Non-convex sparse regularisation (Q847742): Difference between revisions

The author investigates the Tikhonov regularization on the space of quadratic summable sequences, with a very general weighted regularization functional \(T_\alpha\), associated to the solution of an operator equation. The penalty term of the regularization functional is a linear combination, with positive weights, of the values of a given positive function \(\Phi\), multiplied by a positive parameter \(\alpha\). Under some imposed conditions on the function \(\Phi\), the penalty functional term is proved to be weak lower semi-continuous and weak coercitive. Many concrete examples of such functions satisfying these conditions, are presented. As a consequence, it follows that the regularization functional \(T_\alpha\) satisfies the main properties of a regularization method and thus the existence of minimisers for this functional. Also, it implies the convergence of minimisers to the solution of the linear operator equation. A central focus of this paper lies on the possible application of the considered regularization method to the recovery of sparse sequences. A sufficient condition is formulated that enforces the minimisers of the regularization functional to be sparse. Linear convergence of minimisers for a non-convex regularization functional is proved in Theorem 5.1.