Solving linear systems of equations is an integral part of most scientific simulations. In recent years, there has been a considerable interest in large scale scientific simulation of complex physical processes. Iterative solvers are usually preferred for solving linear systems of such magnitude due to their lower computational requirements. Currently, computational scientists have access to a multitude of iterative solver options available as "plug-and- play" components in various problem solving environments. Choosing the right solver configuration from the available choices is critical for ensuring convergence and achieving good performance, especially for large complex matrices. However, identifying the "best" preconditioned iterative solver and parameters is challenging even for an expert due to issues such as the lack of a unified theoretical model, complexity of the solver configuration space, and multiple selection criteria. Therefore, it is desirable to have principled practitioner-centric strategies for identifying solver configuration(s) for solving large linear systems. The current dissertation presents a general practitioner-centric framework for (a) problem independent retrospective analysis, and (b) problem-specific predictive modeling of performance data. Our retrospective performance analysis methodology introduces new metrics such as area under performance-profile curve and conditional variance-based finetuning score that facilitate a robust comparative performance evaluation as well as parameter sensitivity analysis. We present results using this analysis approach on a number of popular preconditioned iterative solvers available in packages such as PETSc, Trilinos, Hypre, ILUPACK, and WSMP. The predictive modeling of performance data is an integral part of our multi-stage approach for solver recommendation. The key novelty of our approach lies in our modular learning based formulation that comprises of three sub problems: (a) solvability modeling, (b) performance modeling, and (c) performance optimization, which provides the flexibility to effectively target challenges such as software failure and multiobjective optimization. Our choice of a "solver trial" instance space represented in terms of the characteristics of the corresponding "linear system", "solver configuration" and their interactions, leads to a scalable and elegant formulation. Empirical evaluation of our approach on performance datasets associated with fairly large groups of solver configurations demonstrates that one can obtain high quality recommendations that are close to the ideal choices.

Dans le contexte de l'analyse numérique en calcul de structures, la génération de maillages conformes sur des modèles à géométrie complexe conduit à des tailles de modèles importantes, et amène à imaginer de nouvelles approches éléments finis. Le temps de génération d'un maillage est directement lié à la complexité de la géométrie, augmentant ainsi considérablement le temps de calcul global. Les processeurs graphiques (GPU) offrent de nouvelles opportunités pour le calcul en temps réel. L'architecture grille des GPU a été utilisée afin d'implémenter une méthode éléments finis sur maillage cartésien. Ce maillage est particulièrement adapté à la parallélisation souhaitée par les processeurs graphiques et permet un gain de temps important par rapport à un maillage conforme à la géométrie. Les formulations de la méthode des éléments finis ainsi que de la méthode des éléments finis étendue ont été reprises afin d'être adaptées à notre méthode. La méthode des éléments finis étendus permet de prendre en compte la géométrie et les interfaces à travers un choix adéquat de fonctions d'enrichissement. Cette méthode discrétise par exemple sans mailler explicitement les fissures, et évite surtout de remailler au cours de leur propagation. Des adaptations de cette méthode sont faites afin de ne pas avoir besoin d'un maillage conforme à la géométrie. La géométrie est définie implicitement par une fonction surfaces de niveau, ce qui permet une bonne approximation de la géométrie et des conditions aux limites sans pour autant s'appuyer sur un maillage conforme. La géométrie est représentée par une fonction surfaces de niveau que nous appelons la densité. La densité est supérieure à 0.5 à l'intérieur du domaine de calcul et inférieure à 0.5 à l'extérieur. Cette fonction densité, définie par ses valeurs aux points noeuds du maillage, est interpolée à l'intérieur de chaque élément. Une méthode d'intégration adaptée à cette représentation géométrique est proposée. En effet, certains éléments sont coupés par la fonction surfaces de niveau et l'intégration de la matrice de raideur ne doit se faire que sur la partie pleine de l'élément. La méthode de quadrature de Gauss qui permet d'intégrer des polynômes de manière exacte n'est plus adaptée. Nous proposons d'utiliser une méthode de quadrature avec des points d'intégration répartis sur une grille régulière et dense. L'intégration peut s'avérer coûteuse en temps de calcul, c'est pour cette raison que nous proposons une technique d'apprentissage donnant la matrice élémentaire de rigidité en fonction des valeurs de la fonction surfaces de niveau aux sommets de l'élément considéré. Cette méthode d'apprentissage permet de grandes améliorations du temps de calcul des matrices élémentaires. Les résultats obtenus après analyse par la méthode des éléments finis standard ou par la méthode des éléments finis sur maillage cartésien ont une taille qui peut croître énormément selon la complexité des modèles, ainsi que la précision des schémas de résolution. Dans un contexte de programmation sur processeurs graphiques, où la mémoire est limitée, il est intéressant d'arriver à compresser ces données. Nous nous sommes intéressés à la compression des modèles et des résultats éléments finis par la transformée en ondelettes. La compression mise en place aidera aussi pour les problèmes de stockage en réduisant la taille des fichiers générés, et pour la visualisation des données. / Generating a conformal mesh on complex geometries leads to important model size of structural finite element simulations. The meshing time is directly linked to the geometry complexity and can contribute significantly to the total turnaround time. Graphics processing units (GPUs) are highly parallel programmable processors, delivering real performance gains on computationally complex, large problems. GPUs are used to implement a new finite element method on a Cartesian mesh. A Cartesian mesh is well adapted to the parallelism needed by GPUs and reduces the meshing time to almost zero. The novel method relies on the finite element method and the extended finite element formulation. The extended finite element method was introduced in the field of fracture mechanics. It consists in enriching the basis functions to take care of the geometry and the interface. This method doesn't need a conformal mesh to represent cracks and avoids refining during their propagation. Our method is based on the extended finite element method, with a geometry implicitly defined, wich allows for a good approximation of the geometry and boundary conditions without a conformal mesh.To represent the model on a Cartesian grid, we use a level set representing a density. This density is greater than 0.5 inside the domain and less than 0.5 outside. It takes 0.5 on the boundary. A new integration technique is proposed, adapted to the geometrical representation. For the element cut by the levet set, only the part full of material has to be integrated. The Gauss quadrature is no longer adapted. We introduce a quadrature method with integration points on a cartesian dense grid.In order to reduce the computational effort, a learning approach is then considered to form the elementary stiffness matrices as function of density values on the vertices of the elements. This learning method reduces the stiffness matrices time computation. Results obtained after analysis by finite element method or the novel finite element method can have important storage size, dependant of the model complexity and the resolution scheme exactitude. Due to the limited direct memory of graphics processing units, the data results are compressed. We compress the model and the element finite results with a wavelet transform. The compression will help for storage issue and also for data visualization.

