J.N. Corcoran and U. Schneider
Journal of Statistical Computation and Simulation, 2005, 75 (6), 459-475.
We describe and examine an imperfect variant of a perfect sampling algorithm based on the Metropolis-Hastings algorithm that appears to perform better than a more traditional approach in terms of speed and accuracy. We then describe and examine an "adaptive" Metropolis-Hasting algorithm which generates and updates a self-target candidate density in such a way that there is no "wrong choice" for an initial candidate density. Simulation examples are provided.
Link to paper: (J. Stat. Comp. Sim.)