[pt] Este trabalho desenvolve um solucionador iterativo baseado no método Resíduo Mínimo Generalizado (GMRES) para solucionar o subproblema linear do problema de fluxo de carga, com destaque para cenários de difícil convergência. O solucionador agrega uma estratégia de reordenamento para minimização do número total de novos elementos não-nulos e um pré-condicionador baseado no algoritmo de Doolittle, com regra de preenchimento de elementos não-nulos baseada no erro resultante. O solucionador foi implementado em um programa computacional de fluxo de carga, a fim de se verificar sua robustez e eficiência em diversos sistemas-teste e diferentes condições de operação. Também é proposto um método para o ajuste dos parâmetros dos solucionadores iterativos, que permite identificar intervalos de valores permissíveis para cada um dos parâmetros, identificando os mais adequados, visando garantir a robustez e melhorar o desempenho do solucionador. / [en] This work develops an iterative solver based on Generalized Minimal Residual method (GMRES) to solve the load flow linear subproblem, especially in scenarios of difficult convergence. The solver combines a reordering strategy to minimise the total number of fill-in terms and a preconditioning strategy based on the Doolittle algorithm with a fill-in dropping strategy based on the resulting error. The solver was implemented into a computational load flow program in order to verify its robustness and efficiency in several test-systems and different operating conditions. It is also proposed a method for adjusting the iterative solver parameters, the method is able to identify intervals of permissible values for each parameter, identifying the most appropriate in order to ensure the robustness and improve the solver performance.

