The restricted strong convexity revisited: analysis of equivalence to error bound and quadratic growth (Q523179)

The author discusses \(*\) the restricted strongly convex property, \(*\) the global error bound property and \(*\) the quadratic growth property of a smooth convex function \(f: \mathbb{R}^n\to\mathbb{R}\) which ensure globally linear convergence rates of descent methods for unconstrained optimization problems. It is shown that these conditions are equivalent. In the main part of the paper, these conditions are extended to different constrained convex optimization problems and it is proved that the equivalences remain true. Based on these notions, a new asymptotically linear convergence result for a proximal gradient method is pointed out. Especially for the problem of minimizing the composition of an affine mapping with a smooth strongly convex function over a polyhedral set, sharper results are obtained.