Nick Gould (Rutherford Appleton Laboratory)

An interior-point L-1-penalty method for nonlinear optimization
Wednesday 14 April 2004 at 18.00, JCMB 5327

Abstract

We discuss the merits of a mixed interior/exterior-point method for nonlinear programming in which all nonlinear constraints are treated by an l_1 penalty function. Inspired by a proposal by Mayne and Polak (1976), a suitable decomposition of the constraints allows us to derive an exact differentiable penalty function involving only inequality constraints, which may then be treated using a logarithmic barrier. Exactness of the exterior penalty function eliminates the need to drive the corresponding penalty parameter to infinity. Global and fast local convergence of the proposed scheme are exposed. The algorithm is implemented as part of the GALAHAD library under the same SUPERB. Initial numerical experience will be discussed.

Seminars by year

Current 2016 2015 2014 2013 2012 2011 2010 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 1998 1997 1996