next up previous contents
Next: Metropolis Hastings Up: Sampling from the Posterior Previous: Markov Chains

Markov Chain Monte Carlo

Let tex2html_wrap_inline2405 be the distribution of interest. Let tex2html_wrap_inline2407 be the Markov matrix to be constructed. Now, what is needed is a method of constructing tex2html_wrap_inline2407 so that it is indeed a Markov Matrix, and that the stationary distribution of this Matrix is tex2html_wrap_inline2405 , the distribution of interest.

definition383

This property yields a method of constructing a suitable matrix, by using the result of the following theorem.

theorem388

Proof: since

displaymath2402

and this is true tex2html_wrap_inline2427 thus tex2html_wrap_inline2429 , that is tex2html_wrap_inline2431 is the stationary distribution for tex2html_wrap_inline2433 . So, given a distribution, tex2html_wrap_inline2435 , it is possible to construct a Markov matrix with tex2html_wrap_inline2435 as the stationary distribution, by imposing the condition of detailed balance.

That is, if tex2html_wrap_inline2407 are chosen so that tex2html_wrap_inline2441 , and of course subject to the constraints that tex2html_wrap_inline2443 and tex2html_wrap_inline2445 , and that the matrix is aperiodic irreducible, then tex2html_wrap_inline2433 is a transition matrix for a Markov chain whose equilibrium distribution is tex2html_wrap_inline2449 . The details of how one might go about such a construction are given in the Metropolis-Hastings Algorithm [30] [19].



Cathal Walsh
Sat Jan 22 17:09:53 GMT 2000