Coupling and monotonicity of Markov processes


Il Dr. Pierre-Yves Louis, dell'Università di Potsdam, terrà il 27 ottobre alle ore 11:00 nell'Aula 2A/45 della Torre Archimede, un seminario dal titolo "COUPLING AND MONOTONICITY OF MARKOV PROCESSES".

- Abstract:
MCMC Algorithms aim at sampling from a given probability distribution.
They approximate it as time-asymptotics of some Markovian dynamics.
Perfect simulation modification of MCMC algorithms allow to sample exactly from this distribution.
When the state space if large, monotonicity properties are fundamental for the applicability of these algorithms.
On partially ordered state space it leads in fact to a stronger monotonicity concept, the one of complete monotonicity. Since it exists handable criteria to test the usual (simple) monotonicity of a Markovian dynamics, it is important to state on which partial order structure these two monotonicity concepts are equivalent.
Results are presented in both the discrete and continuous time framework.

Rif.: P. Dai Pra