Numerical software is a term that is used rather liberally today to describe a range of activities. It has two principle characteristics: it deals with approximations to real numbers and it is usable on a range of computers that have different approximation capabilities. The nature of numerical software is examined and the chapter then discusses what is particularly difficult about it. Numerical software production is viewed by many people either as a routine programming task or as a by-product of that dull subject, numerical analysis that itself falls somewhere between mathematics and computer science, too applied for the one and too irrelevant to the other. The chapter explains that some of the attributes needed in numerical software are reliability, efficiency, and broad applicability. Numerical software is very much a part of computer science. It is concerned with automatic problem solving—that is, with analyzing methods and synthesizing techniques. It has some solid scientific foundations, but still involves a great deal of judgment because the idealized goals are not achievable. The chapter concludes that the numerician is partially reduced to alchemy; for brewing new concoctions and testing the results, sometimes heuristics are used.
All Science Journal Classification (ASJC) codes
- Computer Science(all)