Faces of the cone of Euclidean distance matrices: characterizations, structure and induced geometry (Q2568355)

This paper deals with the faces of the cone of Euclidean distance matrices. A known result tells us how to identify every distance matrix in a particular face using the column space of the coordinate matrices. The author changes from characterizing faces based on ranges of matrices to a new characterization based on null spaces. For every Euclidean distance matrix \(D\) the author defines a new subspace, \(LGS(D)\), in the following way: \(LGS(D)=N(D)\) if \(D\) is spherical and \(LGS(D)=N(D)\oplus \langle x\rangle\) otherwise, with \(x\) solution of \(Dx=e\), where \(N(D)\) denotes the null space of \(D\), \(\langle x\rangle\) the subspace generated by \(x\) and \(e\) the vector with all ones. This subspace contains information about the location of the matrix and the geometrical structure of the configuration of points. In addition, this subspace contains the null space of \(D\), and becomes a tool to describe faces and determine their structure. The subspace \(LGS(D)\) is also related to the centered preimage of \(D\), as well as to any centered coordinate matrix. Given a Euclidean distance matrix \(D\) in the boundary of the cone, \(LGS(D)\) allows us to identify a set of supporting hyperplanes and to obtain a characterization of the minimal face that contains \(D\). A new face characterization based in coordinate matrices is related also to \(LGS(D)\) and state in term of homogeneous linear constraints, a description for optimization problems. Finally on a different direction, from \(LGS(D)\) the author determines basic geometric properties for matrices in the minimal face containing \(D\). In other words, given \(D\), the corresponding \(LGS(D)\) provides basic geometric properties of every configuration of points, whose distance matrix is on that face.