Levy Flight: Power-Law Random Walks

Q: What makes a Levy flight different from ordinary Brownian motion?

Brownian motion draws step lengths from a Gaussian distribution, which has a finite variance and effectively no extreme outliers. A Levy flight draws step lengths from a heavy-tailed power law, which has infinite variance for stability exponents alpha less than 2, so occasional very long jumps are common enough to dominate the overall displacement.

Q: What happens at alpha = 2 and alpha close to 1?

At alpha = 2 the step-length distribution's tail is thin enough that the process reduces exactly to ordinary Brownian motion. As alpha approaches 1, the tail gets heavier and heavier, long jumps become more frequent relative to short ones, and the walk starts to look like a sequence of short local clusters punctuated by occasional huge leaps.

Q: Is there solid evidence that real animals actually forage using Levy flights?

Levy-like movement patterns have been reported in wandering albatrosses, marine predators and some insects, and mathematically an optimal Levy exponent near alpha ≈ 2 does maximise search efficiency for sparse, randomly distributed targets. But later re-analyses showed some original datasets and statistical methods were flawed, and simpler alternatives, such as correlated random walks combined with the patchy real-world distribution of prey, can produce similar-looking trajectories, so the debate over which mechanism truly explains most animal movement remains active.

One change to a random walk

An ordinary random walk picks a random direction and a step length drawn from a distribution with a finite, well-behaved variance — usually a Gaussian. Over many steps this produces the familiar, gently meandering path of Brownian motion, where the average distance travelled grows with the square root of the number of steps and no single step ever dominates the whole trajectory. A Levy flight makes exactly one change: the step-length distribution is replaced with a heavy-tailed power law, and that single change is enough to produce qualitatively different-looking motion — long stretches of small, tightly clustered steps interrupted by sudden, disproportionately large jumps.

live demo · clustered steps and occasional long jumps● LIVE

The stability exponent alpha

The step-length distribution is a power law with a tunable parameter, the stability (or Levy) exponent alpha, typically restricted to the range 0 < alpha ≤ 2:

P(step length >= L)  ~  L^(-alpha)      for large L

alpha = 2   → tail thin enough that variance is finite
              → the walk reduces exactly to ordinary Brownian motion
1 < alpha < 2 → infinite variance, heavy tail
              → rare very long jumps dominate the total displacement
alpha → 1   → the heaviest practical tail, jumps most extreme

The key mathematical fact is that for alpha < 2, the step-length distribution has infinite variance — not because any single step is infinite, but because as you sample more and more steps, the largest one keeps growing without settling toward any fixed typical scale. That is exactly what makes the walk look fractal and self-similar: zoom into any cluster of short steps and, statistically, it resembles the whole trajectory.

Why the histogram looks like a straight line

A defining signature of a power-law distribution is that plotting the frequency of step lengths against step length on log-log axes gives a straight line, with slope −alpha. A Gaussian distribution, plotted the same way, curves sharply downward because its tail decays exponentially fast rather than polynomially — which is exactly why extreme events are so much rarer under a Gaussian than under a power law with the same average step size. This log-log straight-line test is the standard way to check, from real movement data, whether a trajectory is consistent with Levy-flight statistics.

Why Levy flights show up in foraging and search

When a searcher is looking for sparse, randomly and unpredictably distributed targets, and has no information about where they are, a Levy walk with alpha close to 2 turns out to be close to mathematically optimal: short steps thoroughly search a local area, and occasional long jumps relocate the search to an entirely new region rather than wasting time revisiting already-searched ground — a much more efficient strategy than either pure Brownian motion (too locally repetitive) or a fixed-length step (blind to the local density of the search area). This is the theoretical basis behind reports of Levy-like foraging patterns in wandering albatrosses and other predators, and behind Levy-flight-inspired variants of metaheuristic optimisation algorithms like Cuckoo Search.

Frequently asked questions

What makes a Levy flight different from ordinary Brownian motion?

Brownian motion draws step lengths from a Gaussian distribution, which has a finite variance and effectively no extreme outliers. A Levy flight draws step lengths from a heavy-tailed power law, which has infinite variance for stability exponents alpha less than 2, so occasional very long jumps are common enough to dominate the overall displacement.

What happens at alpha = 2 and alpha close to 1?

At alpha = 2 the step-length distribution's tail is thin enough that the process reduces exactly to ordinary Brownian motion. As alpha approaches 1, the tail gets heavier and heavier, long jumps become more frequent relative to short ones, and the walk starts to look like a sequence of short local clusters punctuated by occasional huge leaps.

Is there solid evidence that real animals actually forage using Levy flights?

Levy-like movement patterns have been reported in wandering albatrosses, marine predators and some insects, and mathematically an optimal Levy exponent near alpha ≈ 2 does maximise search efficiency for sparse, randomly distributed targets. But later re-analyses showed some original datasets and statistical methods were flawed, and simpler alternatives, such as correlated random walks combined with the patchy real-world distribution of prey, can produce similar-looking trajectories, so the debate over which mechanism truly explains most animal movement remains active.

Try it live

Everything above runs in your browser — open Levy Flight and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.

▶ Open Levy Flight simulation