next up previous
Next: Finite Point Processes Up: Markov Chain Monte Carlo Previous: State-Dependent Mixing

Why it Works

It is enough to verify that elementary MHG update is reversible. Define

displaymath2893

the probability of accepting a proposed step from x to y and

displaymath2899

the probability of staying at x when the proposal tex2html_wrap_inline2877 is used. Then

displaymath2905

is the transition kernel of the Metropolis-Hastings-Green update, and

multline246

The first term on the right hand side is obviously symmetric under interchange of u and v. The second term

multline248

is symmetric under interchange of u and v because of the symmetry of tex2html_wrap_inline2869 and because of the MHG update has been defined to make

  equation250

hold for all i, x, and y. To see this suppose tex2html_wrap_inline2923 in (1.11). Then

displaymath2925

and (1.12) holds. (1.12) also holds when tex2html_wrap_inline2927 by the same argument.



Charles Geyer
Fri Jul 5 15:26:21 CDT 1996