Variable-precision exponentiation
Paul L. Richman · Communications of the ACM · 1973
A previous paper presented an efficient algorithm, called the Recomputation Algorithm, for evaluating a rational expression to within any desired tolerance on a computer which performs variable-precision arithmetic operations. The Recomputation Algorithm can be applied to expressions involving any variable-precision operations having O (10 - p + ∑ | ε i |) error bounds, where p denotes the operation's precision and ε i denotes the error in the operation's i th argument. This paper presents an efficient variable-precision exponential operation with an error bound of the above order. Other operations, such as log, sin, and cos, which have simple series expansions, can be handled similarly.