Matrix subspaces and determinantal hypersurfaces (Q2380812): Difference between revisions

Let \(\mathcal V\) be a subspace of \(\mathbb{C}^{n\times n}\) having invertible elements, and denote by \(\text{Inv}({\mathcal V})\) the set of the inverses of such matrices. If its closure \({\mathcal W}\) is a subspace, then call \({\mathcal V}\) invertible and denote \({\mathcal V}^{-1}={\mathcal W}\). In the author's earlier paper [BIT 47, 793--808 (2007; Zbl 1186.15011)], this concept was introduced in connection with factoring a matrix into a product of two matrices from prescribed subspaces. The author characterizes invertible subspaces as follows: If \(\mathcal V\) is invertible, then \[ {\mathcal V}^{-1}=V^{-1}{\mathcal V}V^{-1}, \tag{1} \] for all invertible \(V\in{\mathcal V}\). Conversely, if the right-hand side of (1) is the same for all invertible \(V\in{\mathcal V}\), then \({\mathcal V}\) is invertible. For noninvertible \({\mathcal V}\), there does not seem to be a general way to characterize \(\text{Inv}({\mathcal V})\). The author studies this task by examples and gives a fairly versatile structure where~\(\text{Inv}({\mathcal V})\) can be viewed as a product of three subspaces. He also defines the spectrum of \({\mathcal V}\) in terms of determinantal hypersurfaces related to singular elements of \({\mathcal V}\). Links between matrix analysis and algebraic geometry are encountered. Matrix subspaces and respective Grassmannians are split into equivalence classes. Finally, the author studies conditioning of \({\mathcal V}\).