Answers are grounded in this paper and cite their sources.
Investigates convergence criteria of Markov chains, indicating potential for accelerated algorithms.
Let π be a probability density proportional to exp - U(x) in S. A convergent Markov process to π(x) may be regarded as a "conceptual" algorithm. Assume that S is a finite set. Let X 0 ,X 1 ,…,X n ,… be a Markov chain with transition matrix P and invariant probability π. Under suitable condition on P, it is known that [Formula: see text] converges to π(f) and the corresponding asymptotic variance v(f, P) depends only on f and P. It is natural to consider criteria v w (P) and v a (P), defined respectively by maximizing and averaging v(f, P) over f. Two families of transition matrices are considered. There are four problems to be investigated. Some results and conjectures are given. As for the continuum case, to accelerate the convergence a family of diffusions with drift ∇U(x) + C(x) with div(C(x)exp - U(x)) = 0 is considered.
No takes yet. Share an insight, caveat, or question.
Chii-Ruey Hwang (2005) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: