Random Walk: Models and Applications

• Random walk is a stochastic process characterized by successive random steps, serving as a foundation for models in probability and statistical physics.
• Key variants such as discrete-time walks, CTRWs, RWRE, and Lévy walks demonstrate phenomena ranging from subdiffusion and recurrence to superdiffusive and ballistic behaviors.
• These models enable practical insights into anomalous diffusion, environmental randomness, and complex dynamics in varied fields like biology, physics, and network theory.

A random walk is a discrete (or continuous) stochastic process describing the evolution of a system through successive random steps in a defined space. Foundational in probability, statistical physics, and numerous applications, random walks generalize to a wide variety of step rules, spatial topologies, and temporal structures. The class encompasses simple symmetric walks, models with site- or environment-dependent transition rules, and forms inducing anomalous diffusion such as Lévy walks.

1. Fundamental Models and Definitions

The standard discrete-time, integer-lattice random walk is the Markov process $X_{n+1} = X_n + \xi_{n+1}$, where $\{\xi_i\}$ are i.i.d. random variables, frequently chosen as symmetric ±1 increments. A continuous-time generalization allows steps to occur at random times, with waiting times drawn from a specified distribution.

A salient extension is the random walk in random environment (RWRE), in which the transition probabilities themselves are random variables sampled from a stationary or i.i.d. law, and held fixed throughout the trajectory (the "quenched" setting). Formally, for a lattice $\mathbb{Z}^d$, each site $x$ is assigned a transition vector $\omega(x)$, drawn from a probability measure $\mathbb{P}$, specifying jump probabilities to neighboring sites. The walk's path is then governed by the Markov kernel associated to the fixed $\omega$ (Drewitz et al., 2013).

Continuous time random walks (CTRWs), introduced by Montroll and Weiss, decouple jump lengths $X_i$ from waiting times $T_i$, permitting subdiffusive or anomalous statistics when waiting times have heavy tails (Arutkin et al., 2023).

2. Classification and Key Phenomena

2.1 Transience, Recurrence, and Ballisticity

Transience and recurrence characterize whether the process returns infinitely often to its starting point. Classical results for RWRE in one dimension invoke the key parameter $\rho(x) = \omega(x, -1)/\omega(x, +1)$. Solomon’s trichotomy gives:

• $\{\xi_i\}$0: Drift to $\{\xi_i\}$1 (right transience)
• $\{\xi_i\}$2: Drift to $\{\xi_i\}$3
• $\{\xi_i\}$4: Recurrence with sub-diffusive scaling: $\{\xi_i\}$5 (“Sinai regime”) (Drewitz et al., 2013)

Ballisticity concerns linear growth: $\{\xi_i\}$6. In higher dimensions, criteria such as Sznitman's conditions (T)$\{\xi_i\}$7, (T′), and polynomial exit decay rates distinguish true ballistic regimes (Drewitz et al., 2013).

2.2 Anomalous Diffusion and Lévy Walks

Brownian motion yields mean squared displacement (MSD) scaling as $\{\xi_i\}$8. In contrast, the Lévy walk model, which couples spatial increments with random durations via a fixed velocity $\{\xi_i\}$9, exhibits superdiffusion for power-law distributed flight-times $\mathbb{Z}^d$0, yielding $\mathbb{Z}^d$1, $\mathbb{Z}^d$2. The velocity constraint regularizes the otherwise divergent moments of Lévy flights (Zaburdaev et al., 2014). The Lévy walk is broadly applicable to photon transport in disordered media, cold atom dynamics, and biological foraging.

3. Generalizations: Random Media, Variable Range, and Non-Ergodic Dynamics

3.1 RWRE and Bounded-Jump Variants

RWRE can be constructed with arbitrary finite-range jumps; for example, the (1,2) RWRE on $\mathbb{Z}^d$3 allows jumps of $\mathbb{Z}^d$4, with site-dependent probabilities $\mathbb{Z}^d$5. In transient regimes ($\mathbb{Z}^d$6), the proportion of sites in $\mathbb{Z}^d$7 ever visited converges almost surely to $\mathbb{Z}^d$8, in contrast to nearest-neighbor walks which visit all sites ($\mathbb{Z}^d$9). This “range gap” arises from the renewal structure in the record process, distinguishing the long-term coverage properties (Wang, 2016).

