This release of QSS implements a new algorithm for solving nonconvex quadratic separable problems. Previously, we implemented nonconvex ADMM for cost nonconvex cost functions that have closed-form proximal operators. Now, we begin by solving a _relaxed_ convex version of the problem, and then we use that solution to start the iterates of nonconvex ADMM. In practice, this method is much better at avoiding suboptimal local minima.