Novel shortcut strategies in copositivity detection: Decomposition for quicker positive certificates

Abstract

Copositivity is a property of symmetric matrices which is NP-hard to check. Nevertheless, it plays a crucial role in tight bounds for conic approaches of several hard optimization problems. In this paper, we present novel promising shortcut strategies to exploit favorable instances in a systematic way, using decomposition strategies based upon the idea to allow for overlapping, smaller blocks, profiting from a beneficial sign structure of the entries of the given matrix. The working hypothesis of this approach is the common empirical observation in the community that for detection of copositivity, a negative certificate is easier to obtain than a positive one. First empirical results on carefully orchestrated randomly generated instances seem to corroborate our approach.