Exact computation of the sign of a finite sum (Q1294292)

The authors study an algorithm which determines the sign of a sum of a bounded number of real quantities. The restriction of being bounded can be dropped later. The algorithm is especially designed to be applied in geometric computations where rounding error free algorithms are particular desiderable due to the strong influence of rounding errors on logical decisions. For this algorithm only computation with fixed word-length is required, no splitting or other mantissa manipulations are necessary, one only needs to know the exponential parts of the represented summands, and it is almost never necessary to compute the value of the whole sum.