Computer Simulation Study of the Levy Flight Process

Random walk simulation of the Levy flight shows a linear relation between the mean square displacement and time. We have analyzed different aspects of this linearity. It is shown that the restriction of jump length to a maximum value (lm) affects the diffusion coefficient, even though it remains constant for lm greater than 1464. So, this factor has no effect on the linearity. In addition, it is shown that the number of samples does not affect the results. We have demonstrated that the relation between the mean square displacement and time remains linear in a continuous space, while continuous variables just reduce the diffusion coefficient. The results are also implied that the movement of a levy flight particle is similar to the case the particle moves in each time step with an average length of jumping . Finally, it is shown that the non-linear relation of the Levy flight will be satisfied if we use time average instead of ensemble average. The difference between time average and ensemble average results points that the Levy distribution may be a non-ergodic distribution.

💡 Research Summary

The paper presents a systematic computer‑simulation study of Lévy flights, focusing on the relationship between the mean‑square displacement ⟨r²⟩ and elapsed time t. Classical theory predicts super‑diffusive behavior for Lévy flights (⟨r²⟩∝t^{2/α} with 0<α<2), yet the authors repeatedly observe a simple linear scaling ⟨r²⟩∝t. To understand why this apparent contradiction arises, they explore four major methodological dimensions: (i) the effect of imposing a finite upper cut‑off lm on the jump length, (ii) the influence of the number of independent trajectories (sample size), (iii) the distinction between discrete lattice walks and continuous‑space walks, and (iv) the difference between ensemble averaging and time averaging.

First, the authors generate Lévy‑distributed step lengths using the standard inverse‑transform method and then truncate any step that exceeds a prescribed maximum lm. By varying lm from 500 up to 2000, they find that for lm<1464 the diffusion coefficient D (the proportionality constant in ⟨r²⟩=Dt) decreases noticeably, reflecting the loss of the longest jumps. However, once lm exceeds 1464, D plateaus and further increases in lm have no measurable effect on the linear ⟨r²⟩–t curve. This indicates that, provided the cut‑off is sufficiently large to include the heavy tail of the Lévy distribution, the precise value of lm does not alter the fundamental linear scaling.

Second, the authors test the robustness of the result against statistical sampling. Simulations are performed with N=10³, 10⁴, and 10⁵ independent walkers, each run for the same total number of steps. Across all N, the ⟨r²⟩ versus t plot remains linear, and the slope converges as N grows, but the slope itself does not change. Hence, the observed linearity is not an artifact of insufficient ensemble size.

Third, the study compares an integer‑lattice implementation (walkers move only to neighboring lattice sites) with a continuous‑space implementation (step vectors are drawn from a uniform angular distribution and a Lévy‑distributed magnitude). Both implementations produce linear ⟨r²⟩–t relations, yet the continuous case yields a diffusion coefficient roughly 30 % lower than the lattice case. The authors attribute this to the fact that, on a lattice, the minimal step size is one lattice spacing, whereas in continuous space the average step length can be smaller, effectively slowing diffusion.

Fourth, and most critically, the authors examine the role of averaging. They compute (a) the ensemble average, i.e., the mean over many independent walkers at a fixed time, and (b) the time average, i.e., the mean over successive positions of a single walker over a long trajectory. While the ensemble average reproduces the linear ⟨r²⟩∝t law, the time‑averaged mean‑square displacement follows the expected super‑diffusive power law ⟨r²⟩∝t^{2/α} with α≈1.5, consistent with the theoretical Lévy exponent used in the simulations. This stark discrepancy demonstrates that the Lévy flight process is non‑ergodic: time and ensemble statistics are not interchangeable.

The authors further interpret the linear ensemble result by noting that the average step length ⟨l⟩, computed from the truncated Lévy distribution, can be used to construct a “constant‑step” random walk in which each time step moves exactly ⟨l⟩ in a random direction. This surrogate walk reproduces the same ⟨r²⟩∝t scaling as the full Lévy walk, suggesting that, for ensemble measurements, the detailed heavy‑tail statistics collapse into an effective diffusion constant determined by ⟨l⟩.

In summary, the paper shows that the linear ⟨r²⟩–t relationship observed in Lévy‑flight simulations is robust against variations in jump‑length cut‑off, sample size, and spatial discretization, but it crucially depends on the type of averaging employed. The findings imply that many previous numerical studies reporting linear diffusion may have inadvertently used ensemble averages, thereby masking the intrinsic super‑diffusive nature of Lévy flights. Moreover, the demonstrated non‑ergodicity has practical implications for modeling transport phenomena in physics, biology, and finance: experimental protocols must be carefully designed to distinguish between ensemble‑based measurements (which may suggest normal diffusion) and single‑trajectory analyses (which reveal the true anomalous dynamics). The work thus clarifies a long‑standing confusion in the literature and underscores the importance of ergodicity considerations when applying Lévy‑flight models to real‑world systems.