arXiv++ Combinatorics

Motivated in part by various sequences of graphs growing under random rules (like internet models), convergent sequences of dense graphs and their limits were introduced by Borgs, Chayes, Lovász, Sós and Vesztergombi and by Lovász and Szegedy. In this paper we use this framework to study one of the motivating class of examples, namely randomly growing graphs. We prove the (almost sure) convergence of several such randomly growing graph sequences, and determine their limit. The analysis is not always straightforward: in some cases the cut distance from a limit object can be directly estimated, in other case densities of subgraphs can be shown to converge.

We consider sequences of graphs and define various notions of convergence related to these sequences: ``left convergence'' defined in terms of the densities of homomorphisms from small graphs into the graphs of the sequence, and ``right convergence'' defined in terms of the densities of homomorphisms from the graphs of the sequence into small graphs; and convergence in a suitably defined metric. In Part I of this series, we show that left convergence is equivalent to convergence in metric, both for simple graphs, and for graphs with nodeweights and edgeweights. One of the main steps here is the introduction of a cut-distance comparing graphs, not necessarily of the same size. We also show how these notions of convergence provide natural formulations of Szemeredi partitions, sampling and testing of large graphs.

It is shown that a graph parameter can be realized as the number of homomorphisms into a fixed (weighted) graph if and only if it satisfies two linear algebraic conditions: reflection positivity and exponential rank-connectivity. In terms of statistical physics, this can be viewed as a characterization of partition functions of vertex models.

The cover time C of a graph G is the expected time for a random walk starting from the worst vertex to cover all vertices in G. Similarly, the blanket time B is the expected time to visit all vertices within a constant factor of number of times suggested by the stationary distribution. (Our definition will be slightly stronger than this.) Obviously, all vertices are covered when the graph is blanketed, and hence C <= B. The blanket time is introduced by Winkler and Zuckerman motivated by applications in Markov estimation and distributed computing. They conjectured B =O(C) and proved B=O(C ln n ). In this paper, we introduce another parameter M motivated by Matthews' theorem and prove M/2 <= C <= B = O(( M ln ln n)^2). In particular, B = O(C (ln ln n)^2). The lower bound is still valid for the cover time C(π) starting from the stationary distribution. We also show that there is a polynomial time algorithm to approximate M within a factor of 2 and so does for C within a factor of O((ln ln n)^2), improving previous bound of O(ln n) of Matthews'.