Numerical noise analysis - A new probabilistic technique

The paper describes a new probabilistic technique called numerical noise analysis. The fundamental idea underlying this new method is an artificially produced 'numeric noise' for use in error analysis. The evaluation for numerical noise analysis as well as examples are presented. A comparison of results obtained by numerical noise analysis with results from the well-known Monte Carlo technique shows an excellent agreement. An attempt is made at comparing the expenditures involved in different probabilistic techniques. According to these studies, numerical noise analysis has a considerable potential, which should be investigated further. The tests performed so far are cause for cautious optimism.