Non-reversible guided Metropolis kernel

Abstract We construct a class of non-reversible Metropolis kernels as multivariate extension the guided-walk kernel proposed by Gustafson ( Statist. Comput. 8 , 1998). The main idea our method is to introduce projection that maps state space totally ordered group. By using Haar measure, we novel Markov termed mixture kernel, which interest in its own right. This achieved inducing topological structure Our method, $\Delta$ -guided Metropolis–Haar constructed proposal kernel. at least 10 times better than random-walk and Hamiltonian Monte Carlo for logistic regression discretely observed stochastic process terms effective sample size per second.