Abstract: Can we assure math computations by automatic verifying floating-point accuracy? We define fast arithmetic (based on Dekker's) over twofold approximations z0+z1 such that z0 is standard result and z1 assesses inaccuracy of z0. We propose on-fly tracking of z1, detecting if it appears too high. We believe permanent tracking is worth its cost. C++ test code for Intel AVX available via web.
2013 (C) Evgeny Latkin, free for academic and non-commercial use
Download:
Text also uploaded to arXive: http://arxiv.org/abs/1401.6235