The accuracy of floating point summation. (English) Zbl 0788.65053

The author studies five summation methods for computing the sum of a number of floating point numbers. The accuracy is compared using rounding error analysis and numerical experiments. Reordering the numbers is a central idea. Statistical estimates are provided and the model with no guard digit is also considered.
No particular method is uniformly more accurate than the others, but some guidelines are given on the choice of method in particular cases.
The presentation of the paper is clear and the arguments are so convincing that the paper is recommended to anyone interested in the problem.
Reviewer: W.Govaerts (Gent)


65G50 Roundoff error
Full Text: DOI Link