Convergence of an inexact majorization-minimization method for solving a class of composite optimization problems (Q2415211)

A new majorization-minimization method for solving nonconvex optimization problems is proposed, where in each iteration a majorizer of the objective function is minimized. This approach is available for some classes of functions such as the directionally differentiable ones, the concave $C^1$ ones, the polynomials or some composite functions. A more easily implementable inexact variant of the method is presented as well and convergence properties and the accumulation points are investigated in both cases. Examples illustrating the applicability of both versions are provided, too, and a concrete application of the inexact method to sparse source localization is presented as well. For the entire collection see [Zbl 1407.90006].