PLACE: 408 Carver Hall

SPEAKER:
Richard Tweedie
Division of Biostatistics, University of Minnesota

TITLE:
Perfect Sampling and its Uses

ABSTRACT:
We introduce several implementations of the Propp-Wilson algorithm for
simulating “perfect” random samples from the invariant measure of a positive
recurrent Markov chain.  We discuss the way in which the coupling from the past
algorithm works; the role of monotonicity; and how to make the sampler work for
uniformly ergodic chains.

For a Harris chain on a continuous space, we also describe a new method which
uses backward coupling of embedded regeneration times, and works most
effectively for stochastically monotone chains, where paths may be sandwiched
between “upper” and “lower” processes.  Applications of the algorithm to
independent Matropolis-Hastings algorithms, to storage models and to inventory
models will be given as time permits.
 

COFFEE: 3:45 p.m., 404 Carver Hall (Commons)