1 \documentclass{article}
2 \usepackage[utf8]{inputenc}
3 \usepackage{amsfonts,amssymb}
7 \title{A scalable multisplitting algorithm for solving large sparse linear systems}
12 \author{Raphaël Couturier \and Lilia Ziane Khodja}
18 In this paper we revist the krylov multisplitting algorithm presented in
19 \cite{huang1993krylov} which uses a scalar method to minimize the krylov
20 iterations computed by a multisplitting algorithm. Our new algorithm is based on
21 a parallel multisplitting algorithm with few blocks of large size using a
22 parallel GMRES method inside each block and on a parallel krylov minimization in
23 order to improve the convergence. Some large scale experiments with a 3D Poisson
24 problem are presented. They show the obtained improvements compared to a
25 classical GMRES both in terms of number of iterations and execution times.
28 \section{Introduction}
30 Iterative methods are used to solve large sparse linear systems of equations of
31 the form $Ax=b$ because they are easier to parallelize than direct ones. Many
32 iterative methods have been proposed and adapted by many researchers. When
33 solving large linear systems with many cores, iterative methods often suffer
34 from scalability problems. This is due to their need for collective
35 communications to perform matrix-vector products and reduction operations.
36 Preconditionners can be used in order to increase the convergence of iterative
37 solvers. However, most of the good preconditionners are not sclalable when
38 thousands of cores are used.
42 On ne peut pas parler de tout...
44 \section{Related works}
47 A general framework for studying parallel multisplitting has been presented in
48 \cite{o1985multi} by O'Leary and White. Convergence conditions are given for the
49 most general case. Many authors improved multisplitting algorithms by proposing
50 for example a asynchronous version \cite{bru1995parallel} and convergence
51 condition \cite{bai1999block,bahi2000asynchronous} in this case or other
52 two-stage algorithms~\cite{frommer1992h,bru1995parallel}
54 In \cite{huang1993krylov}, the authors proposed a parallel multisplitting
55 algorithm in which all the tasks except one are devoted to solve a sub-block of
56 the splitting and to send their local solution to the first task which is in
57 charge to combine the vectors at each iteration. These vectors form a Krylov
58 basis for which the first tasks minimize the error function over the basis to
59 increase the convergence, then the other tasks receive the update solution until
60 convergence of the global system.
64 In \cite{couturier2008gremlins}, the authors proposed practical implementations
65 of multisplitting algorithms that take benefit from multisplitting algorithms to
66 solve large scale linear systems. Inner solvers could be based on scalar direct
67 method with the LU method or scalar iterative one with GMRES.
69 \bibliographystyle{plain}