Growth of nonconvex functionals at strict local minimizers

Abstract

In this paper, we present new equivalent conditions for the growth of proper lower semicontinuous functionals at strict local minimizers. The main conditions are a variant of the so-called tilt stability property of local minimizers and an analog of the classic Polyak-Łojasiewicz condition, where the gradient is replaced by linear perturbations. We derive the following tilting principle: stability of minimizers under linear perturbations can infer their stability under nonlinear ones. We show how growth conditions can be used to give convergence rates for the proximal point algorithm. Finally, we give an application to elliptic tracking problems, establishing a novel equivalence between second-order conditions and the sensitivity of solutions with respect to uncertainty in data.