The polynomial eigenvalue problem appears in many areas of scientific and technical computing. It can be seen as a generalization of the linear eigenvalue problem in which the equation P(lambda)x = 0, that defines the problem, involves a polynomial P(lambda), of degree d, in the parameter lambda (the eigenvalue), and d+1 matrix coefficients. In its turn, the polynomial eigenvalue problem is a particular case of the nonlinear eigenvalue problem, T(lambda)x = 0, in which T is a nonlinear matrix function. These problems appear in diverse areas of application such as acoustics, fluid mechanics, structure analysis, or photonics. This thesis focuses on the study of methods for the numerical resolution of the polynomial eigenvalue problem, as well as the adaptation of such methods to the most general nonlinear case. Mainly, we consider methods of projection, that are appropriate for the case of sparse matrices of large dimension, where only a small percentage of eigevalues and eigenvectors are required. The algorithms are studied from the point of view of high-performance computing, considering issues like (computational and memory) efficiency and parallel computation. SLEPc, Scalable Library for Eigenvalue Problem Computations, is a software library for the parallel solution of large-scale eigenvalue problems. It is of general purpose and can be used for standard and generalized problems, both symmetric and nonsymmetric, with real or complex arithmetic. As a result of this thesis, we have developed several solvers for polynomial an nonlinear eigenproblems, which have included in the last versions of this software. On one hand, we consider methods based on the linearization of the polynomial problem, that solves an equivalent linear eigenproblem of dimension several times the initial size. Among them, the TOAR method stands out, that repre- sents the search subspace basis in an efficient way in terms of memory, and is appropriate to handle the increase of dimension of the linear problem. The thesis also proposes specific variants for the particular case of symmetric matrices. In all these methods we consider several aspects to provide the implementations with robustness and flexibility, such as spectral transformations, scaling, and techniques of extraction. In addition to the methods of linearization, we propose methods of Newton type, such as the method of Jacobi-Davidson with deflation and the method of Newton for invariant pairs. Due to its characteristics, the latter is not usually employed as a proper method, but as a technique for refinement of the solutions obtained with another method. The previous methods can also be applied to the resolution of the nonlinear problem, using techniques like polynomial or rational interpolation, being necessary in some cases to adapt the algorithms. This thesis covers also these cases. For all the considered algorithms we have made parallel implementations in SLEPc, and have studied its numerical behaviour and its parallel performance in problems coming from real applications. / El problema polinómico de valores propios aparece en muchas áreas de la computación científica y técnica. Puede verse como una generalización del problema lineal de valores propios en el que la ecuación P(lambda)x=0, que define el problema, involucra un polinomio P(lambda), de grado d, en el parámetro lambda del autovalor, y d+1 coeficientes matriciales. A su vez, el problema polinómico de valores propios es un caso particular del problema no lineal de valores propios, T(lambda)x=0, en el que T es una función matricial no lineal. Estos problemas aparecen en diversas áreas de aplicación como acústica, mecánica de fluidos, análisis de estructuras, o fotónica. Esta tesis se centra en el estudio de métodos para la resolución numérica del problema polinómico de valores propios, así como la adaptación de dichos métodos al caso más general no lineal. Principalmente, se consideran métodos de proyección, que son apropiados para el caso de matrices dispersas de gran dimensión cuando se requiere solo un pequeño porcentaje de los valores y vectores propios. Los algoritmos se estudian desde el punto de vista de la computación de altas prestaciones, teniendo en consideración aspectos como la eficiencia (computacional y de memoria) y la computación paralela. SLEPc, Scalable Library for Eigenvalue Problem Computations, es una biblioteca software para la resolución de problemas de valores propios de gran dimensión en paralelo. Es de propósito general y puede usarse para problemas estándares y generalizados, simétricos y no simétricos, con aritmética real o compleja. Como fruto de esta tesis, se han desarrollado diversos solvers para problemas polinómicos y no lineales, los cuales se han incluido en las últimas versiones de este software. Por un lado, se abordan métodos basados en la linealización del problema polinómico, que resuelven un problema lineal equivalente de dimensión varias veces la del inicial. Entre ellos se destaca el método TOAR, que representa la base del subespacio de búsqueda de una forma eficiente en términos de memoria, y es adecuado para manejar el aumento de dimensión del problema lineal. La tesis también propone variantes específicas para el caso particular de matrices simétricas. En todos estos métodos se consideran diversos aspectos para dotar a las implementaciones de robustez y flexibilidad, tales como transformaciones espectrales, escalado, y técnicas de extracción. Además de los métodos de linealización, se proponen métodos de tipo Newton, como el método de Jacobi-Davidson con deflación y el método de Newton para pares invariantes. Por sus características, este último no suele utilizarse como un método en sí mismo sino como técnica de refinamiento de las soluciones obtenidas con otro método. Los métodos anteriores pueden aplicarse a la resolución del problema no lineal, utilizando técnicas como la interpolación polinómica o racional, siendo necesario en algunos casos adaptar los algoritmos. La tesis cubre también estos casos. Para todos los algoritmos considerados se han realizado implementaciones paralelas en SLEPc, y se ha estudiado su comportamiento numérico y sus prestaciones paralelas en problemas procedentes de aplicaciones reales. / El problema polinòmic de valors propis apareix en moltes àrees de la computació científica i tècnica. Pot veure's com una generalització del problema lineal de valors propis en el qual l'equació P(lambda)x=0, que defineix el problema, involucra un polinomi P(lambda), de grau d, en el paràmetre lambda de l'autovalor, i d+1 coeficients matricials. Al seu torn, el problema polinòmic de valors propis és un cas particular del problema no lineal de valors propis, T(lambda)x=0, en el qual T és una funció matricial no lineal. Aquests problemes apareixen en diverses àrees d'aplicació com a acústica, mecànica de fluids, anàlisis d'estructures, o fotònica. Aquesta tesi se centra en l'estudi de mètodes per a la resolució numèrica del problema polinòmic de valors propis, així com l'adaptació d'aquests mètodes al cas més general no lineal. Principalment, es consideren mètodes de projecció, que són apropiats per al cas de matrius disperses de gran dimensió quan es requereix solament un reduït percentatge dels valors i vectors propis. Els algorismes s'estudien des del punt de vista de la computació d'altes prestacions, tenint en consideració aspectes com l'eficiència (computacional i de memòria) i la computació paral·lela. SLEPc, Scalable Library for Eigenvalue Problem Computations, és una biblioteca software per a la resolució de problemes de valors propis de gran dimensió en paral·lel. És de propòsit general i pot usar-se per a problemes estàndards i generalitzats, simètrics i no simètrics, amb aritmètica real o complexa. Com a fruit d'aquesta tesi, s'han desenvolupat diversos solvers per a problemes polinòmics i no lineals, els quals s'han inclòs en les últimes versions d'aquest software. D'una banda, s'aborden mètodes basats en la linealització del problema polinòmic, que resolen un problema lineal equivalent de dimensió diverses vegades la de l'inicial. Entre ells es destaca el mètode TOAR, que representa la base del subespai de cerca d'una forma eficient en termes de memòria, i és adequat per a manejar l'augment de dimensió del problema lineal. La tesi també proposa variants específiques per al cas particular de matrius simètriques. En tots aquests mètodes es consideren diversos aspectes per a dotar a les implementacions de robustesa i flexibilitat, tals com a transformacions espectrals, escalat, i tècniques d'extracció. A més dels mètodes de linealització, es proposen mètodes de tipus Newton, com el mètode de Jacobi-Davidson amb deflació i el mètode de Newton per a parells invariants. Per les seues característiques, aquest últim no sol utilitzar-se com un mètode en si mateix sinó com a tècnica de refinament de les solucions obtingudes amb un altre mètode. Els mètodes anteriors poden aplicar-se a la resolució del problema no lineal, utilitzant tècniques com la interpolació polinòmica o racional, sent necessari en alguns casos adaptar els algorismes. La tesi cobreix també aquests casos. Per a tots els algorismes considerats s'han realitzat implementacions paral·leles en SLEPc, i s'ha estudiat el seu comportament numèric i les seues prestacions paral·leles en problemes procedents d'aplicacions reals. / Campos González, MC. (2017). Implementación paralela de métodos iterativos para la resolución de problemas polinómicos de valores propios [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/86134 / TESIS

