The geometry of ill-conditioning (Q1100848)

The problems of finding a solution to a linear system of equations, finding the roots of a polynomial equation, and finding the eigenvalues and eigenvectors of a matrix are considered. The condition number, which limits the accuracy obtainable in the solution, as well as the speed of convergence of certain numerical algorithms, is proportional to the distance of the given problem to the set of degenerate problems. Upper and lower bounds for the probability that a problem chosen at random is within a given distance to the set of degenerate problems are given. Each problem is considered as a vector in an Euclidean space of data vectors, and probabilities are then ratios of volumes of the sets under consideration.