Orientador: Roberto Andreani / Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Matematica, Estatistica e Computação Cientifica / Made available in DSpace on 2018-08-10T22:54:25Z (GMT). No. of bitstreams: 1 Ramirez_VivianaAnalia_M.pdf: 2563612 bytes, checksum: db80aac5c845975430fe4820638c7a46 (MD5) Previous issue date: 2008 / Resumo: Os sistemas ponto de sela surgem em uma grande quantidade de áreas de investiga¸c¿ao, como física, química, engenharia, reconstrução de imagens, etc. Portanto, s¿ao objeto de pesquisa, tanto as propriedades presentes neles como os métodos utilizados para a sua resolução. Diversos métodos foram desenvolvidos dependendo das características do sistema, alguns deles com a propriedade de preservar a estrutura da matriz do sistema. Neste trabalho utilizamos umo destes métodos para melhorar a precisão obtida pelo método ALGENCAN (Lagrangiano Aumentado usando GENCAN) em problemas de Programação Não Linear (PNL). Este método é muito robusto, ele obtém uma boa aproximação da solução com poucas iterações, mas perto da solução não consegue obter uma precisão muito exigente. Para melhorar esta precisão, aplicamos o método de Newton a um sistema KKT reduzido no ponto obtido por ALGENCAN, gerando um sistema ponto de sela. Para esta implementação utilizamos o método conhecido como fatoração LDLT , escolhido por sua propriedade de preservar a estrutura esparsa do sistema / Abstract: Saddle point systems arise in wide areas of research fields like physics, chemistry and engineering and images reconstructions, etc. Then, the properties of these systems and solving methods have been subjects of intense study in the last years. Depending upon the system properties, several methods were developed; some of these, exhibit the property of preserving the matrix structure system, like the sparsity. In this work, we have used one of these methods to improve the accuracy by using ALGECAN (Augmented Lagrangian using GENCAN) applied to Non-linear Programming (NLP) problems. This is a robust method which helps to get a good approximation to the solution. However, in several cases, it is not possible to get the desired accuracy. In order to improve the precision, we have applied Newton¿s method in a reduced KKT system, starting from a point given by ALGENCAN, which is a saddle point. We employ the so called LDLT factorization in order to implement Newton¿s method, which give us better accuracy / Mestrado / Otimização / Mestre em Matemática Aplicada

Cette thèse vise à concevoir des solveurs efficaces pour résoudre des systèmes linéaires, résultant des problèmes d'optimisation sous contraintes dans certaines applications de dynamique des structures et vibration (la corrélation calcul-essai, la localisation d'erreur, le modèle hybride, l'évaluation des dommages, etc.). Ces applications reposent sur la résolution de problèmes inverses, exprimés sous la forme de la minimisation d'une fonctionnelle en énergie. Cette fonctionnelle implique à la fois, des données issues d'un modèle numérique éléments finis, et des essais expérimentaux. Ceci conduit à des modèles de haute qualité, mais les systèmes linéaires point-selle associés, sont coûteux à résoudre. Nous proposons deux classes différentes de méthodes pour traiter le système. La première classe repose sur une méthode de factorisation directe profitant de la topologie et des propriétés spéciales de la matrice point-selle. Après une première renumérotation pour regrouper les pivots en blocs d'ordre 2. L'élimination de Gauss est conduite à partir de ces pivots et en utilisant un ordre spécial d'élimination réduisant le remplissage. Les résultats numériques confirment des gains significatifs en terme de remplissage, jusqu'à deux fois meilleurs que la littérature pour la topologie étudiée. La seconde classe de solveurs propose une approche à double projection du système étudié sur le noyau des contraintes, en faisant une distinction entre les contraintes cinématiques et celles reliées aux capteurs sur la structure. La première projection est explicite en utilisant une base creuse du noyau. La deuxième est implicite. Elle est basée sur l'emploi d'un préconditionneur contraint avec des méthodes itératives de type Krylov. Différentes approximations des blocs du préconditionneur sont proposées. L'approche est implémentée dans un environnement distribué parallèle utilisant la bibliothèque PETSc. Des gains significatifs en terme de coût de calcul et de mémoire sont illustrés sur plusieurs applications industrielles. / This thesis aims at designing efficient numerical solution methods to solve linear systems, arising in constrained optimization problems in some structural dynamics and vibration applications (test-analysis correlation, model error localization,hybrid model, damage assessment, etc.). These applications rely on solving inverse problems, by means of minimization of an energy-based functional. This latter involves both data from a numerical finite element model and from experimental tests, which leads to high quality models, but the associated linear systems, that have a saddle-point coefficient matrices, are long and costly to solve. We propose two different classes of methods to deal with these problems. First, a direct factorization method that takes advantage of the special structures and properties of these saddle point matrices. The Gaussian elimination factorization is implemented in order to factorize the saddle point matrices block-wise with small blocks of orders 2 and using a fill-in reducing topological ordering. We obtain significant gains in memory cost (up to 50%) due to enhanced factors sparsity in comparison to literature. The second class is based on a double projection of the generated saddle point system onto the nullspace of the constraints. The first projection onto the kinematic constraints is proposed as an explicit process through the computation of a sparse null basis. Then, we detail the application of a constraint preconditioner within a Krylov subspace solver, as an implicit second projection of the system onto the nullspace of the sensors constraints. We further present and compare different approximations of the constraint preconditioner. The approach is implemented in a parallel distributed environment using the PETSc library. Significant gains in computational cost and memory are illustrated on several industrial applications.

