Abstract: We prove a formula for the speed of distance stationary random sequences. A particular case is the classical formula for the largest Lyapunov exponent of an i.i.d. product of two by two matrices in terms of a stationary measure on projective space. We apply this result to Poisson-Delaunay random walks on Riemannian symmetric spaces. In particular, we obtain sharp estimates for the asymptotic behavior of the speed of hyperbolic Poisson-Delaunay random walks when the intensity of the Poisson point process goes to zero. This allows us to prove that a dimension drop phenomena occurs for the harmonic measure associated to these random walks. With the same technique we give examples of co-compact Fuchsian groups for which the harmonic measure of the simple random walk has dimension less than one.
Introduction
In probability and geometry, a fundamental question is: "How fast does a random process move away from its starting point?" This rate is known as the speed or linear drift. While simple additive processes are well-understood, more complex sequences—such as products of random matrices—require sophisticated tools like the classical Furstenberg formula. This formula relates the Lyapunov exponent (the speed) to a stationary measure on a projective space.
This paper generalizes these ideas by deriving a Furstenberg-type formula for a much broader class of "distance stationary" random sequences. These are sequences where the distribution of distances between points remains invariant under shifts. The authors apply this framework to random walks on Poisson-Delaunay graphs in hyperbolic space, revealing striking asymptotic behaviors as the point density (intensity) vanishes.
A Generalized Furstenberg Formula
The core theoretical contribution is a new integral formula for the speed $\ell$ of a distance stationary sequence:
$$\mathbb{E}(\ell) = -\mathbb{E}(\xi(x_0) - \xi(x_1))$$
In this expression, $\xi$ represents a random horofunction. In geometric terms, a horofunction measures distance relative to a point at "infinity." The key insight is that the speed of a sequence is fundamentally captured by the average increment of a horofunction that tracks the sequence's trajectory toward the boundary of the space.
To apply this formula to random walks on graphs, the authors demonstrate that random walks on Poisson-Delaunay graphs are distance stationary when viewed under a specific "degree-biased" probability measure. This connection allows the geometric properties of the graph to be translated into statistical properties of the walk.
Poisson-Delaunay Random Walks
The authors test their formula on Poisson-Delaunay graphs constructed within $d$-dimensional hyperbolic space $\mathbb{H}^d$. These graphs are generated from a Poisson point process with intensity $\lambda$. Points are connected if they share a Voronoi boundary (i.e., they are "neighbors" in terms of territory), creating a random graph that reflects the underlying hyperbolic geometry.
As the intensity $\lambda$ approaches zero, the graph becomes increasingly sparse. In this "low-intensity" regime, the distance to the nearest neighbors—and thus the step size of the random walk—grows logarithmically. The authors use hyperbolic geometric estimates to characterize the behavior of horofunctions and neighbor distances in this limit.
Key Results: Speed and Dimension Drop
The application of the Furstenberg-type formula yields several significant results:
-
Speed Asymptotics: For a Poisson-Delaunay random walk with intensity $\lambda$, the speed $\ell_\lambda$ satisfies: $$\lim_{\lambda \to 0} \frac{\ell_\lambda}{\frac{2}{d-1} \log(\lambda^{-1})} = 1$$ This provides a sharp estimate of how point density affects the walk's progress.
-
Graph Speed: The authors show that as intensity goes to zero, the "graph speed" (steps per unit of time) approaches its maximum possible value of 1.
-
Quantitative Dimension Drop: One of the most striking findings is that the "harmonic measure" (the distribution of where the walk eventually "hits" the boundary at infinity) doesn't spread out over the whole boundary. Instead, it concentrates on a fractal-like subset with a lower dimension: $$\limsup_{\lambda \to 0} \dim(\nu_\lambda) \le \frac{d-1}{2}$$ Since the full boundary has dimension $d-1$, this is a significant "drop" and provides a rare rigorous proof of a phenomenon often observed in complex random walks.
-
Fuchsian Groups: The techniques also extend to co-compact Fuchsian groups, providing new examples where the harmonic measure of a simple random walk has a dimension strictly less than one.
Discussion and Future Work
While the paper establishes a powerful general formula, several questions remain. The proof of the dimension drop for Poisson-Delaunay walks is specifically for the low-intensity regime; whether this drop occurs for all intensities remains an open question. Additionally, while the Furstenberg-type formula is general, the specific applications in the paper focus heavily on hyperbolic spaces. Extending these sharp asymptotic results to other Riemannian symmetric spaces or broader classes of Fuchsian groups is a natural next step.
Summary updated