t + τ) − X(t) < x} = F(x, τ).
  • Lévy processes: A Lévy process has stationary independent increments, with each increment independent of the prior process and distributed like X(τ), while X(0) = 0.Its increment decomposition places the process within the infinitely divisible-distribution framework.

III. SELF-SIMILARITY (SCALING)

Self-similarity imposes scale invariance on Lévy processes, linking temporal and spatial rescaling through an exponent. Combined with stable random-variable scaling, this identifies Lévy motion as a special Lévy-process subfamily.

  • Scaling: Self-similarity requires scale changes in coordinates and time to be compensated by a homothetic transformation of the process or its density.The scaling transformation is expressed for a function u(x, t) under x → kx and t → lt.
  • Scaling: For the probability density P(x, t), normalization and self-similarity impose the relation 1 + δ = 1/α.This yields the corresponding self-similar representation with m = 1.
  • Process self-similarity: A process with stationary increments is self-similar when X(t + κτ) − X(t) d= κ^H[X(t + τ) − X(t)].The scaling exponent satisfies H = 1/α in the Lévy-motion construction.
  • Stable scaling: Strictly stable random variables scale sums of independent copies as t1/αY(α), producing the Lévy-motion subfamily of Lévy processes.The paper notes that Lévy flights is often used as a synonym for Lévy motion.

IV. STABLE RANDOM VARIABLES

Stable random variables are characterized through their scaling property and stable characteristic functions, with index α controlling tail behavior and variance. These laws underpin Lévy-flight statistics and explain departures from ordinary averaging.

  • Characteristic functions: The paper derives stable characteristic functions from either infinitely divisible characteristic functions or the stability property, choosing the latter route.The derivation starts from the second characteristic and imposes continuity and characteristic-function constraints.
  • Stable-law parameters: The stability parameter range is 0 < α ≤ 2, with α = 2 giving finite variance and α < 2 giving infinite variance.The parameter β controls asymmetry, and β = 0 gives symmetric stable distributions.
  • Stable-law parameters: The characteristic function of the strictly stable distribution is specified by the parameters α and β, covering symmetric and asymmetric stable laws.The symmetric class includes both Gaussian and Cauchy distributions.
  • Distributional properties: Stable distributions have heavy power-law tails and infinite variance except for the Gaussian case, while α determines the decay rate of large-value probabilities.For α < 1 and extreme asymmetry, distributions can be supported on a single semiaxis.
  • Distributional properties: Summing n identically distributed stable variables broadens the distribution as n1/α, so arithmetic means need not become more concentrated.For α = 1 the mean width remains constant, and for α < 1 it widens; the law of large numbers fails when the expectation does not exist.
  • Distributional properties: Stable laws are precisely the limiting distributions in the generalized central limit theorem.The paper presents this limiting-property characterization as their most important advantage.

V. STABLE PROCESSES AND L´EVY MOTION

The section defines stable processes and Lévy motion through self-similarity, stability, and independent increments, then contrasts Lévy trajectories with Brownian motion. It also characterizes their scaling, infinite variance, and fractal dimensionality.

  • Definitions: Self-similarity designates L(α,β)(t) as a Lévy process whose stable random-variable structure generalizes Gaussian processes.Stable processes are defined through stable finite-dimensional distributions, while Lévy motion is introduced with parameters 0 < α ≤ 2 and −1 ≤ β ≤ 1.
  • Definitions: Stable random vectors generalize stability to multivariate distributions, with strict stability obtained when the centering vector vanishes.Symmetric stable vectors form a related class, although symmetry does not generally imply strict stability in the reverse direction.
  • Trajectory properties: L(2,0) is the only L(α,β) process with continuous trajectories, distinguishing Brownian motion from processes with α < 2.The section connects this result to the Lindeberg condition and the small-time behavior of increments.
  • Scaling and diffusion: The index α is the fractal dimensionality of Lévy-process trajectories, while increments scale as t1/α and have infinite variance.Alternative width measures remain usable despite the divergent variance, and Lévy flights differ from Gaussian diffusion in distributional shape.
  • Scaling and diffusion: The section introduces Lévy flight superdiffusion as a stochastic model whose anomalous behavior differs from ordinary Brownian diffusion.The discussion motivates Lévy flights through their stable-process formulation and anomalous scaling.

