📝 SummarXiv

Abstract

Matrix completion results deal with the question of when a partially specified symmetric matrix can be completed to a member of certain matrix cones. Results from positive semidefinite matrix completion and completely positive matrix completion have been successfully applied in optimization to greatly reduce the number of variables in conic optimization problems in the space of symmetric matrices. In this text, we go the other direction and show that we can use tools from conic optimization (more precisely: from copositive optimization) to establish a new completion result that complements the existing literature in two regards: firstly, we consider set-completely positive matrix completion, which generalizes completion with respect to the traditional completely positive matrix cone. Secondly, we consider a specification pattern that is not in the scope of classical results for completely positive matrix completion. Namely, we consider arrow-head specification patterns where the width is equal to one. Our theory is applied to a class of quadratic optimization problems.