An attractive algorithm for repulsive spatial point processes
Time: October 1, 2007, 1pm - 2pm
Place: D344, LSRC
Speaker:Mark Huber

Repulsive point processes model underdispersed phenomena such as the location of a particular species of tree in a forest, where bombs fall, oil deposit locations, and hard core gas molecules. These models are stochastic: they put a density on a particular configuration of points. The primary method for generating variates from these distributions is to use Markov chains. In this talk I will describe the basic birth-death chains for generating random variates from these models. Then I will extend these basic methods to include a new type of "swap" move that speeds up convergence to the correct distribution. I will also discuss perfect sampling methods for these methods that improve greatly on the simple Markov chain approach.