Using Random Butterfly Transformations to Avoid Pivoting in Sparse Direct Methods
نویسندگان
چکیده
We consider the solution of sparse linear systems using direct methods via LU factorization. Unless the matrix is positive definite, numerical pivoting is usually needed to ensure stability, which is costly to implement especially in the sparse case. The Random Butterfly Transformations (RBT) technique provides an alternative to pivoting and is easily parallelizable. The RBT transforms the original matrix into another one that can be factorized without pivoting with probability one. This approach has been successful for dense matrices; in this work, we investigate the sparse case. In particular, we address the issue of fill-in in the transformed system. Key-words: Sparse direct methods, LU factorization, randomized algorithms. ∗ Inria and Université Paris-Sud, Orsay, France ([email protected]). † Lawrence Berkeley National Laboratory, Berkeley, CA, USA ([email protected]). ‡ Lawrence Berkeley National Laboratory, Berkeley, CA, USA ([email protected]). Utiliser les Random Butterfly Transformations pour éviter de pivoter dans les méthodes directes creuses Résumé : On considère la solution de systèmes linéaires creux en utilisant des méthodes directes de factorisation LU. Sauf si la matrice est définie positive, le pivotage numérique est en général nécessaire pour assurer la stabilité, ce qui est coûteux à implémenter en particulier pour le cas creux. La technique de Random Butterfly Transformations (RBT) fournit une alternative au pivotage et est facilement parallélisable. Le RBT transforme la matrice originale en une matrice qui peut être factorisée sans pivoter avec une probabilité de 1. Cette approche s’est avérée efficace pour les matrices denses. Dans ce travail nous étudions le cas creux. En particulier nous abordons le problème lié au fill-in dans le système transformé. Mots-clés : Méthodes directes creuses, factorisation LU, algorithmes randomisées. Using Random Butterfly Transformations to Avoid Pivoting in Sparse Direct Methods 3
منابع مشابه
A Randomized LU-based Solver Using GPU and Intel Xeon Phi Accelerators
We present a fast hybrid solver for dense linear systems based on LU factorization. To achieve good performance, we avoid pivoting by using random butterfly transformations for which we developed efficient implementations on heterogeneous architectures. We used both Graphics Processing Units and Intel Xeon Phi as accelerators. The performance results show that the pre-processing due to randomiz...
متن کاملThe randomizing FFT : an alternative to pivoting in
Recently, randomizing input transformations have been developed for streamlining Gaussian elimination. Multiplying by special sparse random matrices, these transformations produce an equivalent linear system which, with probability 1, Gaussian elimination can solve without any augmentation by pivoting. In this paper we study properties of the unitary Randomizing Discrete Fourier Transform (RDFT...
متن کاملCompressed threshold pivoting for sparse symmetric indefinite systems
A key technique for controlling numerical stability in sparse direct solvers is threshold partial pivoting. When selecting a pivot, the entire candidate pivot column below the diagonal must be up-to-date and must be scanned. If the factorization is parallelized across a large number of cores, communication latencies can be the dominant computational cost. In this paper, we propose two alternati...
متن کاملA Scalable Sparse Direct Solver Using Static Pivoting
We propose several techniques as alternatives to partial pivoting to stabilize sparse Gaussian elimination. From numerical experiments we demonstrate that for a wide range of problems the new method is as stable as partial pivoting. The main advantage of the new method over partial pivoting is that it permits a priori determination of data structures and communication pattern, which makes it mo...
متن کاملSolution of the coupled Navier - Stokes equations
A selection of these reports is available in PostScript form at the Faculty's anonymous ftp-Abstract In this paper the incompressible Navier-Stokes equations are discretized by the nite element method. After linearization large, sparse systems of linear equations have to be solved. A well known problem is the occurrence of zero elements on the main diagonal. We describe ordering techniques of t...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 2014