Fisher information in the design of computer simulation experiments

Abstract

The concept of Fisher information is conveniently used as a basis for designing efficient experiments. However, if the output stems from computer simulations they are often approximated as realizations of correlated random fields. Consequently, the conditions under which Fisher information may be suitable must be restated. In the paper we intend to give some simple but illuminating examples for these cases.'Random phenomena have increasing importance in Engineering and Physics, therefore theoretical results are strongly needed. But there is a gap between the probability theory used by mathematicians and practitioners. Two very different languages have been generated in this way...' (Paul Krée, Paris 1995)