東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

The development of a computationally...

Lemieux, Jean Francois.

Linked to FindBook

Google Book

Amazon

博客來

The development of a computationally efficient high-resolution viscous-plastic sea ice model.

Record Type:	Language materials, printed : Monograph/item
Title/Author:	The development of a computationally efficient high-resolution viscous-plastic sea ice model./
Author:	Lemieux, Jean Francois.
Description:	161 p.
Notes:	Source: Dissertation Abstracts International, Volume: 71-05, Section: B, page: 2920.
Contained By:	Dissertation Abstracts International71-05B.
Subject:	Applied Mathematics. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=NR61865
ISBN:	9780494618653

The development of a computationally efficient high-resolution viscous-plastic sea ice model.
Lemieux, Jean Francois.

The development of a computationally efficient high-resolution viscous-plastic sea ice model. - 161 p.

Source: Dissertation Abstracts International, Volume: 71-05, Section: B, page: 2920.

Thesis (Ph.D.)--McGill University (Canada), 2009.

This thesis presents the development of a high-resolution viscous-plastic (VP) sea ice model. Because of the fine mesh and the size of the domain, an efficient and parallelizable numerical scheme is desirable. In a first step, we have implemented the nonlinear solver used in existing VP models (referred to as the standard solver). It is based on a linear solver and an outer loop (OL) iteration. For the linear solver, we introduced the preconditioned Generalized Minimum RESidual (pGMRES) method. The preconditioner is a line successive overrelaxation solver (SOR). When compared to the SOR and the line SOR (LSOR) methods, two solvers commonly used in the sea ice modeling community, pGMRES increases the computational efficiency by a factor of 16 and 3 respectively. For pGMRES, the symmetry of the system matrix is not a prerequisite. The Coriolis term and the off-diagonal part of the water drag can then be treated implicitly. Theoretical and simulation results show that this implicit treatment eliminates a numerical instability present with an explicit treatment. During this research, we have also observed that the approximate nonlinear solution converges slowly with the number of OL iterations. Furthermore, simulation results reveal: the existence of multiple solutions and occasional convergence failures of the nonlinear solver. For a time step comparable to the forcing time scale, a few OL iterations lead to errors in the velocity field that are of the same order of magnitude as the mean drift. The slow convergence is an issue at all spatial resolutions but is more severe as the grid is refined. It is attributed in part to the standard VP formulation that leads to a momentum equation that is not continuously differentiable. To obtain a smooth formulation, we replaced the standard viscous coefficient expression with capping by a hyperbolic tangent function. This provides a unique solution and reduces the computational time and failure rate. To further improve the computational efficiency, the Jacobian-free Newton-Krylov (JFNK) method was implemented. Results show that JFNK is 3.5-6.6 times faster than the standard solver (depending on the resolution and termination criterion). As the standard solver, the JFNK method exhibits a relatively small number of convergence failures. A globalization "trick", which involves a switch in the linearization approach, eliminates the failures for the standard solver. Globalization approaches such as the line search did not improve the failure rate of the JFNK method. Future work will focus on the implementation of an effective globalization approach for JFNK.

ISBN: 9780494618653Subjects--Topical Terms:

1669109
Applied Mathematics.

The development of a computationally efficient high-resolution viscous-plastic sea ice model.
LDR:03481nam 2200265 4500 001 1391058
005 20101222085301.5
008 130515s2009 ||||||||||||||||| ||eng d
020 $a 9780494618653
035 $a (UMI)AAINR61865
035 $a AAINR61865
040 $a UMI $c UMI
100 1 $a Lemieux, Jean Francois. $3 1669433
245 1 4 $a The development of a computationally efficient high-resolution viscous-plastic sea ice model.
300 $a 161 p.
500 $a Source: Dissertation Abstracts International, Volume: 71-05, Section: B, page: 2920.
502 $a Thesis (Ph.D.)--McGill University (Canada), 2009.
520 $a This thesis presents the development of a high-resolution viscous-plastic (VP) sea ice model. Because of the fine mesh and the size of the domain, an efficient and parallelizable numerical scheme is desirable. In a first step, we have implemented the nonlinear solver used in existing VP models (referred to as the standard solver). It is based on a linear solver and an outer loop (OL) iteration. For the linear solver, we introduced the preconditioned Generalized Minimum RESidual (pGMRES) method. The preconditioner is a line successive overrelaxation solver (SOR). When compared to the SOR and the line SOR (LSOR) methods, two solvers commonly used in the sea ice modeling community, pGMRES increases the computational efficiency by a factor of 16 and 3 respectively. For pGMRES, the symmetry of the system matrix is not a prerequisite. The Coriolis term and the off-diagonal part of the water drag can then be treated implicitly. Theoretical and simulation results show that this implicit treatment eliminates a numerical instability present with an explicit treatment. During this research, we have also observed that the approximate nonlinear solution converges slowly with the number of OL iterations. Furthermore, simulation results reveal: the existence of multiple solutions and occasional convergence failures of the nonlinear solver. For a time step comparable to the forcing time scale, a few OL iterations lead to errors in the velocity field that are of the same order of magnitude as the mean drift. The slow convergence is an issue at all spatial resolutions but is more severe as the grid is refined. It is attributed in part to the standard VP formulation that leads to a momentum equation that is not continuously differentiable. To obtain a smooth formulation, we replaced the standard viscous coefficient expression with capping by a hyperbolic tangent function. This provides a unique solution and reduces the computational time and failure rate. To further improve the computational efficiency, the Jacobian-free Newton-Krylov (JFNK) method was implemented. Results show that JFNK is 3.5-6.6 times faster than the standard solver (depending on the resolution and termination criterion). As the standard solver, the JFNK method exhibits a relatively small number of convergence failures. A globalization "trick", which involves a switch in the linearization approach, eliminates the failures for the standard solver. Globalization approaches such as the line search did not improve the failure rate of the JFNK method. Future work will focus on the implementation of an effective globalization approach for JFNK.
590 $a School code: 0781.
650 4 $a Applied Mathematics. $3 1669109
650 4 $a Physical Oceanography. $3 1019163
650 4 $a Atmospheric Sciences. $3 1019179
690 $a 0364
690 $a 0415
690 $a 0725
710 2 $a McGill University (Canada). $b Department of Atmospheric and Oceanic Sciences. $3 1669434
773 0 $t Dissertation Abstracts International $g 71-05B.
790 $a 0781
791 $a Ph.D.
792 $a 2009
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=NR61865