The central theme of this thesis is the design of optimal balanced black-box stopping criteria in iterative solvers of symmetric positive-definite, symmetric indefinite, and nonsymmetric linear systems arising from finite element approximation of stochastic (parametric) partial differential equations. For a given stochastic and spatial approximation, it is known that iteratively solving the corresponding linear(ized) system(s) of equations to too tight algebraic error tolerance results in a wastage of computational resources without decreasing the usually unknown approximation error. In order to stop optimally-by avoiding unnecessary computations and premature stopping-algebraic error and a posteriori approximation error estimate must be balanced at the optimal stopping iteration. Efficient and reliable a posteriori error estimators do exist for close estimation of the approximation error in a finite element setting. But the algebraic error is generally unknown since the exact algebraic solution is not usually available. Obtaining tractable upper and lower bounds on the algebraic error in terms of a readily computable and monotonically decreasing quantity (if any) of the chosen iterative solver is the distinctive feature of the designed optimal balanced stopping strategy. Moreover, this work states the exact constants, that is, there are no user-defined parameters in the optimal balanced stopping tests. Hence, an iterative solver incorporating the optimal balanced stopping methodology that is presented here will be a black-box iterative solver. Typically, employing such a stopping methodology would lead to huge computational savings and in any case would definitely rule out premature stopping. The constants in the devised optimal balanced black-box stopping tests in MINRES solver for solving symmetric positive-definite and symmetric indefinite linear systems can be estimated cheaply on-the- fly. The contribution of this thesis goes one step further for the nonsymmetric case in the sense that it not only provides an optimal balanced black-box stopping test in a memory-expensive Krylov solver like GMRES but it also presents an optimal balanced black-box stopping test in memory-inexpensive Krylov solvers such as BICGSTAB(L), TFQMR etc. Currently, little convergence theory exists for the memory-inexpensive Krylov solvers and hence devising stopping criteria for them is an active field of research. Also, an optimal balanced black-box stopping criterion is proposed for nonlinear (Picard or Newton) iterative method that is used for solving the finite dimensional Navier-Stokes equations. The optimal balanced black-box stopping methodology presented in this thesis can be generalized for any iterative solver of a linear(ized) system arising from numerical approximation of a partial differential equation. The only prerequisites for this purpose are the existence of a cheap and tight a posteriori error estimator for the approximation error along with cheap and tractable bounds on the algebraic error.

