Feasibility of “perfect” function evaluation

This paper casts doubt on the feasibility of perfect subroutines or perfect hardware for mathematical functions, in particular by conversion from double precision. It suggests a slightly relaxed standard.