James Gregory
Abstract
In the June 1970 issue of Communications, Linz [1] proposed a method of pairwise summing of numbers to reduce accumulated roundoff error. Linz compared his method with only a simple recursive summation. The Linz method should have also been compared with a method published in the January 1965 issue by Kahan [2], which is more accurate.
This note compares the schemes by Linz and Kahan with the straight recursive summation. Comparisons of accuracy, speed and storage were carried out on an IBM 360/75 with various compiler optimization levels. (Linz' restriction to a binary exponent is not essential to his method.)
- 1 Linz, P. Accurate floating-point summation. Comm. ACM13, 6 (June 1970), 361-362. Google Scholar
Digital Library
- 2 Kahan, W. Further remarks on reducing truncation errors. Comm. ACM 8, 1 (Jan. 1965), 40. Google ScholarDigital Library
Recommendations
Accurate floating-point summation
This paper describes an alternate method for summing a set of floating-point numbers. Comparison of the error bound for this method with that of the standard summation method shows that it is considerably less sensitive to propagation of round-off ...
Fast Reproducible Floating-Point Summation
ARITH '13: Proceedings of the 2013 IEEE 21st Symposium on Computer ArithmeticReproducibility, i.e. getting the bit wise identical floating point results from multiple runs of the same program, is a property that many users depend on either for debugging or correctness checking in many codes. However, the combination of dynamic ...
Error estimates for the summation of real numbers with application to floating-point summation
AbstractStandard Wilkinson-type error estimates of floating-point algorithms involve a factor $$\gamma _k:=k\mathbf {u}/(1-k\mathbf {u})$$ for $$\mathbf {u}$$ denoting the relative rounding error unit of a floating-point number system. Recently, it was ...
Comments