Abstract:Propp & Wilson (1996) described a protocol, called coupling from the past, for exact sampling from a target distribution using a coupled Markov chain Monte Carlo algorithm. In this paper we extend coupling from the past to various MCMC samplers on a continuous state space; rather than following the monotone sampling device of Propp & Wilson, our approach uses methods related to gamma-coupling and rejection sampling to simulate the chain, and direct accounting of sample paths.
“…In some cases, even running the algorithm for months on the world's fastest computers would not provide a remotely reasonable approximation to π. Overcoming such problems has often necessitated new and more complicated MCMC algorithms; see, e.g., Bélisle et al (1993), Neal (2003), Jain and Neal (2004), and Hamze and de Freitas (2012). In a different direction, detecting convergence of MCMC to π is so challenging that some authors have developed perfect sampling algorithms which guarantee complete convergence at the expense of a more complicated algorithm; see, e.g., Propp and Wilson (1996), Murdoch and Green (1998) or Fill et al (2000). However, such perfect sampling algorithms are often infeasible to run, so we do not discuss them further here.…”
“…In some cases, even running the algorithm for months on the world's fastest computers would not provide a remotely reasonable approximation to π. Overcoming such problems has often necessitated new and more complicated MCMC algorithms; see, e.g., Bélisle et al (1993), Neal (2003), Jain and Neal (2004), and Hamze and de Freitas (2012). In a different direction, detecting convergence of MCMC to π is so challenging that some authors have developed perfect sampling algorithms which guarantee complete convergence at the expense of a more complicated algorithm; see, e.g., Propp and Wilson (1996), Murdoch and Green (1998) or Fill et al (2000). However, such perfect sampling algorithms are often infeasible to run, so we do not discuss them further here.…”
“…The bounds (8) are not needed to implement the simple field coupler. If they are available, however, we can obtain an estimate of the speed of convergence to a finitely coupled field within > 0.…”
Section: A Markov Chain On Field Spacementioning
confidence: 99%
“…Note that we have not assumed that q(·) is symmetric. For symmetric and unimodal increment distributions, an alternative field coupling method for X exists, proposed by Murdoch and Green [8]. Their method, called the bisection coupler, uses translations and reflections to construct a finitely coupled field, and appears to require fewer computations in general.…”
Given a Markov chain, a stochastic flow that simultaneously constructs sample paths started at each possible initial value can be constructed as a composition of random fields. Here, a method is described for coupling flows by modifying an arbitrary field (consistent with the Markov chain of interest) by an independence MetropolisHastings iteration. The resulting stochastic flow is shown to have many desirable coalescence properties, regardless of the form of the original flow.
“…The difficulty is that unless the dynamic is monotone, it is necessary to effectively couple all (noncountable) initial states. Murdoch and Green (1998) proposed various procedures to transform the infinite set of initial states into a more tractable finite subset.…”
The target measure µ is the distribution of a random vector in a box B, a Cartesian product of bounded intervals. The Gibbs sampler is a Markov chain with invariant measure µ. A 'coupling from the past'construction of the Gibbs sampler is used to show ergodicity of the dynamics and to perfectly simulate µ. An algorithm to sample vectors with multinormal distribution truncated to B is then implemented.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.