Parallel Solution of Large Sparse Linear Systems by the Conjugate Gradient Algorithm PDF Download

Author: P.J. Elger
Publisher:
ISBN:
Category :
Languages : en
Pages : 84

View

Book Description

Author: Yousef Saad
Publisher: SIAM
ISBN: 0898715342
Category : Mathematics
Languages : en
Pages : 537

View

Book Description
Mathematics of Computing -- General.

Author: Gerard Meurant
Publisher: Elsevier
ISBN: 0080529518
Category : Mathematics
Languages : en
Pages : 777

View

Book Description
This book deals with numerical methods for solving large sparse linear systems of equations, particularly those arising from the discretization of partial differential equations. It covers both direct and iterative methods. Direct methods which are considered are variants of Gaussian elimination and fast solvers for separable partial differential equations in rectangular domains. The book reviews the classical iterative methods like Jacobi, Gauss-Seidel and alternating directions algorithms. A particular emphasis is put on the conjugate gradient as well as conjugate gradient -like methods for non symmetric problems. Most efficient preconditioners used to speed up convergence are studied. A chapter is devoted to the multigrid method and the book ends with domain decomposition algorithms that are well suited for solving linear systems on parallel computers.

Author: Daniele Bertaccini
Publisher: CRC Press
ISBN: 1498764177
Category : Mathematics
Languages : en
Pages : 375

View

Book Description
This book describes, in a basic way, the most useful and effective iterative solvers and appropriate preconditioning techniques for some of the most important classes of large and sparse linear systems. The solution of large and sparse linear systems is the most time-consuming part for most of the scientific computing simulations. Indeed, mathematical models become more and more accurate by including a greater volume of data, but this requires the solution of larger and harder algebraic systems. In recent years, research has focused on the efficient solution of large sparse and/or structured systems generated by the discretization of numerical models by using iterative solvers.

Author:
Publisher:
ISBN:
Category :
Languages : en
Pages : 11

View

Book Description
The efficiency of a parallel implementation of the conjugate gradient method preconditioned by an incomplete Cholesky factorization can vary dramatically depending on the column ordering chosen. One method to minimize the number of major parallel steps is to choose an ordering based on a coloring of the symmetric graph representing the nonzero adjacency structure of the matrix. In this paper, we compare the performance of the preconditioned conjugate gradient method using these coloring orderings with a number of standard orderings on matrices arising from applications in structural engineering. Because optimal colorings for these systems may not be a priori known: we employ several graph coloring heuristics to obtain consistent colorings. Based on lower bounds obtained from the local structure of these systems, we find that the colorings determined by these heuristics are nearly optimal. For these problems, we find that the increase in parallelism afforded by the coloring-based orderings more than offsets any increase in the number of iterations required for the convergence of the conjugate gradient algorithm.

Author: Institute of Mathematics and Its Applications
Publisher: Academic Press
ISBN:
Category : Mathematics
Languages : en
Pages : 304

View

Book Description

Author: David E. Keyes
Publisher: Springer Science & Business Media
ISBN: 9401154120
Category : Mathematics
Languages : en
Pages : 403

View

Book Description
In this volume, designed for computational scientists and engineers working on applications requiring the memories and processing rates of large-scale parallelism, leading algorithmicists survey their own field-defining contributions, together with enough historical and bibliographical perspective to permit working one's way to the frontiers. This book is distinguished from earlier surveys in parallel numerical algorithms by its extension of coverage beyond core linear algebraic methods into tools more directly associated with partial differential and integral equations - though still with an appealing generality - and by its focus on practical medium-granularity parallelism, approachable through traditional programming languages. Several of the authors used their invitation to participate as a chance to stand back and create a unified overview, which nonspecialists will appreciate.

Author: Richard Barrett
Publisher: SIAM
ISBN: 9781611971538
Category : Mathematics
Languages : en
Pages : 141

View

Book Description
In this book, which focuses on the use of iterative methods for solving large sparse systems of linear equations, templates are introduced to meet the needs of both the traditional user and the high-performance specialist. Templates, a description of a general algorithm rather than the executable object or source code more commonly found in a conventional software library, offer whatever degree of customization the user may desire. Templates offer three distinct advantages: they are general and reusable; they are not language specific; and they exploit the expertise of both the numerical analyst, who creates a template reflecting in-depth knowledge of a specific numerical technique, and the computational scientist, who then provides "value-added" capability to the general template description, customizing it for specific needs. For each template that is presented, the authors provide: a mathematical description of the flow of algorithm; discussion of convergence and stopping criteria to use in the iteration; suggestions for applying a method to special matrix types; advice for tuning the template; tips on parallel implementations; and hints as to when and why a method is useful.

Author: James M. Ortega
Publisher: Springer Science & Business Media
ISBN: 1489921125
Category : Computers
Languages : en
Pages : 309

View

Book Description
Although the origins of parallel computing go back to the last century, it was only in the 1970s that parallel and vector computers became available to the scientific community. The first of these machines-the 64 processor llliac IV and the vector computers built by Texas Instruments, Control Data Corporation, and then CRA Y Research Corporation-had a somewhat limited impact. They were few in number and available mostly to workers in a few government laboratories. By now, however, the trickle has become a flood. There are over 200 large-scale vector computers now installed, not only in government laboratories but also in universities and in an increasing diversity of industries. Moreover, the National Science Foundation's Super computing Centers have made large vector computers widely available to the academic community. In addition, smaller, very cost-effective vector computers are being manufactured by a number of companies. Parallelism in computers has also progressed rapidly. The largest super computers now consist of several vector processors working in parallel. Although the number of processors in such machines is still relatively small (up to 8), it is expected that an increasing number of processors will be added in the near future (to a total of 16 or 32). Moreover, there are a myriad of research projects to build machines with hundreds, thousands, or even more processors. Indeed, several companies are now selling parallel machines, some with as many as hundreds, or even tens of thousands, of processors.

Author: Stojce Nakov
Publisher:
ISBN:
Category :
Languages : en
Pages : 0

View

Book Description
In the context of this thesis, our focus is on numerical linear algebra, more precisely on solution of large sparse systems of linear equations. We focus on designing efficient parallel implementations of MaPHyS, an hybrid linear solver based on domain decomposition techniques. First we investigate the MPI+threads approach. In MaPHyS, the first level of parallelism arises from the independent treatment of the various subdomains. The second level is exploited thanks to the use of multi-threaded dense and sparse linear algebra kernels involved at the subdomain level. Such an hybrid implementation of an hybrid linear solver suitably matches the hierarchical structure of modern supercomputers and enables a trade-off between the numerical and parallel performances of the solver. We demonstrate the flexibility of our parallel implementation on a set of test examples. Secondly, we follow a more disruptive approach where the algorithms are described as sets of tasks with data inter-dependencies that leads to a directed acyclic graph (DAG) representation. The tasks are handled by a runtime system. We illustrate how a first task-based parallel implementation can be obtained by composing task-based parallel libraries within MPI processes throught a preliminary prototype implementation of our hybrid solver. We then show how a task-based approach fully abstracting the hardware architecture can successfully exploit a wide range of modern hardware architectures. We implemented a full task-based Conjugate Gradient algorithm and showed that the proposed approach leads to very high performance on multi-GPU, multicore and heterogeneous architectures.