Distributional lattices on Riemannian symmetric spaces

Elliot PaquetteContemporary 2018 · DOI · arXiv

This paper, authored by Elliot Paquette and published in 2017, introduces a new framework for studying random structures on symmetric spaces. By bridging the gap between smooth Riemannian geometry and discrete probability, the work has significant implications for our understanding of random walks in curved environments.

Abstract

A Riemannian symmetric space is a manifold in which it is possible to reflect all geodesics through a point by an isometry of the space. On such spaces, we introduce the notion of a distributional lattice, generalizing the notion of a classical lattice. Distributional lattices exist in any Riemannian symmetric space; for instance, the Voronoi tessellation of a stationary Poisson point process serves as a primary example. We show that for an appropriate notion of amenability, the amenability of a distributional lattice is equivalent to the amenability of the ambient space. Using this equivalence, we demonstrate that the simple random walk on any non-amenable distributional lattice has positive embedded speed. For non-positively curved, simply connected spaces, we further show that the simple random walk on a Poisson–Voronoi tessellation has positive graph speed.

The Problem: From Rigid to Random Lattices

In mathematics, a lattice is a discrete set of points that "fills" a space evenly, such as the integer grid $\mathbb{Z}^d$ in Euclidean space $\mathbb{R}^d$. While classical lattices are well-studied, they are rigid and deterministic.

The author addresses a more flexible question: Can we define a "random lattice" that captures the statistical essence of a classical lattice? Furthermore, how do the geometric properties of the underlying smooth manifold—like its curvature or "amenability"—translate to these random discrete structures? Specifically, the paper investigates whether a random walk on a random graph "feels" the large-scale geometry of the space it inhabits.

Methodology

The paper employs several sophisticated tools from probability and geometry to construct and analyze these structures:

  • Distributional Lattices: Defined as the Palm process of a stationary point process (like a Poisson process) where the expected volume of a Voronoi cell and the expected degree of the Delaunay graph are both finite.
  • Unimodular Random Networks: The author uses the framework of unimodularity—a property where a random rooted graph "looks the same" from any vertex on average—to apply the mass-transport principle.
  • Voronoi and Delaunay Geometry: The paper constructs graphs by partitioning the manifold into Voronoi cells (regions closest to a specific point) and connecting points whose cells share a boundary to form a Delaunay triangulation.
Figure 1: A Poisson-Voronoi tessellation and its dual Delaunay graph in the Poincaré disk model of hyperbolic space. The red lines denote cell boundaries, while blue lines represent the graph edges.

Key Insights: The Amenability Equivalence

The most significant insight is that a manifold is amenable (roughly, its volume does not grow "too fast" relative to its radius) if and only if every distributional lattice on it is invariantly amenable. This provides a rigorous link between the smooth world of manifolds and the discrete world of random graphs.

In non-amenable spaces, such as hyperbolic spaces, there is "so much room" at the edges that a random walk is essentially pushed away from its starting point. The paper proves this intuition holds even for highly irregular, random graphs.

Figure 3: Two distinct realizations of distributional lattices in hyperbolic space, demonstrating how the statistical properties remain invariant despite local random fluctuations.

Key Results

  • Theorem 1.7 (Amenability Equivalence): For a Riemannian symmetric space $M$, $M$ is amenable if and only if every distributional lattice on $M$ is invariantly amenable.
  • Corollary 1.10 (Positive Speed): In any non-amenable symmetric space, a simple random walk on a distributional lattice has positive speed. This means the distance from the starting point grows linearly with time ($d(X_n, o) \approx sn$ for some $s > 0$), rather than the sub-linear growth (like $\sqrt{n}$) seen in Euclidean space.
  • Graph Speed vs. Embedded Speed: For non-positively curved, simply connected spaces, the author proves positive speed not just in terms of the manifold distance (embedded speed), but also in terms of the number of steps taken on the graph (graph speed).

Limitations and Future Work

While groundbreaking, the results are specifically tailored to Riemannian symmetric spaces. While these include many important examples (Euclidean, hyperbolic, spheres), the behavior on more general, less symmetric manifolds remains an open question. Additionally, some results regarding "graph speed" require the space to be non-positively curved and simply connected. Extending these to spaces with mixed or positive curvature would require new techniques to handle the "looping" nature of geodesics in those environments.

Summary updated

Abstract

A Riemannian symmetric space is a Riemannian manifold in which it is possible to reflect all geodesics through a point by an isometry of the space. On such spaces, we introduce the notion of a distributional lattice, generalizing the notion of lattice. Distributional lattices exist in any Riemannian symmetric space: the Voronoi tessellation of a stationary Poisson point process is an example. We show that for an appropriate notion of amenability, the amenability of a distributional lattice is equivalent to the amenability of the ambient space. Using this equivalence, we show that the simple random walk on any nonamenable distributional lattice has positive embedded speed. For nonpositively curved, simply connected spaces, we show that the simple random walk on a Poisson–Voronoi tessellation has positive graph speed by developing some additional structure for Poisson–Voronoi tessellations.
BibTeX
@inproceedings{paquette2018distributional,
  author = {Elliot Paquette},
  title = {{Distributional lattices on Riemannian symmetric spaces}},
  booktitle = {Contemporary},
  year = {2018},
  doi = {10.1090/conm/719/14470},
}
Figure