Complexity of Bezout's theorem. III: Condition number and packing (Q2365839): Difference between revisions

[For part I see J. Am. Math. Soc. 6, No. 2, 459-501 (1993; Zbl 0821.65035).] The best known example of a condition number is that for the problem of solving a linear system \(Ax= b\). The condition number is \(\kappa_A= |A||A^{- 1}|\), wehre an operator norm on the matrix \(A\) is used. It measures the sensitivity of output error relative to input error. Since \(\kappa_A\geq 1\), \(\log \kappa_A\) is positive and measures the ``loss of precision''. For finding a root of a polynomial \(f\), or solving \(f(\zeta)= 0\), the condition number has been taken as \(W_{f, \zeta}= 1/|f'(\zeta)|\). Again it measures sensitivity relative to input error. This condition number lacks certain naturality properties as invariance under the transformations \(f\to \lambda f\), or \(\zeta\to \alpha\zeta\) for \(\lambda\), \(\alpha\in \mathbb{C}- 0\). Moreover, \(W_{f, \zeta}\) does not satisfy \(W_{f, \zeta}\geq 1\). For these and other reasons, we are motivated to define a new notion of condition number \(\mu(f, \zeta)\) of a polynomial at \(\zeta\in \mathbb{C}\).