Computation Anomalies Detection in Implementations of Algorithms

D. A. Orlov , e-mail: orlovdmal@gmail.com

In the majority of program implementations of algorithms, real numbers are substituted with floating-point numbers. In most cases the result obtained is near to exact result, and good enough for practical purposes. However, in some cases one can obtain result qualitatively different from the exact result. Such situations are called computation anomalies.

In this paper the definition of computation anomaly is given, the mechanism of occurrence of computation anomalies is discussed. The methods of finding cases where computation anomaly can occur and algorithm implementation testing are proposed.

Keywords: computation geometry, floating-point numbers, calculation errors, computation anomalies, software testing

pp. 16–27