This thesis presents numerical techniques for solving problems of incompressible flow coupled to scalar transport equations using finite element discretizations in space. The two applications considered in this thesis are multi-phase flow, modeled by level set or phase field methods, and planetary mantle convection based on the Boussinesq approximation. A systematic numerical study of approximation errors in evaluating the surface tension in finite element models for two-phase flow is presented. Forces constructed from a gradient in the same discrete function space as used for the pressure are shown to give the best performance. Moreover, two approaches for introducing contact line dynamics into level set methods are proposed. Firstly, a multiscale approach extracts a slip velocity from a micro simulation based on the phase field method and imposes it as a boundary condition in the macro model. This multiscale method is shown to provide an efficient model for the simulation of contact-line driven flow. The second approach combines a level set method based on a smoothed color function with a the phase field method in different parts of the domain. Away from contact lines, the additional information in phase field models is not necessary and it is disabled from the equations by a switch function. An in-depth convergence study is performed in order to quantify the benefits from this combination. Also, the resulting hybrid method is shown to satisfy an a priori energy estimate. For the simulation of mantle convection, an implementation framework based on modern finite element and solver packages is presented. The framework is capable of running on today's large computing clusters with thousands of processors. All parts in the solution chain, from mesh adaptation over assembly to the solution of linear systems, are done in a fully distributed way. These tools are used for a parallel solver that combines higher order time and space discretizations. For treating the convection-dominated temperature equation, an advanced stabilization technique based on an artificial viscosity is used. For more efficient evaluation of finite element operators in iterative methods, a matrix-free implementation built on cell-based quadrature is proposed. We obtain remarkable speedups over sparse matrix-vector products for many finite elements which are of practical interest. Our approach is particularly efficient for systems of differential equations.

In this thesis, we develop preconditioned iterative methods for the solution of matrix systems arising from PDE-constrained optimization problems. In order to do this, we exploit saddle point theory, as this is the form of the matrix systems we wish to solve. We utilize well-known results on saddle point systems to motivate preconditioners based on effective approximations of the (1,1)-block and Schur complement of the matrices involved. These preconditioners are used in conjunction with suitable iterative solvers, which include MINRES, non-standard Conjugate Gradients, GMRES and BiCG. The solvers we use are selected based on the particular problem and preconditioning strategy employed. We consider the numerical solution of a range of PDE-constrained optimization problems, namely the distributed control, Neumann boundary control and subdomain control of Poisson's equation, convection-diffusion control, Stokes and Navier-Stokes control, the optimal control of the heat equation, and the optimal control of reaction-diffusion problems arising in chemical processes. Each of these problems has a special structure which we make use of when developing our preconditioners, and specific techniques and approximations are required for each problem. In each case, we motivate and derive our preconditioners, obtain eigenvalue bounds for the preconditioners where relevant, and demonstrate the effectiveness of our strategies through numerical experiments. The goal throughout this work is for our iterative solvers to be feasible and reliable, but also robust with respect to the parameters involved in the problems we consider.