The matricial relaxation of a linear matrix inequality (Q1949260)

This paper presents a natural relaxation of a linear matrix inequality (LMI), by replacing matrices for the variables \({x_j}\). Thus, the domination questions ((\({{\text{Q}}_1}\)) When does one LMI dominate the other?; (\({{\text{Q}}_2}\)) When are two LMIs mutually dominant, i.e., have the same solution?) may be reduced to constructible semidefinite programs. The development of the relaxation algorithms is based on the following two main results giving precise algebraic characterizations of (matriceal) LMI domination. Let \({L_j} \in {\mathbb S}{{\mathcal R}^{{d_j} \times {d_j}}}\left\langle x \right\rangle\), \(j = 1,2,\) be the monic linear pencils, let \({D_{{L_1}}}\) and \({D_{{L_2}}}\) be their matricial positivity domains, respectively, and assume that \({D_{{L_1}}}\) is bounded. (linear Positivstellansatz). ``\({D_{{L_1}}} \subseteq {D_{{L_2}}}\) if and only if there is a \(\mu \in {\mathbb N}\) and an isometry \(V \in {\mathbb S}{{\mathcal R}^{\mu {d_1} \times {d_2}}}\) such that \({L_2}(x) = {V^ * }({I_\mu } \otimes {L_1}(x))V\).'' (linear Gleichstellensatz). ``\({D_{{L_1}}} = {D_{{L_2}}}\) if and only if minimal defining pencils \({\tilde L_1}\) and \({\tilde L_2}\) for \({D_{{L_1}}}\) and \({D_{{L_2}}}\), respectively, are unitarily equivalent. That is, there is a unitary matrix \(U\) such that \({\tilde L_2}(x) = {U^ * }{\tilde L_1}(x)U\).''