A2_1 The fast inverse square root in scientific computing

Abstract

Scientific computing often requires the calculation of square roots - a process which is very expensive in terms of processing time compared to most other floating point operations. In this paper we examine the use of a faster but less accurate square root algorithm in scientific computing using N-body simulations as an example. The results show that the the speed increase afforded by the faster algorithm is not offset by the reduction in accuracy, and hence an alternate approach is suggested.