  


Variational Problems in QuasiNewton Methods
Osman Guler (gulermath.umbc.edu) Abstract: It has been known since the early 1970s that the Hessian matrices in quasiNewton methods can be updated by variational means, in several different ways. The usual formulation of these variational problems uses a coordinate system, and the symmetry of the Hessian matrices are enforced as explicit constraints. As a result, the variational problems seem complicated. In this paper, we give a very simple, coordinatefree, and unified treatment of these variational problems by working directly in the vector space of symmetric matrices. Most of our variational problems become simple, least squares problems. All variational problems are convex programming problems, and we consider their duals. A novel feature of our work is that we construct several new variational problems whose optimal solutions coincide with quasiNewton update matrices. These new variational problems may be useful in suggesting new quasiNewton methods in the future. Keywords: QuasiNewton, DFP, BFGS, variational problems, duality, sparse problems Category 1: Nonlinear Optimization (Unconstrained Optimization ) Citation: Department of Mathematics and Statistics University of Maryland, Baltimore County 1000 Hilltop Circle, Baltimore, MD 21250 July, 2006 Download: [PDF] Entry Submitted: 07/18/2006 Modify/Update this entry  
Visitors  Authors  More about us  Links  
Subscribe, Unsubscribe Digest Archive Search, Browse the Repository

Submit Update Policies 
Coordinator's Board Classification Scheme Credits Give us feedback 
Optimization Journals, Sites, Societies  