A formula for computation of the real stability radius (Q1898494)

The authors present a method for approximation of the maximal perturbations of a real matrix. Its eigenvalues have to stay in a prescribed area in the complex plane. Such a measure of robustness of the eigenvalue clustering property is called a real stability radius. The main result given reduces the problem to the minimization of a unimodal function in the semiopen unit interval. The theorem which states this result has a long and involved proof based on the classical Schmidt-Mirsky theorem of approximating a matrix by one of lower rank [see e.g. \textit{G. W. Stewart} and \textit{J. G. Sun}, Matrix perturbation theory (Academic Press (1990; Zbl 0706.65013)] and some properties of the singular vectors of a specific complex matrix. The theorem is followed by a computable procedure to construct a worst perturbation matrix. The authors discuss also the problem of sensitivity of the real stability radius to changes of the matrix. More precisely it is proved that the stability radius may be discontinuous only for matrices with rank not greater than one. The more general definition of the structured stability radius which allows for a linear fractional dependence of the perturbed matrix on the perturbation is also discussed.