关键词: 精度测试, 最后一位表示的单位, 牛顿迭代, 连分式, 收敛阶

Abstract: Poor adaptability and slow convergence rate are the two main disadvantages of the existing precision test methods. To solve this problem, an elementary functions precision test method based on continued-fraction approximation was proposed by analyzing Unit in the Last Place (ULP) error and comparing several different functions true values calculations. The different calculations were analyzed and tested in the following two ways: time complexity and convergence degree. The experimental results show that the precision test method based on continued-fraction approximation is more effective, less complex and achieves faster convergence.

Key words: precision test, Unit in the Last Place (ULP), Newton iteration, continued-fraction, convergence degree