Experiments with data perturbations to study condition numbers and numerical stability

Numerical computation of matrix norms, conditioning, scaling (65F35) Roundoff error (65G50) Complexity and performance of numerical algorithms (65Y20)

Review text

The condition number of general problems is investigated. Experiments with data perturbations are used to study condition numbers, regularity and numerical stability. By varying the size of these perturbations and under some reasonable assumptions on the rounding errors, an interval of perturbations is determined, where a log-linear least squares fit provides an estimation of the condition number and the regularity. Three examples (computation of the value of a \(C^ 1\)-function, solution of a linear system, calculation of the roots of a quadratic function) using stable algorithms and an unstable algorithm are considered.

Recommendations

Cites work

Cited in

(2)

Describes a project that uses

Uses Software

This page was built for publication: Experiments with data perturbations to study condition numbers and numerical stability

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1308502)