Static and dynamic numerical characteristic of floating point arithmetic

Abstract

The appearance of hexadecimal floating point arithmetic systems has prompted a continuing discourse on the relative numerical merits of various choices of base. Until lately this discourse has centered around the static properties of the floating point representation of numbers, and has primarily concerned only binary and hexadecimal representations. Recent events may change this discourse considerably. A third numerically attractive alternative for the choice of base has been proposed, and a comparison of the dynamic numerical properties of floating point arithmetic systems has been completed. This paper surveys these recent events and summarizes our current knowledge of the numerical characteristics of floating point arithmetic systems.