VI. FRACTIONAL EQUATION FOR L´EVY FLIGHT SUPERDIFFUSION

This section derives differential equations for Lévy motion using Fourier representations of characteristic functions. Symmetric motion leads to a Riesz fractional operator, while asymmetric motion produces a corresponding probability-distribution equation.

  • Brownian limit: For α = 2, Lévy motion reduces to Brownian motion and obeys the ordinary diffusion equation.The Fourier image of the one-dimensional Laplace operator supplies the corresponding differential-equation form.
  • Symmetric motion: For symmetric Lévy motion, the characteristic function contains −|k|α, whose Fourier correspondence is the Riesz fractional operator.This correspondence yields a fractional differential equation for the probability distribution.
  • Symmetric motion: Integral representations of the Riesz derivative provide an alternative formulation of the fractional spatial operator.The representations are obtained through direct Fourier transformation.
  • Asymmetric motion: Asymmetric Lévy motion requires a separate probability-distribution equation reflecting the asymmetry of the process.The section introduces the associated relation for the spatial variable and distribution.
  • Context: Fractional differential equations provide the mathematical framework used to describe Lévy motion and its superdiffusive behavior.The section situates the derivation within the broader theory of Lévy-flight fractional equations.

VII. L´EVY WHITE NOISE

The section treats the time derivative of a Lévy process as stationary, delta-correlated non-Gaussian white noise. A functional approach derives its characteristic functional and generalizes Gaussian correlation formulas.

  • Lévy white noise: The time derivative of a Lévy process is a stationary random process analogous to Gaussian white noise and a generalized Wiener process.This establishes Lévy white noise as the driving source for subsequent stochastic dynamics.
  • Characteristic functional: The characteristic functional of Lévy white noise is derived by discretizing time and using the statistical independence of Lévy-process increments.The construction uses increment characteristic functions over subintervals.
  • Functional method: The Furutsu–Novikov formula for Gaussian noise is extended to arbitrary functionals of non-Gaussian random processes.The section invokes Klyatskin’s generalization on the observation interval.
  • Functional method: The resulting variational expressions are substituted into the functional evolution equations to obtain formulas for Lévy-driven systems.These substitutions prepare the derivation of the generalized Kolmogorov equation.

VIII. DERIVATION OF KOLMOGOROV’S EQUATION

This section derives a generalized Kolmogorov equation for nonlinear Langevin systems driven by Lévy white noise. Gaussian, additive, and symmetric α-stable noise choices recover ordinary, jump-process, and fractional Fokker–Planck equations.

  • Derivation: The Langevin model is Ẋ = f(X,t) + g(X,t)ξ(t), with ξ(t) serving as the Lévy white-noise source.Functional differentiation of the probability density leads from the stochastic dynamics to an evolution equation.
  • Derivation: The variational derivative with respect to ξ(t) becomes an ordinary spatial differential operator involving g(x,t).This operator substitution is a key step in converting the functional formulation into a probability-density equation.
  • General equation: The functional approach yields Kolmogorov’s equation for nonlinear systems driven by Lévy white noise and different non-Gaussian sources.The derivation uses the previously obtained functional relations and kernel representations.
  • Special cases: Gaussian white noise with kernel ρ(x) = 2Dδ(x) reduces the generalized equation to the ordinary Fokker–Planck equation.This identifies the Gaussian case as a particular limit of the broader noise-driven framework.
  • Special cases: For additive noise, g(X,t) = 1 and the exponential operator becomes a space-shift operator, producing a Kolmogorov–Feller-type equation.This case describes purely discontinuous Markovian processes.
  • Special cases: Symmetric α-stable additive noise yields an equation describing anomalous diffusion in the form of symmetric Lévy flights.Writing the equation with the Riesz derivative gives the fractional Fokker–Planck equation for Lévy flights.

IX. STATIONARY PROBABILITY DISTRIBUTIONS FOR L´EVY FLIGHTS

The paper derives stationary probability distributions for Lévy flights in symmetric smooth monostable potentials, obtaining exact results for selected potentials and characterizing their non-unimodal confinement behavior. It also relates the distribution shape to the Lévy index, potential steepness, and transient bifurcations.

