Abstract

A number of iterative techniques have recently been developed which are extremely efficient at solving systems of linear equations. Of these methods probably the most recognized is the Conjugate Gradient Method (CG). This is an extremely efficient solver and has been used successfully for a number of years now. A newer method proposed initially by Davidson [1] is studied in this paper. This method has proven itself in terms of efficiency by solving the same system (of order 2000) that was solved by the CG method. It converged in approximately 40 iterations, taking less than five minutes to do so[5], compared to the CG method which took nearly 100 iterations, converging after about 1 5 minutes. Very little documentation about the derivation or development of Davidson's method exists, and his paper was written in terms of an eigenvalue problem. A portion of a program developed by NASA Ames Research Center uses a variation of Davidson's method as a linear solver. Davidson's method was explored and derived using his paper and the FORTRAN code from NASA. The purpose of this thesis is to provide some insight into the analytical aspect of Davidson's method, using the CG method for comparison.

Library of Congress Subject Headings

Iterative methods (Mathematics)--Evaluation; Equations--Numerical analysis--Evaluation; Conjugate gradient methods--Evaluation

Publication Date

1-1-1998

Document Type

Thesis

Department, Program, or Center

Mechanical Engineering (KGCOE)

Advisor

Ghosh, Amitabha

Advisor/Committee Member

Torok, Josef

Advisor/Committee Member

Kandlikar, Satish

Comments

Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: QA297.8 .C486 1998

Campus

RIT – Main Campus

Download

Share

COinS