[论文解读] Dynamics on Unimodular Random Graphs

François Baccelli, Mir-Omid Haji-Mirsadeghi+1 more|arXiv (Cornell University)|Aug 21, 2016

Stochastic processes and statistical mechanics被引用 3

一句话总结

本文引入了顶点位移——在单模随机网络上具有图同构不变性的协变动力系统——并基于分量与轨线叶状结构（foils）的基数，对其轨道与稳定流形进行了分类。关键结果是在连通分量内对叶状结构进行三类分类：F/F（有限多个有限叶状结构）、I/F（无限多个有限叶状结构）和 I/I（无限多个无限叶状结构），并为每类推导出相应的结构性质。

ABSTRACT

This paper is centered on covariant dynamics on random graphs and random networks (marked graphs), which can be described as rules to navigate the vertices that are preserved by graph/network isomorphisms. Such dynamics are referred to as vertex-shifts here. Unimodular random networks can be heuristically described as networks seen from a vertex chosen uniformly at random, both in the finite and in infinite network cases. Some general ways of constructing unimodular random networks are reviewed together with instances of vertex-shifts. The main result of the paper is a classification of vertex-shifts on unimodular random networks. Each such vertex-shift partitions the vertices into a collection of connected components and foils, which correspond to the orbits and the stable manifold of the dynamics, respectively. The classification is based on the cardinality of the connected components and foils. It is shown that up to an event of zero probability, there are three types of foliations in a connected component: F/F (with finitely many finite foils), I/F (infinitely many finite foils), and I/I (infinitely many infinite foils). Distinctive properties of each type are also derived. An infinite connected component of the graph of a vertex-shift is a tree which, in the unimodular case, shares some similarities with a critical branching process. Such trees are referred to as Eternal Family Trees here. Construction techniques of such trees are also discussed. These lead to additional structural results on the graphs of vertex shifts in each case of the classification. The results are illustrated by concrete examples stemming from classical domains of probability theory: random graphs, branching processes and stationary point processes.

研究动机与目标

• 形式化并分析在图同构下不变的随机网络上的协变动力系统（称为顶点位移）。
• 基于轨道与稳定流形（叶状结构）的基数，对单模随机网络上的顶点位移动力系统结构进行分类。
• 表征顶点位移图中无限连通分量的几何与概率特性，特别是在单模网络背景下的表现。
• 建立顶点位移动力系统与经典概率模型（如分支过程与平稳点过程）之间的联系。

提出的方法

• 将顶点位移定义为在带标记图上具有图同构不变性的导航规则，以建模随机网络上的动力系统。
• 利用单模性性质，将网络视为从均匀随机顶点视角观察的结构，从而实现概率分析。
• 将顶点按顶点位移作用下的轨道（连通分量）与稳定流形（叶状结构）进行划分。
• 根据叶状结构的数量与大小的有限性或无限性，对每个连通分量内的叶状结构进行分类。
• 将无限连通分量视为“永恒家谱树”——其结构类似于临界分支过程的单模树。
• 通过随机图、分支过程与平稳点过程构造具体例子，以说明分类结果与结构性质。

实验结果

研究问题

• RQ1 在单模随机网络上，顶点位移的轨道内可能存在的叶状结构（稳定流形）的结构性质有哪些类型？
• RQ2 轨道与叶状结构的基数如何影响顶点位移图的整体几何结构？
• RQ3 在单模设定下，顶点位移图的无限连通分量具有哪些特征性质？
• RQ4 单模网络上的顶点位移动力系统在多大程度上类似于临界分支过程？
• RQ5 如何利用经典概率模型（如随机图与点过程）来构造并说明顶点位移动力系统？

主要发现

• 在概率为零的事件之外，连通分量内恰好存在三种叶状结构类型：F/F（有限多个有限叶状结构）、I/F（无限多个有限叶状结构）和 I/I（无限多个无限叶状结构）。
• 顶点位移图的无限连通分量是树结构，其结构与临界分支过程相似，被称为“永恒家谱树”。
• 每类叶状结构——F/F、I/F 与 I/I——均表现出独特的概率与几何特性，例如常返性或常返性特征。
• 顶点位移的分类在图同构下保持不变，确保了在单模框架下的稳健性。
• 基于随机图、分支过程与平稳点过程的构造技术，可产生实现全部三类叶状结构的具体例子。
• 单模设定确保了动力系统在保持局部与全局网络结构平衡的同时，能够推导出普适的结构性结果。