Test Construction for Mathematical Functions

The article deals with problems of testing implementations of mathematical functions working with floating-point numbers. It considers current standards’ requirements to such implementations and demonstrates that those requirements are not sufficient for correct operation of modern systems using sophisticated mathematical modeling. Correct rounding requirement is suggested to guarantee preservation of all important properties of implemented functions and to support high level of interoperability between different mathematical libraries and modeling software using them. Test construction method is proposed for conformance test development for current standards supplemented with correct rounding requirement. The idea of the method is to use three different sources of test data: floating-point numbers satisfying specific patterns, boundaries of intervals of uniform function behavior, and points where correct rounding requires much higher precision than in average. Some practical results obtained by using the method proposed are also presented.