Optimal acceptance rates for Metropolis algorithms: Moving beyond 0.234

Abstract

Recent optimal scaling theory has produced a condition for the asymptotically optimal acceptance rate of Metropolis algorithms to be the well-known 0.234 when applied to certain multidimensional target distributions. These d-dimensional target distributions are formed of independent components, each of which is scaled according to its own function of d. We show that when the condition is not met the limiting process of the algorithm is altered, yielding an asymptotically optimal acceptance rate which might drastically differ from the usual 0.234. Specifically, we prove that as d → ∞ the sequence of stochastic processes formed by say the i * th component of each Markov chain usually converges to a Langevin diffusion process with a new speed measure υ, except in particular cases where it converges to a one-dimensional Metropolis algorithm with acceptance rule α *. We also discuss the use of inhomogeneous proposals, which might reveal essential in specific cases.

Optimal Proposal Distributions and Adaptive MCMC

“…In a related direction, Bédard (2007Bédard ( , 2008Bédard ( , 2006; see also Bédard and Rosenthal, 2008) considered the case where the target distribution π has independent coordinates with vastly different scalings (i.e., different powers of d as d → ∞). She proved that if each individual component is dominated by the sum of all components, then the optimal acceptance rate of 0.234 still holds.…”

Optimal Scaling of Random Walk Metropolis Algorithms with Non-Gaussian Proposals

“…Thus the simple rule of thumb of tuning σ d such that one in four proposed moves are accepted holds quite generally. In [4] and [17], examples where the aoar is strictly less than 0.234 are given. These correspond to different orders of magnitude being appropriate for the scaling of the proposed moves in different components.…”

Some of Statistics Canada’s Contributions to Survey Methodology

“…A number of authors have attempted to weaken and generalize the original strong assumptions; see e.g., Bédard (2007Bédard ( , 2008, Bédard and Rosenthal (2008), Beskos et al (2009), andSherlock and. Corresponding results have been developed for Langevin MCMC algorithms (Roberts and Rosenthal, 1998), and for simulated tempering algorithms (Atchadé et al, 2011;Roberts and Rosenthal, 2013).…”