摘要 |
The method of the present invention is to modify an initial target distribution it pi by combining it with a point mass concentrated on an "artificial atom" alpha which is outside the state-space X. A Markov chain may then be constructed using any known technique (for example, using the Metropolis-Hastings Algorithm) with the new target distribution. For this chain, the state alpha is Harris-recurrent (i.e. with probability one, it occurs infinitely many times). By the Markov property, the times at which the new chain hits alpha are regeneration times. To recover an ergodic chain with limiting distribution pi, it is sufficient simply to delete every occurrence of the state alpha from the new chain. The points immediately after the (deleted) occurrences of the state alpha are then regeneration times in a Markov chain with limiting distribution pi.
|