No description available.

This thesis explores the connection between Crouzeix's conjecture and the convergence of the GMRES algorithm. GMRES is a popular iterative method for solving linear systems and is one of the many Krylov methods. Despite its popularity, the convergence of GMRES is not completely understood. While the spectrum can in some cases be a good indicator of convergence, it has been shown that in general, the spectrum does not provide sufficient information to fully explain the behavior of GMRES iterations. Other sets associated with a matrix that can also help predict convergence are the pseudospectrum and the numerical range. This work focuses on convergence bounds obtained by considering the latter. In particular, it focuses on the application of Crouzeix's conjecture, which relates the norm of a matrix polynomial to the size of that polynomial over the numerical range, to describing GMRES convergence.

No description available.

No description available.

Krylov subspace iterative methods provide an effective tool for reducing the solution of large linear systems to a size for which a direct solver may be applied. However, the problems of limited storage and speed are still a concern. Therefore, in this dissertation work, we present iterative Krylov subspace algorithms for non-Hermitian systems which do have fixed memory requirements and have favorable convergence characteristics. This dissertation describes three projects. The first project concerns short-term recurrence Krylov subspace methods for nearly-Hermitian linear systems. In 2008, Beckermann and Reichel introduced a short-term recurrence progressive GMRES algorithm for nearly-Hermitian linear systems. However, we have found this method to be unstable. We document the instabilities and introduce a different fixed-memory algorithm to treat nearly-Hermitian problems. We present numerical experiments demonstrating that the performance of this algorithm is competitive. The other two projects involve extending a strategy called Krylov subspace recycling, introduced by Parks and colleagues in 2005. This method requires more overhead than other subspace augmentation methods but offers the ability to recycle subspace information between cycles for a single linear system and recycle information between related linear systems. In the first project, we extend subspace recycling to the block Krylov subspace setting. A block Krylov subspace is a generalization of Krylov subspace where a single starting vector is replaced with a block of linearly independent starting vectors. We then apply our method to a sequence of matrices arising in a Newton iteration applied to fluid density functional theory and present some numerical experiments. In the second project, we extend the methods of subspace recycling to a family of linear systems differing only by multiples of the identity. These problems arise in the theory of quantum chromodynamics, a theory of the behavior of subatomic particles. We wish to build on the class of Krylov methods which allow the simultaneous solution of all shifted linear systems while generating only one subspace. However, the mechanics of subspace recycling complicates this situation and interferes with our ability to simultaneously solve all systems using these techniques. Therefore, we introduce an algorithm which avoids this complication and present some numerical experiments demonstrating its effectiveness. / Mathematics