Over the last 25 years, interior-point methods (IPMs) have emerged as a viable class of algorithms for solving various forms of conic optimization problems. Most IPMs use a modified Newton method to determine the search direction at each iteration. The system of equations corresponding to the modified Newton system can often be reduced to the so-called normal equation, a system of equations whose matrix ADA' is positive definite, yet often ill-conditioned. In this thesis, we first investigate the theoretical properties of the maximum weight basis (MWB) preconditioner, and show that when applied to a matrix of the form ADA', where D is positive definite and diagonal, the MWB preconditioner yields a preconditioned matrix whose condition number is uniformly bounded by a constant depending only on A. Next, we incorporate the results regarding the MWB preconditioner into infeasible, long-step, primal-dual, path-following algorithms for linear programming (LP) and convex quadratic programming (CQP). In both LP and CQP, we show that the number of iterative solver iterations of the algorithms can be uniformly bounded by n and a condition number of A, while the algorithmic iterations of the IPMs can be polynomially bounded by n and the logarithm of the desired accuracy. We also expand the scope of the LP and CQP algorithms to incorporate a family of preconditioners, of which MWB is a member, to determine an approximate solution to the normal equation. For the remainder of the thesis, we develop a new preconditioning strategy for solving systems of equations whose associated matrix is positive definite but ill-conditioned. Our so-called adaptive preconditioning strategy allows one to change the preconditioner during the course of the conjugate gradient (CG) algorithm by post-multiplying the current preconditioner by a simple matrix, consisting of the identity matrix plus a rank-one update. Our resulting algorithm, the Adaptive Preconditioned CG (APCG) algorithm, is shown to have polynomial convergence properties. Numerical tests are conducted to compare a variant of the APCG algorithm with the CG algorithm on various matrices.

Title: Numerical Methods in Discrete Inverse Problems Author: Marie Kubínová Department: Department of Numerical Mathematics Supervisor: RNDr. Iveta Hnětynková, Ph.D., Department of Numerical Mathe- matics Abstract: Inverse problems represent a broad class of problems of reconstruct- ing unknown quantities from measured data. A common characteristic of these problems is high sensitivity of the solution to perturbations in the data. The aim of numerical methods is to approximate the solution in a computationally efficient way while suppressing the influence of inaccuracies in the data, referred to as noise, that are always present. Properties of noise and its behavior in reg- ularization methods play crucial role in the design and analysis of the methods. The thesis focuses on several aspects of solution of discrete inverse problems, in particular: on propagation of noise in iterative methods and its representation in the corresponding residuals, including the study of influence of finite-precision computation, on estimating the noise level, and on solving problems with data polluted with noise coming from various sources. Keywords: discrete inverse problems, iterative solvers, noise estimation, mixed noise, finite-precision arithmetic - iii -

No description available.

Une méthode numérique précise et efficace est proposée pour la simulation de corps déformables interagissant avec un écoulement incompressible. Les équations de Navier-Stokes, considérées dans leur formulation vorticité fonction de courant, sont discrétisées temporellement et spatialement à l'aide respectivement d'un schéma d'ordre 4 de Runge-Kutta et par des différences finies compactes. Grâce à l'utilisation d'un maillage uniforme, nous proposons un nouveau solveur direct au quatrième ordre pour l'équation de Poisson, permettant de garantir l'incompressibilité au zéro machine sur une grille optimale. L'introduction d'un corps déformable dans l'écoulement de fluide est réalisée au moyen d'une méthode de pénalisation de volume. La déformation du corps est imposée par l'utilisation d'un maillage lagrangien structuré mobile qui interagit avec le fluide environnant en raison des forces hydrodynamiques et du moment (calculés sur le maillage eulérien de référence). Une loi de contrôle efficace de la courbure d'un poisson anguilliforme nageant vers une cible prescrite est proposée. La méthode numérique développée prouve son efficacité et précision tant dans le cas de la nage du poisson mais aussi plus d'un grand nombre de problèmes d'interactions fluide-structure. / We present an efficient algorithm for simulation of deformable bodies interacting with two-dimensional incompressible flows. The temporal and spatial discretizations of the Navier--Stokes equations in vorticity stream-function formulation are based on classical fourth-order Runge--Kutta and compact finite differences, respectively. Using a uniform Cartesian grid we benefit from the advantage of a new fourth-order direct solver for the Poisson equation to ensure the incompressibility constraint down to machine zero over an optimal grid. For introducing a deformable body in fluid flow, the volume penalization method is used. A Lagrangian structured grid with prescribed motion covers the deformable body which is interacting with the surrounding fluid due to the hydrodynamic forces and the torque calculated on the Eulerian reference grid. An efficient law for controlling the curvature of an anguilliform fish, swimming toward a prescribed goal, is proposed which is based on the geometrically exact theory of nonlinear beams and quaternions. Validation of the developed method shows the efficiency and expected accuracy of the algorithm for fish-like swimming and also for a variety of fluid/solid interaction problems.

