P. Langlois et F. Nativel, REDUCTION AND BOUNDING OF THE ROUNDING ERROR IN FLOATING-POINT ARITHMETIC, Comptes rendus de l'Academie des sciences. Serie 1, Mathematique, 327(8), 1998, pp. 781-786
This Note deals with a method of reducing and controlling rounding err
ors in floating point computation. This method is based on the first o
rder linearization numerical evaluation of the global error using the
absolute elementary rounding errors computation and the partial deriva
tives computation of the result with respect to these elementary round
ing errors. (C) Academie des Sciences/Elsevier, Paris.