On Solution Uniqueness and Robust Recovery for Sparse Regularization with a Gauge: from Dual Point of View

Abstract: In this paper, we focus on the exploration of solution uniqueness, sharpness, and robust recovery in sparse regularization with a gauge $J$. Based on the criteria for the uniqueness of Lagrange multipliers in the dual problem, we give a characterization of the unique solution via the so-called radial cone. We establish characterizations of the isolated calmness (i.e., uniqueness of solutions combined with the calmness properties) of two types of the solution mappings and prove that they are equivalent under the assumption of metric subregularity of the subdifferentials for regularizers. Furthermore, we present sufficient and necessary conditions for a sharp solution, and show that these conditions guarantee robust recovery with a linear rate and imply local upper Lipschitzian properties of the solution mapping. Some applications of the polyhedral regularizer case such as sparse analysis regularization, weighted sorted $\ell_1$-norm sparse regularization, and non-polyhedral regularizer cases such as nuclear norm optimization, conic gauge optimization, and semidefinite conic gauge optimization are given.