摘要: We consider “approximate stationary” Markov chains in which the entries of one-step transition probability matrix are known to be different orders magnitude and whose structure that is, probabilities does not change with time. For such we present a method for generating order estimates t-step probabilities, any t. then notice algorithms simulated annealing type may represented by approximately stationary over fairly long time intervals. Using our results obtain characterization convergent “cooling” schedules most general class type.