Bounded normal approximation in simulations of highly reliable Markovian systems

From MaRDI portal





Multi-component systems which tolerate fault and also fault propagation are often represented by Markov chains. For highly reliable systems the computation of dependability measures, such as the mean time to failure, is very expensive. One has to rely on approximate numerical methods such as importance sampling techniques. In order to get valid confidence intervals the approximating scheme must satisfy some properties. Previously the literature has focused on a property called `bounded relative error'. In the present paper the concept of `bounded normal approximation' is defined and its importance for the evaluation of dependability measures is emphasized. This property is strictly stronger than `bounded relative error'. It is shown that several schemes used in practice have the bounded normal approximation property.











This page was built for publication: Bounded normal approximation in simulations of highly reliable Markovian systems

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4954230)