3.2 Variable-Range Random Walks

When walkers step among randomly distributed sites, with jump rates dependent on inter-site distances $x$0, the diffusive or localized regime depends on the decay of $x$1 and site density $x$2. In one dimension, exponentially decaying $x$3 yields a non-diffusive to diffusive transition at $x$4, where $x$5 is the decay length. In three dimensions, the transition is only observed if $x$6 decays super-Gaussianly at large $x$7; otherwise, the system remains diffusive (Odagaki, 2021).

4. Non-Classical and Structured Random Walks

4.1 Random Walks with Hyperbolic Transition Laws

Random walks with hyperbolic probabilities assign site-dependent transition probabilities $x$8 for parameter $x$9. These processes interpolate between simple symmetric and globally biased walks, exhibit ballistic drift far from the origin, and are non-ergodic: not all statistics are accessible from a single trajectory due to a positive probability of non-visit to certain sites. The hyperbolic structure arises from embedding the integer lattice in hyperbolic geometry, with transition probabilities inversely proportional to the projected Euclidean distances (Montero, 2019).

4.2 Lattice-Preserving and Fixed-Point Walks

Discrete walks on $\omega(x)$0, such as the Jacobi–walk, can be constructed to preserve a prescribed stationary distribution, e.g., $\omega(x)$1. For step set $\omega(x)$2 and specific transition rules derived from Jacobi theta-series identities, the stationary measure is the Gaussian. For $\omega(x)$3, at least 96.39% of steps are of size ±1; the rest are zero-steps. Extensions can include step-sets $\omega(x)$4 with the same fixed-point property. These constructions connect to online partial-coloring and discrepancy minimization algorithms (Liu et al., 2021).

5. Continuous Time and Doubly Stochastic Random Walks

The classical continuous-time random walk (CTRW) framework is extended in doubly stochastic models by allowing the rate of jumps to itself be random, possibly varying in time as a nontrivial stochastic process. This “DSCTRW” generalizes standard CTRW via a mixture over Poisson processes with random rates, producing solutions for the propagator,

$\omega(x)$5

where $\omega(x)$6 is the path-wise integrated jump rate. This leads to Brownian MSD scaling, but possibly with non-Gaussian propagators at finite times, subsuming diffusing-diffusivity models and allowing analytic computation of non-trivial PDFs, parameterized by the statistics of $\omega(x)$7 (Arutkin et al., 2023).

6. Open Problems and Research Directions

For RWRE, outstanding questions include the zero–one law for directional transience in $\omega(x)$8; whether directional transience implies ballisticity under uniform ellipticity; the equivalence of various ballisticity conditions, particularly Sznitman's (T) and (T′); and the existence of invariant measures in non-elliptic, non-balanced, or higher-dimensional environments (Drewitz et al., 2013).

For Lévy walks, central unresolved directions include rigorous characterization in dimensions $\omega(x)$9, the effect of correlated step sizes and quenched disorder, detailed first-passage statistics, systematic fractional kinetic descriptions, and quantification of ergodicity breaking in systems exhibiting aging (Zaburdaev et al., 2014).

Further, the exact spectral gap optimization, minimal-support step-sets preserving nontrivial measures, and precise rules for coverage properties in variable-range or bounded-jump RWRE remain open for analytical or numerical investigation (Liu et al., 2021, Wang, 2016, Odagaki, 2021).

7. Comparative Table: RWRE, Lévy Walks, and DSCTRW

Model	Key Mechanism	Characteristic Phenomena
RWRE	Site-random transition laws	Subdiffusion, Sinai regime, ballisticity, non-reversibility in $\mathbb{P}$0 (Drewitz et al., 2013)
Lévy Walk	Power-law timed flights, velocity coupling	Superdiffusion ($\mathbb{P}$1), finite speed, ergodicity breaking (Zaburdaev et al., 2014)
DSCTRW	Random time-dependent rate	Brownian yet non-Gaussian diffusion, exact propagator characterization, universality at long times (Arutkin et al., 2023)

Each model typifies a distinct mechanism for random transport, producing unique scaling laws, ergodic properties, and response to environmental disorder.