zbMATH — the first resource for mathematics

Quasi-Newton bundle-type methods for nondifferentiable convex optimization. (English) Zbl 0927.65074
The authors present a method for solving the nondifferentiable convex programming problem: \(\min f(x)\) subject to \(x\in\mathbb{R}^n\), where \(f: \mathbb{R}^n\to\mathbb{R}\) is a nondifferentiable convex function. The method is developed using the ideas of the Moreau-Yosida regularization [K. Yosida, Functional analysis (1965; Zbl 0126.11504)], the bundle method, and the quasi-Newton method. It is shown that in this method the minimizing solution of the given programming problem is exactly the solution of a subproblem of minimizing the Moreau-Yosida regularization of \(f\) over \(\mathbb{R}^n\).
The entire discussion in the paper is devoted to the solution of the subproblem, and a globally and superlinearly convergent quasi-Newton bundle-type algorithm is described for solving the subproblem.
No numerical results are given.

65K05 Numerical mathematical programming methods
90C25 Convex programming
90C53 Methods of quasi-Newton type
52A41 Convex functions and convex programs in convex geometry
PDF BibTeX Cite
Full Text: DOI