ERROR-BOUNDS FOR ARBITRARY APPROXIMATIONS OF NEARLY REVERSIBLE MARKOV-CHAINS AND A COMMUNICATIONS EXAMPLE

A condition is provided to conclude error bounds when using an arbitra ry steady state approximation of a ''nearly reversible'' Markov chain. The error bound is of the form Delta R where (i) Delta can be compute d by the approximation in order (ii) R can be obtained analytically by the system of interest. The results will be illustrated for a communi cation system with different source characteristics. An approximation is suggested based on truncating the corresponding Mobius-function. An R-value is obtained by an inductive Markov reward equation. Numerical illustration indicates that the error bound can be useful for practic al purposes.