The case of equality in the Dobrushin–Deutsch–Zenger bound

Abstract

Suppose that A = ( a i , j ) is an n × n real matrix with constant row sums μ . Then the Dobrushin–Deutsch–Zenger (DDZ) bound on the eigenvalues of A other than μ is given by Z ( A ) : = 1 2 max 1 ⩽ s , t ⩽ n ∑ r = 1 n | a s , r - a t , r | . When A a transition matrix of a finite homogeneous Markov chain so that μ = 1 , Z ( A ) is called the coefficient of ergodicity of the chain as it bounds the asymptotic rate of convergence, namely, max { | λ | | λ ∈ σ ( A ) ⧹ { 1 } } , of the iteration x i T = x i - 1 T A , to the stationary distribution vector of the chain. In this paper we study the structure of real matrices for which the DDZ bound is sharp. We apply our results to the study of the class of graphs for which the transition matrix arising from a random walk on the graph attains the bound. We also characterize the eigenvalues λ of A for which | λ | = Z ( A ) for some stochastic matrix A .