zbMATH — the first resource for mathematics

Large deviations for Markov chains in the positive quadrant. (English. Russian original) Zbl 1068.60034
Russ. Math. Surv. 56, No. 5, 803-916 (2001); translation from Usp. Mat. Nauk 56, No. 5, 3-116 (2001).
This paper deals with \(N\)-partially space-homogeneous time-homogeneous Markov chains \(X(y,n)\), \(n=0,1,2,\dots, X(y,0)=y,\) in the positive quadrant \(R^{2+}=\{x=(x_1,x_2):x_1\geq 0,x_2\geq 0\}.\) The main object of study is the asymptotic behaviour of the probabilities \(P(X(y,n)\in \delta(x))\) as \(n\to \infty\) and \(x=x(n)\to \infty,\) where \(\delta(x)=\{y\in R^2:x_i\leq y_i<x_i+\delta\), \(i=1,2\}.\) One of the main results says that \[ \ln P(X(y,n)\in \delta(x))\sim-sD(\gamma,\beta,T), \] where the derivation function \(D(\gamma,\beta,T)\) is written out in closed form, time \(n\) depends on \(s\) in such a way that there are limits \(\gamma,\beta,\) and \(T\) for which one has \(y\sim s\gamma,\;x\sim s\beta,\;n\sim sT.\) Results on the asymptotics of large deviation probabilities can be used in system reliability analysis to estimate overflow probabilities or the probability of reaching a high critical level, and as well in control theory to prove the existence of various exponential moments of stationary distributions.

60F10 Large deviations
60J05 Discrete-time Markov processes on general state spaces
60G05 Foundations of stochastic processes
60K25 Queueing theory (aspects of probability theory)
60K30 Applications of queueing theory (congestion, allocation, storage, traffic, etc.)
Full Text: DOI