Consistency of Markov chain quasi-Monte Carlo on continuous state spaces (Q548531)

Markov chain Monte Carlo (MCMC) algorithms are considered in which quasi random numbers are used instead of true random sequences. This technique is called quasi Monte Carlo MCMC (QMC-MCMC). The authors derive conditions of Metropolis and Gibbs sampler consistency in QMC-MCMC setting. The quasi-random numbers (vectors) \(u_1,\dots,u_n\dots\) used to generate the chain \(x(u_1),\dots,x(u_n),\dots\) are assumed to be completely uniformly distributed. It is assumed that MCMC is regular in the sense that for any continuous function \(f\) the mapping \((u_1,\dots,u_n)\to f(x(u_1),\dots,x(u_n))\) is Riemann integrable. The consistency proof is based on coupling arguments. A Gibbs sampler for sampling from posterior of Bayesian probit regression is considered as an example.