| 摘要 |
With every numerical data there is associated an item of truncation information (TA1, TB1, . . . ) which is true or false according to whether its value is or is not truncated in the representation employed, the calculation is carried out a number of times while disturbing the development of at least one procedure independently of the others, the disturbance consisting during the course of each elementary operation of adding on occasion and at random a bit of the lowest significance to the mantissa of each number the associated item of truncation information of which is positive, and the mean value and the standard deviation of the different results provided are worked out in order to determine the number of exact significant figures in one of these results. |
