No abstract

When a system of linear equations is ill-conditioned, regularization techniques provide a quite useful tool for trying to overcome the numerical inherent difficulties: the ill-conditioned system is replaced by another one whose solution depends on a regularization term formed by a scalar and a matrix which are to be chosen. In this paper, we consider the case of several regularizations terms added simultaneously, thus overcoming the problem of the best choice of the regularization matrix. The error of this procedure is analyzed and numerical results prove its efficiency

No abstract

This paper presents a general framework for Shanks transformations of sequences of elements in a vector space. It is shown that the Minimal Polynomial Extrapolation (MPE), the Modified Minimal Polynomial Extrapolation (MMPE), the Reduced Rank Extrapolation (RRE), the Vector Epsilon Algorithm (VEA), the Topological Epsilon Algorithm (TEA), and Anderson Acceleration (AA), which are standard general techniques designed for accelerating arbitrary sequences and/or solving nonlinear equations, all fall into this framework. Their properties and their connections with quasi-Newton and Broyden methods are studied. The paper then exploits this framework to compare these methods. In the linear case, it is known that AA and GMRES are 'essentially' equivalent in a certain sense while GMRES and RRE are mathematically equivalent. This paper discusses the connection between AA, the RRE, the MPE, and other methods in the nonlinear case.

In this paper, the regularized solutions of an ill–conditioned system of linear equations are computed for several values of the regularization parameter lambda. Then, these solutions are extrapolated at lambda = 0 by various vector rational extrapolations techniques built for that purpose. These techniques are justified by an analysis of the regularized solutions based on the singular value decomposition and the generalized singular value decomposition. Numerical results illustrate the effectiveness of the procedures

No abstract

Lanczos type algorithms form a wide and interesting class of iterative methods for solving systems of linear equations. One of their main interest is that they provide the exact answer in at most n steps where n is the dimension of the system. However a breakdown can occur in these algorithms due to a division by a zero scalar product. After recalling the so-called method of recursive zoom (MRZ) which allows to jump over such breakdown we propose two new variants. Then the method and its variants are extended to treat the case of a near-breakdown due to a division by a scalar product whose absolute value is small which is the reason for an important propagation of rounding errors in the method. Programming the various algorithms is then analyzed and explained. Numerical results illustrating the processes are discussed. The subroutines corresponding to the algorithms described can be obtained via netlib

An important problem in Web search is determining the importance of each page. After introducing the main characteristics of this problem, we will see that, from the mathematical point of view, it could be solved by computing the left principal eigenvector (the PageRank vector) of a matrix related to the structure of the Web by using the power method. We will give expressions of the PageRank vector and study the mathematical properties of the power method. Various Padé-style approximations of the PageRank vector will be given. Since the convergence of the power method is slow, it has to be accelerated. This problem will be discussed. Recently, several acceleration methods were proposed. We will give a theoretical justification for these methods. In particular, we will generalize the recently proposed Quadratic Extrapolation, and we interpret it on the basis of the method of moments of Vorobyev, and as a Krylov subspace method. Acceleration results are given for the various -algorithms, and for the E-algorithm. Other algorithms for this problem are also discussed.