Summary: How to Combine Fast Heuristic
Markov Chain Monte Carlo with
Slow Exact Sampling
(with Antar Bandyopadhyay).
See www.stat.berkeley.edu/users/aldous for slides
MCMC is huge field, many different aspects of
theory and of applications. We will ``zoom in''
to describe a very special question -- one point
on the boundary between rigour and heuristics.
Target prob. dist. ß on space S.
Seek to sample from ß.
Design a Markov transition matrix P (x; y) to
have ß as its stationary distribution. Use the
Markov chain X t to get samples; naively, take
large t 0 and regard
X t 0