Tight Error Analysis in Fixed-point Arithmetic

We consider the problem of estimating numerical accuracy programs with operations in fixed-point arithmetic and variables arbitrary, mixed precision, possibly non-deterministic value. By applying a set parameterised rewrite rules, we transform relevant fragments program under consideration into sequences integer over vectors bits, thereby reducing as to whether error enclosures initial can ever exceed given order magnitude simple reachability queries on transformed program. describe possible verification flow prototype analyser that implements our technique. present an experimental evaluation particularly complex industrial case study, including preliminary comparison between bit-level word-level decision procedures.