The main reasons for matrices multiplying to zero
Let \(A_1, A_2, \cdots, A_n\) be a sequence of matrices, with \(A_k\in \mathbb{C}^{d_{k}\times d_{k-1}}\) for a sequence of dimensions \(\underline{d}=\{d_0, d_1, \cdots, d_n\}\), and the variety \(\Sigma_{\underline{d}}=\{A_nA_{n-1}\cdots A_1=0\}\). Describing all components of \(\Sigma_{\underline{d}}\), even just their number and dimensions is an ambitious problem. Instead, the authors focus on the maximal-dimensional components and their codimension \(C\) and the numbers \(\theta\).\N\NIn [\textit{S. P. Lehalleur} and \textit{R. Rimányi}, ``Geometry of fibers of the multiplication map of deep linear neural networks, Preprint, \url{arXiv:2411.19920}], the authors provided explicit formulas for \(C\) and \(\theta\) for any dimension vector \(\underline{d}\). In the paper under review, the authors provide an explicit description of the maximal-dimensional components of \(\Sigma_{\underline{d}}\) for arbitrary \(\underline{d}\), thereby identifying the main reasons for matrices multiplying to zero.\N\NFor any dimension vector \(\underline{d}\), the reference above can construct a quadratic integer program QIP and prove the minimal value of QIP is \(C\), and the number of times QIP takes its minimal value is \(\theta\). In this paper, the authors constructed another quadratic integer program on the set of raising vectors QIP' and showed QIP and QIP' are equivalent. They also described an algorithm that, for each of the \(\theta\) optimal solutions, constructs a lace diagram, a picture of dots and lines. By using these techniques, the authors established a bijection between the set of solutions of QIP' and the set of minimal codimension orbits in \(\Sigma_{\underline{d}}\).
- A formula for non-equioriented quiver orbits of type 𝐴
- Algebraic Geometry and Statistical Learning Theory
- Classes of degeneracy loci for quivers: the Thom polynomial point of view.
- scientific article; zbMATH DE number 3702881 (Why is no real title available?)
- Induzierte Darstellungen in der Theorie der endlichen, algebraischen Gruppen
- Positivity of quiver coefficients through Thom polynomials
- Quiver representations and quiver varieties
- Recent advances in algebraic geometry and Bayesian statistics
- Unzerlegbare Darstellungen. I. (Indecomposable representations. I)
This page was built for publication: The main reasons for matrices multiplying to zero
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6894363)
