The polar slice sampler

From MaRDI portal





The slice sampler is a specialized type of Markov chain Monte Carlo (MCMC) auxiliary method. This interesting paper focuses closely on the relationship between the dimension of the target density and certain level set conditions. In particular, the authors analyze simple slice sampler factorization on log-concave densities. The investigation leads to the introduction of the \textit{polar slice sampler}, an algorithm that effectively preserves the robust convergence properties of the slice sampler for the log-concave densities and for arbitrary dimensional distributions. Furthermore, a~simple comparison of computing times is presented that shows that the polar slice sampler performs well in comparison with the rejection sampler and the simple slice sampler. A~major limitation of all slice sampler algorithms is the fact that their implementation is often difficult. However, the authors demonstrate that the polar slice sampler has very good convergence properties, provided the \textbf{X}-updating step can be feasibly implemented. Section~2 introduces the slice sampler and describes some of its basic properties. In Section~3, the performance of the \textit{uniform} simple slice sampler in high dimensions is investigated, and it is shown empirically that even for spherically symmetric, log-concave densities, the algorithm's performance deteriorates markedly as dimension increases. The polar slice sampler is motivated and constructed in Section~4, and studied theoretically in Section~5. Section~6 gives empirical study of the performance of the polar slice sampler for a~particular example and gives comparisons with the uniform simple slice sampler and a~corresponding rejection sampling algorithm. Section~7 offers some concluding remarks.











This page was built for publication: The polar slice sampler

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3147438)