Totally positive matrices and cyclic polytopes (Q1107599)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Totally positive matrices and cyclic polytopes |
scientific article |
Statements
Totally positive matrices and cyclic polytopes (English)
0 references
1988
0 references
As the main result of the present note it is proved that there is a natural correspondence between alternating polytopes and totally positive matrices. A real (n-d)\(\times d\) matrix A, where \(n>d>1\) is totally positive (totally nonnegative) if all subdeterminants of A are positive (nonnegative). A polytope \(P\subseteq E^{d-1}\) is cyclic if it is isomorphic to the convex hull of a finite subset of the moment curve \(\{(t,t^ 2,...,t^{d-1}):\) \(t\in R\}\). The vertices of a cyclic polytope are canonically labelled with respect to succession on the moment curve. A cyclic polytope P is alternating if all subpolytopes of P are cyclic, and if the corresponding isomorphisms to cyclic polytopes are all induced by the canonical labelling of P. Considering the Euclidean space \(E^{d-1}\) embedded as an affine hyperplane in the vector space \(R^ d\) the author proves: Theorem 1. Let \(x_ 1,...,x_ n\in E^{d-1}\) be the (canonically labelled) vertices of an alternating (d-1)-polytope, and let A be the (n- d)\(\times d\) matrix of homogeneous coordinates of \(x_{d+1},...,x_ n\) versus the ordered basys \(((-1)^{d+1}x_ d,(-1)^ dx_{d-1},...,- x_ 2,x_ 1)\) of \(R^ d\). Then A is totally positive, and conversely, every totally positive matrix A is obtained from an alternating polytope in this way. Some remarks to and some applications of this result are also given, among them the following Corollary 5: The space of totally positive \(m\times n\) matrices is homeomorphic to \(R^{mn}\).
0 references
matroids
0 references
alternating polytopes
0 references
totally positive matrices
0 references
cyclic polytope
0 references
0 references