Exact sampling from non-attractive distributions using summary states

Abstract

Propp and Wilson's method of coupling from the past allows one to efficiently generate exact samples from attractive statistical distributions (e.g., the ferromagnetic Ising model). This method may be generalized to non-attractive distributions by the use of summary states, as first described by Huber. Using this method, we present exact samples from a frustrated antiferromagnetic triangular Ising model and the antiferromagnetic q=3 Potts model. We discuss the advantages and limitations of the method of summary states for practical sampling, paying particular attention to the slowing down of the algorithm at low temperature. In particular, we show that such a slowing down can occur in the absence of a physical phase transition.

5 pages, 6 EPS figures, REVTeX; additional information at http://wol.ra.phy.cam.ac.uk/mackay/exact

References in corpus (3)

• Antiferromagnetic Potts Models on the Square Lattice: A High-Precision Monte Carlo Study
• Antiferromagnetic Potts Models on the Square Lattice
• Inference for Belief Networks Using Coupling From the Past

Cited by in corpus (8)

• MCMC for doubly-intractable distributions
• Estimating the granularity coefficient of a Potts-Markov random field within an MCMC algorithm
• Pseudo-Marginal Slice Sampling
• Renormalization group approach to exact sampling
• Convergence and coupling for spin glasses and hard spheres
• Free energy differences : Representations, estimators, and sampling strategies
• Estimation and uncertainty quantification for the output from quantum simulators
• Bayesian Learning in Undirected Graphical Models: Approximate MCMC algorithms