Abstract
We prove a formula for the speed of distance stationary random sequences generalizing the law of large numbers of Karlsson and Ledrappier. A particular case is the classical formula for the largest Lyapunov exponent of i.i.d. matrix products, but our result has applications in various different contexts. In many situations it gives a method to estimate the speed, and in others it allows to obtain results of dimension drop for escape measures related to random walks. We show applications to stationary reversible random trees with conductances, Bernoulli bond percolation of Cayley graphs, and random walks on cocompact Fuchsian groups.
Introduction
In probability theory and geometric group theory, a fundamental question concerns the "speed" or linear drift of a random process—the rate at which a sequence of points moves away from its origin in a metric space. While the Law of Large Numbers (LLN) often guarantees that this speed exists, deriving an explicit formula to calculate or estimate it has historically been a difficult, context-specific task.
This paper addresses this gap by introducing a unified integral formula for the speed of distance-stationary sequences. This class of sequences generalizes standard random walks to include processes where steps are not necessarily independent, provided the joint distribution of distances between points remains shift-invariant. By leveraging the geometry of "points at infinity," the authors provide a framework that unifies classical results, such as the Furstenberg formula for matrix products, and extends them to complex structures like percolation clusters and hyperbolic tilings.
Theoretical Framework
Distance-Stationary Sequences
The authors define a sequence $(x_n)$ in a metric space as distance-stationary if the joint distribution of the distances $d(x_m, x_n)$ is invariant under the shift operator. This definition is broad enough to encompass not only i.i.d. random walks but also more complex dependent processes.
Horofunctions and the Geometry of Infinity
To describe the "direction" of a sequence's progression, the research utilizes the horofunction compactification of metric spaces. Horofunctions ($\xi$) serve as a mathematical representation of points at infinity. As a distance-stationary sequence progresses, it converges toward a specific horofunction. The authors demonstrate that this random horofunction is essentially constructed from the "past" of the sequence (negative time indices).
The General Speed Formula
The core contribution of the paper is Theorem A, which establishes a universal identity for the speed $\ell$:
$$\mathbb{E}(\ell) = -\mathbb{E}(\xi(x_1) - \xi(x_0))$$
This formula relates the global, long-term speed to the expected change in the horofunction over a single step. The primary insight is that the interaction between the "past" (the direction $\xi$) and the very first step of the "future" ($x_1$) is sufficient to determine the average drift. This result is proven using Kingman’s subadditive ergodic theorem and Komlós’ theorem to manage the convergence of random probability measures on the space of horofunctions.
Applications and Case Studies
The authors evaluate the utility of this formula across several distinct mathematical models:
1. Lyapunov Exponents
The paper shows that the classical Furstenberg formula for the largest Lyapunov exponent of i.i.d. matrix products is a specific realization of this general distance-stationary formula. This provides a geometric unification of previously disparate probabilistic results.
2. Stationary Reversible Random Trees
For random trees with conductances ($A, B, C$), the authors derive a specific speed formula based on effective conductances: $$\ell = \mathbb{E}\left[\frac{AC}{AB + AC + BC}\right]$$ This offers a direct method for calculating drift in branching structures and canopy trees.
3. Percolation on Cayley Graphs
In the context of Bernoulli bond percolation, the authors identify a "hyperbolicity" condition. They prove that for percolation parameters $p$ close to 1, the speed of the walk on the infinite cluster remains positive, provided certain conditions regarding the sum of horofunctions over generators are met.
4. Hyperbolic Tilings and Dimension Drop
In regular tilings of the hyperbolic plane ($T_{P,Q}$), the authors investigate the relationship between speed and the harmonic measure (the distribution of the walk's limit point at infinity). They prove that as the number of polygons per vertex $Q$ increases, the speed $\ell_p$ scales as $2 \log(Q) - O(\log \log Q)$.
Crucially, they establish a "dimension drop," showing that the Hausdorff dimension of the harmonic measure satisfies: $$\limsup_{Q \to \infty} \text{dim}(\nu_{P,Q}) \leq 1/2$$ This confirms that for large $Q$, the random walk is "inefficient," and its endpoint distribution becomes highly concentrated and singular relative to the visual measure of the boundary.
Conclusion and Future Directions
While the established formula is universal, the authors acknowledge that the distribution of the random horofunction $\xi$ can be extremely difficult to compute explicitly for non-i.i.d. sequences. Many of the most striking results, such as the dimension drop, are currently asymptotic in nature.
Future research may apply these methods to other exotic models, such as Poisson-Delaunay walks or Planar Stochastic Hyperbolic Infinite Triangulations (PSHT). These models are of significant interest in modern probability and theoretical physics, and the distance-stationary framework provides a robust new tool for their analysis.
Summary updated