
 Markov Chain Monte Carlo (Site not responding. Last check: 20071104) 
  Generally speaking, MCMC provides a mechanism for taking dependent samples in situations where regular sampling is difficult, if not completely intractable. 
  To implement this algorithm, you need a reversible Markov chain to propose new states from any specified given state, the ability to compute the ratio of the posterior densities (probabilities) for any pair of states, and a random number generator. 
  (There are also ways to handle nonreveersible Markov chains.) Notice that if the posterior density at state x, p(x), equals h(x) / C where C is hard to compute, but h(x) is computable, the ratio of posterior densities at x and y equals p(x) / p(y) = h(x) / h(y). 
 www.mathcs.duq.edu /larget/math496/mcmc.html (351 words) 
