  


An Extension of the Conjugate Directions Method With Orthogonalization to LargeScale Problems With Bound Constraints
Edouard Boudinov (edouard.boudinovmail.ru) Abstract: In our reports on GAMM04 and ECCOMAS04 there has been presented a new conjugate directions method for large scale unconstrained minimization problems. High efficiency of this method is ensured by employing an orthogonalization procedure: when constructing the next conjugate vector the component of the gradient is used that is orthogonal to the subspace of preceding conjugate vectors (instead of the gradient itself). Such a procedure does not require the current point to be the minimum point along . This makes it possible to perform only one step on each iteration, without any linear searches along the vector . As very high accuracy in determining conjugate directions is achieved, the algorithm ensures considerable reduction in the number of iterations. Now this algorithm is extended to minimization problems with bound constraints. Because the linear minimization along the newly found conjugate vector is not needed for constructing the next conjugate vector on the following iteration, we are able to incorporate naturally the bound constraints into the algorithm. The algorithm is supplemented with a simple rule for determining boundary variables and a criterion for restarts. Results of a numerical experiment on several quadratic testfunctions with the number of variables N up to 100'000 are presented. The influence of the bound constraints on convergence of the method is studied. Comparison with other methods, in particular, the limited memory BFGS algorithm, is given. Keywords: optimization, unconstrained minimization, bound constrained optimization, conjugate direction methods, largescale problems, orthogonalization Category 1: Nonlinear Optimization (Boundconstrained Optimization ) Category 2: Nonlinear Optimization (Unconstrained Optimization ) Category 3: Nonlinear Optimization (Quadratic Programming ) Citation: Edouard Boudinov and Arkadiy Manevich, "An Extension of the Conjugate Directions Method With Orthogonalization to LargeScale Problems With Bound Constraints", PAMM, Proceedings in Applied Mathematics and Mechanics, 5, pp 737738, DOI: 10.1002/pamm.200510343 Download: Entry Submitted: 05/30/2005 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  