Precision-aware deterministic and probabilistic error bounds for floating point summation
From MaRDI portal
(Redirected from Publication:6093390)
Abstract: We analyze the forward error in the floating point summation of real numbers, for computations in low precision or extreme-scale problem dimensions that push the limits of the precision. We present a systematic recurrence for a martingale on a computational tree, which leads to explicit and interpretable bounds without asymptotic big-O terms. Two probability parameters strengthen the precision-awareness of our bounds: one parameter controls the first order terms in the summation error, while the second one is designed for controlling higher order terms in low precision or extreme-scale problem dimensions. Our systematic approach yields new deterministic and probabilistic error bounds for three classes of mono-precision algorithms: general summation, shifted general summation, and compensated (sequential) summation. Extension of our systematic error analysis to mixed-precision summation algorithms that allow any number of precisions yields the first probabilistic bounds for the mixed-precision FABsum algorithm. Numerical experiments illustrate that the probabilistic bounds are accurate, and that among the three classes of mono-precision algorithms, compensated summation is generally the most accurate. As for mixed precision algorithms, our recommendation is to minimize the magnitude of intermediate partial sums relative to the precision in which they are computed.
Recommendations
Cites work
- A Class of Fast and Accurate Summation Algorithms
- A New Approach to Probabilistic Rounding Error Analysis
- Accuracy and Stability of Numerical Algorithms
- Accurate and Efficient Floating Point Summation
- Concentration Inequalities and Martingale Inequalities: A Survey
- Error estimation of floating-point summation and dot product
- Faster stochastic trace estimation with a Chebyshev product identity
- scientific article; zbMATH DE number 1033192 (Why is no real title available?)
- Improved error bounds for inner products in floating-point arithmetic
- Mixed precision algorithms in numerical linear algebra
- On relative errors of floating-point operations: optimal bounds and applications
- Probabilistic Error Analysis for Inner Products
- Probability and Computing
- Rigorous roundoff error analysis of probabilistic floating-point computations
- Sharp estimates for perturbation errors in summations
- Sharper probabilistic backward error analysis for basic linear algebra kernels with random data
- Simulating Low Precision Floating-Point Arithmetic
- Stochastic rounding and its probabilistic backward error analysis
Cited in
(12)- Introducing SummerTime: a package for high-precision computation of sums appearing in DRA method
- Fast and accurate floating point summation with application to computational geometry
- Algorithm 908
- Probabilistic analysis of floating-point addition
- scientific article; zbMATH DE number 3682957 (Why is no real title available?)
- A Distillation Algorithm for Floating-Point Summation
- A priori worst case error bounds for floating-point computations
- Linear-Time Approximation Algorithms for Computing Numerical Summation with Provably Small Errors
- Bounds on nonlinear errors for variance computation with stochastic rounding
- Error analysis of sum-product algorithms under stochastic rounding
- Stochastic rounding implicitly regularizes tall-and-thin matrices
- A Second-Moment Theory for Floating-Point Reduction Trees
This page was built for publication: Precision-aware deterministic and probabilistic error bounds for floating point summation
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6093390)
