arXiv subjects

Russell Lyons

Publications and source records attributed to Russell Lyons.

At least 37 records · Page 2Linked to original sources

arXiv2017

Sharp Bounds on Random Walk Eigenvalues via Spectral Embedding

Spectral embedding of graphs uses the top k non-trivial eigenvectors of the random walk matrix to embed the graph into R^k. The primary use of this embedding has been for practical spectral clustering algorithms [SM00,NJW02]. Recently, spectral embedding was studied from a theoretical perspective to prove higher order variants of Cheeger's inequality [LOT12,LRTV12]. We use spectral embedding to provide a unifying framework for bounding all the eigenvalues of graphs. For example, we show that for any finite graph with n vertices and all k >= 2, the k-th largest eigenvalue is at most 1-Omega(k^3/n^3), which extends the only other such result known, which is for k=2 only and is due to [LO81]. This upper bound improves to 1-Omega(k^2/n^2) if the graph is regular. We generalize these results, and we provide sharp bounds on the spectral measure of various classes of graphs, including vertex-transitive graphs and infinite graphs, in terms of specific graph parameters like the volume growth. As a consequence, using the entire spectrum, we provide (improved) upper bounds on the return probabilities and mixing time of random walks with considerably shorter and more direct proofs. Our work introduces spectral embedding as a new tool in analyzing reversible Markov chains. Furthermore, building on [Lyo05], we design a local algorithm to approximate the number of spanning trees of massive graphs.

math.PR↗

arXiv2016

Comparing Graphs of Different Sizes

We consider two notions describing how one finite graph may be larger than another. Using them, we prove several theorems for such pairs that compare the number of spanning trees, the return probabilities of random walks, and the number of independent sets, among other combinatorial quantities. Our methods involve inequalities for determinants, for traces of functions of operators, and for entropy.

math.CO↗

arXiv2016

Factors of IID on Trees

Classical ergodic theory for integer-group actions uses entropy as a complete invariant for isomorphism of IID (independent, identically distributed) processes (a.k.a. product measures). This theory holds for amenable groups as well. Despite recent spectacular progress of Bowen, the situation for non-amenable groups, including free groups, is still largely mysterious. We present some illustrative results and open questions on free groups, which are particularly interesting in combinatorics, statistical physics, and probability. Our results include bounds on minimum and maximum bisection for random cubic graphs that improve on all past bounds.

math.DS↗

arXiv2015

Cycle density in infinite Ramanujan graphs

We introduce a technique using nonbacktracking random walk for estimating the spectral radius of simple random walk. This technique relates the density of nontrivial cycles in simple random walk to that in nonbacktracking random walk. We apply this to infinite Ramanujan graphs, which are regular graphs whose spectral radius equals that of the tree of the same degree. Kesten showed that the only infinite Ramanujan graphs that are Cayley graphs are trees. This result was extended to unimodular random rooted regular graphs by Abért, Glasner and Virág. We show that an analogous result holds for all regular graphs: the frequency of times spent by simple random walk in a nontrivial cycle is a.s. 0 on every infinite Ramanujan graph. We also give quantitative versions of that result, which we apply to answer another question of Abért, Glasner and Virág, showing that on an infinite Ramanujan graph, the probability that simple random walk encounters a short cycle tends to 0 a.s. as the time tends to infinity.

math.PR↗

arXiv2015

Hyperbolic Space Has Strong Negative Type

It is known that hyperbolic spaces have strict negative type, a condition on the distances of any finite subset of points. We show that they have strong negative type, a condition on every probability distribution of points (with integrable distance to a fixed point). This implies that the function of expected distances to points determines the probability measure uniquely. It also implies that the distance covariance test for stochastic independence, introduced by Székely, Rizzo and Bakirov, is consistent against all alternatives in hyperbolic spaces. We prove this by showing an analogue of the Cramér-Wold device.

math.MG↗

arXiv2014

Determinantal Probability: Basic Properties and Conjectures

We describe the fundamental constructions and properties of determinantal probability measures and point processes, giving streamlined proofs. We illustrate these with some important examples. We pose several general questions and conjectures.

math.PR↗

arXiv2014

Invariant Coupling of Determinantal Measures on Sofic Groups

To any positive contraction $Q$ on $\ell^2(W)$, there is associated a determinantal probability measure ${\mathbf P}^Q$ on $2^W$, where $W$ is a denumerable set. Let $Γ$ be a countable sofic finitely generated group and $G = (Γ, \mathsf{E})$ be a Cayley graph of $Γ$. We show that if $Q_1$ and $Q_2$ are two $Γ$-equivariant positive contractions on $\ell^2(Γ)$ or on $\ell^2(\mathsf{E})$ with $Q_1 \le Q_2$, then there exists a $Γ$-invariant monotone coupling of the corresponding determinantal probability measures witnessing the stochastic domination ${\bf P}^{Q_1} \preccurlyeq {\bf P}^{Q_2}$. In particular, this applies to the wired and free uniform spanning forests, which was known before only when $Γ$ is residually amenable. In the case of spanning forests, we also give a second more explicit proof, which has the advantage of showing an explicit way to create the free uniform spanning forest as a limit over a sofic approximation. Another consequence of our main result is to prove that all determinantal probability measures ${\bf P}^Q$ as above are ${\bar d}$-limits of finitely dependent processes. Thus, when $Γ$ is amenable, ${\bf P}^Q$ is isomorphic to a Bernoulli shift, which was known before only when $Γ$ is abelian. We also prove analogous results for sofic unimodular random rooted graphs.

