The Distance of Potentially Stable Sign Patterns to the Unstable Matrices (Q4785778)

A sign pattern is a matrix \(\{{\mathcal A}=[\alpha_{ij}]\}\) with \(\alpha_{ij}\in \{+,0,-\}\); a real (numerical) matrix belongs to this sign pattern (\(A \in \{{\mathcal A}\}\)) if \(\text{sign}(a_{ij})=\alpha_{ij}\). A matrix is stable if all its eigenvalues have negative real parts. A sign pattern is sign stable if all matrices \(A \in \{\mathcal A\}\) are stable; it is potentially stable if there is at least one \(A \in \{\mathcal A\}\) that is stable. It is minimally potentially stable if it is potentially stable, irreducible and if replacing any \(+\) or \(-\) entry by \(0\) results in a pattern that is not potentially stable. The authors define a relative distance \(\delta_{us}(A)\) to the unstable matrices for a stable \(n\times n\) matrix \(A\) and extend it to a potentially stable sign pattern. They show that \(\delta_{us}(A)\in (0,1]\) and \(|\delta_{us}(A)-\delta_{us}(B)|\) is bounded above in terms of \(\|A-B\|_2\). For \(n=2\) there is an essentially unique minimally potentially stable sign pattern and a matrix with maximal relative distance to the unstable matrices can be determined. For \(n=3\) and \(n=4\) a complete list of all minimally potentially stable tree sign patterns is given. For larger matrices the authors provide some approximations and estimates.