Coupling from the past

Among Markov chain Monte Carlo (MCMC) algorithms, coupling from the past is a method for sampling from the stationary distribution of a Markov chain. Contrary to many MCMC algorithms, coupling from the past gives in principle a perfect sample from the stationary distribution. It was invented by James Propp and David Wilson in 1996.

The basic idea edit

Consider a finite state irreducible aperiodic Markov chain   with state space   and (unique) stationary distribution   (  is a probability vector). Suppose that we come up with a probability distribution   on the set of maps   with the property that for every fixed  , its image   is distributed according to the transition probability of   from state  . An example of such a probability distribution is the one where   is independent from   whenever  , but it is often worthwhile to consider other distributions. Now let   for   be independent samples from  .

Suppose that   is chosen randomly according to   and is independent from the sequence  . (We do not worry for now where this   is coming from.) Then   is also distributed according to  , because   is  -stationary and our assumption on the law of  . Define

 

Then it follows by induction that   is also distributed according to   for every  . However, it may happen that for some   the image of the map   is a single element of  . In other words,   for each  . Therefore, we do not need to have access to   in order to compute  . The algorithm then involves finding some   such that   is a singleton, and outputting the element of that singleton. The design of a good distribution   for which the task of finding such an   and computing   is not too costly is not always obvious, but has been accomplished successfully in several important instances.[1]

The monotone case edit

There is a special class of Markov chains in which there are particularly good choices for   and a tool for determining if  . (Here   denotes cardinality.) Suppose that   is a partially ordered set with order  , which has a unique minimal element   and a unique maximal element  ; that is, every   satisfies  . Also, suppose that   may be chosen to be supported on the set of monotone maps  . Then it is easy to see that   if and only if  , since   is monotone. Thus, checking this becomes rather easy. The algorithm can proceed by choosing   for some constant  , sampling the maps  , and outputting   if  . If   the algorithm proceeds by doubling   and repeating as necessary until an output is obtained. (But the algorithm does not resample the maps   which were already sampled; it uses the previously sampled maps when needed.)

References edit

  1. ^ "Web Site for Perfectly Random Sampling with Markov Chains".
  • Propp, James Gary; Wilson, David Bruce (1996), Proceedings of the Seventh International Conference on Random Structures and Algorithms (Atlanta, GA, 1995), pp. 223–252, MR 1611693
  • Propp, James; Wilson, David (1998), "Coupling from the past: a user's guide", Microsurveys in discrete probability (Princeton, NJ, 1997), DIMACS Ser. Discrete Math. Theoret. Comput. Sci., vol. 41, Providence, R.I.: American Mathematical Society, pp. 181–192, doi:10.1090/dimacs/041/09, ISBN 9780821808276, MR 1630414, S2CID 2781385