math.PR↗

arXiv2013

Unimodular Random Trees

We consider unimodular random rooted trees (URTs) and invariant forests in Cayley graphs. We show that URTs of bounded degree are the same as the law of the component of the root in an invariant percolation on a regular tree. We use this to give a new proof that URTs are sofic, a result of Elek. We show that ends of invariant forests in the hyperbolic plane converge to ideal boundary points. We also prove that uniform integrability of the degree distribution of a family of finite graphs implies tightness of that family for local convergence, also known as random weak convergence.

math.PR↗

arXiv2012

Random Orderings and Unique Ergodicity of Automorphism Groups

We show that the only random orderings of finite graphs that are invariant under isomorphism and induced subgraph are the uniform random orderings. We show how this implies the unique ergodicity of the automorphism group of the random graph. We give similar theorems for other structures, including, for example, metric spaces. These give the first examples of uniquely ergodic groups, other than compact groups and extremely amenable groups, after Glasner and Weiss's example of the group of all permutations of the integers. We also contrast these results to those for certain special classes of graphs and metric spaces in which such random orderings can be found that are not uniform.

math.DS↗

arXiv2011

Poisson splitting by factors

Given a homogeneous Poisson process on ${\mathbb{R}}^d$ with intensity $λ$, we prove that it is possible to partition the points into two sets, as a deterministic function of the process, and in an isometry-equivariant way, so that each set of points forms a homogeneous Poisson process, with any given pair of intensities summing to $λ$. In particular, this answers a question of Ball [Electron. Commun. Probab. 10 (2005) 60--69], who proved that in $d=1$, the Poisson points may be similarly partitioned (via a translation-equivariant function) so that one set forms a Poisson process of lower intensity, and asked whether the same is possible for all $d$. We do not know whether it is possible similarly to add points (again chosen as a deterministic function of a Poisson process) to obtain a Poisson process of higher intensity, but we prove that this is not possible under an additional finitariness condition.

math.PR↗

arXiv2011

Fixed Price of Groups and Percolation

We prove that for every finitely generated group $Γ$, at least one of the following holds: (1) $Γ$ has fixed price; (2) each of its Cayley graphs $G$ has infinitely many infinite clusters for some Bernoulli percolation on $G$.

math.DS↗

arXiv2011

Perfect Matchings as IID Factors on Non-Amenable Groups

We prove that in every bipartite Cayley graph of every non-amenable group, there is a perfect matching that is obtained as a factor of independent uniform random variables. We also discuss expansion properties of factors and improve the Hoffman spectral bound on independence number of finite graphs.

math.PR↗

arXiv2011

The Spread of Evidence-Poor Medicine via Flawed Social-Network Analysis

The chronic widespread misuse of statistics is usually inadvertent, not intentional. We find cautionary examples in a series of recent papers by Christakis and Fowler that advance statistical arguments for the transmission via social networks of various personal characteristics, including obesity, smoking cessation, happiness, and loneliness. Those papers also assert that such influence extends to three degrees of separation in social networks. We shall show that these conclusions do not follow from Christakis and Fowler's statistical analyses. In fact, their studies even provide some evidence against the existence of such transmission. The errors that we expose arose, in part, because the assumptions behind the statistical procedures used were insufficiently examined, not only by the authors, but also by the reviewers. Our examples are instructive because the practitioners are highly reputed, their results have received enormous popular attention, and the journals that published their studies are among the most respected in the world. An educational bonus emerges from the difficulty we report in getting our critique published. We discuss the relevance of this episode to understanding statistical literacy and the role of scientific review, as well as to reforming statistics education.

stat.ME↗

arXiv2009

Identities and Inequalities for Tree Entropy

The notion of tree entropy was introduced by the author as a normalized limit of the number of spanning trees in finite graphs, but is defined on random infinite rooted graphs. We give some new expressions for tree entropy; one uses Fuglede-Kadison determinants, while another uses effective resistance. We use the latter to prove that tree entropy respects stochastic domination. We also prove that tree entropy is non-negative in the unweighted case, a special case of which establishes Lueck's Determinant Conjecture for Cayley-graph Laplacians. We use techniques from the theory of operators affiliated to von Neumann algebras.

math.CO↗

arXiv2009

Random Complexes and l^2-Betti Numbers

Uniform spanning trees on finite graphs and their analogues on infinite graphs are a well-studied area. On a Cayley graph of a group, we show that they are related to the first $\ell^2$-Betti number of the group. Our main aim, however, is to present the basic elements of a higher-dimensional analogue on finite and infinite CW-complexes, which relate to the higher $\ell^2$-Betti numbers. One consequence is a uniform isoperimetric inequality extending work of Lyons, Pichot, and Vassout. We also present an enumeration similar to recent work of Duval, Klivans, and Martin.

math.PR↗

arXiv2009

A Measurable-Group-Theoretic Solution to von Neumann's Problem

We give a positive answer, in the measurable-group-theory context, to von Neumann's problem of knowing whether a non-amenable countable discrete group contains a non-cyclic free subgroup. We also get an embedding result of the free-group von Neumann factor into restricted wreath product factors.

math.GR↗

arXiv2008

Ends in Uniform Spanning Forests

It has hitherto been known that in a transitive unimodular graph, each tree in the wired spanning forest has only one end a.s. We dispense with the assumptions of transitivity and unimodularity, replacing them with a much broader condition on the isoperimetric profile that requires just slightly more than uniform transience.

math.PR↗

arXiv2008

A transient Markov chain with finitely many cutpoints

We give an example of a transient reversible Markov chain that almost surely has only a finite number of cutpoints. We explain how this is relevant to a conjecture of Diaconis and Freedman and a question of Kaimanovich. We also answer Kaimanovich's question when the Markov chain is a nearest-neighbor random walk on a tree.

math.PR↗