Automatic forward error analysis for floating point algorithms (Q5946850)

The authors study worst case bounds (in terms of both absolute and relative bounds) for propagation of uncertainties in floating point operations. Explicit formulae for the bounds are given, first for arithmetic operations, then for elementary functions. Most proofs are omitted; the reader is referred to preprints available from the authors' web site. The authors report a great success in using their method for implementation of a fast interval library for elementary functions.