Authors
Radford M Neal
Description
Peskun’s theorem shows that the asymptotic variance of an MCMC estimator based on a reversible Markov chain will not increase if the matrix of transition probabilities for the chain is modified so as to increase the off-diagonal terms. I present a new proof of this result, which is more intuitive than Peskun’s original proof, and which may provide hints for how to prove other results of this nature.