Software for Interval Arithmetic: A Reasonably Portable Package.

摘要

One means of bounding the error in digital computation is through the use of interval, or range, arithmetic; instead of computing with approximate real numbers, one calculates with pairs of approximate real numbers -- the first member of a pair being a lower bound for the true result, and the second an upper bound. By this method, one can take into account such varied sources of error as uncertainty in input data, inaccuracies in mathematical formulae, and errors in approximation of real numbers and the operations on them. The theory of interval arithmetic is developed extensively elsewhere. The major obstacle to the use of interval arithmetic is the unavailability of software. Interval is not a standard data type in any production language that we know of; preparation of a package of subprograms to handle interval data is a nontrivial task.