The Classical Relative Error Bounds for Computing $\sqrt(a^2 + b^2)$ and $c/\sqrt(a^2 + b^2)$ in Binary Floating-Point Arithmetic are Asymptotically Optimal

International audienceWe study the accuracy of classical algorithms for evaluating expressions of the form √a2 + b2 and c/ √a2 + b2 in radix-2, precision-p floating-point arithmetic, assuming that the elementary arithmetic operations ±, ×, /, √ are rounded to nearest, and assuming an unbounded exponent range. Classical analyses show that the relative error is bounded by 2u + O(u2 ) for √a2 + b2 , and by 3u + O(u2 ) for c/ √a2 + b2 , where u = 2−p is the unit round off. Recently, it was observed that for √a2 + b2 the O(u2 ) term is in fact not needed [1]. We show here that it is not needed either for c/√a2 + b2 . Furthermore, we show that these error bounds are asymptotically optimal. Finally, we show that both the bounds and their asymptotic optimality remain valid when an FMA instruction is used to evaluate a2 + b2