Author:  Chen, Jinhai 
Title:  Globalization techniques for solving nonlinear problems and applications 
Degree:  Ph.D. 
Year:  2009 
Subject:  Hong Kong Polytechnic University  Dissertations. Nonsmooth optimization. 
Department:  Dept. of Applied Mathematics 
Pages:  xii, 140 p. : ill. ; 30 cm. 
Language:  English 
InnoPac Record:  http://library.polyu.edu.hk/record=b2321026 
URI:  http://theses.lib.polyu.edu.hk/handle/200/4849 
Abstract:  This thesis is concentrated on the study of global techniques to the numerical solution of nonsmooth problems. These problems are directly relevant to the variational inequality problems, complementarity problems, heat transmission problems in medium, parabolic obstacle problems within financial mathematics, and controlstate constrained optimal control problems. These global approaches are discussed deeply on convergence theories and computational results. A vital point in the implementation of the global approaches to the minimization of a nonsmooth (merit) function, such as Newton methods based on path search, line search, and trust region algorithms, is the calculation of a (generalized) Jacobian matrix equation effectively, especially for largescale problems. In Chapter 2, we consider a Krylov subspace strategy for the underlying global methods to solving nonsmooth equations. Such strategy has a main advantage of computing a generalized Jacobian matrix equation, especially, in many applications where the (generalized) Jacobian matrix is not practically computable, or is expensive to obtain. Another global strategy is to consider avoiding the minimization of a nonsmooth (merit) function, pseudotransient continuation in Chapter 3 may be a nice choice for this purpose. Many practical problems have certain structures; if we could find these structures, we may design more suitable algorithms. These algorithms have not only global convergence, but also have special properties, for example: finite termination, monotonicity. Indeed, the Newtontype methods considered to solve piecewise systems in Chapter 4 have a remarkable monotone convergence. These piecewise systems arise from the discretizations of heat transmission problems in a medium, parabolic obstacle problems in financial mathematics. Other strategies, for example: smoothing strategy, nonmonotone strategy, etc., have also rather good effects in most of the practical implementations, even if in the infinite dimensional spaces. To show this, numerical solution of optimal control problems subject to mixed controlstate constraints has been investigated in Chapter 5. The necessary conditions of the optimal control problems are stated in terms of a local minimum principle. By use of the FischerBurmeister function, the local minimum principle is transformed into an equivalent nonlinear and nonsmooth equation in appropriate Banach spaces. This nonlinear and nonsmooth equation is solved by inexact nonsmooth and smoothing Newton methods. The globalized methods are developed in a very general setting that allows for nonmonotonicity of squared residual norm values at subsequent iterates. Numerical examples are presented to demonstrate the efficiency of these presented approaches. 
Files  Size  Format 

b23210266.pdf  18.42Mb 


As a bona fide Library user, I declare that:  


By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms. 