Efficient minimization of numerical summation errors

Abstract

Given a multiset X={x 1,... x n } of real numbers, the floating-point set summation (FPS) problem asks for S n = x 1+...+ x n , and the floating point prefix set summation problem (FPPS) asks for S k =x 1+...+x k for all k = 1, ... n. Let E k * denote the minimum worst-case error over all possible orderings of evaluating S k . We prove that if X has both positive and negative numbers, it is NP-hard to compute S n with the worst-case error equal to E n * . We then give the first known polynomial-time approximation algorithm for computing Sn that has a provably small error for arbitrary X. Our algorithm incurs a worstcase error at most 2([log(n−1)]+1) E n * . After X is sorted, it runs in O(n) time, yielding an O(n 2)-time approximation algorithm for computing S k for all k = 1, ..., n such that the worst-case error for each S k is less than 2(⌈ log(k−1)⌋+ 1)E k * .