A copositivity probe (Q5954132)

A real symmetric matrix \(A\) of order \(p\) is said to be copositive if \(x^TAx \geq 0\) for \(x \geq 0\) (strictly copositive if it is copositive and equality holds only for \(x=0\)). The aim of the paper is to find whether a real symmetric matrix \(A\) is equal to the sum of a positive semidefinite matrix \(S\) and a nonnegative matrix \(P\). Every such matrix \(A\) is copositive and it is strictly copositive if \(S\) is positive definite. A very simple algorithm based on the perturbation theory of eigenvalues and eigenvectors in quantum mechanics is used here. It is shown that if a tested matrix \(A\) is not copositive then in many such cases the algorithm produces a proof that \(A\) is not copositive by producing a positive vector \(v\) such that \(v^TAv < 0\). This byproduct increases the value of the procedure. If the procedure succeeds, one has a proof of copositivity and if it fails and produces the mentioned vector \(v\), one has a proof of noncopositivity.