[pt] A otimização topológica tem se desenvolvido bastante e possui potencial para revolucionar diversas áreas da engenharia. Este método pode ser implementado a partir de diferentes abordagens, tendo como base o Método dos Elementos Finitos. Ao se utilizar uma abordagem baseada no elemento, potencialmente, cada elemento finito pode se tornar um vazio ou um sólido, e a cada elemento do domínio é atribuído uma variável de projeto, constante, denominada densidade. Do ponto de vista Euleriano, a topologia obtida é um subconjunto dos elementos iniciais. No entanto, tal abordagem está sujeita a instabilidades numéricas, tais como conexões de um nó e rápidas oscilações de materiais do tipo sólido-vazio (conhecidas como instabilidade de tabuleiro). Projetos indesejáveis podem ser obtidos quando elementos de baixa ordem são utilizados e métodos de regularização e/ou restrição não são aplicados. Malhas poliédricas não estruturadas naturalmente resolvem esses problemas e oferecem maior flexibilidade na discretização de domínios não Cartesianos. Neste trabalho investigamos a otimização topológica em malhas poliédricas por meio de um acoplamento entre malhas. Primeiramente, as malhas poliédricas são geradas com base no conceito de diagramas centroidais de Voronoi e posteriormente otimizadas para uso em análises de elementos finitos. Demonstramos que o número de condicionamento do sistema de equações associado pode ser melhorado ao se minimizar uma função de energia relacionada com a geometria dos elementos. Dada a qualidade da malha e o tamanho do problema, diferentes tipos de resolvedores de sistemas de equações lineares apresentam diferentes desempenhos e, portanto, ambos os resolvedores diretos e iterativos são abordados. Em seguida, os poliedros são decompostos em tetraedros por um algoritmo específico de acoplamento entre as malhas. A discretização em poliedros é responsável pelas variáveis de projeto enquanto a malha tetraédrica, obtida pela subdiscretização da poliédrica, é utilizada nas análises via método dos elementos finitos. A estrutura modular, que separa as rotinas e as variáveis usadas nas análises de deslocamentos das usadas no processo de otimização, tem se mostrado promissora tanto na melhoria da eficiência computacional como na qualidade das soluções que foram obtidas neste trabalho. Os campos de deslocamentos e as variáveis de projeto são relacionados por meio de um mapeamento. A arquitetura computacional proposta oferece uma abordagem genérica para a solução de problemas tridimensionais de otimização topológica usando poliedros, com potencial para ser explorada em outras aplicações que vão além do escopo deste trabalho. Finalmente, são apresentados diversos exemplos que demonstram os recursos e o potencial da abordagem proposta. / [en] Topology optimization has had an impact in various fields and has the potential to revolutionize several areas of engineering. This method can be implemented based on the finite element method, and there are several approaches of choice. When using an element-based approach, every finite element is a potential void or actual material, whereas every element in the domain is assigned to a constant design variable, namely, density. In an Eulerian setting, the obtained topology consists of a subset of initial elements. This approach, however, is subject to numerical instabilities such as one-node connections and rapid oscillations of solid and void material (the so-called checkerboard pattern). Undesirable designs might be obtained when standard low-order elements are used and no further regularization and/or restrictions methods are employed. Unstructured polyhedral meshes naturally address these issues and offer fl exibility in discretizing non-Cartesians domains. In this work we investigate topology optimization on polyhedra meshes through a mesh staggering approach. First, polyhedra meshes are generated based on the concept of centroidal Voronoi diagrams and further optimized for finite element computations. We show that the condition number of the associated system of equations can be improved by minimizing an energy function related to the element s geometry. Given the mesh quality and problem size, different types of solvers provide different performances and thus both direct and iterative solvers are addressed. Second, polyhedrons are decomposed into tetrahedrons by a tailored embedding algorithm. The polyhedra discretization carries the design variable and a tetrahedra subdiscretization is nested within the polyhedra for finite element analysis. The modular framework decouples analysis and optimization routines and variables, which is promising for software enhancement and for achieving high fidelity solutions. Fields such as displacement and design variables are linked through a mapping. The proposed mapping-based framework provides a general approach to solve three-dimensional topology optimization problems using polyhedrons, which has the potential to be explored in applications beyond the scope of the present work. Finally, the capabilities of the framework are evaluated through several examples, which demonstrate the features and potential of the proposed approach.