Communication and regeneration Pseudo atoms推荐.pdfVIP

Chapter 5 Pseudo-atoms Much Markov chain theory on a general state space can be developed in complete analogy with the countable state situation when X contains an atom for the chain Φ. Atoms A set α ∈ B(X) is called an atom for Φ if there exists a measure ν on B (X) such that P (x, A) = ν (A), x ∈ α. If Φ is ψ-irreducible and ψ (α) 0 then α is called an accessible atom. A single point α is always an atom. Clearly, when X is countable and the chain is irreducible then every point is an accessible atom. On a general state space, accessible atoms are less frequent. For the random walk on a half line as in (RWHL1), the set {0} is an accessible atom when Γ(−∞, 0) 0: as we have seen in Proposition 4.3.1, this chain has ψ ({0}) 0. But for the random walk on R when Γ has a density, accessible atoms do not exist. It is not too strong to say that the single result which makes general state space Markov chain theory as powerful as countable space theory is that there exists an “artificial atom” for ϕ-irreducible chains, even in cases such as the random walk with absolutely continuous increments. The highlight of this chapter is the development of this result, and some of its immediate consequences. ˇ Atoms are found for “strongly aperiodic” chains by constructing a “split chain” Φ ˇ evolving on a split state space X = X0 ∪ X1 , where X0 and X1 are copies of the state space X, in such a way that ˇ ˇ (i) the chain Φ is the marginal chain of Φ, in the sense that P(Φk ∈ A) = P(Φk ∈ A ∪ A ) for appropriate i