Computer errors in rounding off products dynamically

Conclusions

• 1. Comparison of maximum errors shows that the smallest errors are obtained in the methods entailing the adding of a unit to the (n+p+1)-th order of the partial product [2]. The first method produces a somewhat larger error. Calculations have shown that for compensating the error in this method it is necessary to have only one more order than in the method described in [2]. Moreover, the first method does not entail carries to higher orders and, therefore, its application tends to raise the arithmetic unit's speed of operation. The second and third rounding-off methods have the largest maximum errors.

• 2. It follows from comparing the M and D errors that the method entailing simple discarding has the minimum dispersion. However, it has the maximum M. The method of [2] occupies an intermediate position with respect to the value of its dispersion and has the same M as the second method.

• 3. The selection of the rounding-off method should depend on the required speed of operation and expenditure on the equipment.