  • Analytical construction: Analytical treatment transforms the stationary equation into a Fourier-space differential equation for symmetric potentials U(x) = γx^(2m)/(2m), but arbitrary potentials and Lévy exponents are not generally solvable analytically.The equation becomes an order-(2m−1) differential equation for these polynomial potentials; exact solutions are obtained for Cauchy noise.
  • Analytical construction: For α ∈ (0, 2], confinement with finite coordinate variance begins at the quartic potential m = 2, and the stationary distributions have non-unimodal shapes with power-law tails.The confinement condition follows from the stationary density-tail analysis, while the exact solvability statement is restricted to α = 1.
  • Quartic potential: Lévy superdiffusion produces a bimodal stationary distribution in a monostable potential, unlike Brownian diffusion, with quartic maxima at x = ±β/2 and a maximum-to-minimum ratio of 4/3.For the quartic case, ⟨X^2⟩_st = β^2, and β = (D/γ)^(1/3), so increasing β broadens the distribution through stronger noise or a less steep potential.
  • Quartic potential: Increasing α smooths the quartic stationary profile from strongly bimodal at α = 1 to unimodal at α = 2, where it recovers the Boltzmann distribution.The profiles are obtained by inverse Fourier transformation for Lévy indices from α = 1 through α = 2.
  • Higher-order potentials: For potentials with exponents m = 3, 4, and 5, the stationary distributions remain bimodal, while the maximum-to-minimum ratio increases with m.The corresponding cases use U(x) = γx^6/6, γx^8/8, and γx^10/10, with curves shown for β = 0.5, 1, and 1.5.
  • Higher-order potentials: The bimodal peaks reflect rapid Lévy-flight transport toward either side of a monostable potential, followed by long residence near symmetric regions around the potential walls.For fixed D and m, peak locations shift with potential steepness γ, indicating stronger or weaker confinement.
  • Time-dependent evolution: A transient trimodal state can occur between initial unimodal and final bimodal states only for monostable potentials with c > 4 and fixed Lévy index α.The transitions are characterized by bifurcation times t13 for unimodal-to-trimodal evolution and t32 for trimodal-to-bimodal evolution.

X. BARRIER CROSSING

The section examines barrier crossing and temporal statistics for Lévy flights, emphasizing first arrival, residence, and crossing times alongside difficulties caused by discontinuous jumps and nonstandard boundaries.

  • Barrier-crossing framework: The main temporal tools are first passage, crossing, arrival, and residence times, with mean first passage times presenting particular difficulties.The section develops residence-time formulas from transition probabilities and their Laplace transforms.
  • Barrier-crossing framework: Lévy-flight barrier crossing differs from classical Kramers behavior because stable noise can produce discontinuous jumps and infinite mean squared displacement.A particle may reach a boundary instantaneously from an arbitrary position, requiring boundary conditions different from ordinary diffusion.
  • Escape times: For low noise intensity, numerical studies recover the classical escape-time form, with μ(α) approximately 1 for 0 < α < 2; at intermediate intensities, escape time depends non-monotonically on α.The intermediate-intensity escape-time density remains exponential.
  • Arrival and passage times: The fractional Fokker-Planck equation with a δ-sink yields a non-normalized density whose sink flux represents the first arrival-time density.The first arrival time is used as an appropriate parameter for analyzing Lévy-flight barrier crossing.
  • Arrival and passage times: First arrival-time densities for free Lévy flights have heavy tails whose exponent depends on the Lévy index α and differs from the universal Sparre Andersen result.For Gaussian motion, first arrival and first passage coincide; Lévy jumps can cross a sink without hitting it.

XI. CONCLUSIONS

The conclusions summarize Lévy flights as self-similar Lévy processes and report analytical and numerical results for stationary distributions and temporal characteristics. They emphasize confinement in sufficiently steep potentials and the suitability of arrival and residence times for Lévy-flight analysis.

  • XI. CONCLUSIONS: The paper introduces Lévy flights as self-similar Lévy processes and derives a fractional Fokker-Planck equation from a Langevin equation with symmetric α-stable Lévy noise.It uses a functional-analysis approach to derive the equation and describes anomalous diffusion in Lévy-flight form.
  • XI. CONCLUSIONS: For Cauchy driving noise in symmetric smooth monostable potentials, the stationary distributions are bimodal and narrow as potential steepness increases or noise intensity decreases.The variance is finite for quartic and steeper potential profiles, indicating confinement of Lévy-flight motion.
  • XI. CONCLUSIONS: Arrival and residence times are more appropriate than mean first passage times for investigating Lévy flights across different potential profiles.The conclusions also highlight difficulties in formulating correct mean-first-passage boundary conditions.
Loading 0810.1492v1…