PROBABILISTIC ERROR ANALYSIS FOR INNER PRODUCTS.

Probabilistic models are proposed for bounding the forward error in the numerically computed inner product (dot product, scalar product) between two real n-vectors. We derive probabilistic perturbation bounds as well as probabilistic roundoff error bounds for the sequential accumulation of the inner product. These bounds are nonasymptotic, explicit, with minimal assumptions, and with a clear relationship between failure probability and relative error. The roundoffs are represented as bounded, zero-mean random variables that are independent or have conditionally independent means. Our probabilistic bounds are based on Azuma's inequality and its associated martingale, which mirrors the sequential order of computations. The derivation of forward error bounds "from first principles" has the advantage of producing condition numbers that are customized for the probabilistic bounds. Numerical experiments confirm that our bounds are more informative, often by several orders of magnitude, than traditional deterministic bounds-even for small vector dimensions n and very stringent success probabilities. In particular the probabilistic roundoff error bounds are functions of n rather than n, thus giving a quantitative confirmation of Wilkinson's intuition. The paper concludes with a critical assessment of the probabilistic approach.