Title:Searching for and quantifying non-convexity of functions

Abstract:Convexity plays a prominent role in a number of problems and their solutions, but practical considerations also frequently give rise to non-convex functions. We suggest a method that helps to determine the regions over which the functions are convex; and when they are not, the method also helps to quantify the extent of their non-convexity. The method relies on a specially constructed decomposition of symmetric matrices, such as the Hessian, into the difference of two positive semidefinite matrices. As such, the method also provides a measure of positive semidefiniteness of symmetric and, more generally, Hermitian matrices. We illustrated these theoretical results with several examples, one of which provides a detailed analysis of a problem pertaining to insurance and financial risk measurement and management.