Unger, Thomas : Acceleration of the Level Method by Exploiting Recurrent Subgradients in Linear Programming Decomposition

Author(s):

Unger, Thomas

Title:

Acceleration of the Level Method by Exploiting Recurrent Subgradients in Linear Programming Decomposition

Electronic source:

application/pdf

Preprint series:

Technische Universität Chemnitz, Fakultät für Mathematik (Germany). Preprint 1, 2002

Mathematics Subject Classification:

90C25			[ Convex programming ]
65K05			[ Mathematical programming ]
90C05			[ Linear programming ]

Abstract:

The level method is a special variant of the well-known cutting plane methods for solving nonsmooth convex optimization problems. It proved to be competitive with other methods in nonsmooth optimization. It has been observed that during the solving process these methods compute cutting planes which are (nearly) parallel to each other. In this paper we investigate what effects appear together with this observation and how they may be exploited to speed up the level method algorithm for some classes of problems.

Keywords:

nonsmooth programming, cutting plane methods, level method, piecewise linear functions

Language:

English

Publication time:

1 / 2002