We consider low-rank solution methods for certain classes of large-scale linear matrix equations. Our aim is to adapt existing low-rank solution methods based on standard, extended and rational Krylov subspaces to solve equations which may viewed as extensions of the classical Lyapunov and Sylvester equations. The first class of matrix equations that we consider are constrained Sylvester equations, which essentially consist of Sylvester's equation along with a constraint on the solution matrix. These therefore constitute a system of matrix equations. The second are generalized Lyapunov equations, which are Lyapunov equations with additional terms. Such equations arise as computational bottlenecks in model order reduction. / Mathematics

No / The four vector extrapolation methods, minimal polynomial extrapolation, reduced rank extrapolation, modified minimal polynomial extrapolation and the topological epsilon algorithm, when applied to linearly generated vector sequences are Krylov subspace methods and it is known that they are equivalent to some well-known conjugate gradient type methods. However, the vector -algorithm is an extrapolation method, older than the four extrapolation methods above, and no similar results are known for it. In this paper, a determinantal formula for the vector -algorithm is given. Then it is shown that, when applied to a linearly generated vector sequence, the algorithm is also a Krylov subspace method and for a class of matrices the method is equivalent to a preconditioned Lanczos method. A new determinantal formula for the CGS is given, and an algebraic comparison between the vector -algorithm for linear systems and CGS is also given.

We provide an algorithm and analysis of a high order projection scheme for time integration of the incompressible Navier-Stokes equations (NSE). The method is based on a projection onto the subspace of divergence-free (incompressible) functions interleaved with a Krylov-based exponential time integration (KBEI). These time integration methods provide a high order accurate, stable approach with many of the advantages of explicit methods, and can reduce the computational resources over conventional methods. The method is scalable in the sense that the computational costs grow linearly with problem size. Exponential integrators, used typically to solve systems of ODEs, utilize matrix vector products of the exponential of the Jacobian on a vector. For large systems, this product can be approximated efficiently by Krylov subspace methods. However, in contrast to explicit methods, KBEIs are not restricted by the time step. While implicit methods require a solution of a linear system with the Jacobian, KBEIs only require matrix vector products of the Jacobian. Furthermore, these methods are based on linearization, so there is no non-linear system solve at each time step. Differential-algebraic equations (DAEs) are ordinary differential equations (ODEs) subject to algebraic constraints. The discretized NSE constitute a system of DAEs, where the incompressibility condition is the algebraic constraint. Exponential integrators can be extended to DAEs with linear constraints imposed via a projection onto the constraint manifold. This results in a projected ODE that is integrated by a KBEI. In this approach, the Krylov subspace satisfies the constraint, hence the solution at the advanced time step automatically satisfies the constraint as well. For the NSE, the projection onto the constraint is typically achieved by a projection induced by the L2 inner product. We examine this L2 projection and an H1 projection induced by the H1 semi-inner product. The H1 projection has an advantage over the L2 projection in that it retains tangential Dirichlet boundary conditions for the flow. Both the H1 and L2 projections are solutions to saddle point problems that are efficiently solved by a preconditioned Uzawa algorithm. / Ph. D.