A statistical analysis of the numerical condition of multiple roots of polynomials. (Q1416369)

This is an interesting paper about measurement of the condition of an \(m\)-tuple root \(x_0\) of a polynomial \(p(x)\). It clearly demonstrates that in many situations the error due to uncertainty in the data is much more significant than the errors resulted from discretization, truncation and roundoff. Thus uncertainty of data should be considered when determining the computational reliability of a numerical result. Assuming that each coefficient of \(p(x)\) is perturbed by an independent zero mean Gaussian random variable, the author introduces a measure of the condition of \(x_0\) as the ratio of the expected values of its relative error to the expected value of the relative error in the coefficients of \(p(x)\), both defined in the componentwise and normwise. This new type of condition number of a multiple root of a polynomial is very good for measuring sensitivities of roots with respect to the measurement and experimental inaccuracies in deriving the polynomial itself. The necessities of both componentwise and normwise condition numbers is illustrated by various examples.