Testing automatic quadrature programs (Q1179431)

This article describes a more efficient method for testing automatic quadrature programs. The key idea involves testing with a very small requested error tolerance and monitoring (through extra code inserted in the program that is being tested) the performance of the tested program as it proceeds to try to produce increasingly more accurate results. This procedure avoids the more usual testing procedure of performing a number of repeated tests with different requested error tolerances. The authors also define several functions which can be used with the tests to measure the tested program's performance. The new testing method is applied to five different quadrature programs (four are adaptive and one is not) and the resulting performance function graphs are displayed and discussed in the article.