On Non-reversible Markov Chains
نویسندگان
چکیده
Reversibility is a suucient but not necessary condition for Markov chains for use in Markov chain Monte Carlo simulation. It is necessary to select a Markov chain that has a pre-speciied distribution as its unique stationary distribution. There are many Markov chains that have such property. We give guidelines on how to rank them based on the asymptotic variance of the estimates they produce. Two questions are addressed. First, how to select a Markov chain among several non-reversible (or reversible) ones. Second, given a non-reversible Markov chain, is there a reversible one with the same (or better) performance?
منابع مشابه
Eigenvalues of non-reversible Markov chains: their connection to mixing times, reversible Markov chains, and Cheeger inequalities
We show a lower bound on mixing time for a non-reversible Markov chain in terms of its eigenvalues. This is used to show a bound on the real part of the complex-valued eigenvalues in terms of the realvalued eigenvalues of a related reversible chain, and likewise to bound the second largest magnitude eigenvalue. A myriad of Cheeger-like inequalities also follow for non-reversible chains, which e...
متن کاملA Modification of Neal’s Algorithm for a Continuous State Space and an Application to the Fokker-planck Equation
The Metropolis-Hastings algorithm generates correlated samples from a target distribution by constructing a Markov chain which has as its stationary distribution the desired target distribution. One property of this algorithm is that it creates reversible Markov chains. As a result, reversible chains are often used in Monte Carlo simulations. Reversible Markov chains also have the added benefit...
متن کاملThree Kinds of Geometric Convergence for Markov Chains and the Spectral Gap Property
In this paper we investigate three types of convergence for geometrically ergodic Markov chains (MCs) with countable state space, which in general lead to different ‘rates of convergence’. For reversible Markov chains it is shown that these rates coincide. For general MCs we show some connections between their rates and those of the associated reversed MCs. Moreover, we study the relations betw...
متن کاملG-PCCA: Spectral Clustering for Non-reversible Markov Chains
Spectral clustering methods are based on solving eigenvalue problems for the identification of clusters, e.g., the identification of metastable subsets of a Markov chain. Usually, real-valued eigenvectors are mandatory for this type of algorithms. The Perron Cluster Analysis (PCCA+) is a well-known spectral clustering method of Markov chains. It is applicable for reversible Markov chains, becau...
متن کاملGeneralized Cheeger inequalities for eigenvalues of non-reversible Markov chains
We show lower bounds for the smallest non-trivial eigenvalue, and smallest real portion of an eigenvalue, of the Laplacian of a non-reversible Markov chain in terms of an Evolving set quantity. A myriad of Cheeger-like inequalities follow for non-reversible chains, which even in the reversible case sharpen previously known results. The same argument also produces a new Cheeger-like inequality f...
متن کامل