The reference sampling algorithm in Bayesian statistics is the Metropolis-Hastings algorithm. The principle is to use an auxiliary transition kernel and to split the sampling into two steps: a proposal step and an acceptance step, so as to guarantee the reversibility condition.