Langevin type limiting processes for adaptive MCMC (Q2520143)

In most Markov chain Monte Carlo (MCMC) algorithms so far, the transition probability \( P\) depends on some tuning parameter \(\theta\) defined on some space \({\Theta}\) which is possibly infinite dimensional. The success or failure of the algorithm depends on the choice of \({\theta}\). Adaptive Markov chain Monte Carlo (AMCMC) is a class of MCMC algorithms where the proposal distribution changes at every iteration of the chain. In this paper, the authors apply the diffusion approximation procedure to the AMCMC for both univariate and multivariate target distribution using the standard univariate normal and standard multivariate normal distribution as the proposal. In both the cases the limiting diffusion is obtained. For the definition of one kind of this diffusion approximation see [\textit{G. O. Roberts} et al., Ann. Appl. Probab. 7, No. 1, 110--120 (1997; Zbl 0876.60015)] where the state space increases in dimension to \({+\infty}\).